Low-voltage multi-channel wireless temperature measurement method and device based on artificial intelligence

Abstract

The invention provides a low-voltage multi-channel wireless temperature measurement method and device based on artificial intelligence, and the method comprises the steps: collecting the multi-point temperature of low-voltage electrical equipment, transmitting the collected multi-point temperature data to a central data processing unit through a wireless protocol, and carrying out the preprocessing of the data; extracting features based on the preprocessed multi-point temperature data; taking positions of sensors in the low-voltage electrical equipment as nodes of a graph, taking features extracted from the preprocessed temperature data as features of the nodes of the graph, taking a connection relation between the sensors as edges, obtaining a graph structure, and training a graph neural network model; and inputting real-time data into the trained graph neural network model, obtaining a predicted value, comparing the difference between the predicted value and the detection value, and if the difference exceeds a preset threshold value, retraining the model. Through the method and the corresponding device, manual intervention can be reduced, the data processing and analysis efficiency can be improved, and the temperature change of the equipment can be quickly responded.

Description

Technical Field

[0001] The present invention proposes a low-voltage multi-channel wireless temperature measurement method and device based on artificial intelligence, and relates to the technical field of temperature measurement circuits. Background Art

[0002] In the temperature monitoring of low-voltage electrical equipment in modern industrial and power systems, traditional methods have many problems. Traditional contact temperature measurement methods have problems such as inconvenient installation, susceptibility to interference and difficult maintenance. Because they require complex wiring, they affect equipment operation and maintenance. Long-distance wire transmission is susceptible to electromagnetic interference and is easily affected by the environment, which increases costs. Traditional wireless single-point temperature measurement methods have limited monitoring range, lack of systematic analysis and unstable performance. They cannot fully monitor large and complex systems, and it is difficult to evaluate the overall thermal state. Communication problems will also affect data transmission. Existing data analysis methods also have limitations, mainly due to lack of predictability and insufficient information processing capabilities. Simple statistics and threshold judgments are used, which make it difficult to predict temperature trends and early warnings, and cannot fully mine the potential information of data. With the development of artificial intelligence technology, its powerful data processing and analysis capabilities make it possible to solve these problems. Therefore, a low-voltage multi-channel wireless temperature measurement method based on artificial intelligence has emerged. It aims to use graph neural networks, deep learning and other technologies to achieve efficient, accurate and comprehensive temperature monitoring and analysis, mine more information, achieve early prediction and active maintenance, and ensure the reliability and safety of equipment operation. Summary of the invention

[0003] The present invention provides a low-voltage multi-channel wireless temperature measurement method and device based on artificial intelligence to solve the above-mentioned problems: The present invention proposes a low-voltage multi-channel wireless temperature measurement method based on artificial intelligence, the method comprising: Collect multi-point temperatures of low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through a wireless protocol, and pre-process the transmitted multi-point temperature data; Extract features based on preprocessed multi-point temperature data; The sensor positions in the low-voltage electrical equipment are used as nodes of the graph, the features extracted from the preprocessed temperature data are used as features of the nodes of the graph, the connection relationships between the sensors are used as edges, the graph structure is obtained, and a graph neural network model is trained based on the graph structure; The data monitored in real time by the sensor is input into the trained graph neural network model to obtain the predicted value. The difference between the predicted value and the detected value is compared at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted.

[0004] Furthermore, the multi-point temperature of the low-voltage electrical equipment is collected, the collected multi-point temperature data is transmitted to the central data processing unit through a wireless protocol, and the transmitted multi-point temperature data is pre-processed, including: Wireless temperature sensors installed at switch contacts, busbar connection points and cable connectors of low-voltage electrical equipment collect temperature data in real time at a preset collection frequency. Transmitting the temperature data collected by the wireless temperature sensor to a central data processing unit via a wireless protocol, wherein the wireless protocol includes LoRa, ZigBee and Wi-Fi; For the transmitted temperature data set , first calculate the first quartile and the third quartile , , will be lower than or higher The data are regarded as outliers and the interpolation method is used to correct the outliers.

[0005] Furthermore, features are extracted based on the preprocessed multi-point temperature data, including: Extracting time features from the preprocessed multi-point temperature data, the time features including: the temperature average value, maximum value, minimum value and standard deviation within each preset time window; Extract the spatial features from the preprocessed multi-point temperature data, which include temperature gradient, temperature difference and relative temperature between different sensors. ,in, represents the temperature gradient, Indicates the low voltage circuit temperature measured by sensor i, represents the low voltage circuit temperature measured by sensor j, represents the distance between sensors i and j, and the relative temperature is ,in, Indicates relative temperature; According to the heat transfer principle, the formula

[0006] extracting thermodynamic features from the multi-point temperature data, wherein m represents the mass of the circuit measured by the sensor, c represents the specific heat capacity of the constituent material of the circuit measured by the sensor, represents the temperature measured by the sensor at time t, represents the temperature measured by the sensor at time t-1, It represents the amount of heat transfer in the circuit measured by the sensor during the time period from t to t-1.

[0007] Furthermore, the sensor positions in the low-voltage electrical equipment are used as nodes of the graph, the features extracted from the preprocessed temperature data are used as features of the nodes of the graph, and the connection relationships between the sensors are used as edges to obtain a graph structure, and a graph neural network model is trained based on the graph structure, including: Each wireless temperature sensor in the low-voltage electrical equipment is regarded as a node in the graph, marked with a unique identifier, i.e., a node number, and the extracted time feature, the spatial feature, and the thermodynamic feature are combined as a node feature vector, and the edge of the graph is constructed based on the connection relationship of the graph, and the node feature vector and the edge information are encapsulated into a graph structure; Construct a graph neural network model, which adopts a graph convolutional network in the GNN architecture. The graph neural network defines a class inherited from torch.nn.Module. The graph convolutional network includes three layers of GCNConv graph convolutional layers for extracting features in the graph structure. The first graph convolutional layer converts the input node features into features of the feature dimension of the hidden layer through a graph convolution operation. The second graph convolutional layer receives the output features of the feature dimension of the hidden layer of the first graph convolutional layer and outputs features of the same dimension. The third graph convolutional layer receives the output features of the feature dimension of the hidden layer of the second graph convolutional layer and converts them into one-dimensional predicted temperature values. Input the graph structure data into the constructed graph neural network model for training to obtain the trained graph neural network model.

