A method and system for temperature inversion in switchgear

Abstract

This invention relates to a method and system for retrieving switchgear temperature. The method includes: Step S1: acquiring historical temperature data of the switchgear and training a deep neural network using the historical temperature data; Step S2: during the training of the deep neural network, searching for the optimal parameters of the model using an enhanced Newton-Raphson optimization algorithm to obtain the trained deep neural network; Step S3: inputting real-time temperature data of the switchgear into the trained deep neural network to predict the future temperature of the switchgear. This invention can effectively retrieve the temperature of switchgear.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of switch cabinet temperature detection in power systems, in particular to a switch cabinet temperature inversion method and system. BACKGROUND

[0002] As a key equipment in power transmission and distribution links, switch cabinets are widely used in various production fields such as industry, energy, and construction, and bear the core functions of power distribution, operation control, and system protection. The running state of the switch cabinet is directly related to the safety and stability of the power supply system. Under the conditions of long-term high-load operation or aging of connecting components, local overheating phenomenon is easy to occur inside the switch cabinet. If it is not found and handled in time, it will not only lead to the decline of equipment performance, but also further cause serious accidents such as short circuit, fire, etc., and even cause major economic losses and social impact. Therefore, real-time and accurate inversion monitoring of the internal temperature of the switch cabinet has very important engineering significance.

[0003] The core method of switch cabinet temperature inversion can be divided into traditional physical model method and data-driven method. The traditional physical model method establishes an accurate geometric model and uses the basic laws of heat transfer and fluid mechanics to construct partial differential equations to simulate the temperature field distribution in the cabinet. This method has clear physical meaning. The core of the data-driven method is to use a large amount of historical operation data to train the model through machine learning or deep learning algorithms to find the complex nonlinear mapping relationship between these measurable parameters and the internal target temperature. This method has strong adaptability.

[0004] In the practical application of high-current switch cabinet temperature inversion, the traditional physical model method is relatively complex and has certain limitations, and the data-driven method is usually used for temperature inversion. However, although the data-driven method has strong nonlinear fitting ability, it is essentially a complex optimization problem, which has high computational complexity and is prone to local optimization. Therefore, in order to improve the reliability and practicality of the inversion result, an optimization algorithm is introduced into the data-driven framework.

[0005] Complex optimization problems refer to optimization problems with high-dimensional nonlinearity, multi-modal, multi-constraint, etc. Such problems often have multiple local optimal solutions, discontinuous or non-differentiable objective functions, complex search spaces, etc. Challenges, such as optimizing the algorithm model when inverting the temperature of a high-current switch cabinet. Efficiently solving complex optimization problems can improve system performance, enabling the model to better fit the temperature changes of a high-current switch cabinet, which is of great significance for correctly inverting the temperature of a high-current switch cabinet.

[0006] In solving complex optimization problems, the current common methods include swarm intelligence optimization algorithms (such as grey wolf algorithm, whale algorithm, particle swarm optimization algorithm) and classical mathematical optimization algorithms (such as Newton-Raphson algorithm) and the like. Swarm intelligence optimization algorithm is a kind of meta-heuristic algorithm inspired by the behavior of biological population in nature. They search for optimization in the solution space by simulating the cooperation and competition among individuals in the population. Its significant advantage lies in its strong global exploration capability, which can effectively handle nonlinear, high-dimensional complex problems, but when solving high-dimensional complex optimization problems, it often needs a large number of function evaluation times to converge, and the calculation efficiency is low, and with the increase of problem dimension, it is easy to appear dimension disaster, which leads to the sharp decline of search performance. The classical mathematical optimization method is based on mathematical theory and analysis, and the gradient, Hessian matrix and other precise mathematical information of the objective function are calculated to iteratively search for optimization. This method has a solid mathematical foundation, and under the ideal conditions of meeting the function smoothness, convexity and the like, it usually has extremely fast local convergence speed, and the convergence is strictly proved, but when facing complex optimization problems such as multi-peak and non-convex, it is easy to fall into local optimal solution, and the global search ability is insufficient, and the continuity and derivability of the objective function are strictly required, and the applicable range is limited.

