CN-121983352-A - Patient monitoring abnormal behavior identification method and system combined with deep learning algorithm

Abstract

The invention provides a patient monitoring abnormal behavior identification method and system combined with a deep learning algorithm, which relates to the technical field of medical monitoring, and comprises the steps of firstly acquiring original activity data streams of patients and medical staff collected by a plurality of sensing units in a target monitoring area, constructing a patient activity track space-time sequence and a medical staff operation track space-time sequence, carrying out monitoring space-time association mapping treatment to generate an interactive event triggering relation chain, and then invoking an abnormal behavior recognition model to conduct abnormal behavior pattern deviation quantitative analysis on the interaction event triggering relation chain to obtain an abnormal behavior type label and a time sequence abnormal behavior fragment set, deducing an abnormal behavior induction propagation path based on the result to generate an abnormal behavior space diffusion characteristic set, and finally generating a personalized monitoring intervention strategy instruction according to the information and sending the personalized monitoring intervention strategy instruction to medical personnel mobile terminal equipment. The invention can identify the abnormal behavior of the patient, analyze the propagation rule of the abnormal behavior and provide accurate intervention strategies for medical staff.

Inventors

• XIANG TING
• Hu Guanting
• LU YUNFEI
• SUN QIANG
• CUI WENYAO

Assignees

• 四川大学华西医院

Dates

Publication Date

20260505

Application Date

20260407

Claims (10)

