Method for determining mapping relation of input-output data of new energy transmission project

Abstract

The invention provides a new energy transmission project input-output data mapping relation determination method, and belongs to the technical field of new energy transmission projects, and the method comprises the steps: carrying out the standardization processing of historical input data (construction, equipment and operation and maintenance) and output data (output electric energy and electric energy quality); then constructing an equipment parameter matrix, and obtaining an equipment feature vector through fast Fourier transform; a principal component analysis method is adopted to extract main characteristic components, and core equipment characteristic vectors after dimension reduction are constructed. Establishing an interaction influence matrix, determining a weight coefficient of each equipment parameter, constructing an equipment parameter comprehensive evaluation index based on the weight coefficient, and establishing an integrated mapping relation with input-output data; and finally, constructing a multi-input multi-output deep neural network, taking the core equipment feature vector as input, taking input-output data as output, and performing network training based on the integrated mapping relation to obtain a final mapping relation capable of fully considering the mutual relation between different data.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy transmission projects, and in particular, relates to a method for determining a mapping relationship between input and output data of a new energy transmission project. Background Art

[0002] As a key hub connecting new energy power generation bases and power grid load centers, the input-output efficiency of new energy transmission projects directly affects the economy and reliability of the entire new energy project. At present, the construction and operation of new energy transmission projects are mainly based on traditional engineering experience and standard specifications, lacking a scientific input-output data mapping relationship analysis method.

[0003] In actual engineering applications, new energy transmission projects involve various types of equipment, including transformers, switchgear, transmission lines, reactive power compensation devices, relay protection devices and secondary equipment. The parameter selection and investment decisions of these equipment often adopt a decentralized management model, with independent decisions made in various professional fields, lacking systematic collaborative optimization. At the same time, due to the intermittent and volatile characteristics of new energy power generation, the operating environment of the transmission project is more complex, and the relationship between equipment parameters and system performance is difficult to accurately grasp.

[0004] In the existing technology, input-output mapping is mainly constructed based on the relationship between various costs and power generation, but the relationship between different data is not fully considered, resulting in the mapping relationship that cannot provide a scientific basis for investment decisions and low resource allocation efficiency.

[0005] This problem makes it difficult to accurately evaluate the investment benefits of new energy transmission projects, the allocation of operation and maintenance resources is not reasonable, and it is difficult to balance system reliability and economy. Therefore, it is urgent to establish a scientific input-output data mapping relationship determination method to guide the planning, design, operation and maintenance of new energy transmission projects. Summary of the invention

[0006] In view of this, the present invention provides a method for determining the mapping relationship of input-output data of a new energy transmission project, which can solve the technical problem that the prior art mainly constructs input-output mapping based on relationships between multiple costs and power generation, but fails to fully consider the relationship between different data.

[0007] The present invention is achieved in that: The present invention provides a method for determining a mapping relationship of input-output data of a new energy transmission project, comprising the following steps: S10, obtaining historical input data and output data of the new energy transmission project, and performing standardization on the historical input data and output data, wherein the historical input data includes construction input data, equipment input data, and operation and maintenance input data, and the output data includes output power data and output power quality data; S20, constructing a device parameter matrix for key parameters of all types of devices, where each row represents a type of device and each column represents a key parameter, performing a fast Fourier transform on the device parameter matrix to obtain a complex feature matrix of device parameters, and converting the complex feature matrix into a real domain feature vector to obtain a device feature vector; S30, using a principal component analysis method to perform dimensionality reduction processing on the device feature vector, extracting the main feature components, and constructing a core device feature vector after dimensionality reduction; S40, establishing an interaction matrix between components within the core device feature vector, calculating the eigenvalue of the interaction matrix, and determining the weight coefficient of each device key parameter; S50, constructing a comprehensive evaluation index of equipment parameters according to the weight coefficient, and establishing an integrated mapping relationship between the comprehensive evaluation index of equipment parameters and the construction investment data, the operation and maintenance investment data, the output power data, and the output power quality data; S60, constructing a multi-input and multi-output deep neural network structure, taking the core equipment feature vector as the input layer, taking the construction investment data, the operation and maintenance investment data, the output power data and the output power quality data as the output layer, and using the integrated mapping relationship to perform network training; S70. Use a verification data set in actual engineering data to verify the prediction accuracy of the deep neural network, adjust the parameters of the deep neural network according to the prediction accuracy, until the prediction accuracy reaches a preset accuracy requirement, and output a unified input-output mapping model.

