A transformer oil micro-water content prediction method and device based on an improved TRANSFORMER

Abstract

This invention discloses a method and apparatus for predicting trace moisture content in transformer oil based on an improved Transformer. The method involves obtaining filtered trace moisture and temperature measurement signals; multiplying the filtered trace moisture measurement signal by (1+ξ) to obtain a compensated trace moisture measurement signal; using the compensated trace moisture, temperature, and weather data as training samples to train a neural network model; acquiring the compensated trace moisture measurement signal in real time and displaying it through a visual interface; inputting the real-time compensated trace moisture, future temperature, and weather data into the trained neural network model to obtain the predicted transformer moisture content; and calculating the transformer's failure probability based on the predicted moisture content. This invention provides support for real-time transformer monitoring, timely detection of potential transformer faults, and a basis for transformer operation and maintenance.

Description

Technical Field

[0001] This invention relates to a method and apparatus for predicting trace water content in transformer oil based on an improved Transformer, belonging to the field of power equipment monitoring technology. Background Technology

[0002] Power transformers, as crucial hubs in power transmission and distribution, are widely distributed throughout the power grid. Among them, oil-immersed transformers account for a large proportion due to their excellent insulation performance and economic efficiency. However, due to their large number and wide geographical distances, timely and effective safety maintenance is difficult.

[0003] Currently, excessive water content in transformer oil leading to reduced insulation performance and further causing partial discharge breakdown faults resulting in equipment damage and downtime is one of the primary problems to be solved in the industry. As a liquid insulating material, transformer oil, when its water content increases, not only reduces the breakdown voltage of the insulation system and increases dielectric loss, but also directly participates in the chemical degradation reaction of polymer materials such as oil paper fibers, promoting the degradation and aging of these materials, thereby accelerating the deterioration of various performance aspects of the insulation system. When the water content in the oil exceeds a certain threshold, the insulation performance of the equipment will be greatly reduced, and in severe cases, it can lead to major accidents such as insulation breakdown and equipment burnout.

[0004] Furthermore, the combination of trace amounts of water and organic acids in the oil not only reduces the insulating capacity of the transformer oil but also leads to the loss of its arc-extinguishing ability. According to relevant Chinese standards (GB / T7595-2000), for transformers with voltage levels of 35kV and below, the breakdown voltage of the operating oil should not be lower than 30kV, while for transformers with voltage levels of 500kV, the breakdown voltage of the operating oil should be 50kV. By monitoring the water content in transformer oil, it is possible not only to prevent the insulation strength of the transformer oil from decreasing to a dangerous level, but also to assess the overall insulation condition of the transformer and judge the sealing performance of the equipment based on the water content.

[0005] Therefore, how to achieve online detection of the trace moisture content of transformers in operation and predict the trend of trace moisture content changes through algorithms is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] Objective: To overcome the shortcomings of existing technologies in measuring trace moisture in transformer oil, such as high cost and poor real-time performance, this invention provides a method and apparatus for predicting trace moisture content in transformer oil based on an improved Transformer.

[0007] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] Firstly, a method for predicting trace water content in transformer oil based on an improved TRANSFORMER includes the following steps:

[0009] Step 1: Acquire the water level measurement signal and temperature measurement signal in the transformer oil, and perform sliding filtering on the water level measurement signal and temperature measurement signal to obtain the filtered water level measurement signal and temperature measurement signal.

[0010] Step 2: With the water content in the transformer oil remaining constant, acquire trace water measurement signals under different set temperatures. Based on the trace water measurement signals and the set temperatures, use the linear least squares method to fit a linear equation between the trace water measurement signals and the set temperatures, and obtain the slope ξ of the linear equation as a compensation coefficient. Multiply the filtered trace water measurement signal by (1+ξ) to obtain the compensated trace water measurement signal.

[0011] Step 3: Obtain weather information, encode the weather information into weather digital information, use the compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and use the training samples to train the neural network model to obtain a trained neural network model.

[0012] Step 4: Acquire the compensated micro-water measurement signal in real time and display the compensated micro-water measurement signal in a visual interface for real-time monitoring of the transformer.

