Maintenance optimization method and device of emergency equipment, electronic equipment and storage medium

Abstract

The invention discloses a maintenance optimization method and device for emergency equipment, electronic equipment and a storage medium, which are used for solving the problem of how to optimize the distribution of maintenance resources of the emergency equipment and improve the maintenance efficiency while satisfying the equipment state maintenance. Acquiring emergency equipment state data of the to-be-maintained area and discrimination text information input by a user; extracting an abnormal feature matrix from the emergency equipment state data according to the discrimination text information; constructing a plurality of associated hypergraphs based on the abnormal feature matrix and preset maintenance example data; and generating region maintenance information according to the plurality of associated hypergraphs. Therefore, the association hypergraph is constructed by extracting the abnormal feature matrix, and the complex causal relationship between the multiple emergency devices and the maintenance factors can be accurately described; and further generating regional maintenance information according to the associated hypergraph, thereby optimizing a maintenance strategy, realizing reasonable scheduling of maintenance resources, reducing maintenance cost, and improving maintenance response efficiency and high efficiency of the emergency equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment management, and in particular to a maintenance optimization method, device, electronic equipment and storage medium for emergency equipment. Background Art

[0002] Emergency equipment refers to various specialized devices and technical equipment used to protect lives, minimize property losses, maintain social order, and assist in rescue operations during emergencies such as sudden natural disasters, accidents, public health incidents, or social security incidents. These devices typically feature rapid response, strong adaptability, and high reliability, capable of operating normally in extreme environments and complex conditions. Modern emergency equipment encompasses a wide range of areas, including communications, lighting, medical equipment, search and rescue equipment, and protective gear. Its technological development relies primarily on advances in modern technologies such as the Internet of Things, artificial intelligence, big data, and satellite communications.

[0003] With the widespread use of emergency equipment in various emergency response scenarios, the demand for status awareness, maintenance, and resource scheduling is increasing. However, managing and maintaining large quantities of emergency equipment often presents scheduling difficulties. Emergency equipment maintenance not only involves the wear and tear of the equipment itself but also involves the scheduling of maintenance personnel, the allocation of maintenance materials, and the interoperability of maintenance equipment.

[0004] Due to limited maintenance resources, how to optimize the allocation of maintenance resources and improve maintenance efficiency while maintaining equipment status is a difficult point in current emergency equipment management. Summary of the Invention

[0005] The present invention provides a maintenance optimization method, device, electronic device and storage medium for emergency equipment, which are used to solve or partially solve the technical problem of how to optimize the allocation of emergency equipment maintenance resources and improve maintenance efficiency while meeting the requirements of equipment status maintenance.

[0006] The present invention provides a maintenance optimization method for emergency equipment, the method comprising:

[0007] Obtain emergency equipment status data of the area to be maintained and the judgment text information entered by the user;

[0008] extracting an abnormal feature matrix from the emergency equipment status data according to the discriminant text information;

[0009] Constructing multiple association hypergraphs based on the abnormal feature matrix and preset maintenance example data;

[0010] Regional maintenance information is generated according to the multiple associated hypergraphs, and the regional maintenance information is used to maintain and allocate resources for emergency equipment in the area to be maintained.

[0011] Optionally, extracting an abnormal feature matrix from the emergency equipment status data according to the discriminant text information includes:

[0012] Performing semantic analysis on the discriminant text information to generate a discriminant feature vector;

[0013] Performing context feature extraction on the emergency equipment status data based on the discriminant feature vector to obtain an initial abnormal feature set;

[0014] Extracting nodes whose feature activation values ​​exceed a preset threshold from the initial abnormal feature set to construct a first abnormal feature tensor;

[0015] Performing cluster analysis on the first abnormal feature tensor to obtain a similar data set;

[0016] Performing time-frequency domain joint feature extraction on the similar data set to generate a second abnormal feature tensor;

[0017] Perform high-order feature fusion on the first abnormal feature tensor and the second abnormal feature tensor to generate an abnormal feature matrix.

[0018] Optionally, performing cluster analysis on the first abnormal feature tensor to obtain a similar data set includes:

[0019] Calculating the cosine similarity of the first abnormal feature tensor and constructing a feature similarity matrix based on the cosine similarity;

[0020] Calculating a dynamic neighborhood radius parameter according to the feature similarity matrix, and screening the feature similarity matrix according to the dynamic neighborhood radius parameter to obtain an adjacency matrix;

[0021] A core point set of the adjacency matrix is ​​extracted, and spectral clustering is performed on the core point set to generate a homogeneous data set.

[0022] Optionally, performing time-frequency domain joint feature extraction on the similar data set to generate a second abnormal feature tensor includes:

[0023] Assign dynamic weights to the similar data sets through a multi-head attention mechanism to generate an attention feature matrix;

[0024] Decomposing the attention feature matrix into low-frequency components and high-frequency components using discrete wavelet transform;

[0025] Mapping the low-frequency component into a first sub-tensor and mapping the high-frequency component into a second sub-tensor;

[0026] Perform orthogonal projection on the first sub-tensor and the second sub-tensor to generate a second abnormal feature tensor.

[0027] Optionally, the constructing a plurality of associated hypergraphs based on the abnormal feature matrix and preset maintenance example data comprises:

[0028] performing vectorization processing on the preset maintenance example data to obtain a first example feature matrix;

[0029] matching the abnormal feature matrix and the first example feature matrix to obtain a second example feature matrix;

[0030] generating maintenance instruction information according to the abnormal feature matrix and the second example feature matrix;

[0031] obtaining map information associated with the abnormal feature matrix;

[0032] performing hypergraph construction according to the map information, the maintenance instruction information and a preset hypergraph construction condition to obtain a plurality of associated hypergraphs.

[0033] Optionally, the performing hypergraph construction according to the map information, the maintenance instruction information and a preset hypergraph construction condition to obtain a plurality of associated hypergraphs comprises:

[0034] performing clustering according to the maintenance instruction information to obtain a maintenance type distribution index;

[0035] determining a target hypergraph construction condition from the preset hypergraph construction condition according to the maintenance type distribution index;

[0036] determining a hypergraph node from the map information according to the target hypergraph construction condition;

[0037] generating edges between the hypergraph nodes according to the target hypergraph construction condition, the maintenance instruction information and the map information to obtain a plurality of associated hypergraphs.

[0038] Optionally, the map information comprises emergency equipment positioning coordinate information; the maintenance instruction information comprises maintenance equipment, supplies and personnel configuration; and the generating edges between the hypergraph nodes according to the target hypergraph construction condition, the maintenance instruction information and the map information to obtain a plurality of associated hypergraphs comprises:

[0039] performing Gaussian kernel space encoding processing on the emergency equipment positioning coordinate information to generate a hypergraph node set;

[0040] calculating a resource sharing rate matrix of the hypergraph node set according to the maintenance equipment, supplies and personnel configuration;

[0041] Based on the traffic distance threshold in the target hypergraph construction condition, a geographic proximity analysis is performed on the hypergraph node set. For each node to be analyzed in the hypergraph node set, nodes whose Euclidean distance to the node to be analyzed is less than or equal to the traffic distance threshold are marked as adjacent nodes, thereby generating a traffic adjacency matrix.

[0042] fusing the resource sharing rate matrix and the traffic adjacency matrix, generating a hybrid similarity matrix by weighted summation, and screening a hyperedge candidate set from the hybrid similarity matrix according to a dynamic threshold;

[0043] Aggregating all adjacent nodes and indirectly associated nodes of each node in the hyperedge candidate set to generate a hyperedge set;

[0044] Multi-level hypergraph convolution optimization is performed on the hyperedge set to generate multiple associated hypergraphs.

[0045] The present invention also provides a maintenance optimization device for emergency equipment, comprising:

[0046] A data acquisition unit, used to acquire the emergency equipment status data of the area to be maintained and the judgment text information input by the user;

[0047] an abnormal feature matrix extraction unit, configured to extract an abnormal feature matrix from the emergency equipment status data according to the discriminant text information;

[0048] A correlation hypergraph construction unit, configured to construct a plurality of correlation hypergraphs based on the abnormal feature matrix and preset maintenance example data;

[0049] The regional maintenance information generating unit is used to generate regional maintenance information according to the multiple associated hypergraphs, wherein the regional maintenance information is used to maintain and allocate resources for emergency equipment in the area to be maintained.

[0050] The present invention further provides an electronic device, comprising a processor and a memory:

[0051] The memory is used to store program code and transmit the program code to the processor;

[0052] The processor is configured to execute the maintenance optimization method for emergency equipment as described above according to instructions in the program code.

[0053] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and the program code is used to execute the maintenance optimization method for emergency equipment as described in any one of the above items.

