Electric energy meter temperature distribution prediction method based on physical information neural network

Through the physical information neural network combining the thermal conduction prediction loss function and the temperature monitoring data of the electricity meter, the problem of insufficient prediction accuracy and speed of the temperature distribution of the electricity meter is solved, and the rapid and accurate prediction of the temperature distribution of the electricity meter is achieved, and the equipment monitoring and control are enhanced. Security.

CN118917201APending Publication Date: 2024-11-08NINGXIA LGG INSTR CO LTD
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411056521.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The accuracy and speed of the temperature distribution prediction of the electricity meter are insufficient, and are affected by ambient temperature changes and internal component failures, resulting in reduced measurement accuracy and potential fire risk.

Method used

Using a physical information neural network, a physical information neural network is built for the prediction of the temperature distribution of the electric energy meter through the thermal conduction prediction loss function and the electric energy meter temperature monitoring data set training model to build a physical information neural network for the prediction of the electric energy meter to ensure that the model follows the physical laws of heat conduction and achieve fast and accurate temperature distribution prediction.

Benefits of technology

It realizes rapid and accurate prediction of the temperature distribution of the power meter, monitors the current terminal temperature of the metering equipment and detects abnormal heating points, improving the normal operation and safety of the equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917201A_ABST
    Figure CN118917201A_ABST
Patent Text Reader

Abstract

The invention discloses an electric energy meter temperature distribution prediction method based on a physical information neural network, and belongs to the technical field of metering equipment, and the method comprises the following steps: obtaining an electric energy meter temperature monitoring data set; selecting a physical information neural network for electric energy meter temperature distribution prediction; based on the physical model, defining a heat conduction prediction loss function of the physical information neural network; according to the heat conduction prediction loss function, using the electric energy meter temperature monitoring data set to train a physical information neural network to obtain a trained physical information neural network; and obtaining temperature monitoring data of the target electric energy meter, and inputting the temperature monitoring data of the target electric energy meter into the trained physical information neural network for temperature distribution prediction to obtain a temperature distribution prediction result of the target electric energy meter. The problem that the temperature distribution prediction accuracy and speed of the electric energy meter are insufficient is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of metering equipment, and in particular relates to a method for predicting temperature distribution of an electric energy meter based on a physical information neural network. Background Art

[0002] In the field of metering equipment technology, especially in application scenarios related to electric energy meters, accurate and reliable measurements are crucial. However, in actual use, the performance and accuracy of electric energy meters are often affected by many factors, such as changes in ambient temperature and failure or overheating of internal components. Changes in ambient temperature will affect the temperature rise of the electric energy meter transformer, thereby affecting its measurement accuracy. At the same time, the components inside the electric energy meter may also overheat due to failure during long-term operation, which will not only further affect the measurement accuracy, but may also cause damage to the equipment or fire risks.

[0003] Physics-Informed Neural Networks (Physics-Informed Neural Networks) can add physical constraints to the neural network, making the fitting results of the neural network more consistent with the laws of physics. In the face of scenarios that require real-time detection and prediction of temperature distribution with high accuracy, especially in electronic devices such as electric energy meters, how to obtain accurate temperature distribution prediction results to ensure the normal operation of the equipment and prevent overheating is still an urgent problem to be solved. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting the temperature distribution of an electric energy meter based on a physical information neural network. A physical information neural network for predicting the temperature distribution of an electric energy meter is constructed to extract the relationship between the temperature monitoring data of the electric energy meter and the temperature distribution of the electric energy meter. The physical information neural network is trained by the heat conduction prediction loss function of the physical information neural network and the acquired temperature monitoring data set of the electric energy meter. The trained physical information neural network can quickly and accurately predict the temperature distribution of any target electric energy meter, thereby solving the problem of insufficient accuracy and speed in predicting the temperature distribution of the electric energy meter.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] The present invention provides a method for predicting temperature distribution of an electric energy meter based on a physical information neural network, comprising the following steps:

[0007] S1. Obtaining the temperature monitoring data set of the electric energy meter;

[0008] S2, selecting a physical information neural network for predicting the temperature distribution of an electric energy meter;

[0009] S3. Based on the physical model, define the heat conduction prediction loss function of the physical information neural network;

[0010] S4. According to the heat conduction prediction loss function, the physical information neural network is trained using the electric energy meter temperature monitoring data set to obtain a trained physical information neural network;

[0011] S5. Acquire temperature monitoring data of a target electric energy meter, and input the temperature monitoring data of the target electric energy meter into a trained physical information neural network to perform temperature distribution prediction, thereby obtaining a temperature distribution prediction result of the target electric energy meter.