1. 1. A method for identifying patient monitoring abnormal behavior in combination with a deep learning algorithm, the method comprising: acquiring original patient activity data streams acquired by a plurality of sensing units deployed in a target monitoring area; Constructing an activity track space-time sequence of a monitored object in the target monitoring area according to time sequence activity data units in the original patient activity data stream, and constructing an operation track space-time sequence of medical staff in the target monitoring area according to medical care operation data units in the original patient activity data stream; performing monitoring time-space association mapping processing on the activity track time-space sequence and the operation track time-space sequence to generate an interaction event trigger relation chain between a monitored object and medical staff; Invoking a pre-built abnormal behavior recognition model to conduct abnormal behavior pattern deviation quantitative analysis on the interaction event triggering relation chain to obtain an abnormal behavior type label of the monitored object in the continuous monitoring time period and a time sequence abnormal behavior fragment set corresponding to the abnormal behavior type label; Performing abnormal behavior induction propagation path deduction processing based on the time sequence abnormal behavior fragment set and the interaction event triggering relation chain, and generating an abnormal behavior space diffusion characteristic set containing a space propagation direction sequence and a space propagation speed change rate of the abnormal behavior data sub-segment in the target monitoring area; Generating a personalized monitoring intervention strategy instruction aiming at the monitored object according to the abnormal behavior type tag, the abnormal behavior space diffusion characteristic set and the time sequence abnormal behavior fragment set, and sending the personalized monitoring intervention strategy instruction to a mobile terminal device worn by the medical staff to trigger a corresponding monitoring operation prompt.
2. 2. The method for identifying abnormal behavior of patient monitoring in combination with deep learning algorithm according to claim 1, wherein the constructing the activity trajectory space-time sequence of the monitored object in the target monitoring area according to the time sequence activity data unit in the original patient activity data stream, and simultaneously constructing the operation trajectory space-time sequence of the medical staff in the target monitoring area according to the medical care operation data unit in the original patient activity data stream comprises: Analyzing time sequence activity data units in the original patient activity data stream, extracting a monitored object identifier, a data acquisition time stamp and a spatial position coordinate of the monitored object in the data acquisition time stamp, wherein the spatial position coordinate is obtained by calculating based on output signals of a positioning sensing unit deployed in the target monitoring area; performing ascending arrangement processing on all the extracted space position coordinates containing the monitored object identifiers according to the data acquisition time stamp to obtain a first space position coordinate sequence of the monitored object, which changes with time in the continuous monitoring time period; Performing track point interpolation encryption processing on the first space position coordinate sequence, and inserting a plurality of interpolation position coordinates between the space position coordinates corresponding to two adjacent data acquisition time stamps according to a preset time interpolation density to obtain a second space position coordinate sequence containing original space position coordinates and interpolation position coordinates; Constructing a moving speed vector set of the monitored object according to the time attribute corresponding to each space position coordinate in the second space position coordinate sequence, wherein each moving speed vector in the moving speed vector set is calculated by dividing a displacement vector between two adjacent space position coordinates by a corresponding time interval; Performing association storage on the second space position coordinate sequence and the moving speed vector set to generate an activity track space-time sequence of the monitored object in the target monitoring area, wherein the activity track space-time sequence comprises each space position coordinate in the second space position coordinate sequence, an instantaneous moving speed vector and an instantaneous moving direction angle which are associated with the space position coordinate; Synchronously analyzing medical care operation data units in the original patient activity data stream, extracting a medical care personnel identifier, an operation execution time stamp and an operation position coordinate of the medical care personnel at the operation execution time stamp, wherein the operation position coordinate is obtained by calculating based on an output signal of a positioning sensing unit or a built-in positioning module of a handheld terminal device of the medical care personnel; performing ascending order arrangement processing on all the extracted operation position coordinates containing the medical staff identifiers according to the operation execution time stamp to obtain a third space position coordinate sequence of the medical staff, which changes along with time in the continuous monitoring time period; Performing track point smoothing filtering processing on the third space position coordinate sequence, eliminating coordinate abnormal jump points caused by instantaneous fluctuation of positioning signals, and generating a smoothed fourth space position coordinate sequence; And constructing a spatial position mapping relation of each medical care operation executed by the medical care personnel in the moving process according to the time attribute corresponding to each operation position coordinate in the fourth spatial position coordinate sequence and the operation type code contained in the medical care operation data unit, and generating an operation track space-time sequence containing the operation position coordinate, a time point reaching the operation position coordinate, the operation type code executed by the operation position coordinate and a time point leaving the operation position coordinate.