[0007] In summary, the existing temperature inversion of high-current switch cabinet is a complex optimization problem, and usually needs to introduce optimization algorithm in the data-driven framework (machine learning or deep learning) to solve the optimal solution, but these optimization algorithms have the problems of low calculation efficiency and easy to fall into local optimum, which finally affects the accuracy of switch cabinet temperature inversion. SUMMARY

[0008] Therefore, the technical problem to be solved by the present application is to overcome the problem in the prior art that the temperature inversion of high-current switch cabinet needs to introduce optimization algorithm in the data-driven framework (machine learning or deep learning) to solve the optimal solution, but these optimization algorithms have the problems of low calculation efficiency and easy to fall into local optimum, which finally affects the accuracy of switch cabinet temperature inversion.

[0009] To solve the above technical problems, the present application provides a switch cabinet temperature inversion method, comprising:

[0010] Step S1: obtaining historical temperature data of the switch cabinet, and training a deep neural network through the historical temperature data;

[0011] Step S2: in the training process of the deep neural network, searching for the model optimal parameters of the deep neural network in the training process through an enhanced Newton-Raphson optimization algorithm, and finally obtaining a trained deep neural network;

[0012] Step S3: inputting real-time temperature data of the switch cabinet into the trained deep neural network, and predicting the future temperature of the switch cabinet through the trained deep neural network.

[0013] In one embodiment of the present invention, step S2, which involves searching for the optimal parameters of a deep neural network model during training using an enhanced Newton-Raphson optimization algorithm, includes:

[0014] Step S21: Initialize the population using historical temperature data to obtain initial samples;

[0015] Step S22: Update the location of each individual in the initial sample;

[0016] Step S23: Determine whether the maximum number of iterations has been reached. If not, return to step S22 to continue the optimization process until the maximum number of iterations is reached and the optimal solution is output.

[0017] In one embodiment of the present invention, the method for updating the position of each individual in the initial sample in step S22 includes:

[0018] Construct the difference step size, expressed as:

[0019] (2);

[0020] in , The first and second random numbers are in the interval [0,1]. , For two distinct individuals randomly selected from the population, use their indices. To be the globally optimal position , For any two individuals in the population;

[0021] In the Newton-Raphson optimization algorithm, the search direction is calculated using the Newton-Raphson search rule (NRSR), expressed as:

[0022] (3);

[0023] in, For the Newton-Raphson search direction, This is the worst position globally. These are random numbers distributed according to a standard normal distribution.

[0024] Based on the Newton-Raphson search direction Sum and difference step size Calculate two candidate positions, represented as:

[0025] (4);

[0026] (5);

[0027] in, , As the first and second candidate positions, express or ;

[0028] Constructing an adaptive decay factor The adaptive attenuation factor Used to control the search step size, expressed as:

[0029] (6);

[0030] in, Represents the current iteration number. Represents the maximum number of iterations;

[0031] Based on the first and second candidate positions , and adaptive attenuation factor Update position composition, represented as:

[0032] (7);

[0033] in, This is the updated composition location.

[0034] In one embodiment of the present invention, the method for updating the position of each individual in the initial sample in step S22 further includes: using an improved golden sine strategy to assist in updating the synthetic position. To further improve update diversity, this will be updated again, as follows:

[0035] (10);

[0036] (11);

[0037] in, Update the strategy for the golden sine wave. This is the position after being updated with the assistance of the golden sine strategy. Scaling factor It is the first The position of the individual in the next iteration. It is the first The iterative method using the golden sine strategy yielded the first... The location of each individual It is the first The individual The global optimal position in the next iteration. yes The random number within the range controls the search distance; yes The random number within the range controls the search direction; These are the first and second golden ratios.

[0038] In one embodiment of the invention, the scaling factor Represented as:

[0039] (9);

[0040] in, The initial scaling factor. Minimum scaling factor Represents the current iteration number. This represents the maximum number of iterations.