[0012] The beneficial effects of the present invention are as follows: the present invention provides a method for predicting the temperature distribution of an electric energy meter based on a physical information neural network, obtains an electric energy meter temperature monitoring data set, selects a physical information neural network for predicting the temperature distribution of the electric energy meter, and defines a heat conduction prediction loss function of the physical information neural network, and trains the physical information neural network based on the heat conduction prediction loss function and the electric energy meter temperature monitoring data set. The heat conduction prediction loss function embeds the physical equation of heat conduction as a hard constraint in the training process of the neural network, ensuring that the physical information neural network not only accurately fits the actually observed temperature distribution data of the electric energy meter, but also follows physical laws such as heat conduction; the trained physical information neural network can simultaneously consider the temperature monitoring data of the electric energy meter and the physical mechanism of heat conduction, thereby realizing rapid and accurate prediction of the temperature distribution on the circuit board of the target electric energy meter; the circuit board temperature distribution of the target electric energy meter predicted by the method of the present invention is of great value for monitoring the current terminal temperature of the metering equipment and detecting possible abnormal heating points.

[0013] Furthermore, the electric energy meter temperature monitoring data set in S1 includes electric energy meter temperature monitoring data of several different time periods, and electric energy meter temperature distribution data corresponding to each electric energy meter temperature monitoring data;

[0014] The temperature monitoring data of the electric energy meter in different time periods include the average ambient temperature of the electric energy meter within a preset time window, the ambient temperature change rate, the temperature on the first board, the temperature on the second board, the temperature change rate on the first board and the temperature change rate on the second board.

[0015] The beneficial effects of adopting the above-mentioned further scheme are as follows: the present invention provides an electric energy meter temperature monitoring data set, and the electric energy meter temperature monitoring data and its corresponding electric energy meter temperature distribution data can provide a basis for training a physical information neural network to accurately extract the mapping relationship between the electric energy meter temperature monitoring data and the electric energy meter temperature distribution data of the same period. Through the monitoring data of the temperature on two boards and the temperature change rate on the boards, it can avoid the situation where the temperature monitoring data on a single board itself has errors, and at the same time, it can also more comprehensively and accurately reflect the temperature monitoring information on the board.

[0016] Furthermore, the physical information neural network for predicting the temperature distribution of the electric energy meter adopts a fully connected neural network or a recurrent neural network.

[0017] The beneficial effect of adopting the above further scheme is that the physical information neural network selected by the present invention for predicting the temperature distribution of the electric energy meter can be either a fully connected neural network or a recurrent neural network, as long as it has sufficient capacity to learn the mapping relationship between the input electric energy meter temperature monitoring data and its corresponding electric energy meter temperature distribution data.

[0018] Furthermore, S3 comprises the following steps:

[0019] S31, selecting a two-dimensional heat conduction model as a physical model;

[0020] The calculation expression of the two-dimensional heat conduction model is as follows:

[0021]

[0022] Among them, u represents the temperature field, t represents the time, α represents the thermal diffusivity, x represents the horizontal spatial coordinate, and y represents the vertical spatial coordinate. represents the first-order partial differential result of the temperature field with respect to time, represents the second-order partial differential result of the temperature field with respect to the transverse space coordinate, It represents the second-order partial differential result of the temperature field with respect to the longitudinal spatial coordinate;

[0023] S32, based on the physical model, calculating the residual of the temperature field to be predicted;

[0024] The calculation expression of the residual error of the temperature field to be predicted is as follows:

[0025]

[0026] Among them, RES represents the residual error of the temperature field to be predicted, u pre represents the temperature field to be predicted, It represents the first-order partial differential result of the temperature field to be predicted with respect to time. It represents the second-order partial differential result of the temperature field to be predicted with respect to the transverse space coordinates, It represents the second-order partial differential result of the temperature field to be predicted with respect to the longitudinal space coordinate;