3. 3. The method for identifying abnormal behavior of patient monitoring in combination with deep learning algorithm according to claim 1, wherein the performing a monitoring time-space correlation mapping process on the activity track time-space sequence and the operation track time-space sequence to generate an interaction event trigger relation chain between the monitored object and the medical staff comprises: Acquiring a first space position coordinate of a monitored object at a specific time point contained in the activity track space-time sequence and a second space position coordinate of a medical staff contained in the operation track space-time sequence at the same time point, wherein the same time point is determined by performing global time axis alignment operation on a data acquisition time stamp and an operation execution time stamp; Calculating Euclidean space distance between the first space position coordinate and the second space position coordinate to obtain an instantaneous space distance value of the monitored object and the medical staff at the same time point; Comparing the instantaneous space distance value with a preset interaction trigger distance threshold, and when the instantaneous space distance value is smaller than or equal to the interaction trigger distance threshold, judging that the monitored object and the medical staff generate a space position overlapping event at the same time point, and generating a space overlapping event record containing the same time point, the first space position coordinate, the second space position coordinate and the instantaneous space distance value; Synchronously analyzing an operation type code corresponding to each operation position coordinate in the operation track space-time sequence, and extracting a medical care operation type set which directly acts on a monitored object from the operation type code, wherein the medical care operation type set which directly acts on the monitored object comprises a vital sign measurement operation type, an intravenous injection operation type, a wound care operation type and a body position adjustment operation type; Determining an occurrence time point of each medical care operation directly acting on a monitored object according to the operation execution time stamp, acquiring a first spatial position coordinate of the monitored object corresponding to the occurrence time point, and calculating an operation contact spatial distance between the first spatial position coordinate of the monitored object corresponding to the occurrence time point and an operation position coordinate of the medical care operation; when the operation contact space distance is smaller than or equal to a preset operation contact distance threshold value, judging that an operation contact event occurs between the monitored object and the medical staff at the occurrence time point, and generating an operation contact event record containing the occurrence time point, the first space position coordinate, the operation position coordinate and the operation type code; Combining and sequencing the space overlapping event records and the operation contact event records according to the time points contained in the space overlapping event records to obtain an initial interaction event sequence, wherein the initial interaction event sequence comprises a plurality of interaction event units arranged according to time sequence, and each interaction event unit corresponds to one space overlapping event or one operation contact event; Calculating time interval parameters according to time points of two adjacent interaction event units in the initial interaction event sequence, and calculating space distance change parameters according to space position coordinates corresponding to the two adjacent interaction event units; and taking each interaction event unit in the initial interaction event sequence as an interaction event node, taking a directional connecting line between two adjacent interaction event nodes as a directional correlation edge, and attaching the time interval parameter and the space distance change parameter to the corresponding directional correlation edge to generate an interaction event trigger relation chain between the monitored object and the medical staff.
4. 4. The method for identifying abnormal behavior of patient monitoring in combination with deep learning algorithm according to claim 1, wherein the invoking the pre-constructed abnormal behavior identification model to quantitatively analyze the abnormal behavior pattern deviation of the interactive event trigger relation chain to obtain an abnormal behavior type tag of the monitored object in the continuous monitoring time period and a time sequence abnormal behavior fragment set corresponding to the abnormal behavior type tag includes: Inputting the interaction event triggering relation chain into an input layer of the abnormal behavior recognition model, wherein the interaction event triggering relation chain comprises an interaction event node sequence, and a time interval parameter and a space distance change parameter carried by a directed correlation edge together form a characteristic input tensor of the abnormal behavior recognition model; Performing space structure coding processing on the feature input tensor through a feature encoder of the abnormal behavior recognition model, wherein the feature encoder comprises a plurality of stacked graph rolling network layers, each graph rolling network layer performs neighborhood feature aggregation operation on the interaction event node to generate a high-order space neighborhood feature representation of each interaction event node in a current layer, and the high-order space neighborhood feature representation fuses topological connection relation between the interaction event node and adjacent interaction event nodes in the interaction event triggering relation chain and parameter information carried by the directed association edges; performing feature stitching processing on the high-order space neighborhood feature representations of different orders output by the plurality of graph convolution network layers to obtain a multi-scale space structure embedded vector of each interaction event node; Performing time sequence dependency modeling processing on the multi-scale space structure embedded vector according to the time sequence of the interaction event node through a time sequence encoder of the abnormal behavior recognition model, wherein the time sequence encoder comprises a two-way long-short-term memory network structure, and the two-way long-short-term memory network structure generates a time sequence context feature vector of the current interaction event node containing forward and backward time sequence context information according to the multi-scale space structure embedded