[0013] Step 5: Obtain the real-time compensated micro-water measurement signal, the future temperature measurement signal, and the weather digital information. Input the real-time compensated micro-water measurement signal, the future temperature measurement signal, and the weather digital information into the trained neural network model to obtain the future predicted transformer water content.

[0014] Step 6: Calculate the failure probability of the transformer based on the predicted water content of the transformer in the future. When the failure probability is greater than the threshold, issue an alarm message and arrange for maintenance.

[0015] Secondly, a device for predicting trace water content in transformer oil based on an improved TRANSFORMER includes the following modules:

[0016] Signal preprocessing module: used to acquire the water content measurement signal and temperature measurement signal in transformer oil, and to perform sliding filtering on the water content measurement signal and temperature measurement signal to obtain the filtered water content measurement signal and temperature measurement signal.

[0017] Compensation for micro-water signal: This module is used to acquire micro-water measurement signals under different set temperatures while maintaining a constant water content in the transformer oil. Based on the micro-water measurement signal and the set temperature, a linear least squares method is used to fit a linear equation between the micro-water measurement signal and the set temperature. The slope ξ of this linear equation is then used as a compensation coefficient. The filtered micro-water measurement signal is multiplied by (1+ξ) to obtain the compensated micro-water measurement signal.

[0018] Neural network model training module: Used to acquire weather information, encode the weather information into weather digital information, use the compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and use the training samples to train the neural network model to obtain a trained neural network model.

[0019] Real-time monitoring module: Used to acquire the compensated micro-water measurement signal in real time and display the compensated micro-water measurement signal with a visual interface for real-time monitoring of the transformer.

[0020] Transformer moisture content prediction module: This module is used to acquire real-time compensated micro-moisture measurement signals, future temperature measurement signals, and weather data. The real-time compensated micro-moisture measurement signals, future temperature measurement signals, and weather data are then input into a trained neural network model to obtain the predicted transformer moisture content.

[0021] Alarm module: It is used to calculate the failure probability of the transformer based on the predicted water content of the transformer in the future. When the failure probability is greater than the threshold, an alarm message is issued and maintenance is arranged.

[0022] As a preferred embodiment, the sliding filter processing includes:

[0023] For each sample value acquired, the N consecutively sampled values ​​are placed in an array and sorted according to the first-in-first-out principle. The two largest and two smallest sample values ​​are removed, and the remaining N-4 sample values ​​are subjected to an arithmetic average to obtain the filtered data. Here, N is a fixed number, and the sample values ​​include micro-water measurement signals and temperature measurement signals.

[0024] As a preferred embodiment, the weather information is encoded into weather digital information, including:

[0025] Text weather information is collected using web crawlers, encoded using OneHotEncoder, and then input into Word2Vector to output numerical weather information.

[0026] As a preferred embodiment, the neural network model employs a Transformer+SOM+LSTM model. This model includes a Transformer model, where the decoder is replaced by a structure of a SOM neural network followed by an LSTM neural network. Specifically, the Transformer's encoder in the Transformer+SOM+LSTM model is used to extract features from the data, while the SOM neural network followed by the LSTM neural network structure is used for feature clustering and prediction.

[0027] As a preferred embodiment, the step of training the neural network model using training samples to obtain a trained neural network model includes:

[0028] Step 3.1: Acquire the compensated micro-water measurement signal, temperature measurement signal, and weather data. Perform Z-Score standardization on the compensated micro-water measurement signal, temperature measurement signal, and weather data to obtain the standardized compensated micro-water measurement signal, temperature measurement signal, and weather data. Z-Score standardization includes: calculating the mean and variance of each data category, subtracting the mean from each data category, and then dividing by the variance to achieve the purpose of standardizing each data category.

[0029] Step 3.2: Use the standardized and compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and divide the training samples into training set, validation set, and test set.

[0030] Step 3.3: Input the training set into the Transformer+SOM+LSTM model to obtain the network parameters of the Transformer+SOM+LSTM model.