[0027] S33, discretizing the spatial domain of the temperature field to be predicted according to the residual of the temperature field to be predicted, and obtaining a spatially discrete temperature distribution residual term;

[0028] The calculation expression of the residual term of the spatial discrete temperature distribution is as follows:

[0029]

[0030] Among them, L pre Represents the residual term of spatial discrete temperature distribution, N t Indicates the preset time window length, N x Indicates the maximum value of the horizontal space coordinate on the circuit board of the energy meter, N y Indicates the maximum longitudinal spatial coordinate on the circuit board of the electric energy meter;

[0031] S34, constructing a data error term between the predicted temperature distribution result of the electric energy meter and the actual temperature distribution data of the electric energy meter;

[0032] The calculation expression of the data error term is as follows:

[0033]

[0034] Among them, L data represents the data error term, represents the predicted temperature distribution result of the electric energy meter, u obs Represents the actual temperature distribution data of the electric energy meter, || || 2 represents the square of the two-norm;

[0035] S35. Based on the spatial discrete temperature distribution residual term and the data error term, define the heat conduction prediction loss function of the physical information neural network;

[0036] The calculation expression of the heat conduction prediction loss function is as follows:

[0037] LOSS=λ data L data +λ pre L pre

[0038] Where LOSS represents the heat conduction prediction loss function, λ data Represents the hyperparameter of the residual term of the spatial discrete temperature distribution, λ pre represents the hyperparameter of the error term for the data.

[0039] The beneficial effects of adopting the above-mentioned further scheme are as follows: the present invention provides a method for defining a heat conduction prediction loss function of a physical information neural network based on a physical model, and trains a physical information neural network based on the heat conduction prediction loss function provided by the present invention. Physical equations can be added as restriction conditions to the training process, so that the fitting results of the physical information neural network are more in line with physical laws, providing a basis for the trained physical information neural network to quickly and accurately predict the distribution prediction results of the target electric energy meter.

[0040] Furthermore, the S4 comprises the following steps:

[0041] S41, randomly initialize the weight of each connection in the physical information neural network;

[0042] S42, inputting the electric energy meter temperature monitoring data in the electric energy meter temperature monitoring data set into the physical information neural network, and performing forward propagation to obtain a predicted electric energy meter temperature distribution result;

[0043] S43, using the electric energy meter temperature distribution data corresponding to the electric energy meter temperature monitoring data as the actual electric energy meter temperature distribution data, and calculating a data error term between the predicted temperature distribution data and the actual temperature distribution data;

[0044] S44, using the predicted temperature distribution result of the electric energy meter as the temperature field to be predicted, and calculating the residual term of the spatial discrete temperature distribution;

[0045] S45, according to the data error term and the spatial discrete temperature distribution residual term, using the heat conduction prediction loss function to perform parameter derivation on the physical information neural network to perform back propagation to obtain a gradient;

[0046] S46, obtaining a first-order moment estimate and a second-order moment estimate of each parameter in the physical information neural network according to the method and gradient of the adaptive learning rate;

[0047] S47, according to the first-order matrix estimation value and the second-order matrix estimation value of each parameter in the physical information neural network, adaptively adjust the learning rate of each parameter to complete the batch training;

[0048] S48. Repeat batch training several times, and adjust the hyperparameters of the residual term of the spatial discrete temperature distribution and the hyperparameters of the data error term before each repeated batch training, until the result of the heat conduction prediction loss function is less than a preset convergence threshold or the number of batch training reaches a preset number of training times, to obtain a trained physical information neural network.

[0049] The beneficial effect of adopting the above further scheme is: this method provides a method for training a physical information neural network based on a heat conduction prediction loss function and using an electric energy meter temperature monitoring data set to obtain a trained physical information neural network. By continuously adjusting the hyperparameters of the spatial discrete temperature distribution residual term and the hyperparameters of the data error term, as well as repeated batch training for multiple times, the trained physical information neural network can quickly and accurately predict the temperature monitoring data of the target electric energy meter.

[0050] Furthermore, the method for adaptive learning rate in S46 adopts Adam algorithm or L-BFGS algorithm.

[0051] The beneficial effect of adopting the above further scheme is: the Adam algorithm or L-BFGS algorithm is used in the present invention to solve the problem of optimizing gradient descent and minimizing the loss function, which effectively improves the accuracy and efficiency of the physical information neural network in predicting the temperature distribution of the electric energy meter.