[0054] It can be seen from the above technical solutions that the present invention has the following advantages:

[0055] A maintenance optimization method for emergency equipment is provided. First, the emergency equipment status data for the area to be maintained and user-entered discriminant text information are obtained. Then, an abnormal feature matrix is ​​extracted from the emergency equipment status data based on the discriminant text information. Multiple association hypergraphs are then constructed based on the abnormal feature matrix and preset maintenance example data. Regional maintenance information is generated based on the multiple association hypergraphs. By extracting the abnormal feature matrix and constructing an association hypergraph, the complex causal relationships between multiple emergency equipment and maintenance factors can be accurately depicted. Furthermore, regional maintenance information for maintenance and resource allocation of emergency equipment within the area to be maintained is generated based on the association hypergraph, thereby optimizing maintenance strategies, achieving reasonable scheduling of maintenance resources, reducing maintenance costs, and improving the maintenance response efficiency and effectiveness of emergency equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 A flowchart of the steps of a maintenance optimization method for emergency equipment;

[0058] Figure 2 A schematic diagram of a maintenance optimization system for emergency equipment;

[0059] Figure 3 The structural block diagram of a maintenance optimization device for emergency equipment. DETAILED DESCRIPTION

[0060] The embodiments of the present invention provide a maintenance optimization method, device, electronic device and storage medium for emergency equipment, which are used to solve or partially solve the technical problem of how to optimize the allocation of emergency equipment maintenance resources and improve maintenance efficiency while maintaining equipment status.

[0061] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] As an example, with the widespread use of emergency equipment in various emergency response scenarios, the demand for status awareness, maintenance, and resource scheduling is increasing. However, managing and maintaining large quantities of emergency equipment often presents scheduling difficulties. Emergency equipment maintenance not only involves monitoring the wear and tear of the equipment itself, but also involves factors such as the scheduling of maintenance personnel, the allocation of repair supplies, and the interoperability of maintenance equipment.

[0063] Due to limited maintenance resources, how to optimize the allocation of maintenance resources and improve maintenance efficiency while maintaining equipment status is a difficult point in current emergency equipment management.

[0064] Therefore, one of the core inventive aspects of the present invention is to provide a maintenance optimization method for emergency equipment. First, based on a pre-set natural language model, an abnormal feature matrix is ​​automatically extracted, converting unstructured data into structured information. Next, a correlation hypergraph is constructed to accurately depict the complex causal relationships between multiple emergency equipment and maintenance factors. Furthermore, regional maintenance information is generated based on the correlation hypergraph, thereby optimizing maintenance strategies, rationally scheduling maintenance resources, reducing maintenance costs, and improving the maintenance response efficiency and effectiveness of emergency equipment.

[0065] In the embodiment of the present invention, the emergency equipment may mainly include various sensors, equipment functions, a programmable logic controller (PLC), an intelligent communication module, and an equipment control panel.

[0066] Emergency equipment will be equipped with various sensors for self-monitoring. These sensors will collect environmental data based on their functions and store this data in the PLC via serial communication. The PLC also connects to various relays and intelligent circuit breakers via serial ports. Relays control the power supply to equipment, lighting, and support legs. The PLC communicates with the equipment control panel and intelligent communication module via a network port using TCP / IP (Transmission Control Protocol / Internet Protocol). The emergency equipment, based on the control panel, reads data from the PLC to perform front-end design and data layout, thereby displaying device data. The emergency equipment control panel also programs control commands for the PLC. These control commands trigger responses from the relays and intelligent circuit breakers corresponding to the PLC addresses, thereby controlling the equipment to perform various operations. The intelligent communication module connects to the PLC via a network port using TCP / IP. The intelligent communication module receives query, write, and control commands from the server over the 4G network and sends them to the PLC using the S7Comm (S7 Communication Protocol) protocol to execute the relevant operations.

[0067] Reference Figure 1 , shows a flowchart of a maintenance optimization method for emergency equipment provided by an embodiment of the present invention, which may specifically include the following steps:

[0068] Step 101: Acquire emergency equipment status data of the area to be maintained and identification text information input by the user;

[0069] In a specific implementation, it is first necessary to obtain the status data of each emergency device in the pre-defined maintenance area.

[0070] For example, emergency equipment may include power supplies, generators, circuit breakers, LED (Light Emitting Diode) lighting systems, advertising lights, sidelights, searchlights, and warning lights. The PLC on the emergency equipment collects status data at each location. This data may include information such as the equipment's startup status, online status, battery charge, tilt angle, obstacle distance, input voltage and current, various power levels, support leg stress, lamp current, equipment height, support leg extension length, pole pressure, tilt angle temperature, equipment location, fuel tank level, input mode, and operating time.

[0071] At the same time, according to the model of the emergency equipment, the database can also collect equipment information such as factory information, function information, and equipment ownership information as a supplement to the status data. This can be used to deal with situations where the person in charge of the equipment cannot be found and to understand the specific configuration information of the equipment.

[0072] It should be noted that the collection of the above information requires the platform's permissions system to ensure legal and compliant information collection. Specifically, the system must grant platform function permissions to specific roles and users, and will record user operations to provide network security for devices.

[0073] The user-entered judgment text information refers to the judgment text information entered in advance by the user, such as "abnormal hydraulic system pressure fluctuation" or "battery module temperature exceeds the limit".

[0074] Step 102: extracting an abnormal feature matrix from the emergency equipment status data according to the discriminant text information;

[0075] This step mainly realizes the extraction of abnormal feature matrix from emergency equipment status data based on the preset natural language model and the pre-input discriminant text information.

[0076] In some embodiments, the process of extracting an abnormal feature matrix from emergency equipment status data based on discriminant text information can be implemented by executing the following sub-steps S01 to S06:

[0077] Step S01: semantically analyze the discriminant text information to generate a discriminant feature vector;

[0078] Step S02: extracting context features from the emergency equipment status data based on the discriminant feature vector to obtain an initial abnormal feature set;

[0079] Step S03: extracting nodes whose feature activation values ​​exceed a preset threshold value from the initial abnormal feature set, and constructing a first abnormal feature tensor;

[0080] Step S04: performing cluster analysis on the first abnormal feature tensor to obtain a similar data set;

[0081] Furthermore, the implementation process of step S04 can mainly be: calculating the cosine similarity of the first abnormal feature tensor, and constructing a feature similarity matrix based on the cosine similarity; calculating the dynamic neighborhood radius parameter based on the feature similarity matrix, and screening the feature similarity matrix based on the dynamic neighborhood radius parameter to obtain an adjacency matrix; extracting the core point set of the adjacency matrix, and performing spectral clustering segmentation on the core point set to generate a similar data set.

[0082] Step S05: performing time-frequency domain joint feature extraction on the same data set to generate a second abnormal feature tensor;

[0083] Furthermore, the implementation process of step S05 can mainly be: assigning dynamic weights to similar data sets through a multi-head attention mechanism to generate an attention feature matrix; using discrete wavelet transform to decompose the attention feature matrix into low-frequency components and high-frequency components; mapping the low-frequency components to a first sub-tensor, and mapping the high-frequency components to a second sub-tensor; performing orthogonal projection on the first sub-tensor and the second sub-tensor to generate a second abnormal feature tensor.

[0084] Step S06: performing high-order feature fusion on the first abnormal feature tensor and the second abnormal feature tensor to generate an abnormal feature matrix.

[0085] For example, the server is equipped with a data analysis module. This module contains a pre-installed natural language model. This pre-installed natural language model primarily includes an embedding layer, a bidirectional long short-term memory (BiLSTM) network, a gated linear unit (GLU), and a tensor fusion layer. This pre-installed natural language model uses multimodal data fusion and deep feature extraction techniques to correlate and analyze pre-input discriminant text information with real-time collected emergency equipment status data, outputting a structured abnormality feature matrix.

[0086] The server can be a standalone server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0087] First, based on the preset natural language model, abnormal data is extracted from the emergency equipment status data. At the same time, the abnormal data needs to be output as structured data.

[0088] The system performs semantic analysis on user-entered discriminative text information (e.g., "Abnormal hydraulic system pressure fluctuation" or "Battery module temperature exceeds limit"). Key words in the text are mapped into high-dimensional vectors using an embedding layer. A self-attention mechanism is then used to extract core semantic features from the text to generate a discriminative feature vector. For example, "Abnormal pressure fluctuation" is parsed into a discriminative feature vector containing the device status, fault type, and severity.

[0089] The generated discriminant feature vector is fed into a bidirectional long short-term memory network (LSTM) to perform contextual modeling on the time series of structured anomaly data corresponding to the state data. The BLSTM captures the temporal dependencies of sensor data through forward and reverse channels, identifying potential anomaly patterns (such as periodic fluctuations or sudden increases or decreases). For example, for hydraulic system pressure data, the model can extract contextual features such as mean, variance, and gradient changes within a continuous time window to form an initial set of anomaly features.

[0090] The initial abnormal feature set is nonlinearly filtered based on a gated linear unit. The gated linear unit dynamically activates feature channels through a gating mechanism, and only retains feature nodes with activation values exceeding a preset threshold (such as the abnormal threshold can be set to 3σ range of historical data distribution). The filtered feature nodes are reorganized according to the space-time dimension to construct a first abnormal feature tensor. The dimensions include device number, sensor type, timestamp, and multiple indexes.