[0008] Furthermore, the data monitored in real time by the sensor is input into the trained graph neural network model to obtain the predicted value, and the difference between the predicted value and the detected value is compared at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continuously predicted, including: Input the data monitored in real time by the sensor into the trained graph neural network model to obtain a first prediction value, and continue to input the first prediction value into the trained graph neural network model to obtain a second prediction value; The second predicted value is compared with the value actually detected by the sensor. If the difference between the second predicted value and the value actually detected by the sensor exceeds the preset threshold, the value actually detected by the sensor is used as incremental data to retrain the model. If it does not exceed the preset threshold, the prediction result is stored and added to the historical data set. The temperature of the low-voltage electrical equipment in the next time window is continued to be predicted based on the historical data set.

[0009] The present invention proposes a low-voltage multi-channel wireless temperature measurement device based on artificial intelligence, the device comprising: The acquisition module is used to collect the multi-point temperature of the low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through the wireless protocol, and pre-process the transmitted multi-point temperature data; A feature extraction module is used to extract features based on preprocessed multi-point temperature data; A training graph neural network model module is used to use the sensor positions in the low-voltage electrical equipment as nodes of the graph, the features extracted from the preprocessed temperature data as features of the nodes of the graph, and the connection relationships between the sensors as edges, to obtain a graph structure, and train a graph neural network model based on the graph structure; The prediction module is used to input the data monitored by the sensor in real time into the trained graph neural network model, obtain the predicted value, and compare the difference between the predicted value and the detected value at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted.

[0010] Furthermore, the acquisition module includes: The temperature data collection module is used to collect temperature data in real time according to a preset collection frequency through wireless temperature sensors installed at switch contacts, busbar connection points and cable connectors of low-voltage electrical equipment; A transmission module, used to transmit the temperature data collected by the wireless temperature sensor to the central data processing unit through a wireless protocol, wherein the wireless protocol includes LoRa, ZigBee and Wi-Fi; Correction module for the transmitted temperature data set , first calculate the first quartile and the third quartile , , will be lower than or higher The data are regarded as outliers and the interpolation method is used to correct the outliers.

[0011] Furthermore, the feature extraction module includes: A time feature extraction module is used to extract time features from the preprocessed multi-point temperature data, wherein the time features include: the temperature average value, maximum value, minimum value and standard deviation within each preset time window; The spatial feature extraction module is used to extract the spatial features in the preprocessed multi-point temperature data. The spatial features include: temperature gradient, temperature difference and relative temperature between different sensors. The temperature gradient is ,in, represents the temperature gradient, Indicates the low voltage circuit temperature measured by sensor i, represents the low voltage circuit temperature measured by sensor j, represents the distance between sensors i and j, and the relative temperature is ,in, Indicates relative temperature; Extract thermodynamic characteristics module, used to extract the heat transfer principle through the formula

[0012] extracting thermodynamic features from the multi-point temperature data, wherein m represents the mass of the circuit measured by the sensor, c represents the specific heat capacity of the constituent material of the circuit measured by the sensor, represents the temperature measured by the sensor at time t, represents the temperature measured by the sensor at time t-1, It represents the amount of heat transfer in the circuit measured by the sensor during the time period from t to t-1.

[0013] Furthermore, the training graph neural network model module includes: A graph construction module is used to regard each wireless temperature sensor in the low-voltage electrical equipment as a node in the graph, mark it with a unique identifier, i.e., a node number, combine the extracted time feature, the spatial feature, and the thermodynamic feature as a node feature vector, construct the edge of the graph based on the connection relationship of the graph, and encapsulate the node feature vector and the edge information into a graph structure; Construct a graph neural network model module, which is used to construct a graph neural network model. The graph neural network model adopts a graph convolutional network in the GNN architecture. The graph neural network defines a class inherited from torch.nn.Module. The graph convolutional network includes three layers of GCNConv graph convolutional layers, which are used to extract features in the graph structure. The first graph convolutional layer converts the input node features into features of the feature dimension of the hidden layer through a graph convolution operation. The second graph convolutional layer receives the output features of the feature dimension of the hidden layer of the first graph convolutional layer and outputs features of the same dimension. The third graph convolutional layer receives the output features of the feature dimension of the hidden layer of the second graph convolutional layer and converts them into one-dimensional predicted temperature values. The training module is used to input graph structure data into the constructed graph neural network model for training and obtain the trained graph neural network model.

[0014] Furthermore, the prediction module includes: A module for obtaining a second prediction value is used to input the data monitored in real time by the sensor into the trained graph neural network model to obtain a first prediction value, and further input the first prediction value into the trained graph neural network model to obtain a second prediction value; The correction prediction module is used to compare the second predicted value with the value actually detected by the sensor. If the difference between the second predicted value and the value actually detected by the sensor exceeds a preset threshold, the value actually detected by the sensor is used as incremental data to retrain the model. If it does not exceed the preset threshold, the prediction result is stored and added to the historical data set, and the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted based on the historical data set.