[0052] Furthermore, the S5 comprises the following steps:

[0053] S51, obtaining the average ambient temperature, ambient temperature change rate, first board temperature, second board temperature, first board temperature change rate, and second board temperature change rate of the target electric energy meter within the target time window as temperature monitoring data of the target electric energy meter;

[0054] S52, inputting the temperature monitoring data of the target electric energy meter into the trained physical information neural network to perform temperature distribution prediction, and obtaining the temperature distribution prediction result of the target electric energy meter.

[0055] The beneficial effect of adopting the above-mentioned further scheme is as follows: the present invention provides a method for predicting the temperature distribution of a target electric energy meter using a trained physical information neural network. By inputting the temperature monitoring data of the target electric energy meter into the trained physical information neural network, the temperature distribution prediction result of the target electric energy meter can be obtained quickly and accurately based on the physical mechanism and the learned mapping relationship.

[0056] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0058] Figure 1 The present invention is a flowchart of a method for predicting temperature distribution of an electric energy meter based on a physical information neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0060] Physics-informed Neural Networks (Physics-informed Neural Networks) are a type of machine learning model that combines deep learning and physics knowledge, using physical laws to guide the neural network model during the learning process.

[0061] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides a method for predicting temperature distribution of an electric energy meter based on a physical information neural network, comprising the following steps:

[0062] S1. Obtaining the temperature monitoring data set of the electric energy meter;

[0063] The electric energy meter temperature monitoring data set in S1 includes electric energy meter temperature monitoring data of several different time periods, and electric energy meter temperature distribution data corresponding to each electric energy meter temperature monitoring data;

[0064] The temperature monitoring data of the electric energy meter in different time periods include the average ambient temperature of the electric energy meter within a preset time window, the ambient temperature change rate, the temperature on the first board, the temperature on the second board, the temperature change rate on the first board and the temperature change rate on the second board.

[0065] In this embodiment, the temperature distribution data of the electric energy meter in the temperature monitoring data set of the electric energy meter can be obtained by using thermal imaging technology on the electric energy meter; the preset time window is 8 consecutive hours; an ambient temperature sensor is installed in the environment where the electric energy meter is located, and the ambient temperature of the electric energy meter can be monitored by the ambient temperature sensor, and the average ambient temperature of the electric energy meter in the preset time window can be obtained by averaging the ambient temperature data of the electric energy meter read within 8 consecutive hours, wherein the ambient temperature data of the electric energy meter for averaging is extracted at a frequency of once every 15 minutes in the preset time window; the ambient temperature change rate is the change amount of the ambient temperature of the electric energy meter per 1 minute, and the unit is: ° C / min, and the frequency of collecting the ambient temperature change rate is also once every 15 minutes;

[0066] The present solution also provides at least two temperature sensors on the circuit board of the electric energy meter; when only two temperature sensors are provided, they are a first-board temperature sensor and a second-board temperature sensor respectively; the temperature of the electric energy meter circuit board can be monitored by both the first-board temperature sensor and the second-board temperature sensor, and the temperature on the first board and the temperature on the second board are obtained correspondingly; the reading frequency of the temperature on the first board and the temperature on the second board is once per minute; by respectively calculating the change per minute of the temperature on the first board and the temperature on the second board, the temperature change rate on the first board and the temperature change rate on the second board are obtained correspondingly, and the reading frequency of the temperature change rate on the first board and the temperature change rate on the second board is once per minute.

[0067] S2, selecting a physical information neural network for predicting the temperature distribution of an electric energy meter;

[0068] The physical information neural network for predicting the temperature distribution of the electric energy meter adopts a fully connected neural network or a recurrent neural network.

[0069] In this embodiment, the fully connected neural network or recurrent neural network has sufficient capacity to learn the relationship between the temperature monitoring data of the electric energy meter and the temperature distribution of the electric energy meter;

[0070] Fully connected neural network is also called dense neural network. In a fully connected neural network, neurons in each layer are connected to all neurons in the previous layer, and input data is processed through weights and activation functions to generate output results. A fully connected neural network is a multi-layer perceptron, consisting of an input layer, a hidden layer, and an output layer. Its basic principle is the back propagation algorithm. A recurrent neural network refers to a fully connected neural network that adds a temporal relationship to the previous and next layers. It can better handle time-related problems such as machine translation. A recurrent neural network is a network with strong processing capabilities for time series data.

