CN-121997020-A - Air conditioner host abnormal movement identification method applying deep learning algorithm

CN121997020ACN 121997020 ACN121997020 ACN 121997020ACN-121997020-A

Abstract

The invention relates to the technical field of air conditioner host monitoring and discloses an air conditioner host abnormal movement identification method applying a deep learning algorithm. The method comprises the steps of constructing a mobile behavior knowledge graph of an air conditioner host, wherein nodes are historical mobile positions, and sides are transfer relations. And receiving a real-time mobile data stream, and extracting a mobile characteristic vector containing position, speed and acceleration. And matching the vector with the knowledge graph to calculate the movement deviation degree, generating a movement abnormal event if the movement deviation degree exceeds a preset threshold value, and extracting the multi-mode space-time feature vector. The space-time features are input into a pre-trained deep learning anomaly detection model, the model weights and fuses the features based on an attention mechanism, and an anomaly probability value is output. And when the probability value is larger than a preset threshold value, generating an alarm instruction, sending the alarm instruction to the user terminal, and updating the knowledge graph. The method realizes efficient and reliable abnormal movement identification.

Inventors

XIA HONG
LI XUE

Assignees

安徽玺夏新型材料有限公司

Dates

Publication Date: 20260508
Application Date: 20260115

Claims (10)

1. An air conditioner host abnormal movement identification method applying a deep learning algorithm is characterized by comprising the following steps: constructing a mobile behavior knowledge graph of the air conditioner host, wherein nodes of the mobile behavior knowledge graph are historical mobile positions of the air conditioner host, and the nodes are transfer relations among the historical mobile positions; receiving a real-time mobile data stream sent by air conditioner host monitoring equipment, and extracting a mobile characteristic vector from the real-time mobile data stream, wherein the mobile characteristic vector comprises a position characteristic, a speed characteristic and an acceleration characteristic; matching the mobile characteristic vector with nodes in a mobile behavior knowledge graph, and calculating the mobile deviation degree; when the movement deviation exceeds a preset threshold, generating a movement abnormal event, and carrying out multi-mode feature extraction on the movement abnormal event to obtain a space-time feature vector; inputting the space-time feature vectors into a pre-trained deep learning anomaly detection model, wherein the deep learning anomaly detection model carries out weighted fusion on the space-time feature vectors based on an attention mechanism; Outputting an abnormal probability value through the deep learning abnormal detection model, and generating an alarm instruction when the abnormal probability value is greater than a preset probability threshold value; and sending the alarm instruction to a user terminal, and simultaneously updating the mobile behavior knowledge graph.
2. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 1, wherein the constructing a movement behavior knowledge graph of the air conditioner host comprises: Collecting moving track data of an air conditioner host in a preset history period, and cleaning and denoising the moving track data; Extracting key mobile nodes from the cleaned movement track data, wherein each key mobile node comprises longitude and latitude coordinates and time stamp information; Constructing an edge of a mobile behavior knowledge graph based on a transfer relation between key mobile nodes in the mobile track data, wherein the edge comprises a transfer probability and a transfer time characteristic; a node feature vector is generated for each key mobile node, the node feature vector including node dwell time, access frequency, and neighbor node relationship features.
3. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 2, wherein the receiving the real-time movement data stream sent by the air conditioner host monitoring device, extracting the movement feature vector from the real-time movement data stream, comprises: Carrying out sliding window segmentation on the real-time mobile data stream to obtain a continuous time window sequence; calculating the statistical characteristics of the moving track in each time window, including average moving speed, speed variance, acceleration extremum and moving direction change rate; Extracting morphological characteristics of a moving track, including track curvature, track length and concave-convex characteristics of a moving path; and carrying out feature fusion on the statistical features and the morphological features to generate a movement feature vector.
4. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 3, wherein the matching the movement feature vector with a node in a movement behavior knowledge graph, calculating a movement deviation degree, comprises: calculating the similarity between the mobile feature vector and each node feature vector in the mobile behavior knowledge graph; selecting a node with highest similarity as a matching node, and calculating Euclidean distance between the mobile feature vector and the feature vector of the matching node; Calculating a movement deviation degree based on the Euclidean distance and the history transition probability of the matching node; calculating a time deviation factor by combining the time feature in the mobile feature vector and the time stamp of the matched node; And carrying out weighted summation on the Euclidean distance, the historical transition probability deviation and the time deviation factor to obtain the final movement deviation degree.
5. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 4, wherein the performing multi-modal feature extraction on the movement abnormal event to obtain a space-time feature vector comprises: Extracting spatial features of the abnormal event of movement, including starting point coordinates, end point coordinates and shape features of a movement path of abnormal movement; Extracting time characteristics of a mobile abnormal event, wherein the time characteristics comprise specific time, duration and time period characteristics of occurrence of the abnormal event; extracting environmental characteristics of mobile abnormal events, including weather conditions, surrounding equipment density and geographic position type characteristics; Performing feature stitching on the spatial features, the temporal features and the environmental features to generate a multi-mode feature matrix; and performing dimension reduction processing on the multi-mode feature matrix to obtain space-time feature vectors.
6. The method for identifying abnormal movement of an air conditioner host applying a deep learning algorithm according to claim 5, wherein the inputting the spatiotemporal feature vector into the pre-trained deep learning abnormal detection model comprises: the deep learning abnormal detection model comprises a feature coding layer, an attention fusion layer and a classification output layer; the feature coding layer carries out nonlinear transformation on the time space feature vector to obtain high-level feature representation; the attention fusion layer calculates importance weights of different feature dimensions, and performs weighted fusion on the high-level feature representation; the classification output layer calculates an abnormal probability value based on the weighted and fused feature representation.
7. The method for recognizing abnormal movement of a host computer of an air conditioner applying a deep learning algorithm according to claim 6, wherein the outputting of the abnormal probability value by the deep learning abnormal detection model comprises: The depth learning abnormal detection model calculates the matching degree of the weighted and fused characteristic representation and the normal moving mode template; Based on the matching degree calculation result, converting the matching degree calculation result into a probability value through a sigmoid function; Setting a dynamic probability threshold value, wherein the dynamic probability threshold value is adaptively adjusted according to the distribution condition of the historical abnormal probability value; and when the abnormal probability value is larger than the dynamic probability threshold value, judging that the abnormal movement event occurs.
8. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 7, wherein the generating an alarm command comprises: When the abnormal movement event is judged, extracting detailed information of the abnormal event, wherein the detailed information comprises abnormal occurrence time, abnormal position coordinates and abnormal movement tracks; Determining alarm grades according to the magnitude of the abnormal probability value, and generating alarm instructions of corresponding grades; The abnormal event abstract and the treatment proposal information are added in the alarm instruction.
9. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 8, wherein the sending the alarm command to the user terminal comprises: According to the receiving preference preset by the user, selecting a sending mode of an alarm instruction, wherein the sending mode comprises short message notification, mobile application pushing and mail notification; Performing format conversion on the alarm instruction, and adapting to display requirements of different receiving terminals; Recording the sending state of the alarm instruction and the confirmation condition of the user.
10. The method for identifying abnormal movement of an air conditioner host using a deep learning algorithm according to claim 9, wherein the updating the movement behavior knowledge graph comprises: according to the confirmed abnormal movement event, the transition probability of the corresponding node in the movement behavior knowledge graph is adjusted; adding a new abnormal movement mode to the knowledge graph, and establishing an abnormal mode node; based on the user feedback information, parameters of the deep learning anomaly detection model are optimized.

