Electrical fire early warning detection and equipment installation self-diagnosis method and system

Abstract

The invention relates to the technical field of electrical safety monitoring, in particular to an electrical fire early warning detection and equipment installation self-diagnosis method and system. The method comprises the following steps: constructing an equipment safety diagnosis model and a learning data set; the electrical fire monitoring equipment is installed for the first time and powered on for operation, multivariate time sequence monitoring data generated in the operation process of the electrical fire monitoring equipment is collected and input into the equipment safety diagnosis model, and a safety evaluation result output by the equipment safety diagnosis model is obtained. According to the electrical fire monitoring equipment installation self-diagnosis method provided by the invention, self-diagnosis of the installation condition of an electrical fire monitoring system is realized, correct installation and operation of the electrical fire monitoring system are ensured, and improvement of the accuracy, the intelligent level and the adaptability of electrical fire monitoring is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical safety monitoring, and in particular to an electrical fire early warning detection and equipment installation self-diagnosis method and system. Background Art

[0002] In the field of electrical fire monitoring, abnormal detection and early warning of electrical lines and electrical equipment are crucial to preventing fire accidents. Electrical fires are one of the main causes of fire accidents in buildings, especially in old buildings. Due to aging electrical lines, non-standard construction, and the use of inferior materials, the risk of electrical fires is extremely high. At present, most electrical fire monitoring systems rely on basic current and temperature monitoring, which can usually only monitor abnormal changes in a single parameter and do not integrate real-time analysis of multiple parameters, resulting in the inability to identify potential risks in a timely manner in complex situations. In addition, traditional electrical fire monitoring systems are prone to problems such as miswiring and improper installation during the installation process, resulting in reduced monitoring accuracy and even false alarms, affecting the reliability and use of the system.

[0003] US Patent US20140028253A1 proposes an electrical fire monitoring method based on residual current. This method uses abnormal changes in current to detect abnormal circuit conditions, but is limited to monitoring a single current parameter and fails to comprehensively analyze the change patterns of multiple electrical parameters, making it difficult to accurately determine more complex abnormal conditions. In addition, Chinese Patent CN102345678B describes an electrical fire alarm device based on temperature monitoring. The device issues an alarm by detecting abnormal increases in circuit temperature. However, in actual applications, it is difficult to distinguish different fault types, such as overload and poor contact, by simply relying on temperature parameters.

[0004] In summary, the existing electrical fire warning technology has the following defects:

[0005] 1. Insufficient monitoring accuracy: Existing electrical fire monitoring systems are mostly based on single parameters (such as current or temperature) for detection, and lack the ability to conduct comprehensive analysis of multiple parameters. This method is prone to misjudgment in complex situations, and it is difficult to accurately distinguish various fault types such as overload, poor contact, and line damage, resulting in inaccurate alarm effects and the risk of false alarms and missed alarms.

[0006] 2. Errors are prone to occur during the installation process: The installation process of traditional electrical fire monitoring systems is prone to incorrect connection problems such as reverse connection of the neutral wire, missing neutral wire, insufficient connection of the live wire, etc., which will cause frequent false alarms or even failure of the system during operation. The existing technology lacks automatic diagnosis function for the installation process and relies on manual inspection, which has the risk of human omissions and affects the reliability of the monitoring system.

[0007] 3. Poor adaptability: Since the monitoring algorithms of existing systems are usually relatively simple, it is difficult to adapt to the complex situations of different environments and electrical load changes. For example, in the case of high load or aging lines, the single parameter alarm threshold of the traditional system may not meet the precise monitoring requirements, thus affecting the scope of application of the system.