[0091] For feature processing of the tensor fusion layer, a dynamic density clustering algorithm is used to group the first abnormal feature tensor. According to similarity measurement (such as cosine similarity) and adaptive neighborhood radius adjustment, the associated emergency equipment data is classified into the same set to generate a second abnormal feature tensor. The root cause is determined from the data features of the first abnormal feature tensor and the second abnormal feature tensor, high-order feature fusion is performed, and an abnormal feature matrix is generated.

[0092] Step 103, based on the abnormal feature matrix and the preset maintenance example data, a plurality of associated hypergraphs are constructed;

[0093] There is an associated influence between multiple interacting devices. For example, it is assumed that the normal fluctuation is determined in the A device, which causes the B device to appear abnormal. In this case, if A device and B device are analyzed separately, it is difficult to find such an influence, thereby causing abnormal analysis error.

[0094] Therefore, a plurality of associated hypergraphs can be constructed based on the previously constructed abnormal feature matrix and the preset maintenance example data to avoid such abnormal analysis error.

[0095] In some embodiments, the implementation process of constructing a plurality of associated hypergraphs based on the abnormal feature matrix and the preset maintenance example data can specifically include the following sub-steps S11 to S15:

[0096] Step S11: The preset maintenance example data is vectorized to obtain a first example feature matrix;

[0097] Step S12: The abnormal feature matrix and the first example feature matrix are matched to obtain a second example feature matrix;

[0098] Step S13: Maintenance instruction information is generated according to the abnormal feature matrix and the second example feature matrix;

[0099] Step S14: Map information associated with the abnormal feature matrix is obtained;

[0100] Step S15: Hypergraph construction is performed according to the map information, the maintenance instruction information, and the preset hypergraph construction condition to obtain a plurality of associated hypergraphs.

[0101] Furthermore, a hypergraph is constructed based on the map information, the maintenance instruction information and the preset hypergraph construction conditions to obtain multiple associated hypergraphs, which may be:

[0102] Clustering is performed based on maintenance instruction information to obtain maintenance type distribution indicators;

[0103] determining a target hypergraph construction condition from preset hypergraph construction conditions according to the maintenance type distribution index;

[0104] Determine hypergraph nodes from map information according to target hypergraph construction conditions;

[0105] According to the target hypergraph construction conditions, maintenance instruction information and map information, edges are generated between hypergraph nodes to obtain multiple associated hypergraphs.

[0106] Furthermore, map information can include emergency equipment location coordinates; maintenance instruction information can include maintenance equipment, supplies, and personnel configurations. Based on the target hypergraph construction conditions, maintenance instruction information, and map information, edges are generated between hypergraph nodes to obtain multiple associated hypergraphs, specifically:

[0107] Perform Gaussian kernel spatial coding on the emergency equipment positioning coordinate information to generate a hypergraph node set;

[0108] Calculate the resource sharing rate matrix of the hypergraph node set based on the maintenance equipment, materials and personnel configuration;

[0109] Based on the traffic distance threshold in the target hypergraph construction condition, the geographical proximity analysis of the hypergraph node set is performed. For each node to be analyzed in the hypergraph node set, the nodes whose Euclidean distance to the node to be analyzed is less than or equal to the traffic distance threshold are marked as adjacent nodes, and a traffic adjacency matrix is ​​generated.

[0110] The resource sharing rate matrix and the traffic adjacency matrix are integrated to generate a hybrid similarity matrix through weighted summation. The hyperedge candidate set is then screened from the hybrid similarity matrix based on a dynamic threshold.

[0111] Aggregate all adjacent nodes and indirectly related nodes of each node in the hyperedge candidate set to generate a hyperedge set;

[0112] Perform multi-level hypergraph convolution optimization on the hyperedge set to generate multiple associated hypergraphs.

[0113] For example, the server may further include a result generation module, which constructs a multi-dimensional association hypergraph by fusing the abnormal feature matrix with preset maintenance example data, and generates a regionalized maintenance plan based on the hypergraph topology.

[0114] First, the large number of emergency equipment requiring maintenance in the maintenance area must be divided into multiple maintenance areas or routes, and maintenance personnel and resources must be assigned. Maintenance planning and reminders must be prepared before maintenance personnel depart. Thus, the embodiments of the present invention address scheduling difficulties and improve maintenance efficiency by automating maintenance configuration.

[0115] Preset maintenance example data (e.g., "Replace hydraulic oil" or "Adjust battery cooling module") is semantically encoded (vectorized) and converted into a first-example feature matrix using a natural language model. Each row of this matrix corresponds to a feature vector for a maintenance operation. This includes information such as the required equipment type, material inventory, and personnel skills.

[0116] The abnormality feature matrix is ​​matched against the first example feature matrix for similarity, and the target maintenance example most relevant to the current abnormality pattern is selected as the second example feature matrix. This matching process utilizes a bidirectional attention mechanism, calculating the association weights between the abnormality features and the maintenance features. For example, for "hydraulic pressure abnormality," the model prioritizes matching highly relevant maintenance items such as "hydraulic system maintenance" and "seal replacement" to generate the second example feature matrix.

[0117] The abnormal feature matrix and the second example feature matrix are input into a pre-set natural language model, and the decoder generates structured maintenance instructions. This information includes maintenance priority, a list of required resources, and operational steps. For example, "Priority 1: Replace the hydraulic pump (requires 20L of hydraulic oil and two technicians); Priority 2: Calibrate the pressure sensor (requires one calibrator)."

[0118] Obtain map information associated with the anomaly feature matrix (such as equipment longitude and latitude coordinates and transportation network). Encode the location coordinates of each emergency device into a hypergraph node using a Gaussian kernel function. Node features include attributes such as geographic location, anomaly level, and resource requirements. For example, a device node might be represented as {longitude X, latitude Y, anomaly level: 3, hydraulic oil required: 30L}.

[0119] Consider resource synergy between hypergraph nodes. Specifically, based on the allocation of maintenance equipment, supplies, and personnel, if the shared ratio of maintenance supplies between two nodes exceeds a threshold (e.g., 60%), the connection weight is increased. This constructs a resource sharing ratio matrix for the hypergraph node set. The resource sharing ratio matrix reflects the resource sharing ratio between every two nodes in the hypergraph node set.

[0120] Hyperedges are dynamically generated based on pre-set hypergraph construction conditions (transportation distance, resource sharing rate, and staffing). Geographic proximity analysis involves the following steps: if the Euclidean distance between two device hypergraph nodes is less than a transportation threshold (e.g., 5 kilometers), they are marked as potentially connected.

[0121] In addition, personnel scheduling constraints can also be considered. Specifically, hyperedge connections are optimized based on the skills and locations of technicians to minimize scheduling costs.

[0122] Multi-level convolution optimization is performed on the hyperedge set, and the feature representation is updated through node-hyperedge bidirectional message passing to generate multiple associated hypergraphs.

[0123] Step 104 : generating regional maintenance information according to the plurality of associated hypergraphs, wherein the regional maintenance information is used for maintaining emergency equipment and allocating resources in the area to be maintained.

[0124] Based on the constructed multiple linked hypergraphs (generally understood as a hypergraph topology composed of multiple linked hypergraphs), a spectral clustering algorithm is used to partition the hypergraph nodes into multiple maintenance groups. Each group meets the following criteria: concentrated geographical distribution of equipment within the group to ensure accessibility; high overlap in resource requirements within the group to improve material sharing efficiency; and well-matched personnel allocation and skills to avoid cross-group scheduling. Generated regional maintenance information includes the geographic scope of each maintenance group, resource allocation tables, and personnel scheduling plans. This regional maintenance information is used to manage maintenance and resource allocation for emergency equipment within the maintenance area. This enables the command center to rapidly manage maintenance, resource allocation, and personnel scheduling for emergency equipment within the area.

[0125] In an embodiment of the present invention, a maintenance optimization method for emergency equipment is provided. Based on a pre-set natural language model, an abnormal feature matrix is ​​automatically extracted, and unstructured data is converted into structured information. By constructing an association hypergraph, the complex causal relationships between multiple emergency equipment and maintenance factors can be accurately depicted. Furthermore, regional maintenance information is generated based on the association hypergraph, thereby optimizing maintenance strategies, enabling the rational scheduling of maintenance resources, reducing maintenance costs, and improving the maintenance response efficiency and effectiveness of emergency equipment.

[0126] In some embodiments, based on a preset natural language model, an abnormal feature matrix is ​​extracted from the emergency equipment status data according to pre-input discriminant text information. The preset natural language model includes an embedding layer, a bidirectional long short-term memory network, a gated linear unit, and a tensor fusion layer. In a specific implementation, based on the preset natural language model, an abnormal feature matrix is ​​extracted from the emergency equipment status data according to pre-input discriminant text information, including the following sub-steps S1021 to S1026:

[0127] S1021: Process the pre-input discriminant text information in the embedding layer to generate a discriminant feature vector.

[0128] For example, in the embedding layer processing stage of the natural language model, the system first performs semantic parsing and vector mapping on the discriminative text information pre-entered by the user (such as "hydraulic system pressure abnormality" and "battery module temperature exceeds the limit").