vector of the current interaction event node and hidden state vectors of the previous moment and the next moment; Inputting the time sequence context feature vector into an abnormal mode classifier of the abnormal behavior recognition model, wherein the abnormal mode classifier comprises a multi-layer perceptron network structure, the multi-layer perceptron network structure carries out nonlinear transformation processing on the time sequence context feature vector, and outputs probability distribution vectors of various preset abnormal behavior types of the monitored object at the moment point corresponding to each interaction event node; determining a candidate abnormal behavior type label of the moment according to the abnormal behavior type corresponding to the maximum probability value in the probability distribution vector, and merging the moment continuously appearing on the time axis by the candidate abnormal behavior type label to form a plurality of candidate abnormal behavior time periods; Extracting a first interaction event node identifier corresponding to a starting moment point and a second interaction event node identifier corresponding to an ending moment point of each candidate abnormal behavior time period; intercepting an original data segment in a corresponding time period from a time sequence activity data unit in the original patient activity data stream according to the first interaction event node identification and the second interaction event node identification, generating an abnormal behavior data sub-segment in the time sequence abnormal behavior segment set, and taking a candidate abnormal behavior type label corresponding to the candidate abnormal behavior time period as an abnormal behavior type label of the abnormal behavior data sub-segment; the same intercepting and labeling operation is carried out on all the candidate abnormal behavior time periods to obtain a time sequence abnormal behavior fragment set containing a plurality of abnormal behavior data subsections and abnormal behavior type labels corresponding to each abnormal behavior data subsection, and meanwhile, the association relation between each abnormal behavior data subsection and the interaction event node identifier corresponding to the time point contained in each abnormal behavior data subsection is established.
5. 5. The method for identifying abnormal behavior in patient monitoring in combination with deep learning algorithm according to claim 2, wherein the performing abnormal behavior induced propagation path deduction processing based on the time sequence abnormal behavior segment set and the interactive event trigger relation chain, generating an abnormal behavior spatial diffusion feature set including a spatial propagation direction sequence and a spatial propagation speed change rate of the abnormal behavior data sub-segment in the target monitoring area, includes: Selecting a first abnormal behavior data sub-segment from the time sequence abnormal behavior segment set as an initial abnormal behavior seed segment, and analyzing a first interaction event node identification set associated with the initial abnormal behavior seed segment, wherein the first interaction event node identification set comprises unique identifiers of all interaction event nodes in a time period corresponding to the initial abnormal behavior seed segment; Extracting the space position coordinates of the corresponding interaction event nodes from the interaction event trigger relation chain according to each interaction event node identifier in the first interaction event node identifier set, and generating an initial abnormal behavior space seed point set; performing convex hull complex calculation processing on the initial abnormal behavior space seed point set to obtain a minimum convex polygon area containing all points in the initial abnormal behavior space seed point set, and marking the minimum convex polygon area as an initial abnormal behavior space influence domain; Selecting a next abnormal behavior data subsection which is adjacent to the initial abnormal behavior seed fragment in time from the time sequence abnormal behavior fragment set as a propagation abnormal behavior target fragment, and analyzing a second interaction event node identification set associated with the propagation abnormal behavior target fragment; Extracting the space position coordinates of the corresponding interaction event nodes from the interaction event trigger relation chain according to each interaction event node identifier in the second interaction event node identifier set, and generating a propagation abnormal behavior space target point set; performing convex hull complex calculation processing on the propagation abnormal behavior space target point set to obtain a minimum convex polygon area containing all points in the propagation abnormal behavior space target point set, and marking the minimum convex polygon area as a propagation abnormal behavior space influence domain; Calculating a space displacement vector between the geometric center point coordinates of the initial abnormal behavior space influence domain and the geometric center point coordinates of the propagation abnormal behavior space influence domain, wherein the direction of the space displacement vector is the space propagation direction of the initial abnormal behavior seed segment to the propagation abnormal behavior target segment, and the time interval between the initial abnormal behavior seed segment and the propagation abnormal behavior target segment is the space propagation speed after dividing the modular length of the space displacement vector; taking the space propagation direction and the space propagation speed as a first group of space diffusion characteristic parameters, and storing the space propagation direction and the space propagation speed in a corresponding record of the initial abnormal behavior seed segment and the propagation abnormal behavior target segment in a correlation manner; Continuing to select a subsequent abnormal behavior data sub-segment from the time sequence abnormal behavior segment set as a new abnormal behavior propagation target segment, and repeatedly executing operations of analyzing the interactive event node identification set, generating a space influence domain of abnormal behavior propagation, calculating a space displacement vector and a space propagation speed until all abnormal behavior data sub-segments in the time sequence abnormal behavior segment set are traversed, so as to obtain a plurality of groups of space diffusion characteristic parameters; arranging the space propagation directions in the plurality of groups of space diffusion characteristic parameters according to a time sequence to generate a space propagation direction sequence of the abnormal behavior data subsections in the target monitoring area; Calculating the change rate of the space propagation speed between adjacent time periods according to the space propagation speed values arranged in time sequence in the plurality of groups of space diffusion characteristic parameters, and generating a space propagation speed change rate sequence; And the space propagation direction sequence and the space propagation speed change rate sequence jointly form the abnormal behavior space diffusion characteristic set.
