Distributed power supply configuration detection method and system based on deep learning fusion model

Abstract

The invention relates to a distributed power supply configuration detection method and system based on a deep learning fusion model, and relates to the field of power monitoring technology, and the method comprises the steps: constructing a detection interval with the width being the detection duration on a time axis, and delimiting a detection point in the detection interval according to a preset unit duration, power generation and utilization data are obtained at the detection points; constructing a configuration detection database according to all the power generation and consumption data, and determining a type identification database and a capacity prediction database according to the configuration detection database; training the type identification database according to a preset time convolution network to determine a type identification output matrix, and determining convolution training parameters according to the type identification output matrix for reservation; and training the capacity prediction database according to a preset bidirectional gating cycle unit to determine a capacity prediction output matrix, and determining a training matrix according to the capacity prediction output matrix for retention. The method has a good effect of predicting the access data of the distributed power supply.

Description

Technical Field

[0001] This application relates to the field of power monitoring technology, and in particular to a method and system for detecting distributed power supply configuration based on a deep learning fusion model. Background Technology

[0002] With the construction of new power systems, distributed power sources (such as distributed photovoltaics, self-owned power plants, and electric vehicle charging stations) are widely connected to the user-side power grid. However, the irregular grid connection of a high proportion of distributed power sources leads to a decline in power quality and increases operation and maintenance risks. Traditional intrusive verification methods require operation and maintenance personnel to collect data on-site or install branch sensors, which are costly, have limited coverage, and are difficult to adapt to the needs of large-scale distributed power source management.

[0003] To address the challenges posed by the large-scale integration of distributed power sources, existing measurement and verification technologies are gradually shifting from traditional methods to data-driven approaches. Currently, dedicated measurement devices for different power source types are primarily employed, combined with shallow machine learning algorithms (such as backpropagation neural networks and extreme learning machines) to predict the integrated capacity and operational status of distributed power sources. Compared to traditional, intrusive verification methods that rely on manual on-site data collection or the installation of branch sensors, these technologies offer significant advantages: reduced manpower and hardware investment, lower deployment costs; non-intrusive remote monitoring, expanding coverage; and the ability to rapidly process and preliminarily identify the operational data of various types of distributed power sources.

[0004] However, existing distributed power generation measurement and prediction technologies still have several shortcomings. First, their ability to extract time-series features is weak; traditional machine learning models struggle to effectively capture the long-term dependencies and dynamic changes in distributed power generation and consumption data. Second, their prediction accuracy and generalization ability are insufficient; when faced with mixed power combinations such as photovoltaics, energy storage, and charging piles, the models exhibit poor adaptability and limited accuracy. Third, their computational efficiency is low; single deep learning models have complex structures and large parameter counts, leading to excessively long training times, making it difficult to meet the practical application requirements of real-time state verification and rapid response in the context of high-proportion distributed power generation. Therefore, there is an urgent need to design a method that can effectively predict the access data of distributed power sources. Summary of the Invention

[0005] To better predict the access data of distributed power sources, this application provides a distributed power source configuration detection method and system based on a deep learning fusion model.

[0006] Firstly, this application provides a distributed power configuration detection method based on a deep learning fusion model, employing the following technical solution: A distributed power configuration detection method based on a deep learning fusion model includes: A detection interval with a preset detection duration is constructed on a preset time axis, and detection points are defined in the detection interval according to a preset unit duration, and power generation and consumption data are obtained at the detection points; A configuration detection database is constructed based on all power generation and consumption data, and a type identification database and a capacity prediction database are determined based on the configuration detection database. The type recognition database is trained according to a pre-defined temporal convolutional network to determine the type recognition output matrix, and the convolutional training parameters are determined and retained based on the type recognition output matrix. The capacity prediction database is trained using a pre-defined bidirectional gated loop unit to determine the capacity prediction output matrix, and the training matrix is ​​determined and retained based on the capacity prediction output matrix.