Description

Air conditioner host abnormal movement identification method applying deep learning algorithm Technical Field The invention relates to the technical field of air conditioner host monitoring, in particular to an air conditioner host abnormal movement identification method applying a deep learning algorithm. Background The abnormal movement identification of the existing air conditioner host mostly adopts a method based on a fixed threshold value or a rule. For example, the position change is monitored by a sensor, and an alarm is triggered when the distance, speed or acceleration of movement exceeds a preset limit. Such methods rely on manually set parameters and cannot learn the historical behavior patterns of the device. Some improvements introduce machine learning algorithms, such as support vector machines or simple neural networks, to classify moving features. However, these methods are still limited to static feature analysis and fail to effectively capture the spatio-temporal continuity of moving data. The prior art generally adopts simple stitching or average fusion when processing multi-modal features, and ignores the dynamic importance of different features. The prior art solutions have drawbacks. The fixed threshold method has poor adaptability and is easy to generate false alarm in the environment change or normal carrying operation. The rule system needs frequent manual adjustment, and the maintenance cost is high. Traditional machine learning models rely on a large number of labeling samples and have difficulty modeling long-term mobile dependencies. The feature fusion mode is single, so that key information is lost, and the abnormality detection precision is affected. The method and the device are used for solving the problems of high false alarm rate and weak model generalization capability in abnormal mobile identification. By integrating historical movement behavior modeling and intelligent feature weighting, the adaptability to complex movement modes is improved. Disclosure of Invention The invention aims to provide an air conditioner host abnormal movement identification method applying a deep learning algorithm so as to solve the problems in the background technology. In order to achieve the above object, the present invention provides a method for identifying abnormal movement of an air conditioner host using a deep learning algorithm, the method comprising: constructing a mobile behavior knowledge graph of the air conditioner host, wherein nodes of the mobile behavior knowledge graph are historical mobile positions of the air conditioner host, and the nodes are transfer relations among the historical mobile positions; receiving a real-time mobile data stream sent by air conditioner host monitoring equipment, and extracting a mobile characteristic vector from the real-time mobile data stream, wherein the mobile characteristic vector comprises a position characteristic, a speed characteristic and an acceleration characteristic; matching the mobile characteristic vector with nodes in a mobile behavior knowledge graph, and calculating the mobile deviation degree; when the movement deviation exceeds a preset threshold, generating a movement abnormal event, and carrying out multi-mode feature extraction on the movement abnormal event to obtain a space-time feature vector; inputting the space-time feature vectors into a pre-trained deep learning anomaly detection model, wherein the deep learning anomaly detection model carries out weighted fusion on the space-time feature vectors based on an attention mechanism; Outputting an abnormal probability value through the deep learning abnormal detection model, and generating an alarm instruction when the abnormal probability value is greater than a preset probability threshold value; and sending the alarm instruction to a user terminal, and simultaneously updating the mobile behavior knowledge graph. Preferably, the building the mobile behavior knowledge graph of the air conditioner host includes: Collecting moving track data of an air conditioner host in a preset history period, and cleaning and denoising the moving track data; Extracting key mobile nodes from the cleaned movement track data, wherein each key mobile node comprises longitude and latitude coordinates and time stamp information; Constructing an edge of a mobile behavior knowledge graph based on a transfer relation between key mobile nodes in the mobile track data, wherein the edge comprises a transfer probability and a transfer time characteristic; a node feature vector is generated for each key mobile node, the node feature vector including node dwell time, access frequency, and neighbor node relationship features. Preferably, the receiving the real-time mobile data stream sent by the air conditioner host monitoring device, extracting a mobile feature vector from the real-time mobile data stream, includes: Carrying out sliding window segmentation on the real-time mobile data stream t