[0008] 4. Low intelligence level: Most of the current monitoring systems lack intelligent analysis and self-learning capabilities, and cannot make adaptive adjustments based on the on-site environment or historical data, making it difficult to achieve real-time and efficient monitoring of electrical fires. The system often performs poorly when faced with new abnormal scenarios and is difficult to adapt to the complex and changeable characteristics of electrical fire risks. Summary of the invention

[0009] In order to overcome the defects of the electrical fire monitoring system in the above-mentioned prior art in terms of monitoring accuracy, installation process, adaptability and intelligence level, the present invention proposes a self-diagnosis method for installation of electrical fire monitoring equipment, which realizes self-diagnosis of the installation status of the electrical fire monitoring system, ensures the correct installation and operation of the electrical fire monitoring system, and is conducive to improving the accuracy, intelligence level and adaptability of electrical fire monitoring.

[0010] The present invention provides an electrical fire monitoring device installation self-diagnosis method, comprising the following steps:

[0011] S1. Construct an equipment safety diagnosis model and a learning data set; train the equipment safety diagnosis model on the learning data set until convergence; the input of the equipment safety diagnosis model is the multivariate time series monitoring data collected during the operation of the electrical equipment, and the output of the equipment safety diagnosis model is the safety assessment result; the safety assessment result is associated with the risk level of electrical fire;

[0012] The equipment safety diagnosis model is improved on the basis of the TFT model. The improvement method is as follows: a weighted attention module and a multi-scale attention module are connected in parallel between the input layer and the time distribution embedding layer, and the multi-head attention module is replaced by a contextual embedding module and a multi-head sparse attention module connected before and after.

[0013] The weighted attention module introduces a multi-parameter weighted attention mechanism to generate dynamic weights for each monitoring object; the monitoring objects correspond one-to-one to the data categories of the multivariate time series monitoring data; the multi-scale attention module captures the complex interactive relationship between short-term abnormal fluctuations and long-term trend changes in the multivariate time series monitoring data;

[0014] The context embedding module embeds and fuses the feature representation of the acquired multivariate time-series monitoring data with the environmental data to generate a feature representation that adapts to the scene requirements; the multi-head sparse attention module integrates the sparse attention mechanism and the multi-head attention mechanism to process the feature representation output by the context embedding module and output it;

[0015] S2. After the electrical fire monitoring equipment is installed for the first time and powered on, the multivariate time series monitoring data generated during the operation of the electrical fire monitoring equipment is collected and input into the equipment safety diagnosis model to obtain the safety assessment results output by the equipment safety diagnosis model.

[0016] Preferably, the weighted attention module includes: a fully connected layer, a feature weighted layer and a first output layer; the input of the fully connected layer is connected to the input layer to obtain multivariate time series monitoring data, the fully connected layer generates dynamic weights for each type of data, and the input of the feature weighted layer performs weighted calculation on each type of data in the multivariate time series monitoring data obtained by the input layer according to the dynamic weights output by the fully connected layer; the first output layer performs convolution processing on the weighted features output by the feature weighted layer.

[0017] Preferably, the multi-scale attention module includes a short-term window module, a first attention module, a second attention module, a feature aggregation module and a second output layer; the multivariate time series monitoring data obtained by the input layer is divided into short window sequences by the short-term window module, and the data attention features of approximate instantaneous features are extracted by the first attention module and output after weighted operation; the multivariate time series monitoring data is also directly input into the second attention module to extract the attention features of the long sequence and output after weighted operation; the weighted features output by the first attention module and the weighted features output by the second attention module are dimensionally superimposed in the feature aggregation layer and then input into the second output layer for convolution processing, and then used as the output data of the multi-scale attention module.

[0018] Preferably, the context embedding module includes a first feature encoding layer, an environmental information input layer, a second feature encoding layer, a feature splicing layer and a third output layer; the first feature encoding layer encodes the feature representation output by the time distribution embedding layer; the environmental information input layer extracts and outputs environmental features in combination with the obtained environmental data, and the environmental data includes environmental conditions and load types; the second feature encoding layer encodes the environmental features output by the environmental information input layer; the feature splicing layer dimensionally splices the feature codes output by the first feature encoding layer and the feature codes output by the second feature encoding layer, and the third output layer performs convolution processing on the spliced ​​features output by the feature splicing layer.