[0041] In one embodiment of the present invention, step S22 further includes: to balance global search and local search capabilities, using adaptive nonlinear control parameters to determine whether to perform individual position updates, wherein,

[0042] Obtaining standard normal distribution random numbers If the standard normal distribution random number Less than adaptive nonlinear control parameters Then the individual's location is updated, if the standard normal distribution random number is used. Greater than the adaptive nonlinear control parameter If the individual location is not updated, then step S23 is executed.

[0043] In one embodiment of the present invention, the adaptive nonlinear control parameters Decision factors based on the original Newton-Raphson optimization algorithm The formula is as follows:

[0044] (15);

[0045] in, For adaptive nonlinear control parameters, yes initial value, It is the current iteration number. It represents the maximum number of iterations the algorithm can run.

[0046] To solve the above-mentioned technical problems, the present invention provides a switchgear temperature inversion system, comprising:

[0047] Training module: Used to acquire historical temperature data of the switchgear and train the deep neural network using the historical temperature data;

[0048] Optimization module: Used to search for the optimal parameters of the deep neural network model during the training process by using the enhanced Newton-Raphson optimization algorithm, and finally obtain the trained deep neural network.

[0049] Inversion module: This module is used to input the real-time temperature data of the switchgear into a trained deep neural network, which then predicts the future temperature of the switchgear.

[0050] To solve the above-mentioned technical problems, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the switch cabinet temperature inversion method described above.

[0051] To solve the above-mentioned technical problems, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the switch cabinet temperature inversion method described above are implemented.

[0052] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0053] The switchgear temperature inversion method described in this invention addresses the challenge of complex operating environments and the interplay of various factors affecting temperature distribution in switchgear. Traditional DNN model parameter optimization methods struggle to fully leverage data features, resulting in limited temperature inversion accuracy. To overcome this bottleneck, this invention introduces an enhanced Newton-Raphson optimization algorithm. This algorithm more efficiently explores optimal solutions within a vast parameter space, precisely adjusts DNN model parameters, and significantly improves the accuracy of switchgear temperature inversion, providing strong assurance for the safe and stable operation of switchgear. Attached Figure Description

[0054] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0055] Figure 1 This is a flowchart of the method of the present invention;

[0056] Figure 2 This is a scatter plot of individual distribution in an embodiment of the present invention;

[0057] Figure 3 This is an individual distribution histogram in an embodiment of the present invention;

[0058] Figure 4 This is a temperature residual map of the DNN model performing temperature inversion on simulation data in this embodiment of the invention;

[0059] Figure 5This is a temperature residual map of the simulation data obtained by the ENRBO-DNN model in this embodiment of the invention. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0061] Example 1

[0062] Reference Figure 1 As shown, this invention relates to a method for temperature inversion in switchgear, comprising:

[0063] Step S1: Obtain historical temperature data of the switchgear, and train the deep neural network using the historical temperature data;

[0064] Step S2: During the training of the deep neural network, the optimal parameters of the deep neural network model are searched using the enhanced Newton-Raphson optimization algorithm to finally obtain the trained deep neural network.

[0065] Step S21: Initialize the population using historical temperature data to obtain initial samples;

[0066] Step S22: Update the location of each individual in the initial sample;

[0067] Step S23: Determine whether the maximum number of iterations has been reached. If not, return to step S22 to continue the optimization process until the maximum number of iterations is reached, and then output the optimal solution (optimal individual).

[0068] Step S3: Input the real-time temperature data of the switchgear into the trained deep neural network, and use the trained deep neural network to predict the temperature data of the switchgear at the next moment.

[0069] It should be noted that each individual represents the set of parameters to be optimized during the training process of the deep neural network model, and the optimal individual obtained at the end represents the optimal set of parameters during the model training process.

[0070] The following is a detailed description of this embodiment:

[0071] This embodiment addresses the problem of premature convergence and failure to find the optimal solution that often occurs in the original Newton-Raphson Based Optimizer (NRBO) when dealing with complex nonlinear optimization problems. The algorithm is improved in three aspects: initial population uniformity, balance between global and local search, and diversity of position updates. By introducing the Latin hypercube sampling method to initialize the population, the uniformity of the initial population distribution in the search space is improved. Adaptive nonlinear control parameters are proposed to enhance the algorithm's global exploration capability in the early stages and its local exploration capability in the later stages. An improved golden sine strategy is used to update variable positions, improving the algorithm's exploration capability and avoiding getting trapped in local optima.