[0015] The invention has the following beneficial effects: accurate temperature monitoring and prediction, through multi-point temperature acquisition and graph neural network model training based on graph structure, fully considers the mutual relationship and change trend of the temperature of each part of low-voltage electrical equipment, and can more accurately predict the equipment temperature and timely discover potential overheating risks compared with the traditional monitoring method of single sensor or simple model, providing strong guarantee for the stable operation of the equipment; self-adaptation and self-optimization, with adaptive ability, when the difference between the predicted value and the detected value exceeds the preset range, it can automatically trigger the model to retrain. This enables the model to continuously optimize its own performance with the change of equipment operation status, environmental factors, etc., and continuously maintain a high prediction accuracy to ensure long-term stable and reliable temperature monitoring; early fault warning, accurate temperature prediction helps to timely discover abnormal temperature changes before the equipment fails, and send out early warning signals, and operation and maintenance personnel can take targeted measures according to the warning information, such as adjusting equipment operation parameters, arranging maintenance, etc., to avoid equipment failure, reduce maintenance costs and downtime, and improve equipment reliability and availability; efficient data processing and analysis, from data acquisition, transmission to preprocessing, feature extraction, model training and prediction, the entire process is automated and efficient. It reduces manual intervention, improves the efficiency of data processing and analysis, can quickly respond to equipment temperature changes, and adapt to the requirements of modern industrial production for real-time and efficient equipment monitoring; the constructed graph structure is not only used for temperature prediction, but also can intuitively display the relationship between various parts of low-voltage electrical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of a low-voltage multi-channel wireless temperature measurement method based on artificial intelligence described in the present invention. DETAILED DESCRIPTION

[0017] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0018] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. The embodiments described are only a part of the embodiments of the present invention, rather than 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.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0020] One embodiment of the present invention is a low-voltage multi-channel wireless temperature measurement method based on artificial intelligence, the method comprising: Collect multi-point temperatures of low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through a wireless protocol, and pre-process the transmitted multi-point temperature data; Extract features based on preprocessed multi-point temperature data; The sensor positions in the low-voltage electrical equipment are used as nodes of the graph, the features extracted from the preprocessed temperature data are used as features of the nodes of the graph, the connection relationships between the sensors are used as edges, the graph structure is obtained, and a graph neural network model is trained based on the graph structure; The data monitored in real time by the sensor is input into the trained graph neural network model to obtain the predicted value. The difference between the predicted value and the detected value is compared at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted.

[0021] The working principle and effect of the above technical solution are: data collection and transmission, temperature sensors are deployed at multiple key parts of low-voltage electrical equipment to collect multi-point temperature data. These sensors are like "scouts" distributed throughout the equipment to sense temperature changes in real time; then, the large amount of collected temperature data is sent to the central data processing unit through wireless protocols to ensure that the data is quickly and accurately aggregated; the original temperature data received by the central data processing unit may have problems such as noise and missing values. The preprocessing stage aims to clean and optimize the data through filtering, denoising, filling missing values ​​and other operations, so as to provide a high-quality data foundation for subsequent analysis and processing; extract the features that can reflect the temperature change law, trend and equipment operation status from the preprocessed multi-point temperature data; construct a graph structure: take the position of the sensor in the low-voltage electrical equipment as the node of the graph, assign the extracted features to these nodes, and at the same time, determine the edges of the graph according to the actual connection relationship between the sensors. In this way, the originally scattered temperature data and sensor information are converted into a structured graph representation, which comprehensively and intuitively displays the temperature and connection relationship of each part of the equipment; model training and prediction, use the constructed graph structure to train the graph neural network model. During the training process, the model continuously learns the patterns and laws contained in the node features and edge relationships in the graph, and grasps the intrinsic connection between the temperature data and the equipment operation status. After the training is completed, the data monitored by the sensor in real time is input into the trained model, and the model will output the predicted value of the temperature of the low-voltage electrical equipment. At the end of the preset time window, the predicted value is compared with the actual detection value. If the difference between the two exceeds the preset threshold, it means that the current model has a large deviation in predicting the temperature change of the equipment, and the model needs to be retrained to optimize its performance; if the difference is within an acceptable range, the model will continue to be used to predict the temperature of the next time window. Accurate temperature monitoring and prediction, through multi-point temperature acquisition and graph neural network model training based on graph structure, fully considers the relationship and change trend of the temperature of various parts of low-voltage electrical equipment. Compared with the traditional single sensor or simple model monitoring method, it can more accurately predict the temperature of the equipment, timely detect potential overheating risks, and provide strong guarantees for the stable operation of the equipment; self-adaptation and self-optimization, with adaptive capabilities, when the difference between the predicted value and the detected value exceeds the preset range, it can automatically trigger model retraining.This enables the model to continuously optimize its own performance as the equipment's operating status changes, environmental factors change, etc., and maintain a high level of prediction accuracy to ensure long-term stable and reliable temperature monitoring; Early fault warning, accurate temperature prediction helps to detect abnormal temperature changes in time before the equipment fails, and issue early warning signals. Operation and maintenance personnel can take targeted measures based on the warning information, such as adjusting equipment operating parameters, arranging maintenance, etc., to avoid equipment failures, reduce maintenance costs and downtime, and improve equipment reliability and availability; Efficient data processing and analysis, from data collection, transmission to preprocessing, feature extraction, and model training and prediction, the entire process is automated and efficient. It reduces manual intervention, improves the efficiency of data processing and analysis, can quickly respond to equipment temperature changes, and adapt to the requirements of modern industrial production for real-time and efficient equipment monitoring;. The constructed graph structure is not only used for temperature prediction, but also can intuitively display the relationship between various parts of low-voltage electrical equipment. Operation and maintenance personnel can use the graph structure to understand the overall operating status of the equipment, discover potential weak links or abnormal connections, and provide comprehensive data support and decision-making basis for the optimization design of equipment and the formulation of maintenance strategies.

[0022] In one embodiment of the present invention, the temperature of multiple points of low-voltage electrical equipment is collected, the collected temperature data of multiple points is transmitted to a central data processing unit through a wireless protocol, and the transmitted temperature data of multiple points is preprocessed, including: Wireless temperature sensors installed at switch contacts, busbar connection points and cable connectors of low-voltage electrical equipment collect temperature data in real time at a preset collection frequency. Transmitting the temperature data collected by the wireless temperature sensor to a central data processing unit via a wireless protocol, wherein the wireless protocol includes LoRa, ZigBee and Wi-Fi; For the transmitted temperature data set , first calculate the first quartile and the third quartile , , will be lower than or higher The data are regarded as outliers and the interpolation method is used to correct the outliers.