[0129] The embedding layer uses pre-trained word vector models (such as Word2Vec (Word to Vector) or BERT (Bidirectional Encoder Representations from Transformers)) to convert key terms in the text (such as "pressure," "temperature," and "abnormal") into high-dimensional dense vectors while preserving the semantic relevance between the terms. For example, "pressure" and "hydraulic pressure" have a high cosine similarity in vector space, while "temperature" and "heat dissipation" are mapped to similar semantic clusters.

[0130] After word-level embedding is completed, the self-attention mechanism is used to perform global semantic modeling of the text sequence. Specifically, the self-attention layer calculates the association weight between each word and the context words, dynamically focusing on core words describing abnormal conditions (such as "fluctuation" and "overlimit"), and suppressing the interference of redundant modifiers (such as "slight" and "possible"). By weightedly aggregating the attention vectors of all words, a discriminant feature vector is generated. This vector not only contains the semantic label of the abnormality type (such as "hydraulic system failure"), but also encodes multi-dimensional information such as the severity of the abnormality (such as "urgent" and "general") and the potential scope of impact (such as "local" and "global"), providing strong semantic guidance for subsequent temporal feature extraction.

[0131] S1022: Input the discriminant feature vector and the emergency equipment status data into a bidirectional long short-term memory network, perform context feature extraction on the emergency equipment status data based on the discriminant feature vector, and obtain an initial abnormal feature set.

[0132] For example, a bidirectional long short-term memory (LSTM) network performs context-aware feature extraction on time series of emergency equipment status data based on semantic priors provided by discriminant feature vectors. The network employs a dual-channel architecture with forward and reverse channels. Specifically, the forward LSTM captures the evolving trends of sensor data (e.g., a gradual increase in pressure values) in chronological order (t = 1 → T), while the reverse LSTM identifies the delayed effects of abnormal events (e.g., a persistently low temperature value after a sudden drop) in reverse order (t = T → 1).

[0133] The dual-channel outputs are concatenated at each time step to form a hidden state vector containing complete contextual information. For example, for pressure sensor data from a hydraulic system, a bidirectional long short-term memory network can simultaneously capture the onset of a pressure surge (forward feature) and its ongoing impact on subsequent time windows (reverse feature). Furthermore, the network introduces residual connections (RC) to mitigate the vanishing gradient problem and ensure effective modeling of long-sequence data. After iterative processing over multiple time steps, the network outputs an initial set of anomaly features. These features include dimensions such as device ID, sensor type, timestamp, and statistics (mean, variance, and gradient), providing a fine-grained data foundation for subsequent feature screening and pattern mining.

[0134] S1023: Based on the gated linear unit, extract nodes whose feature activation values ​​exceed a preset threshold in the initial abnormal feature set to construct a first abnormal feature tensor.

[0135] For example, a gated linear unit (GLU) uses a gating mechanism to dynamically filter an initial set of abnormal features, separating abnormal signals from normal noise. Specifically, the GLU divides the input features into two sets of channels. One set generates feature representations through a linear transformation, while the other generates gating weights (ranging from 0 to 1) through a sigmoid activation function. These two sets of channels are element-wise multiplied, retaining only significant abnormal features with weights close to 1 (e.g., a sudden pressure increase exceeding three standard deviations from the historical mean) while suppressing irrelevant fluctuations with weights approaching 0 (e.g., minor perturbations caused by environmental noise).

[0136] The gating threshold is adaptively set based on the statistical distribution of historical data. For example, for temperature sensor data, if the temperature value within a certain time window continuously exceeds the normal range (such as ±5°C) and the duration exceeds the preset threshold (such as 10 seconds), the activation value of the corresponding feature node is determined to be a valid anomaly.

[0137] The filtered feature nodes are reorganized according to the three dimensions of device, sensor, and time to construct the first anomaly feature tensor. The high-order structure of this tensor can characterize cross-device coordinated anomalies (such as simultaneous pressure anomalies in multiple hydraulic presses) or cross-sensor correlated faults (such as temperature rise accompanied by current fluctuations), providing structured input for subsequent cluster analysis.

[0138] S1024: For the feature processing of the tensor fusion layer, cluster analysis is performed on the first abnormal feature tensor, and a similar data set with the same or similar abnormal pattern is obtained by adjusting the dynamic radius parameter.

[0139] For example, a density clustering algorithm (such as OPTICS (Ordering Points To Identify the Clustering Structure) or DBSCAN (Density-Based Spatial Clustering of Applications with Noise)) is used to identify clusters of devices or sensors with similar abnormal patterns based on the spatial distribution characteristics of the first abnormal feature tensor.

[0140] Specifically, the algorithm first calculates the cosine similarity between feature vectors to measure the semantic consistency of anomaly patterns (for example, "pressure fluctuations" and "flow anomalies" may belong to the same hydraulic system failure). Unlike traditional clustering methods with a fixed neighborhood radius, this step introduces a dynamic radius adjustment mechanism. That is, the search range is adaptively scaled based on the temporal rate of change of feature similarity (such as the difference in similarity between adjacent time windows). For example, when similarity fluctuates dramatically (such as sudden multi-device failures), the neighborhood radius is expanded to capture a wider range of anomaly associations. When similarity is stable (such as periodic minor anomalies), the radius is reduced to improve clustering accuracy.

[0141] Through iterative optimization, the algorithm partitions the feature tensor into multiple homogeneous data sets. Each set represents a specific anomaly pattern (e.g., "periodic hydraulic system leakage" or "decreased battery pack heat dissipation efficiency"). The algorithm also automatically removes isolated noise points (e.g., single false alarm data) to ensure that the clustering results only contain statistically significant anomaly patterns, providing high-quality input for subsequent time-frequency domain joint analysis and maintenance strategy generation.

[0142] In some embodiments, the execution process of performing cluster analysis on the first abnormal feature tensor in step S1024 and obtaining similar data sets with the same or similar abnormal patterns by adjusting the dynamic radius parameter may further include the following sub-steps S241 to S244:

[0143] S241: Calculate the cosine similarity of the first abnormal feature tensor, and construct a feature similarity matrix based on the cosine similarity.

[0144] For example, the cosine similarity between different first abnormal feature tensors is calculated by the following formula: :

[0145]

[0146] in, and are different first abnormal feature tensors respectively.

[0147] Then according to the cosine similarity Construct the feature similarity matrix S. Where S∈R N×N , indicating that S is a real number matrix with N rows and N columns, R represents a real number set, and N is the preset feature dimension.

[0148] S242: Calculate the dynamic neighborhood radius parameter according to the feature similarity matrix.

[0149] Among them, the dynamic neighborhood radius parameter The specific calculation formula is as follows:

[0150]

[0151] Where t represents the time; λ is the attenuation coefficient; δ is the smoothing factor; ||·|| F The exponential decay coefficient λ is used to control the parameter adjustment rate and prevent clustering instability caused by sudden changes in similarity.

[0152] S243: Screening the feature similarity matrix according to the dynamic neighborhood radius parameter to obtain an adjacency matrix.

[0153] For example, to satisfy The eigenvectors of construct the adjacency matrix A∈{0,1} n×n When processing data, the neighborhood radius of the adjacency matrix is ​​dynamically adjusted based on the distribution and characteristics of the dynamic neighborhood radius parameter to adapt to the local characteristics of the data. This dynamic adjustment can improve the adaptability and accuracy of the algorithm.

[0154] S244: extracting a core point set of the adjacency matrix, and performing spectral clustering segmentation on the core point set to generate a similar data set.

[0155] Through graph connectivity analysis, the core point set C={c k ∣deg(c k )≥θ}. Among them, deg(c k ) represents the node degree, and θ is the core point threshold.

[0156] Perform spectral clustering segmentation on the core point set, divide the feature space into K subclusters by Laplace matrix eigendecomposition, and generate and The same data set {D1, D2, ..., D K}, and remove the noise point set N=C c .

[0157] It should be noted that spectral clustering segmentation includes the following steps:

[0158] Construct the normalized Laplace matrix ; Where D is the degree matrix;

[0159] Calculate the eigenvector U∈R corresponding to the first K smallest eigenvalues ​​of L n×K ; Where U is a real matrix with n rows and K columns, and R represents a set of real numbers;

[0160] Perform K-means clustering on the row vectors of U to complete data partitioning.

[0161] S1025: Perform time-frequency domain joint feature extraction on the same data set to generate a second abnormal feature tensor.

[0162] In some embodiments, the implementation process of performing time-frequency domain joint feature extraction on the similar data set in step S1025 to generate the second abnormal feature tensor may further include the following sub-steps S251 to S253:

[0163] S251: Assign dynamic weights to similar data sets through a multi-head attention mechanism to generate an attention feature matrix.

[0164] For example, a multi-head attention mechanism is used to assign dynamic weights to similar data sets (i.e., time series data from multiple sensor channels) to generate an attention feature matrix with channel correlation. Data from multiple sensor channels typically exhibit strong temporal dependencies and inter-channel correlations, making it difficult for traditional methods to fully exploit these complex relationships. By introducing a multi-head attention mechanism, the model can capture multi-angle feature information from different subspaces, avoiding the information omission caused by a single weight assignment.