[0072] The Enhanced Newton-Raphson Based Optimizer (ENRBO) algorithm in this embodiment searches for the optimal parameters of the deep neural network model during training, including the following steps:

[0073] Step S1: To obtain a uniformly distributed initial population, Latin hypercube sampling is used for population initialization. Latin hypercube sampling is a high-dimensional spatial statistical sampling method that can generate uniformly distributed samples within the search space. Specifically, the interval [0,1] is uniformly divided into n equal parts (n sub-intervals). Sampling points are randomly selected and permuted within each sub-interval to ensure that each parameter axis is uniformly distributed and non-repeating. The expression is shown in Equation 1:

[0074] (1)

[0075] in, This represents the final sample point in the actual search space. This represents a uniformly distributed sample generated by Latin hypercube sampling in the interval [0,1]. and These represent the upper and lower bounds of the search space, respectively.

[0076] Assuming a population size of 50 and a variable dimension of 1, the initial population is generated using random initialization, tent chaotic mapping, and Latin hypercube sampling methods, respectively. Figure 2 This is a scatter plot of individual distribution obtained from three sampling methods. The vertical axis represents the mapping value of each point between 0 and 1, and the horizontal axis represents the individual ID of the population. Figure 3 This is a histogram of individual distributions when the population size is set to 50. Comparing random initialization, TENT chaotic sequence, and Latin hypercube sampling, it is easy to see that the Latin hypercube sampling method yields a more uniform population distribution. Therefore, this embodiment uses the Latin hypercube sampling method to initialize the population, improving the diversity of optimization solutions and the algorithm's global search capability.

[0077] Step two (S2) involves updating the position information using the vector synthesis formula of the Newton-Raphson search rule combined with a golden sine strategy to achieve efficient position updates. The NRSR (Newton-Raphson Search Rule) is the core search mechanism of the NRBO (Newton-Raphson Based Optimizer) algorithm, which utilizes the principles of Newton's method combined with random perturbations to achieve a global search.

[0078] Differential step size calculation:

[0079] (2)

[0080] in , The first and second random numbers are in the interval [0,1]. , For two distinct individuals randomly selected from the population, use their indices. To be the globally optimal position , Let be the positions of any two individuals in the population.

[0081] The Newton-Raphson Search Rule (NRSR) calculates the search direction:

[0082] (3)

[0083] in, For the Newton-Raphson search direction, This is the worst position globally. These are random numbers that follow a standard normal distribution.

[0084] Based on the Newton-Raphson search direction Sum and difference step size Calculate two candidate positions:

[0085] (4)

[0086] (5)

[0087] in, , As the first and second candidate positions, express or .

[0088] Constructing an adaptive decay factor Controlling the search step size:

[0089] (6)

[0090] in, Represents the current iteration number. This represents the maximum number of iterations during the iteration process. It updates as the number of iterations changes, and its value varies between 1 and -1 as the number of iterations increases.

[0091] Based on the first and second candidate positions , and adaptive attenuation factor Update position composition, represented as:

[0092] (7)

[0093] in, This is the updated composition location.

[0094] This invention introduces an improved Golden Sine Algorithm (GSA) after updating the basic position. It utilizes the periodic oscillation characteristics of the sine function and the uniform division characteristics of the golden ratio to increase the diversity of search paths and improve the algorithm's ability to escape local optima.

[0095] The original formula for the golden sine strategy is as follows:

[0096] (8)

[0097] However, directly introducing the golden sine strategy will result in an excessively large maximum perturbation, which may disrupt the search stability and lead to the loss of high-quality solutions.

[0098] This embodiment improves the golden sine strategy by introducing a scaling factor. This allows the perturbation amplitude to be larger in the early stages of the algorithm to facilitate global exploration, and smaller in the later stages to facilitate fine-grained search.

[0099] (9)

[0100] in, The initial scaling factor. Minimum scaling factor Represents the current iteration number. This represents the maximum number of iterations.