[0071] S3. Based on the physical model, define the heat conduction prediction loss function of the physical information neural network;

[0072] The S3 comprises the following steps:

[0073] S31, selecting a two-dimensional heat conduction model as a physical model;

[0074] The calculation expression of the two-dimensional heat conduction model is as follows:

[0075]

[0076] Among them, u represents the temperature field, t represents the time, α represents the thermal diffusivity, x represents the horizontal spatial coordinate, and y represents the vertical spatial coordinate. represents the first-order partial differential result of the temperature field with respect to time, represents the second-order partial differential result of the temperature field with respect to the transverse space coordinate, It represents the second-order partial differential result of the temperature field with respect to the longitudinal spatial coordinate;

[0077] S32, based on the physical model, calculating the residual of the temperature field to be predicted;

[0078] The calculation expression of the residual error of the temperature field to be predicted is as follows:

[0079]

[0080] Among them, RES represents the residual error of the temperature field to be predicted, u pre represents the temperature field to be predicted, It represents the first-order partial differential result of the temperature field to be predicted with respect to time. It represents the second-order partial differential result of the temperature field to be predicted with respect to the transverse space coordinates, It represents the second-order partial differential result of the temperature field to be predicted with respect to the longitudinal space coordinate;

[0081] S33, discretizing the spatial domain of the temperature field to be predicted according to the residual of the temperature field to be predicted, and obtaining a spatially discrete temperature distribution residual term;

[0082] The calculation expression of the residual term of the spatial discrete temperature distribution is as follows:

[0083]

[0084] Among them, L pre Represents the residual term of spatial discrete temperature distribution, N t Indicates the preset time window length, N x Indicates the maximum value of the horizontal space coordinate on the circuit board of the energy meter, N y Indicates the maximum longitudinal spatial coordinate on the circuit board of the electric energy meter;

[0085] S34, constructing a data error term between the predicted temperature distribution result of the electric energy meter and the actual temperature distribution data of the electric energy meter;

[0086] The calculation expression of the data error term is as follows:

[0087]

[0088] Among them, L data represents the data error term, represents the predicted temperature distribution result of the electric energy meter, u obs Represents the actual temperature distribution data of the electric energy meter, |||| 2 represents the square of the two-norm;

[0089] S35. Based on the spatial discrete temperature distribution residual term and the data error term, define the heat conduction prediction loss function of the physical information neural network;

[0090] The calculation expression of the heat conduction prediction loss function is as follows:

[0091] LOSS=λ data L data +λ pre L pre

[0092] Where LOSS represents the heat conduction prediction loss function, λ data Represents the hyperparameter of the residual term of the spatial discrete temperature distribution, λ pre represents the hyperparameter of the error term for the data.

[0093] S4. According to the heat conduction prediction loss function, the physical information neural network is trained using the electric energy meter temperature monitoring data set to obtain a trained physical information neural network;

[0094] The S4 comprises the following steps:

[0095] S41, randomly initialize the weight of each connection in the physical information neural network;

[0096] S42, inputting the electric energy meter temperature monitoring data in the electric energy meter temperature monitoring data set into the physical information neural network, and performing forward propagation to obtain a predicted electric energy meter temperature distribution result;

[0097] S43, using the electric energy meter temperature distribution data corresponding to the electric energy meter temperature monitoring data as the actual electric energy meter temperature distribution data, and calculating a data error term between the predicted temperature distribution data and the actual temperature distribution data;

[0098] S44, using the predicted temperature distribution result of the electric energy meter as the temperature field to be predicted, and calculating the residual term of the spatial discrete temperature distribution;

[0099] S45, according to the data error term and the spatial discrete temperature distribution residual term, using the heat conduction prediction loss function to perform parameter derivation on the physical information neural network to perform back propagation to obtain a gradient;

[0100] S46. According to the adaptive learning rate method and gradient, obtain the first-order moment estimate and the second-order moment estimate of each parameter in the physical information neural network; in this embodiment, the first-order moment estimate is the mean of the gradient, and the second-order moment estimate is the uncentered variance of the gradient;

[0101] The method for adaptive learning rate in S46 adopts Adam algorithm or L-BFGS algorithm.