[0165] Specifically, when processing multi-source sensor channel data, the data from each sensor channel is treated as an independent feature stream. The model first embeds the time series data for each channel to extract basic features. Subsequently, a multi-head attention mechanism adaptively assigns dynamic weights to each channel based on the importance of each channel's features. These dynamic weights can be adjusted based on the fluctuation characteristics, abnormal change trends, or periodic patterns of the channel signal, thereby improving the model's adaptability to complex signals. A weighted aggregation operation is then used to integrate the channel features to generate a channel-correlated attention feature matrix. This ensures that the model fully captures the correlations and differences between sensor channels, laying the foundation for subsequent feature extraction and anomaly detection.

[0166] S252: Use discrete wavelet transform to decompose the attention feature matrix into low-frequency approximate components and high-frequency detail components, and map the low-frequency approximate components to the first sub-tensor and the high-frequency detail components to the second sub-tensor.

[0167] Exemplarily, the attention feature matrix is decomposed by using a discrete wavelet transform (DWT) to convert it into a low-frequency approximation component and a high-frequency detail component, so as to construct a first sub-tensor and a second sub-tensor. Since the attention feature matrix has integrated the time-series data and channel correlation features of the multi-source sensor, the internal signals often contain both smooth change components and sudden abnormal signals. As a time-frequency analysis method, the discrete wavelet transform can capture the local features of the signals at multiple scales and has strong separation ability for sudden signals and noise.

[0168] In a specific implementation, multi-layer decomposition is performed on the attention feature matrix to divide the signal into a low-frequency approximation component and a high-frequency detail component. The low-frequency approximation component retains the overall trend and long-term change features of the original signal and can represent the smooth information in the sensor channel. The high-frequency detail component mainly contains sudden abnormal signals, noise or other short-term dramatic change features and can highlight the short-time dynamic changes in the sensor signal. Further, the low-frequency approximation component is mapped to the first sub-tensor and the high-frequency detail component is mapped to the second sub-tensor, realizing multi-scale decomposition processing of the features and providing more discriminative data representation for subsequent feature orthogonalization and abnormality detection.

[0169] S253: Orthogonal projection is performed on the first sub-tensor and the second sub-tensor to generate a second abnormal feature tensor.

[0170] Exemplarily, the first sub-tensor and the second sub-tensor are subjected to orthogonal projection based on the Hilbert-Schmidt orthogonalization criterion to eliminate the collinearity interference between the features and generate the second abnormal feature tensor. Since the first sub-tensor and the second sub-tensor extracted in the previous steps may have a certain degree of feature redundancy, especially in multi-source sensor data, the features of each signal channel have a certain similarity and overlap, which easily leads to the collinearity problem in model training.

[0171] The collinearity interference reduces the generalization ability of the model and affects the accuracy of abnormality detection. Therefore, the embodiment of the application introduces the Hilbert-Schmidt orthogonalization criterion to perform orthogonal projection on the first sub-tensor and the second sub-tensor. Specifically, first, normalization processing is performed on the two types of feature sub-tensors to eliminate the influence of amplitude difference on the projection result. Then, based on the Hilbert-Schmidt criterion, the optimal projection matrix between the first sub-tensor and the second sub-tensor is calculated to make the projection results of the two as orthogonal as possible in the feature space, so as to reduce the feature redundancy to the maximum extent. Through the orthogonalized feature fusion operation, the second abnormal feature tensor is generated, so that the feature tensor has stronger discriminability and robustness and provides more accurate feature description for abnormal pattern recognition and device state perception.

[0172] In one embodiment, the first sub-tensor T∈Rn×m and the second subtensor F∈R n×m , construct the Gram matrix G=T T F. Where T and F are both real matrices with n rows and m columns, and R represents a real number set. Then, by decomposing G through QR decomposition (QRDecomposition), we can obtain the orthogonal basis matrix Q and the upper triangular matrix R. After the QR decomposition operation, the matrix Q is generated. T The orthogonalized feature tensor Q of Q=I (that is, the orthogonal basis matrix Q) is used as the second abnormal feature tensor.

[0173] S1026: Perform high-order feature fusion on the first abnormal feature tensor and the second abnormal feature tensor to generate an abnormal feature matrix.

[0174] In some embodiments, the implementation process of performing high-order feature fusion of the first abnormal feature tensor and the second abnormal feature tensor to generate an abnormal feature matrix in step S1026 may further include the following sub-steps S261 to S263:

[0175] S261. Calculate the tensor product of the first abnormal feature tensor and the second abnormal feature tensor to generate a fusion tensor.

[0176] For example, the first and second anomaly feature tensors are derived from different feature extraction processes, capturing different aspects of the signal's anomaly characteristics in the time and frequency domains, respectively. To more comprehensively represent these anomaly characteristics, a high-order feature fusion is performed on the two. By calculating the tensor product of these two tensors, the time and frequency domain features can be combined in a high-order space.

[0177] The tensor product is a computational method that can fuse multidimensional data. It not only considers the product relationship between the elements of two tensors but also preserves more multidimensional feature information. This helps to more deeply reveal potential abnormal patterns in time series data. The resulting fused tensor is multidimensional. It incorporates information from different feature spaces, ensuring the effective integration of time and frequency domain features, laying the foundation for subsequent feature dimensionality reduction and anomaly detection. By fusing high-order features, a rich fused tensor can be obtained, enabling a more accurate description of anomalies in multi-source sensor data.

[0178] S262: Perform layer-by-layer singular value decomposition on the fused tensor, retain the first r principal component eigenvectors, and perform dimensionality reduction mapping through the projection matrix to generate a dimensionality-reduced feature matrix.

[0179] For example, although the fused tensor obtained by tensor product contains rich feature information, in practical applications, the high dimensionality of the data may lead to excessive computational complexity and contain a lot of redundant information. Therefore, dimensionality reduction is required.

[0180] Singular Value Decomposition (SVD), a common dimensionality reduction method, maps high-dimensional data to a lower-dimensional space by performing SVD on the fused tensor. Specifically, SVD is used to decompose the fused tensor into multiple singular values ​​and corresponding eigenvectors. By selecting the first r largest singular values ​​and their corresponding eigenvectors, the key information in the data is retained. These principal component eigenvectors are then projected into a new, lower-dimensional space using a projection matrix to generate a reduced-dimensional feature matrix.

[0181] The reduced matrix not only removes redundant information but also retains the core features of the data, making subsequent calculations more efficient. This step, through effective dimensionality reduction, helps extract the most representative anomaly features, providing a refined and efficient data representation for anomaly detection.

[0182] S263: Applying a spectral norm constraint to the feature matrix after dimensionality reduction to generate an abnormal feature matrix.

[0183] For example, the reduced feature matrix significantly reduces redundant information while preserving the core features of the data. However, in practical applications, some noise or instability may still exist, affecting subsequent anomaly detection performance. To address this issue, embodiments of the present invention introduce spectral norm constraints to further optimize the feature matrix.

[0184] The spectral norm constraint is a regularization method that restricts the spectral norm of a matrix (i.e., the maximum singular value of the matrix) to prevent overfitting during training and improve the model's generalization ability.

[0185] In this step, applying a spectral norm constraint to the reduced feature matrix further suppresses noise components in the matrix, ensuring that the model focuses more on key anomaly features in the data and reducing misjudgments caused by data noise. Furthermore, the spectral norm constraint ensures that the generated anomaly feature matrix has good numerical stability and robustness, making it more adaptable to complex and dynamically changing environments. The optimized anomaly feature matrix using the spectral norm constraint has higher accuracy and can be more effectively used for anomaly detection and device status awareness.

[0186] In some embodiments, the implementation process of constructing multiple association hypergraphs based on the abnormal feature matrix and the preset maintenance example data may include the following sub-steps S1031 to S1035:

[0187] S1031: Based on a preset natural language model, vectorize the preset maintenance example data to obtain a first example feature matrix.

[0188] Maintenance example data is vectorized using a pre-built natural language model, converting key elements of the maintenance information into vector representations. For example, assuming the pre-built maintenance example data includes "replace hydraulic oil filter," "check battery voltage," and "calibrate temperature control equipment," a pre-trained word embedding model (such as BERT or Word2Vec) is used to convert this text information into a high-dimensional dense vector.

[0189] This process maps similar words to similar locations in the vector space based on their semantic relationships. For example, "hydraulic" and "oil filter" might have high similarity in the vector space, while "battery" and "voltage" would form relatively independent vector representations.

[0190] By performing this processing on each maintenance example, the semantic features of the maintenance task are captured, generating a first example feature matrix. This matrix includes information such as the maintenance task type, the components involved, and the urgency of the operation. This provides the foundation for subsequent feature matching and hypergraph construction.

[0191] S1032: Match the abnormal feature matrix with the first example feature matrix to obtain a second example feature matrix.