[0101] The improved golden sine position update formula in this embodiment is:

[0102] (10)

[0103] (11)

[0104] in, Update the strategy for the golden sine wave. This is the position after being updated with the assistance of the golden sine strategy. Scaling factor It is the first The position of the individual in the next iteration. It is the first The iterative method using the golden sine strategy yielded the first... The location of each individual It is the first The individual The global optimal position in the next iteration. yes The random number within the range controls the search distance; yes The random number within the range controls the search direction; These are the first and second golden ratios.

[0105] Step 3 (S3) balances global and local search capabilities by using adaptive nonlinear control parameters to determine whether to employ a trap avoidance operator to update position information. The original NRBO uses a trap avoidance operator to help escape local optima. Random numbers are generated. ,Will With decision factors Compare:

[0106] like : Triggering trap avoidance operator, position update performs jump update;

[0107] like Maintain the basic update position (update the position normally according to the original algorithm);

[0108] Furthermore, triggering the trap avoidance operator, for and The formula for performing a skip update is as follows:

[0109] (12)

[0110] (13)

[0111] (14)

[0112] in, This is the difference between the current individual position and the global optimal position. This is the difference between the population average position and the current individual position. This represents the average population location. The meaning is that random numbers will affect the update position of the algorithm, and the two in formula (14) They are the same, but the algorithm will only take one of the paths; To indicate or , Represents a random number. and These are uniformly random numbers between (-1, 1) and (-0.5, 0.5), respectively. It is a decision factor. and It is a random number:

[0113] In the original NRBO algorithm, decision factors The algorithm switches between global and local search modes. In the original NRBO algorithm, The fact that the parameter is a constant leads to a weak balance between global and local search capabilities in the original NRBO algorithm, making it prone to insufficient search or premature convergence. Therefore, an adaptive nonlinear control parameter is proposed to improve the NRBO algorithm's global exploration capability in the early stages and its local exploration capability in the later stages. Adaptive nonlinear control parameter The expression is as follows:

[0114] (15)

[0115] in, For adaptive nonlinear control parameters, yes initial value, It is the current iteration number. It represents the maximum number of iterations the algorithm can run.

[0116] Step 4 (S4) To ensure algorithm convergence, determine whether the current iteration count has reached the maximum iteration count. If the maximum number of iterations has not been reached, then the current iteration number will be... Increment by 1, return to step two and continue the optimization process; if the maximum number of iterations has been reached, the algorithm terminates and outputs the current global optimal solution as the final optimization result.

[0117] The invention will be further illustrated below with specific examples. One example is the parameter optimization of a DNN temperature inversion model for a high-current switchgear.

[0118] Deep Neural Networks (DNNs) are an extension of multilayer perceptrons. Their core principle is to automatically learn the complex nonlinear relationship between input and output through the stacking of multiple neurons. DNN models achieve a complex mapping from input features to the output target through an ordered combination of input layers, hidden layers, and output layers. The input layer, as the network's entry point, formally receives the input data. The hidden layers include fully connected layers and activation function layers. The fully connected layers are the core computational part of the network, responsible for nonlinear transformations and combinations of features; the activation function layers introduce nonlinear transformations, enabling the model to fit complex problems. Finally, the output layer outputs the model's temperature inversion results.

[0119] Training a DNN model involves three processes: forward propagation, back propagation, and parameter update. Forward propagation begins at the input layer, feeding a batch of training data into the network. The data propagates layer by layer, with each neuron undergoing weighted summation and activation function operations until the output layer's prediction is obtained. A loss function is then needed to measure the difference between the model's predictions and the true values. Back propagation minimizes this difference. It starts at the output layer and calculates the gradient of the loss function with respect to each weight and bias in the network. Finally, the parameter update process uses gradient descent to update all parameters, learning from the data to form an optimal parameter model.