[0007] Optionally, the steps of training the type recognition database according to a preset temporal convolutional network to determine the type recognition output matrix, and determining and preserving the convolutional training parameters based on the type recognition output matrix include: The type recognition database input is used to perform residual connection calculations to determine the convolution process parameters, and the type recognition output matrix is ​​determined based on the convolution process parameters; The original type identification matrix is ​​determined from the configuration detection database, and the comparison network output matrix is ​​determined from the type identification output matrix. The type recognition rate is determined by comparing the original type recognition matrix with the comparison network output matrix. Determine if the type recognition rate is greater than the preset required recognition rate; If the type recognition rate is greater than the required recognition rate, then the convolution process parameters corresponding to the currently determined type recognition output matrix are defined as convolution training parameters and retained. If the type recognition rate is not greater than the required recognition rate, then retrain based on the type recognition database until the convolution training parameters are determined and retained.

[0008] Optionally, the steps of training the capacity prediction database according to a preset bidirectional gated recurrent unit to determine the capacity prediction output matrix, and determining and retaining the training matrix based on the capacity prediction output matrix, include: The type feature matrix is ​​determined from the type identification database, and the capacity input database is constructed by combining the capacity prediction database and the type feature matrix. The capacity input database is input to the preset forward gate control loop unit to output forward gate data, and the capacity input database is input to the preset reverse gate control loop unit to output reverse gate data; The forward gate data and the reverse gate data are fused to determine the output parameters of the loop unit, and the transition matrix is ​​determined based on the analysis of all loop unit output parameters. An attention mechanism transformation is performed on the input database based on the capacity to determine the query matrix, key matrix, and numerical matrix; The weight matrix is ​​determined by calculating the query matrix, key matrix, and numerical matrix, and then the weight matrix and transition matrix are combined to determine the time-weighted feature matrix. The time-weighted feature matrix is ​​compressed in one dimension to determine the capacity prediction output matrix, and the capacity prediction vector is output based on the capacity prediction output moment. The original capacity prediction matrix is ​​determined in the capacity prediction database of the constructed configuration detection database, and the capacity prediction vector and the original capacity prediction matrix are compared to determine the capacity prediction error. Determine whether the capacity prediction error is less than the preset permissible prediction error; If the capacity prediction error is less than the permissible prediction error, then the transition matrix corresponding to the currently determined capacity prediction output matrix is ​​defined as the training matrix and retained. If the capacity prediction error is not less than the permissible prediction error, then retrain based on the current capacity prediction database until a training matrix is ​​determined to be retained.

[0009] Secondly, this application provides a distributed power configuration detection system based on a deep learning fusion model, which adopts the following technical solution: A distributed power configuration detection system based on a deep learning fusion model includes: The acquisition module is used for acquiring information; The processing module, connected to the acquisition module, is used for information storage and processing; The processing module constructs a detection interval with a preset detection duration on a preset time axis, and defines detection points in the detection interval according to a preset unit duration, and enables the acquisition module to acquire power generation and consumption data at the detection points; The processing module constructs a configuration detection database based on all power generation and consumption data, and determines the type identification database and capacity prediction database based on the configuration detection database; The processing module trains the type recognition database according to the preset temporal convolutional network to determine the type recognition output matrix, and determines and retains the convolutional training parameters according to the type recognition output matrix; The processing module trains the capacity prediction database according to the preset bidirectional gated loop unit to determine the capacity prediction output matrix, and determines and retains the training matrix based on the capacity prediction output matrix.

[0010] In summary, this application includes at least one of the following beneficial technical effects: Through a hybrid architecture of temporal convolutional network, bidirectional gated recurrent unit, and attention mechanism, only user-side total electricity meter data is required, eliminating the need for hardware deployment. It can also extract long-term time-series features and make better predictions on the access data of distributed power sources. It can extract long-term time-series features, capture cross-time-period dependencies through TCN-expanded causal convolution, cover all time points throughout the day, and use BiGRU to fuse historical and future power generation and consumption features of distributed power sources to improve capacity prediction accuracy. Attached Figure Description

[0011] Figure 1 This is a flowchart of a distributed power configuration detection method based on a deep learning fusion model.

