CN-122025055-A - Intelligent management method and device for mobile emergency unit cluster based on Internet of things

CN122025055ACN 122025055 ACN122025055 ACN 122025055ACN-122025055-A

Abstract

The invention discloses an intelligent management method and device for a mobile emergency unit cluster based on the Internet of things, and the intelligent management method comprises the steps of collecting sensing data of vehicles and equipment, preprocessing the sensing data to generate a multi-source data set, constructing a response dynamic field based on multi-source data modeling task gravitation, road damping and space mutual exclusion, inputting a time sequence prediction model generation result, calculating risk influence, constructing a causal chain and a prediction sequence, generating an event chain causal force field, fusing the response dynamic field to form a global behavior driving field, constructing a plurality of behavior projection bodies, executing synchronous scene deduction, generating a future behavior prediction result, calculating comprehensive scores, selecting an optimal behavior scheme, issuing a scheduling instruction and controlling execution action. According to the invention, the intelligent, prospective and collaborative efficient scheduling management of the mobile emergency unit is realized by fusing the multi-source internet of things data, the time sequence prediction model and the cluster behavior deduction.

Inventors

SHEN XIEDONG
LIU HONGCHAO
ZOU SILI
QU LEFENG
HONG HANHAN
WU JIANJIN
CAO NA
JIANG QINGJUN
JI XIANGGUO
JIANG HAILIN

Assignees

中国人民解放军海军军医大学

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (9)