[0120] In the research on achieving accurate temperature inversion of switchgear using a DNN model, the key and challenging aspect lies in the parameter optimization stage of the DNN model. The operating environment of switchgear is complex, and temperature distribution is influenced by a variety of factors. Traditional parameter optimization methods struggle to fully exploit data features, resulting in limited temperature inversion accuracy. To overcome this bottleneck, an enhanced Newton-Raphson optimization algorithm is introduced. This algorithm can more efficiently explore the optimal solution within a vast parameter space, precisely adjust the DNN model parameters, and thus significantly improve the accuracy of the model's temperature inversion for switchgear, providing a strong guarantee for the safe and stable operation of the switchgear.

[0121] Experiments and Analysis

[0122] The high-current switchgear was modeled and simulated using COMSOL software to obtain simulation data under different conductivities, which was then used in the DNN model for temperature inversion of the high-current switchgear. Figure 4 The graph shows the temperature residuals obtained by the DNN model from the simulation data. The solid line represents the baseline, and the dashed line represents the difference between the temperature inversion result of the DNN model on the test set samples and the actual temperature value. The closer the black dashed line is to the solid line, the better the simulation effect of the model.

[0123] Subsequently, the ENRBO optimization algorithm was introduced to optimize the parameters of the DNN model during training, enabling the DNN model to obtain better feature parameters from the simulation data and thus improve the model's inversion accuracy. The temperature inversion residual plot of the DNN model optimized using the ENRBO algorithm is shown below. Figure 5 As shown. Comparison Figure 4 and Figure 5 It can be seen that, Figure 5 The variation of the dashed line is smaller and closer to the solid line, indicating that the performance of the model optimized by the ENRBO optimization algorithm is improved, and it can perform more accurate temperature inversion for high-current switchgear.

[0124] Mean squared error (MSE) is the sum of the squares of the differences between predicted and true values ​​divided by the sample size. It reflects the average squared error; the smaller the value, the better the model. Root mean squared error (RMSE) is the square root of the mean squared error, restoring the error to the same order of magnitude as the original data. The smaller the value, the higher the prediction accuracy. Mean absolute error (MAE) is the average of the absolute differences between predicted and true values, reflecting the average linear level of the error. The smaller the value, the more stable the model's predictions. The coefficient of determination (R²) represents the proportion of the dependent variable variation that the model can explain. It measures the goodness of fit of the model to the data; the closer R² is to 1, the better the model fit.

[0125] Table 1 shows a comparison of the performance evaluation metrics of the DNN model before and after optimization. The table shows that the optimized model's MSE decreased from 0.0868 to 0.0140, a reduction of 83.87%, significantly lowering the average squared error. RMSE decreased from 0.2946 to 0.1182, a reduction of 60.08%, indicating a significant reduction in the average deviation between the model's predicted and actual values. MAE decreased from 0.2512 to 0.1012, a reduction of 59.71%, demonstrating a significant improvement in the model's robustness and predictive stability. Furthermore, the original model's R² was already 0.9999, indicating a good fit. However, the optimized model's R² increased to 1.0000, indicating that the ENRBO-DNN model better captures the changing patterns in the data, achieving an ideal fit. From the data comparison in the table, we can conclude that the optimized ENRBO-DNN model outperforms the original DNN model in all performance metrics, with significantly improved prediction accuracy and fit.

[0126] Table 1 Comparison of DNN model performance evaluation metrics before and after optimization

[0127]

[0128] Example 2

[0129] This embodiment provides a switchgear temperature inversion system, including:

[0130] Training module: Used to acquire historical temperature data of the switchgear and train the deep neural network using the historical temperature data;

[0131] Optimization module: Used to search for the optimal parameters of the deep neural network model during the training process by using the enhanced Newton-Raphson optimization algorithm, and finally obtain the trained deep neural network.

[0132] Inversion module: This module is used to input the real-time temperature data of the switchgear into a trained deep neural network, which then predicts the future temperature of the switchgear.

[0133] Example 3

[0134] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the switch cabinet temperature inversion method described in Embodiment 1.

[0135] Example 4

[0136] This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the switchgear temperature inversion method described in Embodiment 1.