[0012] Figure 2 This is a flowchart of the modules for a distributed power configuration detection method based on a deep learning fusion model. Detailed Implementation

[0013] To make the purpose, technical solution, and advantages of this application clearer, the following is combined with Figures 1-2 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0014] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0015] This application discloses a distributed power configuration detection method based on a deep learning fusion model, referring to... Figure 1 The method flow of the distributed power configuration detection method based on deep learning fusion model includes the following steps: Step S100: Construct a detection interval with a preset detection duration on a preset time axis, and define detection points in the detection interval according to a preset unit duration, and obtain power generation and consumption data at the detection points; The time axis is a coordinate axis formed by combining various time points. This coordinate axis points from the time points that have already passed to the time points that have not yet been reached. The time points that have already passed are on the left side of the coordinate axis, and the left side of the coordinate axis is defined as the front side of the time axis. The detection duration is the historical duration of the user-side electricity consumption data that the staff needs to collect. In this application, one day is used as an example. By constructing a detection interval, the data within the detection duration can be acquired and analyzed. The unit duration is the interval between two adjacent data points set by the staff. In this application, one minute is used as an example. That is, there will be 1440 time nodes in the detection interval, and these time nodes are the detection points. The power generation and consumption data are the data collected from the smart energy meters at the grid connection of the distributed power source. This includes data such as the power generation and consumption of distributed power sources, type, installed capacity, daily time series, and annual time series. The distributed power source types included are distributed power sources with a single unit installed capacity of less than 6MW.

[0016] Step S200: Construct a configuration detection database based on all power generation and consumption data, and determine the type identification database and capacity prediction database based on the configuration detection database; Configure the detection database, i.e., a database with a time scale of 1440 nodes, using the expression: ; Where D a Let D be the configuration detection database for detection object a, D be the configuration detection database composed of all detection objects, S be the power generation and consumption matrix, K be the type matrix, and C be the installed capacity matrix. Convert the English character labels in the original data to numeric labels. The value is not greater than the number of distributed power source types to be detected. Sort by the date of data collection; For data collection nodes within a day, with a time scale of 1440, ; For the first Day Distributed power at each time point Apparent power output; For the detection object The installed capacity of various types of distributed power sources is included. This method uses linear interpolation to fill in missing data in the original power generation and consumption data caused by communication fiber optic interruptions, signal loss, etc. If there is only one missing data point in the original distributed power generation and consumption data, the average value of the data before and after the missing data point is used to replace the missing part in the original data. If multiple missing data points occur consecutively, the average value of the power generation and consumption data at the same time on the two consecutive days immediately preceding the missing data point is used to replace them.

[0017] To adapt the network architecture to both type identification and capacity prediction tasks, reduce the amount of invalid data occupying server memory, and relieve the computing power pressure on hardware devices, it is necessary to configure the detection database. By combining them appropriately, a distributed power source type identification database and a capacity prediction database were constructed separately; among them in, A database for identifying the type of all objects. For object Type identification database, For capacity prediction database, For object A capacity prediction database.

[0018] Step S300: Train the type recognition database according to the preset temporal convolutional network to determine the type recognition output matrix, and determine and retain the convolutional training parameters according to the type recognition output matrix.

[0019] Temporal convolutional networks can acquire the dependency relationship between distributed power generation and consumption data and long-term series, and effectively avoid the network learning ability degradation problem caused by type recognition. At the same time, temporal convolutional networks require fewer hyperparameters than traditional convolutional neural networks, making the network more efficient. Therefore, this method uses temporal convolutional networks as the detection system for distributed power source type recognition. The specific steps are as follows: Step S301: Perform residual connection calculation on the type recognition database input to determine the convolution process parameters, and determine the type recognition output matrix based on the convolution process parameters.

[0020] Using a type recognition database as input to a temporal convolutional network, dilated causal convolution can be performed after residual connection calculation through residual components in the network. The specific residual connection formula is as follows: In the formula, To express relationships for network runtime functions. Data at a certain moment The output obtained after the residual connection transformation, i.e., the convolution process parameters, specifically, will be obtained through the dilated causal convolution contained in the residual connections. It is about Power generation and consumption data at any time The function is shown in the following formula: In the formula, The dilation factor in dilated causal convolution is a function of minute-level time. For Starting from a moment and covering forward Data at each moment; The exponent with base 2 in causal convolution The coefficient of change, i.e. In this invention Take an integer between 0 and 5, that is , For the kernel size, in this application, when In a short time, it can meet the requirement of entering 1440 time periods of data per day.

[0021] Among them, the parameters of the convolution process After processing by the fully connected layer, the required type recognition output matrix is ​​obtained, as shown below: in, That is, object The type identification output matrix.