[0102] The Adam algorithm is an optimization algorithm with adaptive learning rate. The L-BFGS algorithm is an unconstrained nonlinear programming method based on Hessian approximation.

[0103] S47, according to the first-order matrix estimation value and the second-order matrix estimation value of each parameter in the physical information neural network, adaptively adjust the learning rate of each parameter to complete batch training; each batch training includes two processes of forward propagation and back propagation;

[0104] S48. Repeat batch training several times, and adjust the hyperparameters of the residual term of the spatial discrete temperature distribution and the hyperparameters of the data error term before each repeated batch training, until the result of the heat conduction prediction loss function is less than a preset convergence threshold or the number of batch training reaches a preset number of training times, to obtain a trained physical information neural network.

[0105] S5. Acquire temperature monitoring data of a target electric energy meter, and input the temperature monitoring data of the target electric energy meter into a trained physical information neural network to perform temperature distribution prediction, thereby obtaining a temperature distribution prediction result of the target electric energy meter;

[0106] The S5 comprises the following steps:

[0107] S51, obtaining the average ambient temperature, ambient temperature change rate, first board temperature, second board temperature, first board temperature change rate, and second board temperature change rate of the target electric energy meter within the target time window as temperature monitoring data of the target electric energy meter;

[0108] In this embodiment, the time length of the target time window is the same as the time length of the preset time window when obtaining the average ambient temperature of the electric energy meter, and the method for obtaining the temperature monitoring data of the target electric energy meter is the same as the method for obtaining the electric energy meter monitoring data in the electric energy meter temperature monitoring data set.

[0109] S52, inputting the temperature monitoring data of the target electric energy meter into the trained physical information neural network to perform temperature distribution prediction, and obtaining the temperature distribution prediction result of the target electric energy meter.

[0110] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting temperature distribution of electric energy meters based on physical information neural network, characterized in that: The steps include: S1. Obtaining the temperature monitoring data set of the electric energy meter; S2, selecting a physical information neural network for predicting the temperature distribution of an electric energy meter; S3. Based on the physical model, define the heat conduction prediction loss function of the physical information neural network; S4. According to the heat conduction prediction loss function, the physical information neural network is trained using the electric energy meter temperature monitoring data set to obtain a trained physical information neural network; S5. Acquire temperature monitoring data of a target electric energy meter, and input the temperature monitoring data of the target electric energy meter into a trained physical information neural network to perform temperature distribution prediction, thereby obtaining a temperature distribution prediction result of the target electric energy meter.

2. The method for predicting temperature distribution of electric energy meters based on physical information neural network according to claim 1, characterized in that: The electric energy meter temperature monitoring data set in S1 includes electric energy meter temperature monitoring data of several different time periods, and electric energy meter temperature distribution data corresponding to each electric energy meter temperature monitoring data; The temperature monitoring data of the electric energy meter in different time periods include the average ambient temperature of the electric energy meter within a preset time window, the ambient temperature change rate, the temperature on the first board, the temperature on the second board, the temperature change rate on the first board and the temperature change rate on the second board.

3. The method for predicting temperature distribution of electric energy meters based on physical information neural network according to claim 1, characterized in that: The physical information neural network for predicting the temperature distribution of the electric energy meter adopts a fully connected neural network or a recurrent neural network.