6. 6. The method for identifying abnormal behavior of patient monitoring in combination with deep learning algorithm according to claim 3, wherein the generating a personalized monitoring intervention policy instruction for the monitored object according to the abnormal behavior type tag, the abnormal behavior space diffusion feature set and the time sequence abnormal behavior fragment set, and sending the personalized monitoring intervention policy instruction to a mobile terminal device worn by the medical staff to trigger a corresponding monitoring operation prompt includes: Analyzing a starting time point and an ending time point of each abnormal behavior data subsection contained in the time sequence abnormal behavior fragment set, and calculating a duration parameter of each abnormal behavior data subsection according to the starting time point and the ending time point; Acquiring a space propagation direction sequence and a space propagation speed change rate sequence contained in the space diffusion feature set of the abnormal behavior, comparing a speed change rate value in the space propagation speed change rate sequence with a preset speed change rate threshold value, and identifying an abnormal propagation acceleration point with the speed change rate value exceeding the speed change rate threshold value and an abnormal propagation deceleration point with the speed change rate value lower than the speed change rate threshold value; generating an abnormal behavior propagation situation dynamic change map according to the corresponding space propagation directions of the abnormal propagation acceleration point and the abnormal propagation deceleration point in the space propagation direction sequence, wherein the abnormal behavior propagation situation dynamic change map marks a space position area in which the propagation speed of the abnormal behavior in the target monitoring area is obviously changed; invoking a preset intervention strategy rule base, wherein the intervention strategy rule base stores mapping relations between various abnormal behavior type labels and various basic intervention strategy templates, and each basic intervention strategy template comprises a medical operation action list to be executed, a recommended execution time window for executing the medical operation action and a medical instrument equipment list required by executing the medical operation action; matching a corresponding basic intervention strategy template from the intervention strategy rule base according to the abnormal behavior type label to serve as an initial intervention strategy template; comparing and fusing the recommended execution time window in the initial intervention strategy template with the duration parameter of the abnormal behavior data subsection and the time point of the abnormal propagation acceleration point marked in the abnormal behavior propagation situation dynamic change graph, adjusting the starting time and the ending time of the recommended execution time window, and generating a personalized adjusted execution time window, so that the personalized adjusted execution time window covers the time period before the abnormal propagation acceleration point; Predicting the possible propagation direction of the abnormal behavior in the future according to the spatial propagation direction sequence in the abnormal behavior spatial diffusion feature set, extracting adjacent monitoring bed position identifiers or adjacent functional area identifiers in the possible propagation direction from a spatial layout database of the target monitoring area by combining with a spatial position area marked in the dynamic change graph of the abnormal behavior propagation situation, and adding the adjacent monitoring bed position identifiers or the adjacent functional area identifiers into a medical instrument list in the initial intervention strategy template to generate a personalized medical instrument list containing an expanded monitoring range; Packaging the personalized adjusted execution time window, the personalized medical instrument equipment list and the medical instrument operation action list in the initial intervention strategy template to generate the personalized monitoring intervention strategy instruction, wherein the personalized monitoring intervention strategy instruction comprises a plurality of sub-instruction units which are arranged according to a time sequence, and each sub-instruction unit corresponds to a medical operation action, a specific time point for executing the medical operation action, medical instrument equipment required by executing the medical operation action and target space position coordinates for executing the medical operation action; the personalized monitoring intervention strategy instruction is sent to mobile terminal equipment worn by medical staff through a hospital internal wireless communication network, after the mobile terminal equipment receives the personalized monitoring intervention strategy instruction, the plurality of sub-instruction units are analyzed, medical operation action prompts, required medical equipment prompts and target space position navigation information corresponding to each sub-instruction unit are sequentially displayed on a display screen according to time sequence, and accordingly the medical staff is triggered to execute corresponding monitoring operation according to the personalized monitoring intervention strategy instruction.