[0019] Preferably, the multi-head sparse attention module includes a sparse pattern generation module, a multi-head attention module and a fourth output layer; the input of the sparse pattern generation module and the input of the multi-head attention module are both connected to the output of the context embedding module. The sparse pattern generation module introduces a sparsity constraint and generates a sparse attention pattern as an output; the multi-head attention module uses a multi-head attention mechanism to process the input data features and generates multi-head attention features; the input of the fourth output layer is the matrix product of the sparse attention pattern and the multi-head attention feature; the fourth output layer performs convolution processing on the input data.

[0020] Preferably, the sparse pattern generation module adopts a weight decay model.

[0021] Preferably, the safety assessment results include correct wiring, reverse insertion of the neutral wire, no insertion of the neutral wire, and insufficient insertion of the live wire.

[0022] Preferably, the data categories of the multivariate time-series monitoring data include at least two of current, voltage, residual current and wire temperature.

[0023] The present application proposes an electrical fire early warning detection method, comprising the following steps:

[0024] Set up the corresponding relationship between security assessment results and risk categories, and some risk categories are associated with corresponding alarm procedures;

[0025] Installing electrical fire monitoring equipment, executing the electrical fire monitoring equipment installation self-diagnosis method, and obtaining a trained equipment safety diagnosis model and a correctly installed electrical fire monitoring equipment;

[0026] Collect the operating data of the monitored equipment through the electrical fire warning equipment, construct multivariate time series monitoring data and input it into the equipment safety diagnosis model, and obtain the safety assessment results of the monitored equipment and the corresponding risk category output by the equipment safety diagnosis model;

[0027] Determine whether the risk category is associated with an alarm program. If so, execute the corresponding alarm program.

[0028] The present invention proposes a system adopting the electrical fire early warning detection method, comprising a memory and a processor, wherein the memory is used to store a device safety diagnosis model and a computer program, and when the computer program is executed, it is used to implement the electrical fire early warning detection method.

[0029] The advantages of the present invention are:

[0030] (1) This application combines the multivariate time-series monitoring data generated during the initial operation of the electrical fire monitoring equipment, and uses a deep learning model to perform real-time analysis of various abnormal scenarios, so as to promptly determine whether the electrical fire monitoring equipment is installed correctly. The system has a self-diagnosis function during the installation process, which can automatically detect installation anomalies such as reverse neutral wire insertion, lack of neutral wire insertion, and insufficient live wire insertion, ensuring installation accuracy and work reliability. This application uses an intelligent algorithm that can adaptively adjust monitoring parameters according to on-site environment and load changes, improve the adaptability and intelligence level of the system, and thus achieve real-time, efficient, and accurate early warning and monitoring of electrical fire risks.

[0031] (2) Multivariate time series monitoring data contains multiple types of data and integrates multiple electrical parameter monitoring and abnormal diagnosis technologies. It can identify multiple faults of electrical lines and equipment in real time to ensure that electrical fire monitoring is safe and reliable during installation and operation, and achieve accurate fault detection and timely alarm.

[0032] (3) The electrical fire early warning detection method proposed in this application uses a multi-parameter comprehensive analysis of current, voltage, temperature, residual current and other data, and uses a machine learning algorithm to build a multi-parameter correlation model, which can accurately identify various fault types such as line overload, poor contact, line damage, and neutral line sharing. Compared with the traditional single parameter monitoring method, this application has higher accuracy in complex electrical fault diagnosis and effectively reduces false alarms and missed alarms.

[0033] (4) The equipment safety diagnosis model proposed in this application can automatically adjust the monitoring parameter threshold according to the electrical load and environmental changes in different places, so that the system can adapt to various load changes, different building environments and line aging conditions, and meet a wide range of application needs. Its high adaptability and scalability effectively broaden the scope of application and is suitable for a variety of electrical fire monitoring scenarios.