[0192] For example, the abnormal feature matrix extracted from device status perception is matched with the first example feature matrix. For example, suppose the emergency equipment's status data indicates "abnormal hydraulic pressure" at a certain moment. First, convert this abnormal information into an abnormal feature matrix and extract the key features associated with it (e.g., "pressure" exceeding the limit, "hydraulic pressure" failure, etc.). These features are then matched with the first example feature matrix. For example, suppose the maintenance example includes the task "Replace the hydraulic oil filter." This task is likely to have a high degree of match with the abnormal feature of "abnormal hydraulic pressure" (since both are related to hydraulics).

[0193] By calculating the similarity between the abnormality feature matrix and the first example feature matrix, the most relevant maintenance example tasks are selected and extracted to form a second example feature matrix. The second example feature matrix contains the maintenance tasks that best match the current equipment abnormality characteristics and provides guidance for subsequent maintenance decisions.

[0194] S1033: Input the abnormal feature matrix and the second example feature matrix into a preset natural language model to generate maintenance instruction information.

[0195] For example, natural language processing techniques can be used to combine abnormal features with maintenance example data to generate specific maintenance recommendations tailored to the current equipment condition. For example, suppose the abnormal feature matrix indicates "hydraulic pressure exceeds the limit," while the second example feature matrix contains tasks related to "hydraulic maintenance," such as "replace hydraulic oil" or "check the pressure regulating valve." This feature information is fed into a natural language model (such as BERT). Based on this input, the model generates specific maintenance instructions, such as "replace hydraulic oil and check the regulating valve."

[0196] The model also generates priority tags, such as "urgent treatment" or "normal treatment," based on the urgency of the maintenance task and the equipment's operating conditions. This intelligently generates detailed maintenance instructions based on the equipment's actual conditions, ensuring timely and appropriate maintenance.

[0197] S1034: Obtain map information associated with the abnormal feature matrix.

[0198] By acquiring map information associated with the abnormality feature matrix, the equipment maintenance decision-making process can be further improved. For example, if the abnormality feature matrix indicates a hydraulic pressure overrun fault, the associated equipment location and geographic information will need to be identified based on this abnormality to optimize maintenance scheduling. In real-world applications, equipment may be distributed across diverse geographic locations. Map information can help technicians understand factors such as the specific location of the equipment, the status of adjacent equipment, and transportation routes.

[0199] S1035: Construct a hypermap based on the map information, the maintenance instruction information, and preset hypermap construction conditions to obtain multiple related hypermaps.

[0200] In some embodiments, the implementation process of constructing a hypergraph based on the map information, the maintenance instruction information, and the preset hypergraph construction conditions in step S1035 to obtain multiple associated hypergraphs may further include the following sub-steps S351-S354:

[0201] S351: Clustering is performed based on the maintenance instruction information to obtain a maintenance type distribution index.

[0202] Through this step, the maintenance type can be classified, thereby determining the maintenance resources corresponding to the maintenance task and determining the scale of the associated hypergraph in advance.

[0203] S352: Determine a target hypergraph construction condition from preset hypergraph construction conditions according to the maintenance type distribution index.

[0204] For example, once the scale of the associated hypergraph is determined, target hypergraph construction conditions can be selected from the hypergraph construction conditions. For example, total transportation distance, total maintenance resource requirements, and maintenance personnel allocation can be considered. This ensures that the construction of the associated hypergraph is aligned with the current maintenance task, improving maintenance efficiency.

[0205] S353: Determine a hypergraph node from the map information according to the target hypergraph construction condition.

[0206] S354: Generate edges between hypergraph nodes according to the target hypergraph construction conditions, maintenance instruction information, and map information to obtain multiple associated hypergraphs.

[0207] In one example, suppose a piece of equipment has a maintenance task called "replace hydraulic oil." The associated equipment nodes and task nodes are connected via hyperedges. These nodes and their relationships are constructed into a hypergraph, which is then adjusted based on pre-defined hypergraph construction conditions (such as shared equipment maintenance resources and task priorities) to ensure the effectiveness and enforceability of the hypergraph. This resulting hypergraph can not only be used to schedule equipment maintenance tasks but also serve as a foundation for global decision support, helping to dynamically adjust maintenance strategies and resource allocation, thereby ensuring efficient and reliable equipment operation.

[0208] In some embodiments, the implementation process of generating edges between hypergraph nodes to obtain multiple associated hypergraphs in step S354 based on the target hypergraph construction conditions, maintenance instruction information, and map information may further include the following sub-steps S3541 to S3546:

[0209] S3541: Perform Gaussian kernel spatial encoding processing on the emergency equipment location coordinates in the map information to generate a hypergraph node set.

[0210] For example, the location coordinates of emergency equipment are extracted from map information and Gaussian kernel spatial encoding is performed on these coordinates. Map information typically includes the precise location of each emergency device, such as the device's GPS (Global Positioning System) coordinates. Gaussian kernel spatial encoding is a technique that maps location coordinates into a high-dimensional space using a Gaussian function. This technique can capture the relative distances between locations and spatial distribution characteristics.

[0211] Specifically, the location coordinates of each emergency device are transformed into a new spatial vector using a Gaussian kernel function. This vector not only preserves the original geospatial information but also effectively maps similarities between locations into a more compact, high-dimensional space. This approach generates a high-dimensional hypergraph node representation for each emergency device. These nodes represent the device's location in space, providing an accurate geographic information foundation for subsequent resource allocation, scheduling, and maintenance tasks.

[0212] S3542: Calculate the resource sharing rate matrix of the hypergraph node set according to the maintenance equipment, materials, and personnel configuration in the maintenance instruction information.

[0213] Illustratively, according to the maintenance equipment, materials, and personnel configuration in the maintenance instruction information, the resource sharing rate between each two nodes is calculated to further construct the resource sharing rate matrix of the hypergraph node set. The resource sharing rate matrix reflects the resource sharing rate between each two nodes in the hypergraph node set.

[0214] The purpose of this step is to evaluate the resource sharing between different emergency equipment nodes to help understand the resources that each device may share during maintenance. Assuming that two nodes represent emergency equipment at different locations, the maintenance tasks of both require the same maintenance equipment or staff. At this time, the degree of resource overlap between the two is measured by calculating the resource sharing rate between the two nodes. For example, if the maintenance tasks of both nodes require "hydraulic oil filter element" replacement, and the filter element can only be operated by the same group of personnel, then the resource sharing rate of the two nodes is high. By analyzing the maintenance equipment, materials, and personnel configuration, a resource sharing rate matrix is generated. This matrix can represent the resource allocation relationship between nodes and provide a basis for the connection between nodes and the generation of hyperedges in the subsequent steps.

[0215] S3543: Perform geographic proximity analysis on the hypergraph node set based on the traffic distance threshold in the target hypergraph construction conditions. For each node to be analyzed in the hypergraph node set, mark the nodes with a Euclidean distance less than or equal to the traffic distance threshold as adjacent nodes, and generate a traffic adjacency matrix.

[0216] Illustratively, geographic proximity analysis is performed on the hypergraph node set based on the traffic distance threshold in the target hypergraph construction conditions. Specifically, the Euclidean distance between each pair of nodes is calculated and compared with the preset traffic distance threshold. If the Euclidean distance between two nodes is less than or equal to the preset threshold, the two nodes are considered "adjacent nodes". This means that the transportation connection between them can be made within a reasonable distance.

[0217] Through this geographic proximity analysis, it can be determined which emergency equipment can be connected to each other through short-distance transportation or walking, and which requires remote scheduling or additional transportation arrangements. The generated traffic adjacency matrix reflects the relative geographical relationship between nodes and provides an important spatial reference for the subsequent generation of hyperedges, making the construction of the hypergraph more consistent with the actual geographical distribution and resource allocation.

[0218] S3544: Fuse the resource sharing rate matrix with the traffic adjacency matrix, generate a hybrid similarity matrix through weighted summation, and screen out a super-edge candidate set from the hybrid similarity matrix according to a dynamic threshold.

[0219] Exemplarily, the resource sharing rate matrix is fused with the traffic adjacency matrix to generate a hybrid similarity matrix through weighted summation. The process of weighted summation combines the information of resource sharing rate and traffic adjacency to comprehensively evaluate the correlation between nodes. For example, two nodes may be considered to have strong correlation even if they are geographically far apart, as long as they share resources highly. The hybrid similarity matrix is a comprehensive evaluation of the two kinds of information and can reflect the overall similarity between nodes under the conditions of geography, resource sharing, etc.

[0220] Subsequently, the hybrid similarity matrix is screened according to a dynamic threshold, and node pairs with similarity higher than the threshold are marked as candidate super-edges. These candidate super-edges represent possible resource sharing and scheduling relationships. The screened super-edge candidate set provides a basis for subsequent super-edge construction and helps determine which devices or tasks need to be scheduled and allocated resources in priority.

[0221] S3545: Aggregate all adjacent nodes and indirectly associated nodes of each node in the super-edge candidate set to generate a super-edge set.