[0023] The working principle and effect of the above technical solution are as follows: wireless temperature sensors are installed at key positions of low-voltage electrical equipment, namely switch contacts, busbar connection points and cable connectors. These positions are parts of electrical equipment that are prone to heat accumulation and temperature anomalies. Temperature data can be collected in a targeted manner through these sensors. The sensors will periodically measure the temperature in real time according to the preset collection frequency to obtain a series of temperature data points; for example, if the preset collection frequency is once every 5 minutes, then at each 5-minute time node, the sensor will record the temperature information at that time to form an original temperature data set; the collected temperature data is sent to the central data processing unit using a wireless protocol. The wireless protocol can be LoRa, ZigBee or Wi-Fi, etc. These protocols have different characteristics and applicable scenarios. LoRa has the advantages of long-distance transmission and low power consumption, and is suitable for large or widely distributed electrical equipment areas; ZigBee is suitable for short-distance, low-rate data transmission, and works better in small devices or relatively concentrated equipment areas; Wi-Fi has the characteristics of high-speed transmission, which facilitates rapid processing and sharing of data. The sensor transmits the collected temperature data in the form of data packets through the selected wireless protocol across the spatial distance to the central data processing unit for subsequent processing, realizing remote centralized management of data; Data preprocessing: For the temperature data set transmitted to the central data processing unit, the first quartile and the third quartile are first calculated to obtain the interquartile range. According to the principle of quartiles, data below or above are regarded as outliers; these outliers may be inaccurate measurements caused by sensor failure, electromagnetic interference or other accidental factors. The interpolation method is used to correct the outliers. The interpolation method can estimate the reasonable range of outliers based on the information of adjacent normal data points, and replace them with values ​​that are more in line with the actual situation, making the entire data set smoother and more accurate for subsequent analysis.

[0024] By deploying temperature sensors at key locations and collecting data in real time, the temperature information of low-voltage electrical equipment can be accurately captured, providing a rich and targeted data source for the temperature monitoring of the equipment; the processing of abnormal values ​​can eliminate erroneous data caused by accidental factors, avoid the interference of these abnormal data on subsequent analysis and judgment, ensure the accuracy and reliability of the data, and provide a more scientific data basis for subsequent equipment status evaluation and fault diagnosis; using a variety of wireless transmission protocols, it can be flexibly selected according to different electrical equipment installation environments and requirements to ensure that data can be stably and efficiently transmitted to the central data processing unit without being restricted by space and wiring, thereby improving the adaptability and scalability of the system; whether it is a large factory, a complex electrical system or a relatively small electrical equipment group, a suitable wireless transmission method can be found to facilitate the deployment and maintenance of the system; through reasonable temperature data collection, flexible transmission methods and effective data preprocessing, it provides strong support for the stable operation and fault prevention of low-voltage electrical equipment, and improves the reliability and safety of equipment management.

[0025] One embodiment of the present invention extracts features based on preprocessed multi-point temperature data, including: Extracting time features from the preprocessed multi-point temperature data, the time features including: the temperature average value, maximum value, minimum value and standard deviation within each preset time window; Extract the spatial features from the preprocessed multi-point temperature data, which include temperature gradient, temperature difference and relative temperature between different sensors. ,in, represents the temperature gradient, Indicates the low voltage circuit temperature measured by sensor i, represents the low voltage circuit temperature measured by sensor j, represents the distance between sensors i and j, and the relative temperature is ,in, Indicates relative temperature; According to the heat transfer principle, the formula

[0026] extracting thermodynamic features from the multi-point temperature data, wherein m represents the mass of the circuit measured by the sensor, c represents the specific heat capacity of the constituent material of the circuit measured by the sensor, represents the temperature measured by the sensor at time t, represents the temperature measured by the sensor at time t-1, It represents the amount of heat transfer in the circuit measured by the sensor during the time period from t to t-1.

[0027] The working principle and effect of the above technical solution are as follows: time feature extraction, for the preprocessed multi-point temperature data, it is divided into different preset time windows, and in each time window, the average, maximum, minimum and standard deviation of the temperature are calculated. Through these statistics, the overall situation, fluctuation range and discreteness of the temperature in different time windows can be clearly understood. For example, in a one-hour time window, the average temperature in this time period can be obtained by calculating the average value, the maximum and minimum values ​​can show the fluctuation range of the temperature, and the standard deviation reflects the discreteness of the temperature, which helps to judge the stability of the temperature. These time features describe the temperature data from the time dimension, which provides a basis for the subsequent analysis of the time change law of temperature, and can be used to observe the trend of temperature change over time, such as whether there is a periodic increase or decrease in temperature, whether there is abnormal temperature fluctuation in a certain time period, etc.; spatial feature extraction, spatial features mainly focus on the temperature relationship between different sensors. By calculating the temperature gradient between different sensors, the rate of change of temperature in space, that is, the trend of temperature change between different positions, can be reflected. Based on this gradient, the direction and speed of heat transfer in the equipment can be understood. At the same time, the relative temperature can be calculated to reflect the relative size relationship of the temperature at different positions, which helps to find the hot and cold spots of temperature, as well as the distribution of temperature in different areas. The extraction of spatial features is based on the positional relationship of multiple sensors and the measured temperature data. By analyzing these features, we can gain an in-depth understanding of whether the temperature distribution inside or on the surface of low-voltage electrical equipment is uniform, and whether there are local overheating or overcooling areas, so as to evaluate the operating status and potential problems of the equipment; Thermodynamic feature extraction, based on the principle of heat transfer, uses a formula to extract thermodynamic features. Through this formula, the amount of heat transfer occurring in the circuit measured by the sensor over a period of time can be calculated. This feature reflects the heat transfer in the circuit. Combined with time information, the rate of heat transfer can be understood, which helps to determine whether the energy conversion and transfer of the equipment is normal, and whether there is abnormal energy loss or excessive local heat accumulation.Comprehensive feature information, extracting time, space and thermodynamic features, can comprehensively characterize the temperature information of low-voltage electrical equipment from multiple dimensions. It can not only grasp the dynamic changes of temperature in time, but also grasp the temperature distribution and transfer trend in space. At the same time, it can grasp the heat transfer from a thermodynamic perspective, providing all-round information for the evaluation of the equipment's operating status. Together, they constitute a comprehensive description of the equipment's temperature, which helps to more accurately judge the health of the equipment, rather than simply judging from a single-dimensional temperature data; for fault diagnosis and prevention, time features can help detect abnormal temperature fluctuations and periodic anomalies, spatial features can find areas with uneven temperatures in the equipment, and thermodynamic features can determine whether the heat transfer is abnormal. By combining these features, temperature anomalies that may lead to equipment failures, such as local overheating and obstructed heat transfer, can be discovered in advance, providing a basis for equipment fault diagnosis and prevention; by observing abnormal changes in temperature gradients, it can be found whether there are poor contact or heat dissipation problems; through abnormal heat transfer, it can be determined whether the circuit has short circuits or overloads, and then measures can be taken in advance to avoid equipment failure or damage and improve the safety and reliability of the equipment; through multi-dimensional feature extraction, rich information is provided for temperature monitoring and management of low-voltage electrical equipment, improving the safety and reliability of the equipment, and also providing a basis for equipment performance optimization and fault diagnosis, which helps to improve the operating efficiency and stability of the entire system.