[0022] Step S302: Determine the original type recognition matrix in the configuration detection database, and determine the comparison network output matrix in the type recognition output matrix.

[0023] The original type identification matrix is Compare the network output matrices. .

[0024] Step S303: Compare the original type recognition matrix with the comparison network output matrix to determine the type recognition rate.

[0025] Type recognition rate is a parameter that reflects the accuracy of class identification of distributed power sources using the current database. It is the number of identical parameter values ​​in both databases divided by the total number of values.

[0026] Step S304: Determine whether the type recognition rate is greater than the preset required recognition rate.

[0027] The requirement recognition rate is the minimum type recognition rate that the trained database set by the staff needs to achieve. The purpose of the judgment is to determine whether the current output database meets the type recognition requirements.

[0028] Step S3041: If the type recognition rate is greater than the required recognition rate, then define the convolution process parameters corresponding to the currently determined type recognition output matrix as convolution training parameters and retain them.

[0029] When the type recognition rate is greater than the required recognition rate, it means that the current database meets the type recognition requirements. In other words, it means that the currently determined convolution process parameters can be well applied to the type recognition of distributed power sources. Therefore, these parameters are defined as convolution training parameters and reserved so that other users can directly use them to achieve distributed power source type recognition.

[0030] Step S3042: If the type recognition rate is not greater than the required recognition rate, then retrain according to the type recognition database until the convolution training parameters are determined and retained.

[0031] When the type recognition rate is not greater than the required recognition rate, it means that the current database cannot meet the type recognition requirements. Therefore, retraining is necessary until a type recognition output matrix that meets the requirements is obtained to complete the type recognition of the distributed power source.

[0032] Step S400: Train the capacity prediction database according to the preset bidirectional gated loop unit to determine the capacity prediction output matrix, and retain the training matrix according to the capacity prediction output matrix.

[0033] The bidirectional gated cyclic unit is an improved network structure based on the classic gated cyclic unit. It can simultaneously take into account the impact of past and future power generation and consumption data on the current data. Therefore, it is necessary to build a bidirectional gated cyclic unit architecture suitable for capacity prediction. The specific steps are as follows: Step S401: Determine the type feature matrix in the type identification database, and construct the capacity input database by combining the capacity prediction database and the type feature matrix.

[0034] The type feature matrix is ​​as described above. The capacity input database is the database input to the bidirectional gated loop unit, and the specific formula is as follows: in, That is, object The capacity is input into the database.

[0035] Step S402: Input the capacity input database to the preset forward gate control loop unit to output forward gate data, and input the capacity input database to the preset reverse gate control loop unit to output reverse gate data.

[0036] Using the capacity prediction database as input to the bidirectional gated recurrent unit, the capacity prediction database first undergoes function calculation by the forward gated recurrent unit, as shown in the following formula: In the formula, Transformation relationship of gated loop unit; and These represent the input data at the previous time step and the output data at the next time step when using the forward propagation algorithm, respectively. The forward gate data can be determined through this function.

[0037] Then, the capacity input database will undergo function calculations by the reverse-gated loop unit, as shown in the following formula: In the formula, and These represent the input data at the previous time step and the output data at the next time step when used in the backpropagation algorithm, respectively. The reverse gate data can be determined through this function.

[0038] Step S403: Fuse the forward gate data and the reverse gate data to determine the output parameters of the loop unit, and analyze all the output parameters of the loop unit to determine the transition matrix.

[0039] Finally, by fusing the forward gate data and the reverse gate data, the output parameters of the loop unit can be obtained, as follows: In the formula, That is, the output parameters of the loop unit. These are the weight values ​​for the forward propagation results during capacity prediction; These are the weight values ​​for the backpropagation results during capacity prediction; This is the correction value for the bidirectional gated recurrent unit, its function being to make the algorithm continuously approach the true value during training. The output parameters of each recurrent unit are... Combine and integrate the installed capacity matrix The transition matrix is ​​formed, and its specific expression is as follows: In the formula, This is the transition matrix.

[0040] Step S404: Perform attention mechanism transformation based on the capacity input database to determine the query matrix, key matrix, and numerical matrix.