[0222] Exemplarily, by analyzing the direct adjacent nodes of each node in the super-edge candidate set and the nodes indirectly associated through other nodes, these nodes are combined into a super-edge set. Specifically, in a hypergraph, a super-edge can connect not only two nodes but also multiple nodes, indicating that there is a certain resource sharing or scheduling association between these nodes. For example, a node shares the same group of maintenance personnel with multiple devices, or is in the same geographic area with multiple devices. These nodes are combined into a super-edge so as to be considered together in subsequent scheduling and maintenance tasks.

[0223] This step, through the aggregation operation of multiple nodes, more finely reflects the complex relationships and mutual dependencies between devices in the entire hypergraph structure, ensuring that the hypergraph can accurately represent the dynamic connections between all devices and resources.

[0224] S3546: Perform multi-level hypergraph convolution optimization on the super-edge set to generate multiple associated hypergraphs.

[0225] Exemplarily, the hypergraph convolution is a way of processing the connection relationship between hyperedges in a hierarchical manner. It can effectively learn the high-order dependency relationship between nodes. Through multi-layer convolution operation, the representation of the hyperedge set is gradually optimized, so as to extract more detailed association information between nodes. Each layer of convolution operation is aimed at aggregating the features of nodes and updating the connection strength between nodes according to the structure of hyperedge, so that the hyperedge can more accurately reflect the complex dependency relationship between nodes.

[0226] In the convolution process, the topological structure of multiple associated hypergraphs is also optimized layer by layer to reduce redundant relationships and enhance the influence of key nodes and hyperedges, thereby generating optimized multiple associated hypergraphs. The hypergraph topological structure can not only effectively represent the relationship between emergency equipment, maintenance tasks and resources, but also provide accurate support for subsequent decision and scheduling, ensuring that resources are reasonably configured and the maintenance and scheduling process is optimized.

[0227] In some embodiments, the present application also provides an emergency equipment maintenance optimization system as shown in Figure 2 The emergency equipment maintenance optimization system can be used to perform the emergency equipment maintenance optimization method of any one of the embodiments of the present application.

[0228] As shown in Figure 2 The emergency equipment maintenance optimization system 200 can mainly include a data acquisition module 201, a data analysis module 202 and a result generation module 203.

[0229] It should be noted that the data acquisition module 201, the data analysis module 202 and the result generation module 203 can be communicatively connected to each other to transmit data or instructions to each other.

[0230] The data acquisition module 201 is configured to acquire state data of the emergency equipment.

[0231] The data analysis module 202 is configured to extract an abnormal feature matrix from the state data based on a preset natural language model.

[0232] Exemplarily, the database of the data analysis module 202 has a maintenance record function. When a user finds an abnormality, the abnormality of the equipment can be filled in through a pre-designed equipment management applet. The equipment maintenance personnel can prepare maintenance related tools according to the abnormality, and record the maintenance after the maintenance is completed. Through this function, the maintenance efficiency of the product can be greatly improved, and the maintenance of the equipment can be traced.

[0233] The database of the data analysis module 202 also includes equipment maintenance record and configuration functions. When equipment parameters exhibit abnormalities or a device reaches maintenance deadline, users are notified via text message or device-side notification, alerting them to the need for maintenance. Users can also choose to perform maintenance themselves or contact the manufacturer based on their specific needs and maintenance requirements. Maintenance records can be filled out and the equipment maintenance checklist updated. Furthermore, customized maintenance configurations, maintenance notifications, maintenance methods, and maintenance records can be used to streamline equipment maintenance and reduce the risk of equipment abnormalities or damage.

[0234] The result generation module 203 is used to construct an abnormality hypergraph according to the abnormality feature matrix and preset maintenance example data, and generate regional maintenance information.

[0235] In some embodiments, the emergency equipment maintenance optimization system 200 also includes emergency equipment. Emergency equipment supports one-touch self-checking, which can execute self-checking procedures and proactively detect abnormal conditions. This is primarily used to monitor the equipment's condition when the user has not used the equipment for an extended period. It also allows non-professionals to easily inspect the equipment.

[0236] Emergency equipment supports device authorization. Users can temporarily borrow other customers' equipment by initiating the device authorization process. In the event of equipment shortage and emergency, equipment can be quickly deployed.

[0237] In some embodiments, the emergency equipment maintenance optimization system 200 may also include a remote control terminal. By remotely controlling the emergency equipment through the remote control terminal and sending control signals to the PLC address points, the power supply of the equipment can be turned on and off, the generator can be turned on and off, the circuit breaker can be turned on and off, the LED light group can be turned on and off, the advertising light can be turned on and off, the side lights can be turned on and off, the searchlight can be turned on and off, the warning light can be turned on and off, the support legs can be raised and lowered, the lighthouse pole can be raised and lowered, and the lighthouse can be tilted and lowered, thereby achieving remote dispatch of the equipment. In addition, the remote control terminal can be configured with batch intelligent control tasks. By connecting multiple devices, multiple devices can be controlled with one click, reducing manpower consumption and increasing equipment utilization efficiency.

[0238] For example, the remote control terminal features device notification configuration. Users can customize the method for identifying abnormal fields and select the notification method (SMS notification or client notification). Specifically, this can be used for abnormal voltage notifications, abnormal current notifications, low battery warnings, low fuel warnings, device obstacle notifications, and device altitude warnings.

[0239] In addition, the data display booth of the remote control terminal can be filtered by region and time, displaying statistics on equipment quantity, equipment usage, equipment abnormality data, and equipment maintenance data. This allows for a more intuitive understanding of the overall equipment status and maintenance information through statistical charts.

[0240] Reference Figure 3 , shows a structural block diagram of a maintenance optimization device for emergency equipment provided by an embodiment of the present invention, which may specifically include:

[0241] The data acquisition unit 301 is used to acquire the emergency equipment status data of the area to be maintained and the identification text information input by the user;

[0242] An abnormal feature matrix extraction unit 302 is used to extract an abnormal feature matrix from the emergency equipment status data according to the discriminant text information;

[0243] A correlation hypergraph construction unit 303 is configured to construct a plurality of correlation hypergraphs based on the abnormal feature matrix and preset maintenance example data;

[0244] The regional maintenance information generating unit 304 is configured to generate regional maintenance information according to the plurality of associated hypergraphs, wherein the regional maintenance information is used to perform maintenance and resource allocation on emergency equipment in the area to be maintained.

[0245] In an optional embodiment, the abnormal feature matrix extraction unit 302 includes:

[0246] A semantic parsing unit, configured to perform semantic parsing on the discriminant text information and generate a discriminant feature vector;

[0247] a context feature extraction unit, configured to extract context features from the emergency equipment status data based on the discriminant feature vector to obtain an initial abnormal feature set;

[0248] A first abnormal feature tensor construction unit is configured to extract nodes whose feature activation values ​​exceed a preset threshold value from the initial abnormal feature set and construct a first abnormal feature tensor;

[0249] a cluster analysis unit, configured to perform cluster analysis on the first abnormal feature tensor to obtain a similar data set;

[0250] A time-frequency domain joint feature extraction unit, configured to perform time-frequency domain joint feature extraction on the similar data set to generate a second abnormal feature tensor;

[0251] The high-order feature fusion unit is used to perform high-order feature fusion on the first abnormal feature tensor and the second abnormal feature tensor to generate an abnormal feature matrix.

[0252] In an optional embodiment, the cluster analysis unit includes:

[0253] a feature similarity matrix construction unit, configured to calculate the cosine similarity of the first abnormal feature tensor and construct a feature similarity matrix based on the cosine similarity;

[0254] an adjacency matrix construction unit, configured to calculate a dynamic neighborhood radius parameter according to the feature similarity matrix, and to screen the feature similarity matrix according to the dynamic neighborhood radius parameter to obtain an adjacency matrix;

[0255] The spectral clustering segmentation unit is used to extract the core point set of the adjacency matrix and perform spectral clustering segmentation on the core point set to generate a homogeneous data set.

[0256] In an optional embodiment, the time-frequency domain joint feature extraction unit includes:

[0257] An attention feature matrix generating unit, configured to assign dynamic weights to the similar data sets through a multi-head attention mechanism to generate an attention feature matrix;

[0258] A discrete wavelet transform unit, configured to decompose the attention feature matrix into low-frequency components and high-frequency components using discrete wavelet transform;

[0259] a mapping unit, configured to map the low-frequency component into a first sub-tensor and map the high-frequency component into a second sub-tensor;

[0260] An orthogonal projection unit is used to perform orthogonal projection on the first sub-tensor and the second sub-tensor to generate a second abnormal feature tensor.

[0261] In an optional embodiment, the association hypergraph construction unit 303 includes:

[0262] A vectorization processing unit, configured to perform vectorization processing on the preset maintenance example data to obtain a first example feature matrix;

[0263] a feature matrix matching unit, configured to match the abnormal feature matrix with the first example feature matrix to obtain a second example feature matrix;

[0264] a maintenance instruction information generating unit, configured to generate maintenance instruction information based on the abnormality feature matrix and the second example feature matrix;

[0265] a map information acquisition unit, configured to acquire map information associated with the abnormal feature matrix;

[0266] The hypergraph construction unit is used to construct a hypergraph according to the map information, the maintenance instruction information and preset hypergraph construction conditions to obtain multiple related hypergraphs.