[0028] In one embodiment of the present invention, the sensor positions in the low-voltage electrical equipment are used as nodes of the graph, the features extracted from the preprocessed temperature data are used as the features of the nodes of the graph, the connection relationships between the sensors are used as edges, the graph structure is obtained, and the graph neural network model is trained based on the graph structure, including: Each wireless temperature sensor in the low-voltage electrical equipment is regarded as a node in the graph, marked with a unique identifier, i.e., a node number, and the extracted time feature, the spatial feature, and the thermodynamic feature are combined as a node feature vector, and the edge of the graph is constructed based on the connection relationship of the graph, and the node feature vector and the edge information are encapsulated into a graph structure; Construct a graph neural network model, which adopts a graph convolutional network in the GNN architecture. The graph neural network defines a class inherited from torch.nn.Module. The graph convolutional network includes three layers of GCNConv graph convolutional layers for extracting features in the graph structure. The first graph convolutional layer converts the input node features into features of the feature dimension of the hidden layer through a graph convolution operation. The second graph convolutional layer receives the output features of the feature dimension of the hidden layer of the first graph convolutional layer and outputs features of the same dimension. The third graph convolutional layer receives the output features of the feature dimension of the hidden layer of the second graph convolutional layer and converts them into one-dimensional predicted temperature values. Input the graph structure data into the constructed graph neural network model for training to obtain the trained graph neural network model.

[0029] The working principle and effect of the above technical solution are as follows: graph structure construction, node definition: each wireless temperature sensor in the low-voltage electrical equipment is regarded as a node in the graph and marked with a unique identifier (node ​​number) to ensure the uniqueness of each node; feature vector combination: the previously extracted time features (such as the average temperature, maximum temperature, minimum temperature and standard deviation in each time window), spatial features (temperature gradient, temperature difference and relative temperature between different sensors) and thermodynamic features (heat transfer amount calculated based on heat transfer formula, etc.) are combined as the feature vector of each node. These feature vectors comprehensively describe the temperature characteristics of each sensor location, covering multiple aspects of information such as time, space, and energy transfer; Edge construction: The edges of the graph are constructed based on the connection relationships of the graph. These connection relationships may reflect the physical or logical connections of the sensors in the equipment, thereby connecting the nodes and their feature vectors with edges to form a complete graph structure. This graph structure can effectively represent the temperature sensor network of low-voltage electrical equipment and their mutual relationships; Graph neural network model construction: Architecture selection: Use the graph convolutional network (GCN) architecture in the graph neural network (GNN) to define a class inherited from torch.nn.Module; Convolutional layer setting: The constructed graph convolutional network contains three layers of GCNConv Graph convolution layer; the first graph convolution layer receives the input node feature vector and converts it into features with hidden layer feature dimensions through graph convolution operation. The graph convolution operation considers the information of the node and its neighboring nodes, fuses the local information between nodes, and mines the relationship between nodes; the second graph convolution layer receives the output of the first graph convolution layer, outputs features of the same dimension, and further deepens the extraction of node features and information fusion; the third graph convolution layer receives the output of the second graph convolution layer, converts it into a one-dimensional predicted temperature value, and finally converts the high-dimensional feature information into a predicted result of the temperature; the constructed graph structure data is input into the graph neural network model for training. During the training process, the model continuously adjusts its own parameters by learning the node features and edge connection information in the graph structure to minimize the error between the predicted temperature value and the actual temperature value, thereby optimizing the performance of the model and finally obtaining a trained graph neural network model.Efficient feature fusion and representation. By combining multi-dimensional features (time, space, thermodynamics) into node feature vectors and constructing a graph structure, the scattered temperature sensor information can be presented in a structured and associated manner, effectively integrating the temperature information of low-voltage electrical equipment at different locations and times, so that the model can better utilize the intrinsic connection between different features. The graph structure can better represent the topological relationship between sensors, which helps to explore the potential relationship between sensors, rather than viewing the information of each sensor in isolation, improving the representation ability and understanding depth of temperature data. Powerful feature extraction capability. The graph neural network with a graph convolutional network architecture extracts features and fuses information on the graph structure through multiple layers of graph convolutional layers. Each layer of graph convolutional layer can make full use of the local information of the node, gradually propagate and fuse the local information of the node through convolution operations, extract deep feature information from the graph structure, and more accurately capture the complex patterns and laws of the temperature data of low-voltage electrical equipment. Compared with traditional machine learning methods, graph neural networks can better process graph structure data, explore nonlinear relationships between nodes, and have stronger feature extraction capabilities for tasks such as temperature prediction. Accurate temperature prediction and equipment monitoring. The trained graph neural network model can accurately predict the temperature based on the input graph structure data. Since the spatial layout, temporal changes and thermodynamic properties of sensors are fully considered, the prediction results will be more accurate and can better adapt to the complex environment and diverse operating conditions of low-voltage electrical equipment; it will help to realize real-time monitoring of low-voltage electrical equipment and detect temperature anomalies in advance, such as local overheating and uneven temperature distribution, providing a more reliable basis for equipment failure prevention and maintenance, and reducing equipment damage risks and maintenance costs; based on graph structure and graph neural network, when adding or adjusting sensor nodes, only the graph structure needs to be updated, and the model can adapt to the new structure through retraining or fine-tuning, with good scalability; at the same time, due to the characteristics of graph structure and graph neural network, the solution can adapt to low-voltage electrical equipment of different scales and layouts, and has certain adaptability to different equipment structures and complex temperature distribution conditions, and can be widely used in temperature monitoring scenarios of various low-voltage electrical equipment.