[0034] (5) This application uses machine learning algorithms to analyze the collected data in real time and has self-learning capabilities. It can continuously optimize the equipment safety diagnosis model based on historical data and improve the accurate monitoring of electrical fire risks. The high degree of intelligence of the system not only improves the accuracy of fault diagnosis, but also simplifies manual operations, reduces the monitoring burden of staff, and improves operational efficiency.

[0035] (6) The equipment safety diagnosis model provided by this application sets the corresponding relationship between safety assessment results and risk levels, improves classification robustness through a layered soft label structure, and achieves high-precision identification of multiple types of faults, providing comprehensive and efficient core technical support for electrical fire monitoring systems. In addition, multi-level early warning prompts are provided according to the urgency of different fault types, so that managers can respond quickly according to emergency situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of a self-diagnosis method for installing an electrical fire monitoring device;

[0037] Figure 2 This is the structure diagram of the equipment safety diagnosis model improved based on the TFT model;

[0038] Figure 3 This is the structural diagram of the TFT model;

[0039] Figure 4 This is the structure diagram of the weighted attention module;

[0040] Figure 5 This is the structure diagram of the multi-scale attention module;

[0041] Figure 6 Embed module structure diagram for context;

[0042] Figure 7 This is the structure diagram of the multi-head sparse attention module;

[0043] Figure 8 It is a flow chart of the electrical fire early warning detection method;

[0044] Fig. 9 For model performance comparison;

[0045] Fig.10 For model delay comparison. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0047] Reference Figure 1 The present application proposes a method for self-diagnosis of electrical fire monitoring equipment installation, comprising the following steps:

[0048] S1. Construct an equipment safety diagnosis model and a learning data set; train the equipment safety diagnosis model on the learning data set until convergence; the input of the equipment safety diagnosis model is the multivariate time series monitoring data collected during the operation of the electrical equipment, and the output of the equipment safety diagnosis model is the safety assessment result, including correct wiring, reverse neutral line, no neutral line, insufficient live line, etc.

[0049] The multivariate time series monitoring data includes: multiple items of current, voltage, residual current and wire temperature.

[0050] S2. After the electrical fire monitoring equipment is installed for the first time and powered on, the multivariate time series monitoring data generated during the operation of the electrical fire monitoring equipment is collected and input into the equipment safety diagnosis model to obtain the safety assessment results output by the equipment safety diagnosis model.

[0051] In this embodiment, after the electrical fire monitoring equipment is installed, it first performs a self-diagnosis of its own installation status, effectively avoiding erroneous detection results of the electrical system caused by incorrect installation of the electrical fire monitoring equipment, improving the reliability of the electrical fire monitoring system in monitoring other electrical equipment, ensuring the correct installation of the electrical fire monitoring equipment in different environments, and improving adaptability.

[0052] Reference Figure 2 The equipment safety diagnosis model includes an input layer, a weighted attention module, a multi-scale attention module, a time distribution embedding layer, a context embedding module, a multi-head sparse attention module and a decoder; the input layer is used to access the multivariate time series monitoring data and transmit it to the weighted attention module and the multi-scale attention module respectively; the output of the weighted attention module and the output of the multi-scale attention module are both connected to the input of the time distribution embedding layer, and the input data of the time distribution embedding layer is the dimensional superposition result of the output data of the weighted attention module and the output data of the multi-scale attention module; the time distribution embedding layer, the context embedding module, the multi-head sparse attention module and the decoder are connected in sequence, and the output of the decoder is the output of the equipment safety diagnosis model.

[0053] It is obvious that the equipment safety diagnosis model proposed in this application is improved on the TFT model. The structure of the TFT model is as follows: Figure 3 Compared with the TFT model, the device safety diagnosis model proposed in this application connects a weighted attention module and a multi-scale attention module in parallel between the input layer and the time distribution embedding layer, and replaces the multi-head attention module with a contextual embedding module and a multi-head sparse attention module that are connected forward and backward.