[0137] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0138] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0141] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for temperature inversion in switchgear, characterized in that, include: Step S1: Obtain historical temperature data of the switchgear, and train the deep neural network using the historical temperature data; Step S2: During the training of the deep neural network, the optimal parameters of the deep neural network model are searched using the enhanced Newton-Raphson optimization algorithm to finally obtain the trained deep neural network. Step S3: Input the real-time temperature data of the switchgear into the trained deep neural network, and use the trained deep neural network to predict the future temperature of the switchgear.

2. The switchgear temperature inversion method according to claim 1, characterized in that: The method for searching the optimal parameters of the deep neural network model during the training process using the enhanced Newton-Raphson optimization algorithm in step S2 includes: Step S21: Initialize the population using historical temperature data to obtain initial samples; Step S22: Update the location of each individual in the initial sample; Step S23: Determine whether the maximum number of iterations has been reached. If not, return to step S22 to continue the optimization process until the maximum number of iterations is reached and the optimal solution is output.

3. The switchgear temperature inversion method according to claim 2, characterized in that: The method for updating the location of each individual in the initial sample in step S22 includes: Construct the difference step size, expressed as: (2)； in , The first and second random numbers are in the interval [0,1]. , For two distinct individuals randomly selected from the population, use their indices. To be the globally optimal position , For any two individuals in the population; In the Newton-Raphson optimization algorithm, the search direction is calculated using the Newton-Raphson search rule (NRSR), expressed as: (3)； in, For the Newton-Raphson search direction, This is the worst position globally. These are random numbers distributed according to a standard normal distribution. Based on the Newton-Raphson search direction Sum and difference step size Calculate two candidate positions, represented as: (4)； (5)； in, , As the first and second candidate positions, express or ; Constructing an adaptive decay factor The adaptive attenuation factor Used to control the search step size, expressed as: (6)； in, Represents the current iteration number. Represents the maximum number of iterations; Based on the first and second candidate positions , and adaptive attenuation factor Update position composition, represented as: (7)； in, This is the updated composition location.

4. The switchgear temperature inversion method according to claim 3, characterized in that: The method for updating the position of each individual in the initial sample in step S22 further includes: using an improved golden sine strategy to assist in updating the synthetic position. To further improve update diversity, this will be updated again, as follows: (10)； (11)； in, Update the strategy for the golden sine wave. This is the position after being updated with the assistance of the golden sine strategy. Scaling factor It is the first The position of the individual in the next iteration. It is the first The iterative method using the golden sine strategy yielded the first... The location of each individual It is the first The individual The global optimal position in the next iteration. yes The random number within the range controls the search distance; yes The random number within the range controls the search direction; These are the first and second golden ratios.

5. The switchgear temperature inversion method according to claim 4, characterized in that: The scaling factor Represented as: (9)； in, The initial scaling factor. Minimum scaling factor Represents the current iteration number. This represents the maximum number of iterations.

6. The switchgear temperature inversion method according to claim 2, characterized in that: Step S22 further includes: to balance global and local search capabilities, using adaptive nonlinear control parameters to determine whether to perform individual position updates, wherein... Obtaining standard normal distribution random numbers If the standard normal distribution random number Less than adaptive nonlinear control parameters Then the individual's location is updated, if the standard normal distribution random number is used. Greater than the adaptive nonlinear control parameter If the individual location is not updated, then step S23 is executed.

7. The switchgear temperature inversion method according to claim 6, characterized in that: The adaptive nonlinear control parameters Decision factors based on the original Newton-Raphson optimization algorithm The formula is as follows: （15）； in, For adaptive nonlinear control parameters, yes initial value, It is the current iteration number. It represents the maximum number of iterations the algorithm can run.

8. A switchgear temperature inversion system, characterized in that, include: Training module: Used to acquire historical temperature data of the switchgear and train the deep neural network using the historical temperature data; Optimization module: Used to search for the optimal parameters of the deep neural network model during the training process by using the enhanced Newton-Raphson optimization algorithm, and finally obtain the trained deep neural network. Inversion module: This module is used to input the real-time temperature data of the switchgear into a trained deep neural network, which then predicts the future temperature of the switchgear.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the switchgear temperature inversion method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the switchgear temperature inversion method as described in any one of claims 1 to 7.

Images