[0030] In one embodiment of the present invention, the data monitored in real time by the sensor is input into a trained graph neural network model to obtain a predicted value, and the difference between the predicted value and the detected value is compared at the end of a preset time window. If the difference exceeds a preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continuously predicted, including: Input the data monitored in real time by the sensor into the trained graph neural network model to obtain a first prediction value, and continue to input the first prediction value into the trained graph neural network model to obtain a second prediction value; The second predicted value is compared with the value actually detected by the sensor. If the difference between the second predicted value and the value actually detected by the sensor exceeds the preset threshold, the value actually detected by the sensor is used as incremental data to retrain the model. If it does not exceed the preset threshold, the prediction result is stored and added to the historical data set. The temperature of the low-voltage electrical equipment in the next time window is continued to be predicted based on the historical data set.

[0031] The working principle and effect of the above technical solution are as follows: data input and prediction, first, the data monitored by the sensor in real time is input into the trained graph neural network model. Since the graph neural network model has been trained before, it can process and extract features of the input data, and output the first prediction value based on the knowledge and patterns it has learned. Then, the first prediction value is input into the trained graph neural network model again to further obtain the second prediction value. This process takes advantage of the characteristics of graph neural networks, that is, it can iterate input multiple times, further mining features and reprocessing information in order to obtain more accurate and in-depth prediction results. For example, the first prediction may be a preliminary prediction based on the original data of the sensor at the current moment, while the second prediction is based on the first prediction result, taking into account the model's further understanding and analysis of the data, and may take into account more feature associations and potential patterns, thereby improving the accuracy and reliability of the prediction. Result comparison and decision-making: After obtaining the second predicted value, it is compared with the actual value detected by the sensor. Here, a preset threshold is set to measure whether the deviation between the predicted value and the actual detection value is within an acceptable range. If the difference between the two exceeds the preset threshold, it means that there is a problem with the current prediction performance of the model, which may be due to a change in the operating status of the equipment, or the previous training data no longer reflects the current situation well. At this time, the actual value detected by the sensor is used as incremental data and added to the data set of the retrained model. The parameters and learned patterns of the model are adjusted through retraining to make it better adapt to the new situation. If the difference does not exceed the preset threshold, it means that the model's prediction is accurate, and the prediction result is stored and added to the historical data set. These historical data sets contain data from previous successful predictions, which accumulate over time and provide more reference information for subsequent predictions, so that the temperature of low-voltage electrical equipment in the next time window can be predicted based on richer data. The solution continuously optimizes the prediction accuracy by inputting the predicted values ​​for multiple iterations and comparing them with the actual detection values. The solution can promptly detect deviations in the model predictions. Once the deviation exceeds the threshold, it will be retrained so that the model can continuously adapt to the changes in the operating status of low-voltage electrical equipment and new data patterns, thereby ensuring the accuracy and reliability of the prediction results. This self-adjustment mechanism ensures that the model will not fail due to changes in the equipment status or long-term changes in the data, and always maintains a good prediction capability for the equipment temperature, providing accurate temperature prediction information for the stable operation of the equipment. The historical data is effectively used. For accurate predictions, the results are stored in the historical data set, enriching the historical data resources of the model.Based on a richer historical data set, the model can better grasp the temperature change trends and laws when predicting the temperature in the next time window, because it can learn more patterns and information from past successful predictions, avoiding starting from scratch every time prediction, and improving the stability and consistency of the prediction; when the operating status of the equipment changes, such as load increase, component aging, etc., resulting in new patterns in the temperature data, the technical solution can adjust the prediction ability by retraining the model, flexibly adapt to these changes, and will not become invalid due to equipment upgrades or changes in the use environment; this flexibility makes the solution more adaptable in long-term equipment monitoring and maintenance, and can provide reliable temperature prediction services for low-voltage electrical equipment in different working conditions and different use stages, improve the operating safety and life of the equipment, and reduce the risk of failures caused by abnormal temperatures.

[0032] One embodiment of the present invention is a low-voltage multi-channel wireless temperature measurement device based on artificial intelligence, the device comprising: The acquisition module is used to collect the multi-point temperature of the low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through the wireless protocol, and pre-process the transmitted multi-point temperature data; A feature extraction module is used to extract features based on preprocessed multi-point temperature data; A training graph neural network model module is used to use the sensor positions in the low-voltage electrical equipment as nodes of the graph, the features extracted from the preprocessed temperature data as features of the nodes of the graph, and the connection relationships between the sensors as edges, to obtain a graph structure, and train a graph neural network model based on the graph structure; The prediction module is used to input the data monitored by the sensor in real time into the trained graph neural network model, obtain the predicted value, and compare the difference between the predicted value and the detected value at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted.

[0033] In one embodiment of the present invention, the acquisition module includes: The temperature data collection module is used to collect temperature data in real time according to a preset collection frequency through wireless temperature sensors installed at switch contacts, busbar connection points and cable connectors of low-voltage electrical equipment; A transmission module, used to transmit the temperature data collected by the wireless temperature sensor to the central data processing unit through a wireless protocol, wherein the wireless protocol includes LoRa, ZigBee and Wi-Fi; Correction module for the transmitted temperature data set , first calculate the first quartile and the third quartile , , will be lower than or higher The data are regarded as outliers and the interpolation method is used to correct the outliers.