[0054] The weighted attention module introduces a multi-parameter weighted attention mechanism to generate dynamic weights for each monitored object; the monitored objects correspond one-to-one to the data categories of the multivariate time series monitoring data, that is, the monitored objects are current, voltage, residual current or wire temperature.

[0055] Reference Figure 4 The weighted attention module proposed in this application includes: a fully connected layer, a feature weighted layer and a first output layer; the input of the fully connected layer is connected to the input layer to obtain multivariate time series monitoring data, the fully connected layer generates dynamic weights for various types of data, and the input of the feature weighted layer performs weighted calculations on various types of data in the multivariate time series monitoring data obtained by the input layer according to the dynamic weights output by the fully connected layer; the first output layer performs convolution processing on the weighted features output by the feature weighted layer and then outputs them. In specific implementation, the first output layer can use a convolutional layer.

[0056] Specifically, let the multivariate time series monitoring data obtained by the input layer be denoted as X;

[0057] X={X1;X2;X3;…;Xt;…;XT}

[0058] Xt={x(t,1);x(t,2);…;x(t,n);…;x(t,N)}

[0059] Where T represents the time step, 1≤t≤T; Xt represents the data collected at the tth time point; x(t,n) represents the nth type of data collected at the tth time point, N is the total number of data types, 1≤n≤N; it can be seen that X1, X2, X3, and XT represent the data collected at the 1st, 2nd, 3rd, and Tth time points respectively; x(t,1), x(t,2), and x(t,N) represent the 1st, 2nd, and Nth types of data collected at the tth time point respectively;

[0060] The fully connected layer generates dynamic weights for each type of data, and the weight of the nth type of data generated by it is a n , then the feature weighting layer performs the following operations:

[0061] X'={a n x(t,n)|1≤t≤T;1≤n≤N}

[0062] X' is the weighted feature output by the feature weighting layer.

[0063] Reference Figure 5 The multi-scale attention module captures the complex interactive relationship between short-term abnormal fluctuations and long-term trend changes of multivariate time series monitoring data; it uses different time windows to segment the input multivariate time series monitoring data, performs hierarchical analysis of electrical parameters at different time scales, prioritizes the capture of sudden anomalies, and further improves the response speed to transient faults.

[0064] Specifically, the multi-scale attention module includes a short-term window module, a first attention module, a second attention module, a feature aggregation module, and a second output layer. The multivariate time series monitoring data obtained by the input layer is divided into short window sequences by the short-term window module, and the first attention module extracts the data attention features of the approximate instantaneous features and outputs them after weighted operation; the multivariate time series monitoring data is also directly input into the second attention module to extract the attention features of the long sequence and output them after weighted operation. The weighted features output by the first attention module and the weighted features output by the second attention module are input into the second output layer for convolution processing after dimension superposition in the feature aggregation layer, and then used as the output data of the multi-scale attention module.

[0065] Specifically, the feature aggregation layer performs dimension superposition on the weighted features output by the first attention module and the weighted features output by the second attention module in the time dimension.

[0066] The input of the time distribution embedding layer is the output data of the weighted attention module and the output data of the multi-scale attention module, which are superimposed on the time dimension; the time distribution embedding layer embeds the selected features into time series to generate a more expressive feature representation.

[0067] Reference Figure 6The context embedding module includes a first feature encoding layer, an environmental information input layer, a second feature encoding layer, a feature concatenation layer, and a third output layer. The first feature encoding layer encodes the feature representation output by the time distribution embedding layer; the environmental information input layer extracts and outputs environmental features in combination with the obtained environmental data, and the environmental data includes environmental conditions such as temperature, humidity, and load type; the second feature encoding layer encodes the environmental features output by the environmental information input layer; the feature concatenation layer concatenates the feature codes output by the first feature encoding layer and the feature codes output by the second feature encoding layer, and the third output layer convolves the concatenated features output by the feature concatenation layer and outputs them as the output data of the context embedding module.