[0041] Because the data to be processed in capacity prediction tasks is extremely complex, this invention introduces an attention mechanism for auxiliary computation to further accelerate the task process. To make the capacity input database conform to the network architecture of the attention mechanism algorithm, the data needs to be transformed, as shown in the following formula. In the formula, The input database is converted into a query matrix adapted for the attention mechanism. The input database is converted into a key matrix adapted for the attention mechanism. The input database is converted into a numerical matrix adapted for the attention mechanism. , , These are the transformation matrices to be learned, which are used to convert the capacity input into the database into a query matrix, a key matrix, and a numerical matrix, respectively.

[0042] Step S405: Calculate the weight value matrix based on the query matrix, key matrix and numerical matrix, and combine the weight value matrix and transition matrix to determine the time-weighted feature matrix.

[0043] In the formula, This is the weight matrix for the capacity prediction attention mechanism; For the mathematical transformation function of the attention mechanism; For the first Line number Attention weight values ​​at each time point, where ; The weight matrix obtained from the attention mechanism With transition matrix By combining and assigning values, we can obtain the combined time-weighted feature matrix, expressed as follows: .

[0044] Step S406: Perform one-dimensional compression processing on the time-weighted feature matrix to determine the capacity prediction output matrix, and output the capacity prediction vector based on the capacity prediction output matrix.

[0045] The two-dimensional time-weighted feature matrix is ​​flattened through a flattening layer in a bidirectional gated recurrent unit network. Compressed to a one-dimensional output vector It also utilizes the fully connected layer to output the capacity prediction vector of the distributed power source. The formula for the one-dimensional output vector is as follows: .

[0046] Step S407: Determine the original capacity prediction matrix in the capacity prediction database of the constructed configuration detection database, and compare the capacity prediction vector with the original capacity prediction matrix to determine the capacity prediction error.

[0047] The original capacity prediction matrix is The capacity prediction error is the difference between the actual installed capacity and the predicted value.

[0048] Step S408: Determine whether the capacity prediction error is less than the preset permissible prediction error.

[0049] The permissible prediction error is the maximum capacity prediction error allowed when the staff considers the current capacity prediction database to be performing well. The purpose of this assessment is to determine whether the current capacity prediction database meets the usage requirements.

[0050] Step S4081: If the capacity prediction error is less than the permissible prediction error, then define the transition matrix corresponding to the currently determined capacity prediction output matrix as the training matrix and retain it.

[0051] When the capacity prediction error is less than the permissible prediction error, it means that the current capacity prediction output matrix meets the usage requirements. This means that the currently determined transition matrix can be well applied to the capacity prediction of distributed power sources. Therefore, it is defined as a training matrix and reserved so that other users can directly use this parameter to achieve capacity prediction of distributed power sources.

[0052] Step S4082: If the capacity prediction error is not less than the permissible prediction error, then retrain based on the current capacity prediction database until a training matrix is ​​determined to be retained.

[0053] When the capacity prediction error is not less than the permissible prediction error, it indicates that the current capacity prediction output matrix does not meet the prediction accuracy requirements. Therefore, retraining is performed until the capacity prediction output matrix meets the requirements.

[0054] Reference Figure 2 Based on the same inventive concept, embodiments of the present invention provide a distributed power configuration detection system based on a deep learning fusion model, comprising: The acquisition module is used for acquiring information; The processing module, connected to the acquisition module, is used for information storage and processing; The processing module constructs a detection interval with a preset detection duration on a preset time axis, and defines detection points in the detection interval according to a preset unit duration, and enables the acquisition module to acquire power generation and consumption data at the detection points; The processing module constructs a configuration detection database based on all power generation and consumption data, and determines the type identification database and capacity prediction database based on the configuration detection database; The processing module trains the type recognition database according to the preset temporal convolutional network to determine the type recognition output matrix, and determines and retains the convolutional training parameters according to the type recognition output matrix; The processing module trains the capacity prediction database according to the preset bidirectional gated loop unit to determine the capacity prediction output matrix, and determines and retains the training matrix according to the capacity prediction output matrix. The type recognition module processes and analyzes the type recognition database through a temporal convolutional network to determine the type recognition output matrix for identifying distributed power source types. The capacity prediction module processes and analyzes the type identification database through a bidirectional gated loop unit to determine the capacity prediction output matrix for distributed power supply capacity.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