7. 7. The method for identifying abnormal behavior of patient monitoring in combination with deep learning algorithm according to claim 3, wherein the performing a monitoring time-space correlation mapping process on the activity track time-space sequence and the operation track time-space sequence generates an interaction event trigger relation chain between the monitored object and the medical staff, further comprising: After the space overlapping event record and the operation contact event record are generated, an operation type code contained in the operation contact event record is obtained, and a corresponding standard operation duration is searched from a preset operation duration database according to the operation type code; calculating a theoretical ending time point of the operation contact event according to the occurrence time point in the operation contact event record and the standard operation duration; Detecting whether the space position coordinates of the medical staff in the operation track space-time sequence are continuously located in a preset distance range near the operation position coordinates in the operation contact event record in a preset time window around the theoretical ending time point, and detecting whether the space position coordinates of the monitored object in the activity track space-time sequence are continuously located in the preset distance range near the operation position coordinates; If the space position coordinates of the medical staff and the space position coordinates of the monitored object are continuously located in a preset distance range near the operation position coordinates within a preset time window before and after the theoretical ending time point, judging that the operation contact event is completely executed, and adding a complete execution mark for the operation contact event record; if the space position coordinate of the medical staff or the space position coordinate of the monitored object leaves a preset distance range near the operation position coordinate before the theoretical ending time point arrives, judging that the operation contact event is interrupted, and adding an interruption execution mark and an interruption occurrence time point for the operation contact event record; Updating the corresponding interaction event node attribute in the interaction event trigger relation chain according to the interrupt execution mark and the interrupt occurrence time point of the operation contact event record, and attaching an operation interrupt information field to the interaction event node, wherein the operation interrupt information field comprises an interrupt identifier and an interrupt time point; And in the subsequent abnormal behavior pattern deviation quantitative analysis process, taking the interaction event node containing the operation interruption information field as a key analysis object, and increasing the initial weight coefficient of the interaction event node in the characteristic input tensor of the abnormal behavior recognition model.
8. 8. The method for identifying abnormal behavior in patient monitoring in combination with deep learning algorithm according to claim 4, wherein the invoking the pre-constructed abnormal behavior identification model performs an abnormal behavior pattern deviation quantization analysis on the interactive event trigger relationship chain, further comprising: before a feature encoder of the abnormal behavior recognition model performs space structure coding processing on the feature input tensor, performing node type classification processing on the interaction event nodes contained in the interaction event triggering relation chain, marking the interaction event nodes corresponding to the space overlapping events as first type nodes, and marking the interaction event nodes corresponding to the operation contact events as second type nodes; Respectively distributing different initial feature mapping matrixes for the first type node and the second type node, wherein the dimension of the initial feature mapping matrix of the first type node is matched with the number of parameters contained in the space overlapping event record, and the dimension of the initial feature mapping matrix of the second type node is matched with the number of parameters contained in the operation contact event record; Mapping the space overlapping event record into a first type node initial embedded vector through the initial feature mapping matrix of the first type node, and mapping the operation contact event record into a second type node initial embedded vector through the initial feature mapping matrix of the second type node; When the graph volume network layer performs neighborhood feature aggregation operation, respectively adopting different aggregation weight parameters according to node types of neighborhood nodes, adopting a first aggregation weight matrix for neighborhood aggregation of a first class of nodes, and adopting a second aggregation weight matrix for neighborhood aggregation of a second class of nodes, so that the high-order spatial neighborhood feature representation can distinguish influence differences of interaction events of different types on abnormal behavior modes; After generating a time sequence context feature vector by the time sequence encoder, inputting the time sequence context feature vector into an attention mechanism module built in the abnormal behavior recognition model, wherein the attention mechanism module calculates an attention score between the time sequence context feature vector of each interaction event node and a preset global abnormal behavior pattern vector; the time sequence context feature vectors are weighted and summed according to the attention scores, and a global abnormal behavior representation vector fused with all interaction event node information is generated; And splicing the global abnormal behavior representation vector with the time sequence context feature vector of each interaction event node to generate an enhanced time sequence context feature vector, and inputting the enhanced time sequence context feature vector into the abnormal mode classifier for classification processing.
9. 9. A patient monitoring abnormal behavior recognition system in combination with a deep learning algorithm, comprising: A processor; a machine-readable storage medium storing machine-executable instructions for the processor; wherein the processor is configured to perform the patient care abnormal behavior identification method in combination with the deep learning algorithm of any one of claims 1 to 8 via execution of the machine executable instructions.
10. 10. A computer program product, characterized in that the computer program product comprises machine executable instructions stored in a computer readable storage medium, from which a processor of a deep learning algorithm-incorporated patient care abnormal behavior recognition system reads the machine executable instructions, which processor executes the machine executable instructions such that the deep learning algorithm-incorporated patient care abnormal behavior recognition system performs the deep learning algorithm-incorporated patient care abnormal behavior recognition method of any one of claims 1 to 8.