[0008] Specifically, step S10 specifically includes the following steps: acquiring historical operation data of the new energy transmission project from a professional power data acquisition system, the historical operation data including construction investment data, equipment investment data, operation and maintenance investment data and output data; the construction investment data extracts infrastructure investment, design and survey investment, environmental impact assessment and water conservation investment, and construction investment from the project management system; the equipment investment data extracts various equipment parameter information and cost information from the equipment management system; the equipment parameter information includes transformers, switchgear, transmission lines, reactive power compensation, relay protection and key technical parameters of secondary equipment; the equipment cost information includes procurement cost and installation and commissioning cost; the operation and maintenance investment data extracts daily maintenance, fault repair, technical transformation and emergency disposal cost data from the operation and maintenance management system; the output data extracts output power data from the power dispatching automation system, the output power data includes annual, monthly and daily power generation data; voltage deviation, frequency deviation, three-phase imbalance and harmonic content data are extracted from the power quality online monitoring system; the maximum and minimum value normalization method is used to map the historical operation data to the range of 0 to 1.

[0009] The step S20 specifically includes the following steps: arranging the key parameters of the equipment according to the equipment type and parameter type to construct an equipment parameter matrix, wherein each row of the equipment parameter matrix represents a type of equipment and each column represents a key parameter, performing a fast Fourier transform on each row sequence of the equipment parameter matrix, converting the time domain information into the frequency domain space, using a butterfly operation structure to reduce the computational complexity, extracting amplitude spectrum and phase spectrum information from the frequency domain signal, wherein the amplitude spectrum reflects the intensity characteristics of parameter changes, and the phase spectrum reflects the phase relationship of parameter changes, and combining the amplitude spectrum and phase spectrum information to form a device feature vector.

[0010] The step S30 specifically includes the following steps: calculating the covariance matrix of the device eigenvector, the covariance matrix reflects the correlation between each eigencomponent, performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors, sorting the eigenvectors according to the size of the eigenvalues, selecting the eigenvectors corresponding to the first k eigenvalues ​​to form a dimensionality reduction transformation matrix, the value of k ensures that the cumulative contribution rate reaches 85%, and projecting the device eigenvector onto the dimensionality reduction transformation matrix to obtain the core device eigenvector.

[0011] The step S40 specifically includes the following steps: constructing an interaction influence matrix based on the core device feature vector, each element of the interaction influence matrix represents the degree of correlation between different feature components, using the Pearson correlation coefficient to calculate the correlation between the feature components, the Pearson correlation coefficient ranges from negative 1 to 1, calculating the eigenvalues ​​and eigenvectors of the interaction influence matrix, and obtaining the feature component weight coefficients by normalizing the eigenvalues, and the sum of the weight coefficients is 1.

[0012] The step S50 specifically includes the following steps: performing weighted summation on the components of the core equipment characteristic vector according to the weight coefficient to obtain a comprehensive evaluation index of equipment parameters, establishing a mapping relationship between the comprehensive evaluation index of equipment parameters and construction investment data, operation and maintenance investment data, output power data and output power quality data, introducing a normally distributed random error term with a mean of 0 into the mapping relationship, and using the least squares method to estimate the coefficients in the mapping relationship.

[0013] The step S60 specifically includes the following steps: constructing a multi-layer perceptron network structure, including an input layer, multiple hidden layers and an output layer, the number of nodes in the input layer is equal to the dimension of the core equipment feature vector, the number of nodes in the output layer is equal to the total dimension of construction investment data, operation and maintenance investment data, output power data and output power quality data, the number of nodes in the hidden layer is the average of the number of nodes in the input layer and the output layer, the hidden layer uses a ReLU activation function, the output layer uses a linear activation function for continuous data and a Sigmoid activation function for probabilistic data, the network is trained using a back-propagation algorithm, the network parameters are updated using an Adam optimizer, batch normalization technology is used to accelerate network convergence, and a dropout rate of 0.5 is used to prevent network overfitting.

[0014] The step S70 specifically includes the following steps: randomly selecting 20% ​​of the data from the actual engineering data as a validation set, inputting the validation set data into the trained network to obtain a prediction result, calculating the root mean square error between the prediction result and the actual value, using a grid search method to adjust the number of hidden layers and the number of nodes to determine the optimal network structure, using a Bayesian optimization method to search for the optimal learning rate, batch size, and dropout rate hyperparameter combination, repeatedly adjusting and verifying until the root mean square error is less than 0.1, and outputting a unified input-output mapping model.