[0031] Step 3.4: Input the validation set into the Transformer+SOM+LSTM model to obtain the adjustment values ​​of the network parameters.

[0032] Step 3.5: Substitute the adjusted network parameters into the Transformer+SOM+LSTM model, and input the test set into the Transformer+SOM+LSTM model to obtain the test prediction value. Calculate the MAPE evaluation index based on the test prediction value, and use the Transformer+SOM+LSTM model that meets the MAPE evaluation index requirements as the trained neural network model.

[0033] Beneficial Effects: This invention provides a method and apparatus for predicting trace water content in transformer oil based on an improved Transformer. It employs a compensation algorithm to accurately detect the trace water content in transformers, providing effective support for real-time transformer monitoring. Predicting the changing trend of transformer trace water content enables timely detection of potential transformer faults, providing a basis for transformer operation and maintenance. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the method of the present invention.

[0035] Figure 2 This is a schematic diagram of the structure of the device of the present invention. Detailed Implementation

[0036] The present invention will be further described below with reference to specific embodiments.

[0037] like Figure 1 As shown, the first embodiment of a method for predicting trace water content in transformer oil based on an improved TRANSFORMER includes the following steps:

[0038] Step 1: Acquire the water level measurement signal and temperature measurement signal in the transformer oil, and perform sliding filtering on the water level measurement signal and temperature measurement signal to obtain the filtered water level measurement signal and temperature measurement signal.

[0039] The sliding filter process includes:

[0040] For each sample value acquired, the N consecutively sampled values ​​are placed in an array and sorted according to the first-in-first-out principle. The two largest and two smallest sample values ​​are removed, and the remaining N-4 sample values ​​are subjected to an arithmetic average to obtain the filtered data. Here, N is a fixed number, and the sample values ​​include micro-water measurement signals and temperature measurement signals.

[0041] Step 2: With the water content in the transformer oil remaining constant, acquire trace water measurement signals under different set temperatures. Based on the trace water measurement signals and the set temperatures, use the linear least squares method to fit a linear equation between the trace water measurement signals and the set temperatures, and obtain the slope ξ of the linear equation as a compensation coefficient. Multiply the filtered trace water measurement signal by (1+ξ) to obtain the compensated trace water measurement signal.

[0042] Considering the effect of temperature on micro-water sensors, an increase in temperature will lead to a decrease in the volume fraction of water in the polymer film, which in turn will cause a decrease in the dielectric constant and thus a decrease in capacitance. Therefore, corresponding compensation measures are taken to compensate for the micro-water data based on the changes in temperature data.

[0043] Step 3: Obtain weather information, encode the weather information into weather digital information, use the compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and use the training samples to train the neural network model to obtain a trained neural network model.

[0044] The weather information is encoded into weather digital information, including:

[0045] Text weather information is collected using web crawlers, encoded using OneHotEncoder, and then input into Word2Vector to output numerical weather information.

[0046] The neural network model adopted is a Transformer+SOM+LSTM model. This model includes a Transformer model, where the decoder is replaced by a structure of a SOM neural network followed by an LSTM neural network. Specifically, the Transformer's encoder in the Transformer+SOM+LSTM model is used to extract features from the data, while the SOM neural network followed by the LSTM neural network structure is used for feature clustering and prediction.

[0047] The step of training the neural network model using training samples to obtain a trained neural network model includes:

[0048] Step 3.1: Acquire the compensated micro-water measurement signal, temperature measurement signal, and weather data. Perform Z-Score standardization on the compensated micro-water measurement signal, temperature measurement signal, and weather data to obtain the standardized compensated micro-water measurement signal, temperature measurement signal, and weather data. Z-Score standardization includes: calculating the mean and variance of each data category, subtracting the mean from each data category, and then dividing by the variance to achieve the purpose of standardizing each data category.

[0049] Step 3.2: Use the standardized and compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and divide the training samples into training set, validation set, and test set.

[0050] Step 3.3: Input the training set into the Transformer+SOM+LSTM model to obtain the network parameters of the Transformer+SOM+LSTM model.