[0034] In one embodiment of the present invention, the feature extraction module includes: A time feature extraction module is used to extract time features from the preprocessed multi-point temperature data, wherein the time features include: the temperature average value, maximum value, minimum value and standard deviation within each preset time window; The spatial feature extraction module is used to extract the spatial features in the preprocessed multi-point temperature data. The spatial features include: temperature gradient, temperature difference and relative temperature between different sensors. The temperature gradient is ,in, represents the temperature gradient, Indicates the low voltage circuit temperature measured by sensor i, represents the low voltage circuit temperature measured by sensor j, represents the distance between sensors i and j, and the relative temperature is ,in, Indicates relative temperature; Extract thermodynamic characteristics module, used to extract the heat transfer principle through the formula

[0035] extracting thermodynamic features from the multi-point temperature data, wherein m represents the mass of the circuit measured by the sensor, c represents the specific heat capacity of the constituent material of the circuit measured by the sensor, represents the temperature measured by the sensor at time t, represents the temperature measured by the sensor at time t-1, It represents the amount of heat transfer in the circuit measured by the sensor during the time period from t to t-1.

[0036] In one embodiment of the present invention, the training graph neural network model module includes: A graph construction module is used to regard each wireless temperature sensor in the low-voltage electrical equipment as a node in the graph, mark it with a unique identifier, i.e., a node number, combine the extracted time feature, the spatial feature, and the thermodynamic feature as a node feature vector, construct the edge of the graph based on the connection relationship of the graph, and encapsulate the node feature vector and the edge information into a graph structure; Construct a graph neural network model module, which is used to construct a graph neural network model. The graph neural network model adopts a graph convolutional network in the GNN architecture. The graph neural network defines a class inherited from torch.nn.Module. The graph convolutional network includes three layers of GCNConv graph convolutional layers, which are used to extract features in the graph structure. The first graph convolutional layer converts the input node features into features of the feature dimension of the hidden layer through a graph convolution operation. The second graph convolutional layer receives the output features of the feature dimension of the hidden layer of the first graph convolutional layer and outputs features of the same dimension. The third graph convolutional layer receives the output features of the feature dimension of the hidden layer of the second graph convolutional layer and converts them into one-dimensional predicted temperature values. The training module is used to input graph structure data into the constructed graph neural network model for training and obtain the trained graph neural network model.

[0037] In one embodiment of the present invention, the prediction module comprises: A module for obtaining a second prediction value is used to input the data monitored in real time by the sensor into the trained graph neural network model to obtain a first prediction value, and further input the first prediction value into the trained graph neural network model to obtain a second prediction value; The correction prediction module is used to compare the second predicted value with the value actually detected by the sensor. If the difference between the second predicted value and the value actually detected by the sensor exceeds a preset threshold, the value actually detected by the sensor is used as incremental data to retrain the model. If it does not exceed the preset threshold, the prediction result is stored and added to the historical data set, and the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted based on the historical data set.

[0038] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A low-voltage multi-channel wireless temperature measurement method based on artificial intelligence, characterized in that: The method comprises: Collect multi-point temperatures of low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through a wireless protocol, and pre-process the transmitted multi-point temperature data; Extract features based on preprocessed multi-point temperature data; The sensor positions in the low-voltage electrical equipment are used as nodes of the graph, the features extracted from the preprocessed temperature data are used as features of the nodes of the graph, the connection relationships between the sensors are used as edges, the graph structure is obtained, and a graph neural network model is trained based on the graph structure; The data monitored in real time by the sensor is input into the trained graph neural network model to obtain the predicted value. The difference between the predicted value and the detected value is compared at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted.

2. According to the artificial intelligence-based low-voltage multi-channel wireless temperature measurement method of claim 1, it is characterized in that: Collect multi-point temperatures of low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through wireless protocol, and pre-process the transmitted multi-point temperature data, including: Wireless temperature sensors installed at switch contacts, busbar connection points and cable connectors of low-voltage electrical equipment collect temperature data in real time at a preset collection frequency. Transmitting the temperature data collected by the wireless temperature sensor to a central data processing unit via a wireless protocol, wherein the wireless protocol includes LoRa, ZigBee and Wi-Fi; For the transmitted temperature data set , first calculate the first quartile and the third quartile , , will be lower than or higher The data are regarded as outliers and the interpolation method is used to correct the outliers.

3. According to the artificial intelligence-based low-voltage multi-channel wireless temperature measurement method of claim 1, it is characterized in that: Extract features based on preprocessed multi-point temperature data, including: Extracting time features from the preprocessed multi-point temperature data, the time features including: the temperature average value, maximum value, minimum value and standard deviation within each preset time window; Extract the spatial features from the preprocessed multi-point temperature data, which include temperature gradient, temperature difference and relative temperature between different sensors. ,in, represents the temperature gradient, Indicates the low voltage circuit temperature measured by sensor i, represents the low voltage circuit temperature measured by sensor j, represents the distance between sensors i and j, and the relative temperature is ,in, Indicates relative temperature; According to the heat transfer principle, the formula extracting thermodynamic features from the multi-point temperature data, wherein m represents the mass of the circuit measured by the sensor, c represents the specific heat capacity of the constituent material of the circuit measured by the sensor, represents the temperature measured by the sensor at time t, represents the temperature measured by the sensor at time t-1, It represents the amount of heat transfer in the circuit measured by the sensor during the time period from t to t-1.

4. According to the artificial intelligence-based low-voltage multi-channel wireless temperature measurement method of claim 1, it is characterized in that: The sensor positions in the low-voltage electrical equipment are used as nodes of the graph, the features extracted from the preprocessed temperature data are used as features of the nodes of the graph, and the connection relationships between the sensors are used as edges to obtain a graph structure, and a graph neural network model is trained based on the graph structure, including: Each wireless temperature sensor in the low-voltage electrical equipment is regarded as a node in the graph, marked with a unique identifier, i.e., a node number, and the extracted time feature, the spatial feature, and the thermodynamic feature are combined as a node feature vector, and the edge of the graph is constructed based on the connection relationship of the graph, and the node feature vector and the edge information are encapsulated into a graph structure; Construct a graph neural network model, which adopts a graph convolutional network in the GNN architecture. The graph neural network defines a class inherited from torch.nn.Module. The graph convolutional network includes three layers of GCNConv graph convolutional layers, which are used to extract features in the graph structure. The first graph convolutional layer converts the input node features into features of the feature dimension of the hidden layer through a graph convolution operation. The second graph convolutional layer receives the output features of the feature dimension of the hidden layer of the first graph convolutional layer and outputs features of the same dimension. The third graph convolutional layer receives the output features of the feature dimension of the hidden layer of the second graph convolutional layer and converts them into one-dimensional predicted temperature values. Input the graph structure data into the constructed graph neural network model for training to obtain the trained graph neural network model.