[0068] Specifically, the first feature encoding layer, the second feature encoding layer and the third output layer all use convolutional networks, and the feature concatenation layer uses a Cancat network.

[0069] Reference Figure 7 The multi-head sparse attention module includes a sparse pattern generation module, a multi-head attention module and a fourth output layer; the input of the sparse pattern generation module and the input of the multi-head attention module are both connected to the output of the context embedding module. The sparse pattern generation module introduces sparsity constraints to reduce redundant parameters. It selects the key positions of the input data features through a lightweight network and generates sparse attention patterns as outputs; the multi-head attention module uses a multi-head attention mechanism to process the input data features and generates multi-head attention features. The input of the fourth output layer is the matrix product of the sparse attention pattern and the multi-head attention feature; the fourth output layer performs convolution processing on the input data as the output of the multi-head sparse attention module.

[0070] Specifically, the sparse pattern generation module adopts a weight decay model, and the fourth output layer adopts a convolutional network.

[0071] The decoder decodes the data features output by the multi-head sparse attention module to obtain a security assessment result.

[0072] In this implementation, the time distributed embedding layer, the multi-head attention module in the multi-head sparse attention module, and the decoder are obtained from the TFT model (Temporal Fusion Transformer).

[0073] The equipment safety diagnosis model provided in this application introduces a multi-parameter weighted attention mechanism through a weighted attention module, generates dynamic weights for each input feature through a learnable fully connected layer, and highlights the impact of key features such as current and temperature on fault detection. The complex interactive relationship between short-term abnormal fluctuations and long-term trend changes is captured through a multi-scale attention mechanism, and the input data is segmented using different time windows. The electrical parameters are graded and analyzed at different time scales to prioritize the capture of sudden anomalies, further improving the response speed to transient faults. Context-aware feature embedding is introduced through the context embedding module, and the input features are embedded and fused with contextual information such as environmental conditions and load types to generate feature representations that adapt to specific scene requirements, improve adaptability to complex application scenarios, and improve the accuracy and robustness of fault detection. The model integrates a sparse attention mechanism to reduce computational complexity, and supports embedded hardware deployment through lightweight design to ensure the real-time and applicability of the model and further optimize system performance.

[0074] Reference Figure 8 The electrical fire warning detection method proposed in this application includes the following steps.

[0075] S1. Set the corresponding relationship between security assessment results and risk categories, and some risk categories are associated with corresponding alarm procedures;

[0076] S2. Install the electrical fire monitoring equipment, execute the above-mentioned electrical fire monitoring equipment installation self-diagnosis method, and obtain the trained equipment safety diagnosis model and the correctly installed electrical fire monitoring equipment;

[0077] S3. Collect the operating data of the monitored equipment through the electrical fire warning equipment, construct the multivariate time series monitoring data and input it into the equipment safety diagnosis model, and obtain the safety assessment results of the monitored equipment and the corresponding risk categories output by the equipment safety diagnosis model;

[0078] S4. Determine whether the risk category is associated with an alarm program. If so, execute the corresponding alarm program.

[0079] The electrical fire early warning detection system proposed in the present application includes an electrical fire monitoring device and an alarm feedback module; the electrical fire monitoring device includes a data acquisition module and a fault diagnosis module.

[0080] The fault diagnosis module is connected to the data acquisition module and the alarm feedback module respectively.

[0081] The data acquisition module monitors the operating data of electrical equipment in real time through sensors, especially key parameters such as current, voltage, temperature and residual current in the line;

[0082] The fault diagnosis module stores the trained equipment safety diagnosis model. The fault diagnosis module obtains the key parameters collected by the data acquisition module and organizes them into a multivariate data monitoring sequence; the fault diagnosis module inputs the multivariate data monitoring sequence into the equipment safety diagnosis model to obtain a safety assessment result.