4. The method for predicting temperature distribution of electric energy meters based on physical information neural network according to claim 2, characterized in that: The S3 comprises the following steps: S31, selecting a two-dimensional heat conduction model as a physical model; The calculation expression of the two-dimensional heat conduction model is as follows: Among them, u represents the temperature field, t represents the time, α represents the thermal diffusivity, x represents the horizontal spatial coordinate, and y represents the vertical spatial coordinate. represents the first-order partial differential result of the temperature field with respect to time, represents the second-order partial differential result of the temperature field with respect to the transverse space coordinate, It represents the second-order partial differential result of the temperature field with respect to the longitudinal spatial coordinate; S32, based on the physical model, calculating the residual of the temperature field to be predicted; The calculation expression of the residual error of the temperature field to be predicted is as follows: Among them, RES represents the residual error of the temperature field to be predicted, u pre represents the temperature field to be predicted, It represents the first-order partial differential result of the temperature field to be predicted with respect to time. It represents the second-order partial differential result of the temperature field to be predicted with respect to the transverse space coordinates, It represents the second-order partial differential result of the temperature field to be predicted with respect to the longitudinal space coordinate; S33, discretizing the spatial domain of the temperature field to be predicted according to the residual of the temperature field to be predicted, and obtaining a spatially discrete temperature distribution residual term; The calculation expression of the residual term of the spatial discrete temperature distribution is as follows: Among them, L pre Represents the residual term of spatial discrete temperature distribution, N t Indicates the preset time window length, N x Indicates the maximum value of the horizontal space coordinate on the circuit board of the energy meter, N y Indicates the maximum longitudinal spatial coordinate on the circuit board of the electric energy meter; S34, constructing a data error term between the predicted temperature distribution result of the electric energy meter and the actual temperature distribution data of the electric energy meter; The calculation expression of the data error term is as follows: Among them, L data represents the data error term, represents the predicted temperature distribution result of the electric energy meter, u obs Represents the actual temperature distribution data of the electric energy meter, || || 2 represents the square of the two-norm; S35. Based on the spatial discrete temperature distribution residual term and the data error term, define the heat conduction prediction loss function of the physical information neural network; The calculation expression of the heat conduction prediction loss function is as follows: LOSS=λ data L data +λ pre L pre Where LOSS represents the heat conduction prediction loss function, λ data Represents the hyperparameter of the residual term of the spatial discrete temperature distribution, λ pre represents the hyperparameter of the error term for the data.

5. The method for predicting temperature distribution of electric energy meters based on physical information neural network according to claim 4 is characterized in that: The S4 comprises the following steps: S41, randomly initialize the weight of each connection in the physical information neural network; S42, inputting the electric energy meter temperature monitoring data in the electric energy meter temperature monitoring data set into the physical information neural network, and performing forward propagation to obtain a predicted electric energy meter temperature distribution result; S43, using the temperature distribution data of the electric energy meter corresponding to the temperature monitoring data of the electric energy meter as the actual temperature distribution data of the electric energy meter, and calculating a data error term between the predicted temperature distribution data and the actual temperature distribution data; S44, using the predicted temperature distribution result of the electric energy meter as the temperature field to be predicted, and calculating the residual term of the spatial discrete temperature distribution; S45, according to the data error term and the spatial discrete temperature distribution residual term, using the heat conduction prediction loss function to perform parameter derivation on the physical information neural network to perform back propagation to obtain a gradient; S46, obtaining a first-order moment estimate and a second-order moment estimate of each parameter in the physical information neural network according to the method and gradient of the adaptive learning rate; S47, according to the first-order matrix estimation value and the second-order matrix estimation value of each parameter in the physical information neural network, adaptively adjust the learning rate of each parameter to complete the batch training; S48. Repeat batch training several times, and adjust the hyperparameters of the spatial discrete temperature distribution residual term and the hyperparameters of the data error term before each repeated batch training, until the result of the heat conduction prediction loss function is less than a preset convergence threshold or the number of batch training reaches a preset number of training times, to obtain a trained physical information neural network.

6. The method for predicting temperature distribution of electric energy meters based on physical information neural network according to claim 5, characterized in that: The method for adaptive learning rate in S46 adopts Adam algorithm or L-BFGS algorithm.

7. The method for predicting temperature distribution of electric energy meters based on physical information neural network according to claim 6, characterized in that: The S5 comprises the following steps: S51, obtaining the average ambient temperature, ambient temperature change rate, first board temperature, second board temperature, first board temperature change rate, and second board temperature change rate of the target electric energy meter within the target time window as temperature monitoring data of the target electric energy meter; S52, inputting the temperature monitoring data of the target electric energy meter into the trained physical information neural network to perform temperature distribution prediction, and obtaining the temperature distribution prediction result of the target electric energy meter.

Citation Information

Cited By

  • Multi-parameter integrated detection method and device for frozen soil, terminal and storage medium

    CN119959517A

  • Multi-parameter integrated detection method, device, terminal and storage medium for frozen soil

    CN119959517B

  • Cryoablation temperature prediction method and device, electronic equipment and storage medium

    CN120549608A

  • Transformer magnetic field prediction method and system based on physical information neural network

    CN120807439A