1. The intelligent management method for the mobile emergency unit cluster based on the Internet of things is characterized by comprising the following steps of: acquiring the sensor data of the Internet of things deployed on a mobile emergency vehicle, an emergency unmanned aerial vehicle and mobile emergency equipment, preprocessing the sensor data of the Internet of things, and generating a multi-source data set; carrying out multi-scale tensor modeling on the attraction of emergency tasks, road traffic damping and the spatial mutual exclusion relation of mobile emergency units according to the multi-source data set, mapping spatial influence factors under different scales into continuous potential energy distribution, and constructing an emergency response dynamic field; Converting the multisource data set into a time sequence form, inputting the time sequence prediction model into an improved PatchTST time sequence prediction model, generating a future prediction result, calculating the influence intensity of each time data on the emergency risk by combining the current time data, constructing an emergency event causal chain and outputting a future state prediction sequence of a causal node; generating an event chain causal force field according to the future state prediction sequence, mapping the future state change of the causal node into causal driving force, and fusing the event chain causal force field and the emergency response dynamic field to form a global behavior driving field; Constructing a plurality of future behavior projection bodies for each mobile emergency unit based on the global behavior driving field, and carrying out synchronous scene deduction on the future behavior projection bodies to generate a prediction result of the future behavior projection bodies; And calculating comprehensive decision scores according to the prediction results of the future behavior projectors, selecting a future behavior scheme with the highest comprehensive decision score from the plurality of future behavior projectors, and taking the future behavior scheme as a scheduling execution strategy of the mobile emergency unit to carry out instruction issuing and action control.
2. The internet of things-based mobile emergency unit cluster intelligent management method of claim 1, wherein the internet of things sensor data comprises location information data, road traffic data, vital sign data, resource status data, and environmental parameter data.
3. The intelligent management method of mobile emergency unit clusters based on the internet of things according to claim 1, wherein the preprocessing of the sensor data of the internet of things comprises performing time synchronization processing, abnormal data rejection processing and format standardization processing on the sensor data of the internet of things.
4. The intelligent management method of mobile emergency unit clusters based on the internet of things according to claim 1, wherein the constructing an emergency response dynamic field comprises: Establishing a multiscale space grid and a road topology index in an urban working area, respectively extracting the geographic position and the emergency degree of an emergency task, the geometric form and the congestion intensity of road elements, and the geographic position and the occupation state of a mobile emergency unit based on a multisource data set, and generating a task feature tensor, a road damping feature tensor and a unit mutual exclusion feature tensor according to a preset local scale, a patch scale and a global scale; For each scale, mapping from characteristics to potential energy by adopting a learnable space kernel, introducing geographic constraint based on road network accessibility and hospital service radius constraint, and respectively generating continuous potential energy layers for a task characteristic tensor, a road damping characteristic tensor and a unit mutual exclusion characteristic tensor; performing dynamic coupling and normalization processing on each continuous potential energy layer, wherein the dynamic coupling weights potential energy influences of different sources based on edge side real-time feedback coefficients, boundary conditions are applied to an obstacle region and a forbidden region, and smooth constraint is performed between adjacent grids to obtain intermediate potential energy distribution of different scales; the method comprises the steps of carrying out multi-scale fusion on intermediate potential energy distribution, adopting scale weight constraint and topology connectivity constraint in the fusion process to keep consistency of local response and global coordination, and carrying out online updating on fusion weight according to recent arrival time deviation and path feasibility error to form an emergency response dynamic field covering a working area; And carrying out space sampling and block sparse storage on the emergency response dynamic field, determining the maximum descending direction of each sampling position for representing the guide vector of the mobile emergency unit, and outputting the emergency response dynamic field and the guide vector set.
5. The internet of things-based mobile emergency unit cluster intelligent management method of claim 1, wherein the constructing an emergency event causal link and outputting a future state prediction sequence of causal nodes comprises: Time sequencing and alignment are carried out on the multi-source data set on a unified time axis, and input samples are constructed according to a preset sampling period, a historical window length and a prediction window length; Processing the input samples using a modified PatchTST timing prediction model, the modified PatchTST model inserts three structures in the topology of the time sliced embedded layer-encoder block sequence-prediction head and concatenates the three structures in the following positions: After the time slicing embedding layer and before the first encoder block, a time-event co-anchoring layer is inserted in series, an anchoring mark of the starting time, the dispatching time, the road state switching time and the resource state changing time of the emergency event is received, and an anchoring fusion feature is output to the first encoder block after being aligned with the slicing feature; Inserting multi-scale expansion slicing coding branches in parallel at the first, third and fifth encoder blocks of the encoder block sequence, carrying out parallel coding on the same slicing characteristic with different expansion step sizes, and fusing the same slicing characteristic with a trunk output in a residual mode at the tail end of the corresponding encoder block; inserting causal condition gating units in series after the second encoder block and the fourth encoder block, receiving risk priori and resource accessibility priori obtained by multi-source data set calculation, performing gating modulation on the characteristics of the encoder blocks, and outputting the encoded characteristics to the next encoder block; Training an improved PatchTST time sequence prediction model, wherein a training process comprises an unconditional pre-training stage and a conditional fine-tuning stage, the unconditional pre-training stage comprises a freezing time-event co-anchoring layer, a causal condition gating unit and a multi-scale expansion sliced coding branch, only main parameters are optimized, the conditional fine-tuning stage unfreezes a newly added structure and is combined with the main to be optimized, and training key parameters comprise a learning rate, a batch size, a history window length, a prediction window length and an expansion step length; And (3) adopting a sliding time window to send a new input sample into the improved PatchTST time sequence prediction model to generate a future state prediction sequence, only performing small-step online updating on the time-event co-anchoring layer and the causal condition gating unit based on the concept drift detection threshold value, keeping the parameters of the time slicing embedded layer and the trunk encoder block unchanged, and calculating the influence intensity of each target variable on the emergency risk according to the future state prediction sequence and the current time data to form an emergency event causal chain.