[0015] Furthermore, the device parameters include at least: Transformer key parameters: rated capacity, impedance percentage, no-load loss rate, load loss rate; Key parameters of switchgear: rated voltage, rated current, breaking capacity; Key parameters of transmission lines: line length, conductor cross section, tower load; Key parameters of reactive power compensation: compensation capacity, response time, and adjustment range; Key parameters of relay protection: action time, return coefficient, reliability coefficient; Key parameters of secondary equipment: measurement accuracy, sampling frequency, and communication bandwidth.

[0016] Furthermore, the device parameter matrix construction and feature extraction process in S20 are specifically expressed as follows: 1) Construction of original matrix of device parameters: ; In the formula, For the The first The normalized values ​​of the key parameters are , ; is the total number of device types; is the total number of key parameters for each device.

[0017] 2) Fast Fourier Transform Process: ; In the formula, is the complex domain representation after Fourier transform; is the frequency component number; is an imaginary unit; Time series index.

[0018] 3) The complex feature matrix is ​​converted to the real number domain: ; ; In the formula, is the amplitude spectrum; is the phase spectrum; and are the real and imaginary parts of the complex feature, respectively.

[0019] 4) Device feature vector construction: ; In the formula, For the The feature vector of the device.

[0020] Furthermore, the principal component analysis dimensionality reduction process in S30 is as follows: 1) Calculate the covariance matrix: ; In the formula, is the mean of the eigenvector; is the sample size.

[0021] 2) Eigenvalue decomposition: ; In the formula, is the characteristic value; is the corresponding feature vector.

[0022] 3) Dimensionality reduction projection: ; In the formula, For the front The matrix composed of the eigenvectors corresponding to the principal components; is the core feature vector after dimensionality reduction; The value is determined by the cumulative contribution rate and is usually required to be greater than 85%.

[0023] Cumulative contribution rate calculation: , in, for No. eigenvalues, is the cumulative contribution rate.

[0024] Furthermore, the construction process of the interaction influence matrix in S40 is as follows: 1) Construct the interaction influence matrix: ; In the formula, For the The characteristic component and The correlation coefficient of the feature components is calculated by Pearson correlation coefficient: .

[0025] 2) Calculate the weight coefficient: ; In the formula, The interaction matrix No. eigenvalues; For the The weight coefficient of the feature component.

[0026] Furthermore, the construction process of the integrated mapping relationship in S50 is as follows: 1) Calculation of comprehensive evaluation index of equipment parameters: ; In the formula, It is a comprehensive evaluation indicator; is a random error term, which follows a normal distribution .

[0027] 2) Mapping relationship construction: ; In the formula, Input vector for construction; Provide vectors for operation and maintenance; is the output electric energy vector; is the power quality vector; are unknown coefficients, obtained by least squares fitting.

[0028] Least squares coefficient estimates: , in, is the sample size of historical data, is the actual observed value, is the comprehensive evaluation index value at the corresponding moment.

[0029] The above parameters are obtained as follows: 1) The original parameters are obtained through the equipment nameplate, test report, and operation record; 2) Normalization processing uses the maximum and minimum method: ; 3) The coefficients to be determined are obtained through historical data training: ; In the formula, is the sample size of historical data; is the actual observed value; is the comprehensive evaluation index value at the corresponding moment.

[0030] Furthermore, the multi-input and multi-output deep neural network adopts an encoder-decoder architecture, the input layer receives the core equipment feature vector after dimensionality reduction processing, the encoder is composed of 3 layers of bidirectional LSTM, the hidden layer dimension of each LSTM layer is 256, and a Dropout layer is added between the LSTM layers to prevent overfitting, and an 8-head self-attention mechanism is set after the encoder output to learn the correlation between different features; the decoder adopts a multi-branch structure, including four independent decoding branches of construction investment, operation and maintenance investment, output power and power quality, and each branch independently learns the corresponding mapping relationship.