5. According to the artificial intelligence-based low-voltage multi-channel wireless temperature measurement method of claim 1, it is characterized in that: The data monitored by the sensor in real time is input into the trained graph neural network model to obtain the predicted value. The difference between the predicted value and the detected value is compared at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained. Otherwise, the temperature of the low-voltage electrical equipment in the next time window is predicted, including: Input the data monitored in real time by the sensor into the trained graph neural network model to obtain a first prediction value, and continue to input the first prediction value into the trained graph neural network model to obtain a second prediction value; The second predicted value is compared with the value actually detected by the sensor. If the difference between the second predicted value and the value actually detected by the sensor exceeds the preset threshold, the value actually detected by the sensor is used as incremental data to retrain the model. If it does not exceed the preset threshold, the prediction result is stored and added to the historical data set. The temperature of the low-voltage electrical equipment in the next time window is continued to be predicted based on the historical data set.

6. A low-voltage multi-channel wireless temperature measurement device based on artificial intelligence, characterized in that: The device comprises: The acquisition module is used to collect the multi-point temperature of the low-voltage electrical equipment, transmit the collected multi-point temperature data to the central data processing unit through the wireless protocol, and pre-process the transmitted multi-point temperature data; A feature extraction module is used to extract features based on preprocessed multi-point temperature data; A training graph neural network model module is used to use the sensor positions in the low-voltage electrical equipment as nodes of the graph, the features extracted from the preprocessed temperature data as features of the nodes of the graph, and the connection relationships between the sensors as edges, to obtain a graph structure, and train a graph neural network model based on the graph structure; The prediction module is used to input the data monitored by the sensor in real time into the trained graph neural network model, obtain the predicted value, and compare the difference between the predicted value and the detected value at the end of the preset time window. If the difference exceeds the preset threshold, the model is retrained, otherwise the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted.

7. According to claim 6, a low-voltage multi-channel wireless temperature measurement device based on artificial intelligence is characterized in that: The acquisition module comprises: The temperature data collection module is used to collect temperature data in real time according to a preset collection frequency through wireless temperature sensors installed at switch contacts, busbar connection points and cable connectors of low-voltage electrical equipment; A transmission module, used to transmit the temperature data collected by the wireless temperature sensor to the central data processing unit through a wireless protocol, wherein the wireless protocol includes LoRa, ZigBee and Wi-Fi; Correction module for the transmitted temperature data set , first calculate the first quartile and the third quartile , , will be lower than or higher The data are regarded as outliers and the interpolation method is used to correct the outliers.

8. According to claim 6, a low-voltage multi-channel wireless temperature measurement device based on artificial intelligence is characterized in that: The feature extraction module comprises: A time feature extraction module is used to extract time features from the preprocessed multi-point temperature data, wherein the time features include: the temperature average value, maximum value, minimum value and standard deviation within each preset time window; The spatial feature extraction module is used to extract the spatial features in the preprocessed multi-point temperature data. The spatial features include: temperature gradient, temperature difference and relative temperature between different sensors. The temperature gradient is ,in, represents the temperature gradient, Indicates the low voltage circuit temperature measured by sensor i, represents the low voltage circuit temperature measured by sensor j, represents the distance between sensors i and j, and the relative temperature is ,in, Indicates relative temperature; Extract thermodynamic characteristics module, used to extract the heat transfer principle through the formula extracting thermodynamic features from the multi-point temperature data, wherein m represents the mass of the circuit measured by the sensor, c represents the specific heat capacity of the constituent material of the circuit measured by the sensor, represents the temperature measured by the sensor at time t, represents the temperature measured by the sensor at time t-1, It represents the amount of heat transfer in the circuit measured by the sensor during the time period from t to t-1.

9. According to claim 6, a low-voltage multi-channel wireless temperature measurement device based on artificial intelligence is characterized in that: The training graph neural network model module includes: A graph construction module is used to regard each wireless temperature sensor in the low-voltage electrical equipment as a node in the graph, mark it with a unique identifier, i.e., a node number, combine the extracted time feature, the spatial feature, and the thermodynamic feature as a node feature vector, construct the edge of the graph based on the connection relationship of the graph, and encapsulate the node feature vector and the edge information into a graph structure; Construct a graph neural network model module, which is used to construct a graph neural network model. The graph neural network model adopts a graph convolutional network in the GNN architecture. The graph neural network defines a class inherited from torch.nn.Module. The graph convolutional network includes three layers of GCNConv graph convolutional layers, which are used to extract features in the graph structure. The first graph convolutional layer converts the input node features into features of the feature dimension of the hidden layer through a graph convolution operation. The second graph convolutional layer receives the output features of the feature dimension of the hidden layer of the first graph convolutional layer and outputs features of the same dimension. The third graph convolutional layer receives the output features of the feature dimension of the hidden layer of the second graph convolutional layer and converts them into one-dimensional predicted temperature values. The training module is used to input graph structure data into the constructed graph neural network model for training and obtain the trained graph neural network model.

10. The low-voltage multi-channel wireless temperature measurement device based on artificial intelligence according to claim 6, characterized in that: The ... prediction module comprises: A module for obtaining a second prediction value is used to input the data monitored in real time by the sensor into the trained graph neural network model to obtain a first prediction value, and further input the first prediction value into the trained graph neural network model to obtain a second prediction value; The correction prediction module is used to compare the second predicted value with the value actually detected by the sensor. If the difference between the second predicted value and the value actually detected by the sensor exceeds a preset threshold, the value actually detected by the sensor is used as incremental data to retrain the model. If it does not exceed the preset threshold, the prediction result is stored and added to the historical data set, and the temperature of the low-voltage electrical equipment in the next time window is continued to be predicted based on the historical data set.

Images