[0083] The fault diagnosis module is also used to identify the risk category of the safety assessment results. Specifically, the safety assessment results other than correct wiring are recorded in the risk category and different risk levels are marked based on experience. When the safety assessment result is a risk category, the fault diagnosis module will notify the alarm feedback module of the assessment result and risk level. The alarm feedback module executes the alarm procedure, such as issuing audible and visual alarms and notifying the corresponding executive personnel.

[0084] In this way, when a fault or installation anomaly is detected, an audible and visual alarm can be immediately issued through the alarm feedback module, and the fault information can be fed back to the background management system through wireless communication, so that managers can respond in time and take protective measures. In addition, the system provides a graded early warning function, adjusting the alarm level according to the urgency of the fault, ensuring rapid and targeted fault handling, thereby improving the safety and management efficiency of electrical fire monitoring.

[0085] The following is a verification of the above equipment safety diagnosis model in conjunction with a specific embodiment. This embodiment uses four existing models, namely LSTM (Long Short-Term Memory Network), Transformer model, TCN model (Temporal Convolutional Network) and TFT model (Temporal Fusion Transformer), as comparison models.

[0086] The learning data set constructed in this application is {voltage, current, residual current, wire temperature; safety assessment results}; the learning data set is divided into a training set and a test set.

[0087] In this embodiment, four comparison models and Figure 2 The equipment safety diagnosis models shown are all trained on the training set until convergence, and then tested on the test set. The test results are shown in Table 1 and Fig. 9 , Fig.10 As shown in the figure, the present application exceeds all the comparison models in terms of accuracy, recall rate and F1 value, and compared with the TFT model, it greatly reduces the delay and improves the diagnostic efficiency of electrical equipment.

[0088] Table 1: Model performance comparison

[0089]

[0090]

[0091] Of course, it is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any reference numerals in the claims should not be regarded as limiting the claims involved.

[0092] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0093] The techniques, shapes, and structural parts not described in detail in the present invention are all well-known techniques.

Claims

1. A method for self-diagnosis of electrical fire monitoring equipment installation, characterized in that: The following steps are involved: S1. Build equipment safety diagnosis model and learning dataset; The equipment safety diagnosis model is trained on the learning data set until convergence; the input of the equipment safety diagnosis model is the multivariate time series monitoring data collected during the operation of the electrical equipment, and the output of the equipment safety diagnosis model is the safety assessment result; the safety assessment result is associated with the risk level of the electrical fire; The equipment safety diagnosis model is improved on the basis of the TFT model. The improvement method is as follows: a weighted attention module and a multi-scale attention module are connected in parallel between the input layer and the time distribution embedding layer, and the multi-head attention module is replaced by a contextual embedding module and a multi-head sparse attention module connected before and after. The weighted attention module introduces a multi-parameter weighted attention mechanism to generate dynamic weights for each monitored object; The monitoring objects correspond one to one with the data categories of the multivariate time series monitoring data; The multi-scale attention module captures the complex interactive relationship between short-term abnormal fluctuations and long-term trend changes in multivariate time series monitoring data; The context embedding module embeds and fuses the feature representation of the acquired multivariate time-series monitoring data with the environmental data to generate a feature representation that adapts to the scene requirements; The multi-head sparse attention module integrates the sparse attention mechanism and the multi-head attention mechanism to process and output the feature representation output by the context embedding module; S2. After the electrical fire monitoring equipment is installed for the first time and powered on, the multivariate time series monitoring data generated during the operation of the electrical fire monitoring equipment is collected and input into the equipment safety diagnosis model to obtain the safety assessment results output by the equipment safety diagnosis model.

2. The method for self-diagnosis of electrical fire monitoring equipment installation according to claim 1, characterized in that: The weighted attention module includes: a fully connected layer, a feature weighted layer and the first output layer; the input of the fully connected layer is connected to the input layer to obtain multivariate time series monitoring data, the fully connected layer generates dynamic weights for various types of data, and the input of the feature weighted layer performs weighted calculations on various types of data in the multivariate time series monitoring data obtained by the input layer according to the dynamic weights output by the fully connected layer; the first output layer performs convolution processing on the weighted features output by the feature weighted layer.