[0031] The four decoder branches have the same hierarchical structure, but the parameters are trained independently. Each branch contains a three-layer network structure: the first layer is a 512-node fully connected layer, followed by a ReLU activation function and dropout (0.2); the second layer is a 256-node fully connected layer, also followed by a ReLU activation function and dropout (0.2); the last layer is the output layer, which uses a single-node fully connected layer to produce the final prediction value. Inside the branch, the output of the encoder is directly connected to each decoding stage through a residual connection, and Layer Normalization is used after each layer for feature normalization. A learnable scaling factor is set at the output of each branch to adaptively adjust the output amplitude.

[0032] The Adam optimizer is used for network training, the initial learning rate is set to 0.001, and the learning rate decay strategy is used. The loss function is designed as the weighted sum of the mean square error losses of the four branches, and the weights are determined according to the importance of different output indicators. During the training process, the Warmup strategy is used to gradually increase the learning rate, and mixed precision training is used to accelerate the calculation. To prevent overfitting, the early stopping strategy is adopted, and the training is stopped when the validation set loss does not decrease for multiple consecutive rounds. At the same time, gradient clipping is used to prevent gradient explosion and limit the gradient norm to not exceed the preset threshold. To improve the prediction stability of the model, the model integration technology is used to integrate the prediction results of the model trained multiple times. In terms of regularization, in addition to using Dropout, the L2 regularization term is also introduced to control the complexity of the model and improve the generalization ability of the model. During the training process, the model performance is regularly evaluated on the validation set, and the best model parameters are saved.

[0033] Furthermore, the step of calculating the power quality data includes: Step 1: Collect real-time operation data of the power system of the new energy transmission project, including: 1) Three-phase voltage effective value , , ,The sampling period is 10ms, and the sampling duration is not less than 24h; 2) System frequency ,The sampling period is 100ms, and the sampling duration is no less than 24h; 3) Effective value of each harmonic voltage (h=2,3,…,25), the sampling period is 3s, and the sampling time is not less than 24h; 4) Phase angle of each phase voltage , , ,The sampling period is 10ms, and the sampling duration is no less than 24h.

[0034] Step 2: Construct matrix groups separately, including: The power quality equation matrix group when new energy is not connected: The first equation matrix (voltage stability equation matrix 1): ; In the formula, and Respectively The voltage at each sampling point Axis and Axis component (pu); is the voltage deviation (pu); is the voltage phase angle (rad); is the error term; is the number of sampling points.

[0035] The second equation matrix (frequency stability equation matrix 1): ; In the formula, For the The measured frequency of each sampling point (Hz); is the frequency deviation (Hz); is the frequency change rate (Hz / s); is the rated frequency (Hz); is the active power deviation (pu); is the frequency adjustment coefficient.

[0036] The third equation matrix (harmonic equation matrix 1): ; In the formula, For the The sampling point Subharmonic voltage (pu); is the total harmonic distortion rate (%); is the harmonic content; and is the harmonic characteristic parameter.

[0037] The fourth equation matrix (three-phase balance equation matrix 1): ; In the formula, , , Respectively Three-phase voltage of sampling points (pu); is the rotation factor; is the degree of imbalance; is the phase sequence parameter.

[0038] The power quality equation matrix group after the integration of new energy: The fifth equation matrix (voltage stability equation matrix 2): ; The meaning of each parameter is the same as that of the matrix ,Adding apostrophes indicates the parameter values ​​after the new energy is connected.

[0039] The sixth equation matrix (frequency stability equation matrix 2): ; The meaning of each parameter is the same as that of the matrix ,Adding apostrophes indicates the parameter values ​​after the new energy is connected.

[0040] Seventh equation matrix (harmonic equation matrix 2): ; The meaning of each parameter is the same as that of the matrix ,Adding apostrophes indicates the parameter values ​​after the new energy is connected.

[0041] The eighth equation matrix (three-phase balance equation matrix 2): ; The meaning of each parameter is the same as that of the matrix ,Adding apostrophes indicates the parameter values ​​after the new energy is connected.

[0042] Step 3: Perform singular value decomposition (SVD) on the eight equation matrices to obtain the stable equation matrix and the variable equation matrix: For Equation Matrix , its SVD decomposition is: ; In the formula, is a left singular matrix; is the singular value matrix; is the transpose of the right singular matrix; For the The matrix singular values; For the matrix rank.

[0043] Stability equation matrix and the transformation equation matrix Calculation: ; ; In the formula, is the number of main singular values, take .