Description

Patient monitoring abnormal behavior identification method and system combined with deep learning algorithm Technical Field The invention relates to the technical field of medical monitoring, in particular to a method and a system for identifying abnormal behaviors of patient monitoring by combining a deep learning algorithm. Background In the field of medical monitoring, abnormal behavior identification and intervention of patients are key links for guaranteeing patient safety and improving medical quality. The traditional patient monitoring mode mainly relies on manual observation and regular inspection of medical staff, and the mode not only consumes a great deal of manpower, but also is difficult to realize real-time and comprehensive monitoring of patient behaviors. With the development of sensor technology, sensor-based patient monitoring systems are increasingly emerging that collect patient activity data by deploying various types of sensors in a monitored area. However, most existing sensor-based monitoring methods focus only on the patient's own activity data, ignoring the impact of the medical personnel's operational behavior on the patient in the monitored area. In fact, many complications may arise during patient interaction with healthcare workers, which are closely related to the occurrence of abnormal patient behavior. For example, improper operation of the healthcare worker may cause an uncomfortable response to the patient, resulting in abnormal behavior, or abnormal behavior of the patient may result from some stimulus during interaction with the healthcare worker. However, the prior art cannot effectively integrate the activity data of the patient and the medical staff, is difficult to accurately identify the abnormal behavior mode in the interaction process of the patient and the medical staff, and cannot deeply analyze the induction propagation path of the abnormal behavior, so that an accurate and personalized monitoring intervention strategy cannot be provided for the medical staff, and the actual effect of medical monitoring is limited. Disclosure of Invention In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for identifying abnormal patient monitoring behavior in combination with a deep learning algorithm, the method comprising: Acquiring an original patient activity data stream acquired by a plurality of sensing units deployed in a target monitoring area, wherein the original patient activity data stream comprises time sequence activity data units generated by a monitored object in the target monitoring area in a continuous monitoring time period and medical care operation data units generated by medical staff executing monitoring tasks in the target monitoring area in the continuous monitoring time period; Constructing an activity track space-time sequence of a monitored object in the target monitoring area according to time sequence activity data units in the original patient activity data stream, and constructing an operation track space-time sequence of medical staff in the target monitoring area according to medical care operation data units in the original patient activity data stream; Performing monitoring time-space correlation mapping processing on the activity track time-space sequence and the operation track time-space sequence to generate an interaction event triggering relation chain between a monitored object and medical staff, wherein the interaction event triggering relation chain comprises a plurality of interaction event nodes arranged according to time sequence and directed correlation edges connected with the interaction event nodes, the interaction event nodes correspond to space position overlapping events or operation contact events of the monitored object and the medical staff at specific time points, and the directed correlation edges correspond to time interval parameters and space distance change parameters between the interaction event nodes; Invoking a pre-built abnormal behavior recognition model to conduct abnormal behavior pattern deviation quantitative analysis on the interaction event trigger relation chain to obtain an abnormal behavior type label of the monitored object in the continuous monitoring time period and a time sequence abnormal behavior fragment set corresponding to the abnormal behavior type label, wherein the time sequence abnormal behavior fragment set comprises a plurality of abnormal behavior data subsections cut from the time sequence activity data unit and interaction event node identifications associated with each abnormal behavior data subsection in the interaction event trigger relation chain; Performing abnormal behavior induction propagation path deduction processing based on the time sequence abnormal behavior fragment set and the interaction event triggering relation chain, and generating an abnormal behavior sp