3. The electrical fire monitoring equipment installation self-diagnosis method according to claim 1, characterized in that: The multi-scale attention module includes a short-term window module, a first attention module, a second attention module, a feature aggregation module and a second output layer; the multivariate time series monitoring data obtained by the input layer is divided into short window sequences by the short-term window module, and the data attention features of approximate instantaneous features are extracted by the first attention module and output after weighted operation; the multivariate time series monitoring data is also directly input into the second attention module to extract the attention features of the long sequence and output after weighted operation; the weighted features output by the first attention module and the weighted features output by the second attention module are input into the second output layer for convolution processing after dimension superposition in the feature aggregation layer, and then used as the output data of the multi-scale attention module.

4. The electrical fire monitoring equipment installation self-diagnosis method according to claim 1, characterized in that: The context embedding module includes a first feature encoding layer, an environmental information input layer, a second feature encoding layer, a feature concatenation layer, and a third output layer; the first feature encoding layer encodes the feature representation output by the time distribution embedding layer; the environmental information input layer extracts and outputs environmental features in combination with the obtained environmental data, and the environmental data includes environmental conditions and load types; the second feature encoding layer encodes the environmental features output by the environmental information input layer; The feature concatenation layer concatenates the feature codes output by the first feature coding layer and the feature codes output by the second feature coding layer, and the third output layer performs convolution processing on the concatenated features output by the feature concatenation layer.

5. The electrical fire monitoring equipment installation self-diagnosis method according to claim 1, characterized in that: The multi-head sparse attention module includes a sparse pattern generation module, a multi-head attention module and a fourth output layer; the input of the sparse pattern generation module and the input of the multi-head attention module are both connected to the output of the context embedding module. The sparse pattern generation module introduces sparsity constraints and generates sparse attention patterns as outputs; the multi-head attention module uses a multi-head attention mechanism to process input data features and generate multi-head attention features; the input of the fourth output layer is the matrix product of the sparse attention pattern and the multi-head attention feature; the fourth output layer performs convolution on the input data.

6. The method for self-diagnosis of electrical fire monitoring equipment installation according to claim 5, characterized in that: The sparse pattern generation module adopts a weight decay model.

7. The electrical fire monitoring equipment installation self-diagnosis method according to any one of claims 1 to 6, characterized in that: The safety assessment results include correct wiring, reverse insertion of the neutral wire, missing neutral wire, and insufficient live wire insertion.

8. The method for self-diagnosis of electrical fire monitoring equipment installation according to any one of claims 1 to 6, characterized in that: The data categories of the multivariate time series monitoring data include at least two of current, voltage, residual current and wire temperature.

9. An electrical fire early warning detection method using the electrical fire monitoring equipment installation self-diagnosis method as claimed in any one of claims 1 to 8, characterized in that: The following steps are involved: Set up the corresponding relationship between security assessment results and risk categories, and some risk categories are associated with corresponding alarm procedures; Installing electrical fire monitoring equipment, executing the electrical fire monitoring equipment installation self-diagnosis method according to any one of claims 1 to 8, and obtaining a trained equipment safety diagnosis model and a correctly installed electrical fire monitoring equipment; Collect the operating data of the monitored equipment through the electrical fire warning equipment, construct multivariate time series monitoring data and input it into the equipment safety diagnosis model, and obtain the safety assessment results of the monitored equipment and the corresponding risk category output by the equipment safety diagnosis model; Determine whether the risk category is associated with an alarm program. If so, execute the corresponding alarm program.

10. A system using the electrical fire early warning detection method according to claim 9, characterized in that: It comprises a memory and a processor, the memory is used to store a device safety diagnosis model and a computer program, and when the computer program is executed, it is used to implement the electrical fire early warning detection method as claimed in claim 9.

Images