Claims

1. A distributed power configuration detection method based on a deep learning fusion model, characterized in that, include: A detection interval with a preset detection duration is constructed on a preset time axis, and detection points are defined in the detection interval according to a preset unit duration, and power generation and consumption data are obtained at the detection points; A configuration detection database is constructed based on all power generation and consumption data, and a type identification database and a capacity prediction database are determined based on the configuration detection database. The type recognition database is trained according to a pre-defined temporal convolutional network to determine the type recognition output matrix, and the convolutional training parameters are determined and retained based on the type recognition output matrix. The capacity prediction database is trained using a pre-defined bidirectional gated loop unit to determine the capacity prediction output matrix, and the training matrix is ​​determined and retained based on the capacity prediction output matrix.

2. The distributed power configuration detection method based on a deep learning fusion model according to claim 1, characterized in that, The steps of training a type recognition database using a pre-defined temporal convolutional network to determine the type recognition output matrix, and then determining and preserving the convolutional training parameters based on the type recognition output matrix, include: The type recognition database input is used to perform residual connection calculations to determine the convolution process parameters, and the type recognition output matrix is ​​determined based on the convolution process parameters; The original type identification matrix is ​​determined from the configuration detection database, and the comparison network output matrix is ​​determined from the type identification output matrix. The type recognition rate is determined by comparing the original type recognition matrix with the comparison network output matrix. Determine if the type recognition rate is greater than the preset required recognition rate; If the type recognition rate is greater than the required recognition rate, then the convolution process parameters corresponding to the currently determined type recognition output matrix are defined as convolution training parameters and retained. If the type recognition rate is not greater than the required recognition rate, then retrain based on the type recognition database until the convolution training parameters are determined and retained.

3. The distributed power configuration detection method based on a deep learning fusion model according to claim 2, characterized in that, The steps of training the capacity prediction database using a pre-defined bidirectional gated recurrent unit to determine the capacity prediction output matrix, and then determining and retaining the training matrix based on the capacity prediction output matrix, include: The type feature matrix is ​​determined from the type identification database, and the capacity input database is constructed by combining the capacity prediction database and the type feature matrix. The capacity input database is input to the preset forward gate control loop unit to output forward gate data, and the capacity input database is input to the preset reverse gate control loop unit to output reverse gate data; The forward gate data and the reverse gate data are fused to determine the output parameters of the loop unit, and the transition matrix is ​​determined based on the analysis of all loop unit output parameters. An attention mechanism transformation is performed on the input database based on the capacity to determine the query matrix, key matrix, and numerical matrix; The weight matrix is ​​determined by calculating the query matrix, key matrix, and numerical matrix, and then the weight matrix and transition matrix are combined to determine the time-weighted feature matrix. The time-weighted feature matrix is ​​compressed in one dimension to determine the capacity prediction output matrix, and the capacity prediction vector is output based on the capacity prediction output moment. The original capacity prediction matrix is ​​determined in the capacity prediction database of the constructed configuration detection database, and the capacity prediction vector and the original capacity prediction matrix are compared to determine the capacity prediction error. Determine whether the capacity prediction error is less than the preset permissible prediction error; If the capacity prediction error is less than the allowable prediction error, then the transition matrix corresponding to the currently determined capacity prediction output matrix is ​​defined as the training matrix and retained. If the capacity prediction error is not less than the permissible prediction error, then retrain based on the current capacity prediction database until a training matrix is ​​determined to be retained.

4. A distributed power configuration detection system based on a deep learning fusion model, characterized in that, include: The acquisition module is used for acquiring information; The processing module, connected to the acquisition module, is used for information storage and processing; The processing module constructs a detection interval with a preset detection duration on a preset time axis, and defines detection points in the detection interval according to a preset unit duration, and enables the acquisition module to acquire power generation and consumption data at the detection points; The processing module constructs a configuration detection database based on all power generation and consumption data, and determines the type identification database and capacity prediction database based on the configuration detection database; The processing module trains the type recognition database according to the preset temporal convolutional network to determine the type recognition output matrix, and determines and retains the convolutional training parameters according to the type recognition output matrix; The processing module trains the capacity prediction database according to the preset bidirectional gated loop unit to determine the capacity prediction output matrix, and determines and retains the training matrix based on the capacity prediction output matrix.

Images