[0051] Step 3.4: Input the validation set into the Transformer+SOM+LSTM model to obtain the adjustment values ​​of the network parameters.

[0052] Step 3.5: Substitute the adjusted network parameters into the Transformer+SOM+LSTM model, and input the test set into the Transformer+SOM+LSTM model to obtain the test prediction value. Calculate the MAPE evaluation index based on the test prediction value, and use the Transformer+SOM+LSTM model that meets the MAPE evaluation index requirements as the trained neural network model.

[0053] Step 4: Acquire the compensated micro-water measurement signal in real time and display the compensated micro-water measurement signal in a visual interface for real-time monitoring of the transformer.

[0054] Step 5: Obtain the real-time compensated micro-water measurement signal, the future temperature measurement signal, and the weather digital information. Input the real-time compensated micro-water measurement signal, the future temperature measurement signal, and the weather digital information into the trained neural network model to obtain the future predicted transformer water content.

[0055] Step 6: Calculate the failure probability of the transformer based on the predicted water content of the transformer in the future. When the failure probability is greater than the threshold, issue an alarm message and arrange for maintenance.

[0056] like Figure 2 As shown, the second embodiment of a device for predicting trace water content in transformer oil based on an improved TRANSFORMER includes the following modules:

[0057] Signal preprocessing module: used to acquire the water content measurement signal and temperature measurement signal in transformer oil, and to perform sliding filtering on the water content measurement signal and temperature measurement signal to obtain the filtered water content measurement signal and temperature measurement signal.

[0058] The sliding filter process includes:

[0059] For each sample value acquired, the N consecutively sampled values ​​are placed in an array and sorted according to the first-in-first-out principle. The two largest and two smallest sample values ​​are removed, and the remaining N-4 sample values ​​are subjected to an arithmetic average to obtain the filtered data. Here, N is a fixed number, and the sample values ​​include micro-water measurement signals and temperature measurement signals.

[0060] Compensation for micro-water signal: This module is used to acquire micro-water measurement signals under different set temperatures while maintaining a constant water content in the transformer oil. Based on the micro-water measurement signal and the set temperature, a linear least squares method is used to fit a linear equation between the micro-water measurement signal and the set temperature. The slope ξ of this linear equation is then used as a compensation coefficient. The filtered micro-water measurement signal is multiplied by (1+ξ) to obtain the compensated micro-water measurement signal.

[0061] Considering the effect of temperature on micro-water sensors, an increase in temperature will lead to a decrease in the volume fraction of water in the polymer film, which in turn will cause a decrease in the dielectric constant and thus a decrease in capacitance. Therefore, corresponding compensation measures are taken to compensate for the micro-water data based on the changes in temperature data.

[0062] Neural network model training module: Used to acquire weather information, encode the weather information into weather digital information, use the compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and use the training samples to train the neural network model to obtain a trained neural network model.

[0063] The weather information is encoded into weather digital information, including:

[0064] Text weather information is collected using web crawlers, encoded using OneHotEncoder, and then input into Word2Vector to output numerical weather information.

[0065] The neural network model adopted is a Transformer+SOM+LSTM model. This model includes a Transformer model, where the decoder is replaced by a structure of a SOM neural network followed by an LSTM neural network. Specifically, the Transformer's encoder in the Transformer+SOM+LSTM model is used to extract features from the data, while the SOM neural network followed by the LSTM neural network structure is used for feature clustering and prediction.

[0066] The step of training the neural network model using training samples to obtain a trained neural network model includes:

[0067] Step 3.1: Acquire the compensated micro-water measurement signal, temperature measurement signal, and weather data. Perform Z-Score standardization on the compensated micro-water measurement signal, temperature measurement signal, and weather data to obtain the standardized compensated micro-water measurement signal, temperature measurement signal, and weather data. Z-Score standardization includes: calculating the mean and variance of each data category, subtracting the mean from each data category, and then dividing by the variance to achieve the purpose of standardizing each data category.

[0068] Step 3.2: Use the standardized and compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and divide the training samples into training set, validation set, and test set.