6. The intelligent management method of mobile emergency unit clusters based on the internet of things according to claim 1, wherein the forming a global behavior driving field comprises: based on a future state prediction sequence, sequentially reading a vital sign predicted value, a road traffic predicted value, a task density predicted value and a resource state predicted value according to a prediction step length to form a state record set; Taking current time data and state records corresponding to the predicted time as input, calculating the influence intensity of vital signs, road traffic, task density and resource states on future emergency risks, and generating an influence intensity set corresponding to each predicted time one by one; setting causal scaling factors for vital signs, road traffic, task density and resource states respectively, setting time reduction weights for different prediction step sizes, and carrying out weighting treatment on influence intensities and corresponding causal scaling factors and time reduction weights to obtain a driving dynamics set for each causal node and each prediction time; mapping the driving dynamics to an urban working area according to the driving dynamics set and a spatial anchoring position, wherein the spatial anchoring position comprises a task geographic position, a road element position, a task hot spot position and a resource node position, and generating continuous event chain causal force field distribution around the corresponding position by adopting a preset spatial mapping core; and fusing the event chain causal force field distribution with the emergency response dynamic field to obtain a global behavior driving field for driving the behavior decision of the mobile emergency unit.
7. The intelligent management method of mobile emergency unit clusters based on the internet of things according to claim 1, wherein the generating the prediction result of the future behavior projection body comprises: acquiring a global behavior driving field, the current position, the speed, the residual resources, the electric quantity or the oil quantity of each mobile emergency unit, the allocated and to-be-allocated tasks, the road and forbidden area constraint, setting a unified prediction time window and a time step, and establishing a strategy template library and a synchronous clock; constructing a plurality of future behavior projectors based on a strategy template library for each mobile emergency unit, wherein the future behavior projectors are instantiated by adopting a ternary structure of a strategy constraint unit, a resource time capsule and a cooperative identification, and the method comprises the following steps: the strategy constraint unit solidifies the direct strategy, the detour strategy, the replenishment priority strategy, the insertion tolerance strategy and the vehicle-machine cooperative strategy rule; The resource time capsule encapsulates available resources, the accessibility of the supplement points and a service time limit window; the collaborative identification marks cross-unit collaborative relation and establishes coupling with other projectors in the same time window; Performing synchronous scene deduction on a plurality of future behavior projectors according to a synchronous clock, updating states in parallel at each time step, calculating a guide vector based on a global behavior driving field, calling a road and forbidden region constraint to perform feasibility verification, performing linkage update on related future behavior projectors according to a cooperative mark when joint arrival, handover or division is required, and creating event branches and generating comparable branch tracks for the corresponding future behavior projectors when road state switching, task emergency change or resource threshold triggering are detected; And generating a track record and an event record in a prediction time window aiming at each future behavior projection body, calculating response time, emergency risk change, resource consumption and cluster coverage according to the track record and the event record to form a prediction result, and performing feasibility screening and deduplication on the future behavior projection body with hard constraint conflict or repeated with similar strategies.
8. The intelligent management method of mobile emergency unit clusters based on the internet of things according to claim 1, wherein the calculating the comprehensive decision score according to the prediction result of the future behavior projection body comprises: For each future behavior projection, reading a prediction result, wherein the prediction result comprises a response time prediction value, an emergency risk change prediction value, a resource consumption prediction value and a cluster coverage prediction value, and forming the prediction values into a vector to be scored according to a preset format; Setting a comprehensive decision scoring function, wherein the comprehensive decision scoring function consists of vector components to be scored and corresponding weights, and the weights comprise response time weights, emergency risk weights, resource consumption weights and cluster coverage weights, and the comprehensive decision scoring function generates comprehensive scores of future behavior projectors in a weighted mode; Calculating comprehensive scores of all future behavior projectors, forming a one-to-one corresponding set of the future behavior projectors and the comprehensive scores, and sequencing all the future behavior projectors according to the comprehensive scores from high to low; Selecting a future behavior projection body with the highest comprehensive score as a future behavior scheme, and extracting a behavior strategy, path information, resource actions and cooperation information in the future behavior scheme to form a scheduling instruction; and sending the scheduling instruction to the corresponding mobile emergency unit, executing path control, speed adjustment, task update or resource replenishment, and feeding back an execution result.
9. The mobile emergency unit cluster intelligent management device based on the internet of things for performing the mobile emergency unit cluster intelligent management method based on the internet of things according to any one of claims 1 to 8, which is characterized by comprising the following modules: The multi-source data acquisition and preprocessing module is used for acquiring the sensor data of the Internet of things and executing preprocessing to generate a multi-source data set; The dynamic field modeling module is used for carrying out multi-scale tensor modeling based on the multi-source data set and constructing an emergency response dynamic field; The time sequence prediction and causal chain construction module is used for converting the multi-source data set into a time sequence and inputting the time sequence into the improved PatchTST model to generate a future prediction result so as to form a causal chain of the emergency event; The causal force field fusion module is used for generating an event chain causal force field according to the future state prediction sequence and fusing the event chain causal force field with the emergency response dynamic field to form a global behavior driving field; The behavior projection body deduction module is used for constructing a plurality of future behavior projection bodies based on the global behavior driving field and executing synchronous scene deduction to generate a prediction result; And the scheduling decision and execution module is used for calculating the comprehensive decision score of the future behavior projection body according to the prediction result, selecting an optimal behavior scheme, and generating and issuing a scheduling execution instruction.