[0267] In an optional embodiment, the hypergraph construction unit includes:

[0268] a maintenance instruction information clustering unit, configured to cluster the maintenance instruction information to obtain a maintenance type distribution index;

[0269] a target hypergraph construction condition determination unit, configured to determine a target hypergraph construction condition from preset hypergraph construction conditions according to the maintenance type distribution index;

[0270] a hypergraph node determination unit, configured to determine a hypergraph node from the map information according to the target hypergraph construction condition;

[0271] The associated hypergraph construction subunit is used to generate edges between the hypergraph nodes according to the target hypergraph construction conditions, the maintenance instruction information and the map information to obtain multiple associated hypergraphs.

[0272] In an optional embodiment, the map information includes emergency equipment location coordinate information; the maintenance instruction information includes maintenance equipment, supplies, and personnel configuration; and the association hypergraph construction subunit includes:

[0273] A Gaussian kernel spatial coding processing unit, configured to perform Gaussian kernel spatial coding processing on the emergency equipment positioning coordinate information to generate a hypergraph node set;

[0274] a resource sharing rate matrix calculation unit, configured to calculate a resource sharing rate matrix of the hypergraph node set according to the configuration of the maintenance equipment, materials, and personnel;

[0275] a traffic adjacency matrix generating unit, configured to perform a geographic proximity analysis on the hypergraph node set based on a traffic distance threshold in the target hypergraph construction condition, and for each node to be analyzed in the hypergraph node set, mark nodes whose Euclidean distance to the node to be analyzed is less than or equal to the traffic distance threshold as adjacent nodes, thereby generating a traffic adjacency matrix;

[0276] a hyperedge candidate set screening unit, configured to fuse the resource sharing rate matrix and the traffic adjacency matrix, generate a hybrid similarity matrix by weighted summation, and screen a hyperedge candidate set from the hybrid similarity matrix according to a dynamic threshold;

[0277] a hyperedge set generating unit, configured to aggregate all adjacent nodes and indirectly associated nodes of each node in the hyperedge candidate set to generate a hyperedge set;

[0278] The multi-level hypergraph convolution optimization unit is used to perform multi-level hypergraph convolution optimization on the hyperedge set to generate multiple associated hypergraphs.

[0279] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the aforementioned method embodiment.

[0280] It should be noted that, in order to enable those skilled in the art to better distinguish data of the same type but with different actual meanings, some technical features are distinguished by the first and second in the embodiments of the present invention. The first and second are only used for data distinction and have no other special meanings. It can be understood that the present invention does not impose any restrictions on this.

[0281] An embodiment of the present invention further provides an electronic device, the device including a processor and a memory:

[0282] The memory is used to store program codes and transmit the program codes to the processor;

[0283] The processor is configured to execute the maintenance optimization method for emergency equipment according to any embodiment of the present invention according to the instructions in the program code.

[0284] An embodiment of the present invention further provides a computer-readable storage medium, which is used to store program code, and the program code is used to execute the maintenance optimization method for emergency equipment according to any embodiment of the present invention.

[0285] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0286] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0287] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0288] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0289] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0290] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A maintenance optimization method for emergency equipment, characterized in that: include: Obtain emergency equipment status data of the area to be maintained and the judgment text information entered by the user; extracting an abnormal feature matrix from the emergency equipment status data according to the discriminant text information; Constructing multiple association hypergraphs based on the abnormal feature matrix and preset maintenance example data; Regional maintenance information is generated according to the multiple associated hypergraphs, and the regional maintenance information is used to maintain and allocate resources for emergency equipment in the area to be maintained.

2. The maintenance optimization method for emergency equipment according to claim 1, characterized in that: The extracting of an abnormal feature matrix from the emergency equipment status data according to the discriminant text information includes: Performing semantic analysis on the discriminant text information to generate a discriminant feature vector; Performing context feature extraction on the emergency equipment status data based on the discriminant feature vector to obtain an initial abnormal feature set; Extracting nodes whose feature activation values ​​exceed a preset threshold from the initial abnormal feature set to construct a first abnormal feature tensor; Performing cluster analysis on the first abnormal feature tensor to obtain a similar data set; Performing time-frequency domain joint feature extraction on the similar data set to generate a second abnormal feature tensor; Perform high-order feature fusion on the first abnormal feature tensor and the second abnormal feature tensor to generate an abnormal feature matrix.

3. The maintenance optimization method for emergency equipment according to claim 2, characterized in that: The performing cluster analysis on the first abnormal feature tensor to obtain a similar data set includes: Calculating the cosine similarity of the first abnormal feature tensor and constructing a feature similarity matrix based on the cosine similarity; Calculating a dynamic neighborhood radius parameter according to the feature similarity matrix, and screening the feature similarity matrix according to the dynamic neighborhood radius parameter to obtain an adjacency matrix; A core point set of the adjacency matrix is ​​extracted, and spectral clustering is performed on the core point set to generate a homogeneous data set.

4. The maintenance optimization method for emergency equipment according to claim 2, characterized in that: The performing time-frequency domain joint feature extraction on the similar data set to generate a second abnormal feature tensor includes: Assign dynamic weights to the similar data sets through a multi-head attention mechanism to generate an attention feature matrix; Decomposing the attention feature matrix into low-frequency components and high-frequency components using discrete wavelet transform; Mapping the low-frequency component into a first sub-tensor and mapping the high-frequency component into a second sub-tensor; Perform orthogonal projection on the first sub-tensor and the second sub-tensor to generate a second abnormal feature tensor.

5. The maintenance optimization method for emergency equipment according to any one of claims 1 to 4, characterized in that: The step of constructing multiple association hypergraphs based on the abnormal feature matrix and preset maintenance example data includes: Performing vectorization processing on the preset maintenance example data to obtain a first example feature matrix; Matching the abnormal feature matrix with the first example feature matrix to obtain a second example feature matrix; generating maintenance instruction information according to the abnormal feature matrix and the second example feature matrix; Acquiring map information associated with the abnormal feature matrix; A hypergraph is constructed according to the map information, the maintenance instruction information and preset hypergraph construction conditions to obtain a plurality of associated hypergraphs.

6. The maintenance optimization method for emergency equipment according to claim 5, characterized in that: The step of constructing a hypergraph according to the map information, the maintenance instruction information, and preset hypergraph construction conditions to obtain a plurality of associated hypergraphs includes: performing clustering based on the maintenance instruction information to obtain a maintenance type distribution index; determining a target hypergraph construction condition from preset hypergraph construction conditions according to the maintenance type distribution index; Determining a hypergraph node from the map information according to the target hypergraph construction condition; According to the target hypergraph construction condition, the maintenance instruction information and the map information, edges are generated between the hypergraph nodes to obtain a plurality of associated hypergraphs.

7. The maintenance optimization method for emergency equipment according to claim 6, characterized in that: The map information includes emergency equipment location coordinate information; the maintenance instruction information includes maintenance equipment, supplies, and personnel configuration; and edges are generated between the hypergraph nodes based on the target hypergraph construction conditions, the maintenance instruction information, and the map information to obtain multiple associated hypergraphs, including: Performing Gaussian kernel spatial encoding processing on the emergency equipment location coordinate information to generate a hypergraph node set; Calculating a resource sharing rate matrix of the hypergraph node set based on the maintenance equipment, materials, and personnel configuration; Based on the traffic distance threshold in the target hypergraph construction condition, a geographic proximity analysis is performed on the hypergraph node set. For each node to be analyzed in the hypergraph node set, nodes whose Euclidean distance to the node to be analyzed is less than or equal to the traffic distance threshold are marked as adjacent nodes, thereby generating a traffic adjacency matrix. fusing the resource sharing rate matrix and the traffic adjacency matrix, generating a hybrid similarity matrix by weighted summation, and screening a hyperedge candidate set from the hybrid similarity matrix according to a dynamic threshold; Aggregating all adjacent nodes and indirectly associated nodes of each node in the hyperedge candidate set to generate a hyperedge set; Multi-level hypergraph convolution optimization is performed on the hyperedge set to generate multiple associated hypergraphs.

8. A maintenance optimization device for emergency equipment, characterized in that: include: A data acquisition unit, used to acquire the emergency equipment status data of the area to be maintained and the judgment text information input by the user; an abnormal feature matrix extraction unit, configured to extract an abnormal feature matrix from the emergency equipment status data according to the discriminant text information; A correlation hypergraph construction unit, configured to construct a plurality of correlation hypergraphs based on the abnormal feature matrix and preset maintenance example data; The regional maintenance information generating unit is used to generate regional maintenance information according to the multiple associated hypergraphs, wherein the regional maintenance information is used to maintain and allocate resources for emergency equipment in the area to be maintained.

9. An electronic device, characterized in that: The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the maintenance optimization method for emergency equipment according to any one of claims 1 to 7 according to instructions in the program code.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store program code, and the program code is used to execute the maintenance optimization method for emergency equipment according to any one of claims 1 to 7.

Images