[0069] Step 3.3: Input the training set into the Transformer+SOM+LSTM model to obtain the network parameters of the Transformer+SOM+LSTM model.

[0070] Step 3.4: Input the validation set into the Transformer+SOM+LSTM model to obtain the adjustment values ​​of the network parameters.

[0071] Step 3.5: Substitute the adjusted network parameters into the Transformer+SOM+LSTM model, and input the test set into the Transformer+SOM+LSTM model to obtain the test prediction value. Calculate the MAPE evaluation index based on the test prediction value, and use the Transformer+SOM+LSTM model that meets the MAPE evaluation index requirements as the trained neural network model.

[0072] Real-time monitoring module: Used to acquire the compensated micro-water measurement signal in real time and display the compensated micro-water measurement signal with a visual interface for real-time monitoring of the transformer.

[0073] Transformer moisture content prediction module: This module is used to acquire real-time compensated micro-moisture measurement signals, future temperature measurement signals, and weather data. The real-time compensated micro-moisture measurement signals, future temperature measurement signals, and weather data are then input into a trained neural network model to obtain the predicted transformer moisture content.

[0074] Alarm module: It is used to calculate the failure probability of the transformer based on the predicted water content of the transformer in the future. When the failure probability is greater than the threshold, an alarm message is issued and maintenance is arranged.

[0075] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0076] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0077] 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.

[0078] 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.

[0079] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting trace water content in transformer oil based on an improved Transformer, characterized in that: Includes the following steps: Step 1: Acquire the water level measurement signal and temperature measurement signal in the transformer oil, and perform sliding filtering on the water level measurement signal and temperature measurement signal to obtain the filtered water level measurement signal and temperature measurement signal; Step 2: Under the condition that the water content in the transformer oil remains constant, obtain the micro-water measurement signal under different set temperature conditions. Based on the micro-water measurement signal and the set temperature, use the linear least squares method to fit the linear equation of the micro-water measurement signal and the set temperature, and obtain the slope ξ of the linear equation as the compensation coefficient. Multiply the filtered micro-water measurement signal by (1+ξ) to obtain the compensated micro-water measurement signal. Step 3: Obtain weather information, encode the weather information into weather digital information, use the compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and use the training samples to train the neural network model to obtain a trained neural network model. Step 4: Acquire the compensated micro-water measurement signal in real time and display the compensated micro-water measurement signal in a visual interface for real-time monitoring of the transformer; Step 5: Obtain the real-time compensated micro-water measurement signal, the future temperature measurement signal, and the weather digital information. Input the real-time compensated micro-water measurement signal, the future temperature measurement signal, and the weather digital information into the trained neural network model to obtain the future predicted transformer water content. Step 6: Calculate the failure probability of the transformer based on the predicted water content of the transformer in the future. When the failure probability is greater than the threshold, issue an alarm message and arrange for maintenance. The sliding filter process includes: For each sample value acquired, the N consecutively sampled values ​​are placed in an array and sorted according to the first-in-first-out principle. The two largest and two smallest sample values ​​are removed, and the remaining N-4 sample values ​​are subjected to an arithmetic average to obtain the filtered data. Here, N is a fixed number, and the sample values ​​include micro-water measurement signals and temperature measurement signals. The step of training the neural network model using training samples to obtain a trained neural network model includes: Step 3.1: Acquire the compensated micro-water measurement signal, temperature measurement signal, and weather digital information. Perform Z-Score standardization on the compensated micro-water measurement signal, temperature measurement signal, and weather digital information to obtain the standardized compensated micro-water measurement signal, temperature measurement signal, and weather digital information. Step 3.2: Use the standardized and compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and divide the training samples into training set, validation set, and test set; Step 3.3: Input the training set into the Transformer+SOM+LSTM model to obtain the network parameters of the Transformer+SOM+LSTM model; Step 3.4: Input the validation set into the Transformer+SOM+LSTM model to obtain the adjustment values ​​of the network parameters; Step 3.5: Substitute the adjusted network parameters into the Transformer+SOM+LSTM model, and input the test set into the Transformer+SOM+LSTM model to obtain the test prediction value. Calculate the MAPE evaluation index based on the test prediction value, and use the Transformer+SOM+LSTM model that meets the requirements of the MAPE evaluation index as the trained neural network model. The neural network model adopts the Transformer+SOM+LSTM model. The Transformer+SOM+LSTM model includes a Transformer model, in which the decoder of the Transformer model is replaced by a structure of SOM neural network followed by LSTM neural network. In the Transformer+SOM+LSTM model, the encoder of the Transformer model is used to extract features from the data, and the structure of SOM neural network followed by LSTM neural network is used for feature clustering and prediction.