Description

Intelligent management method and device for mobile emergency unit cluster based on Internet of things Technical Field The invention relates to the technical field of Internet of things and intelligent scheduling decision, in particular to an intelligent management method and device for a mobile emergency unit cluster based on the Internet of things. Background The existing mobile emergency system generally relies on manual dispatching rules and linear decision-making processes to dispatch rescue vehicles, emergency unmanned aerial vehicles and mobile rescue equipment, and task assignment is mainly carried out based on static elements such as calling for help positions, vehicle distances, road traffic conditions and the like. However, as urban traffic environments become increasingly complex, emergency task densities continue to rise, and emergency unit types become increasingly diverse, traditional scheduling approaches have been struggled to handle multi-unit synergy and dynamic risk changes. Particularly, under the condition that the multisource internet of things equipment continuously generates log, state, positioning and monitoring data, the existing system cannot effectively fuse mass data distributed in time and space, so that understanding of the whole emergency situation is lagged. Current emergency dispatch techniques generally lack the ability to predict future trends. Most systems can only make decisions based on the current state, and cannot sense the evolution of road congestion, the change trend of vital signs of patients or the resource consumption condition in advance, so that a scheduling result is delayed from a real condition. Although some researches have introduced machine learning or traffic prediction models, they mainly focus on single tasks or single resources, lack a unified time sequence prediction framework, further fail to form a causal chain for explaining the emergency event development mechanism, and are difficult to support the prospective command of the whole urban emergency system. In a large scale scenario of multiple emergency unit parallel responses, traditional scheduling strategies lack the ability to systematically derive multiple behavioral path, multiple strategy response schemes. The current system can only give a single path or a single scheduling scheme, can not evaluate dynamic performance of multiple schemes under different task urgency, road state switching and resource constraint, and can not select an optimal scheme from multiple future behavior schemes. The lack of modeling of the power field of the multi-unit cooperative response further causes low cluster scheduling efficiency and low resource utilization rate, thereby affecting the success rate of emergency tasks. Therefore, how to provide a mobile emergency unit cluster intelligent management method and device based on the internet of things is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide an intelligent management method and device for a mobile emergency unit cluster based on the Internet of things, which fully utilizes technologies such as multi-source Internet of things perception, a time sequence depth prediction model, causal force field reasoning, behavior scene deduction and the like, constructs a fusion mechanism of an emergency response dynamic force field and an event chain causal force field, and realizes cooperative scheduling and optimal policy selection of mobile emergency vehicles, emergency unmanned aerial vehicles and mobile emergency equipment. The invention can fuse multiple types of emergency related data in real time in a complex urban environment, predict emergency situation change in advance, evaluate cluster-level response schemes through the multi-strategy future behavior projection body, and has the advantages of high response speed, high scheduling result stability, remarkable resource utilization rate improvement and emergency success rate improvement. According to the embodiment of the invention, the intelligent management method for the mobile emergency unit cluster based on the Internet of things comprises the following steps: acquiring the sensor data of the Internet of things deployed on a mobile emergency vehicle, an emergency unmanned aerial vehicle and mobile emergency equipment, preprocessing the sensor data of the Internet of things, and generating a multi-source data set; carrying out multi-scale tensor modeling on the attraction of emergency tasks, road traffic damping and the spatial mutual exclusion relation of mobile emergency units according to the multi-source data set, mapping spatial influence factors under different scales into continuous potential energy distribution, and constructing an emergency response dynamic field; Converting the multisource data set into a time sequence form, inputting the time sequence prediction model into an improved PatchTST time sequence prediction model