2. A device for predicting trace water content in transformer oil based on an improved Transformer, characterized in that: Includes the following modules: Signal preprocessing module: used to acquire the water content measurement signal and temperature measurement signal in transformer oil, and to perform sliding filtering on the water content measurement signal and temperature measurement signal to obtain the filtered water content measurement signal and temperature measurement signal; Compensation for micro-water signal: This module is used to acquire micro-water measurement signals under different set temperatures while keeping the water content in transformer oil constant. Based on the micro-water measurement signal and the set temperature, the linear least squares method is used to fit a linear equation between the micro-water measurement signal and the set temperature, and the slope ξ of the linear equation is obtained as the compensation coefficient. The filtered micro-water measurement signal is multiplied by (1+ξ) to obtain the compensated micro-water measurement signal. Neural network model training module: used to acquire weather information, encode the weather information into weather digital information, use the compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and use the training samples to train the neural network model to obtain a trained neural network model; Real-time monitoring module: used to acquire the compensated micro-water measurement signal in real time and display the compensated micro-water measurement signal with a visual interface for real-time monitoring of the transformer; Transformer moisture content prediction module: It is used to acquire real-time compensated micro-moisture measurement signals, future temperature measurement signals and weather digital information, and input the real-time compensated micro-moisture measurement signals, future temperature measurement signals and weather digital information into the trained neural network model to obtain the future predicted transformer moisture content; Alarm module: It is used to calculate the failure probability of the transformer based on the predicted water content of the transformer in the future. When the failure probability is greater than the threshold, an alarm message is issued and maintenance is arranged. The sliding filter process includes: For each sample value acquired, the N consecutively sampled values ​​are placed in an array and sorted according to the first-in-first-out principle. The two largest and two smallest sample values ​​are removed, and the remaining N-4 sample values ​​are subjected to an arithmetic average to obtain the filtered data. Here, N is a fixed number, and the sample values ​​include micro-water measurement signals and temperature measurement signals. The step of training the neural network model using training samples to obtain a trained neural network model includes: Step 3.1: Acquire the compensated micro-water measurement signal, temperature measurement signal, and weather digital information. Perform Z-Score standardization on the compensated micro-water measurement signal, temperature measurement signal, and weather digital information to obtain the standardized compensated micro-water measurement signal, temperature measurement signal, and weather digital information. Step 3.2: Use the standardized and compensated micro-water measurement signal, temperature measurement signal, and weather digital information as training samples, and divide the training samples into training set, validation set, and test set; Step 3.3: Input the training set into the Transformer+SOM+LSTM model to obtain the network parameters of the Transformer+SOM+LSTM model; Step 3.4: Input the validation set into the Transformer+SOM+LSTM model to obtain the adjustment values ​​of the network parameters; Step 3.5: Substitute the adjusted network parameters into the Transformer+SOM+LSTM model, and input the test set into the Transformer+SOM+LSTM model to obtain the test prediction value. Calculate the MAPE evaluation index based on the test prediction value, and use the Transformer+SOM+LSTM model that meets the requirements of the MAPE evaluation index as the trained neural network model. The neural network model adopts the Transformer+SOM+LSTM model. The Transformer+SOM+LSTM model includes a Transformer model, in which the decoder of the Transformer model is replaced by a structure of SOM neural network followed by LSTM neural network. In the Transformer+SOM+LSTM model, the encoder of the Transformer model is used to extract features from the data, and the structure of SOM neural network followed by LSTM neural network is used for feature clustering and prediction.

Images