[0044] Step 4: Based on the change equation matrix of the eight matrices, the improved BP neural network is used to calculate the power quality vector: 1) For the eight variable equation matrices Perform feature extraction and obtain feature vector: ; In the formula, For the The matrix of the equations of change The characteristic value, is the number of features selected.

[0045] 2) Construct the input vector of BP neural network: .

[0046] 3) The output layer of the BP neural network uses the Softmax function: ; In the formula, For the The input value of the output node; For the The output value of the output node.

[0047] 4) Finally, the power quality vector is obtained, which is [voltage deviation, frequency deviation, three-phase unbalance, harmonic content]: ; In the formula, is the voltage deviation (%), the normal range is ±7%; is the frequency deviation (Hz), the normal range is ±0.2Hz; Is the three-phase imbalance (%), the normal range is ≤2%; is the total harmonic distortion rate (%), and the normal range is ≤5%.

[0048] Furthermore, an improved BP neural network is used to calculate the power quality vector based on the change equation matrix.

[0049] The improved BP neural network adopts a five-layer neural network structure, including an input layer, three hidden layers and an output layer. The number of input layer nodes is 8q, corresponding to the q eigenvalues ​​extracted from each of the 8 variable equation matrices. The first hidden layer is set with 6q nodes, the second hidden layer is set with 4q nodes, the third hidden layer is set with 2q nodes, and the output layer is set with 4 nodes corresponding to the four indicators of power quality. The hidden layer adopts the ReLU activation function to enhance the nonlinear expression ability of the network, and the output layer adopts the Softmax function for probability normalization. To prevent overfitting, a Dropout layer is added after each hidden layer, and the dropout rate is set to 0.3. At the same time, a batch normalization layer is introduced to accelerate the training process and improve the stability of the model.

[0050] The training data set of the improved BP neural network: The training data set consists of two parts. The first part comes from the actual power system operation data, and selects historical operation data under different time periods and different load conditions, including complete sampling data before and after the new energy is connected to the grid. Each set of data contains the sampling values ​​of three-phase voltage, system frequency, harmonic voltage and phase angle, and marks the corresponding true value of the power quality index. The second part is generated by professional simulation software (such as PSCAD, PowerFactory, etc.) to simulate data under various working conditions, especially some extreme situations that rarely occur in actual systems but may occur. The simulation data should cover various situations with new energy penetration rates from 0 to 100% to enhance the generalization ability of the model. The total data set size is not less than 10,000 groups, of which 80% are used for training, 10% are used for verification, and 10% are used for testing. All input data are standardized so that their mean is 0 and their standard deviation is 1.

[0051] Training steps: 1. Data preprocessing: Clean the original sampled data to remove outliers and missing values. Then, construct eight matrices according to the above method and perform SVD decomposition to extract eigenvalues ​​to form training samples. Normalize the label data (four power quality indicators).

[0052] 2. Parameter initialization: Use the He initialization method to initialize the network weights so that the variance of each layer output is roughly equal. The bias term is initialized to a small positive number (such as 0.01) to break the symmetry.

[0053] 3. Improved training algorithm: Adopt Adam optimizer with momentum term, set the initial learning rate to 0.001, and use cosine annealing strategy to dynamically adjust the learning rate. The loss function uses weighted mean square error function, assigning different weights to the four output indicators to reflect their different importance in practical applications.

[0054] 4. Training process: The mini-batch gradient descent method is used, and the batch size is set to 64. The early stopping strategy is introduced during training, and the training is stopped when the loss on the validation set has not improved for 10 consecutive epochs. At the same time, the learning rate decay strategy is used, and the learning rate is reduced to 0.1 times the original every 50 epochs.

[0055] 5. Model evaluation and optimization: Use k-fold cross validation (k=5) to evaluate model performance. Dynamically adjust network structure and hyperparameters based on validation results, including the number of nodes, number of layers, dropout rate, etc. At the same time, introduce L2 regularization to prevent overfitting, and the regularization coefficient is determined by grid search.

[0056] 6. Test verification: Use a test set that was not used in training to evaluate the performance of the final model. Quantify the model accuracy by calculating indicators such as mean square error and mean absolute error, and pay special attention to the performance of the model under extreme conditions.

[0057] Furthermore, the power quality vector is specifically [voltage deviation, frequency deviation, three-phase unbalance, harmonic content].

[0058] Compared with the prior art, the method for determining the mapping relationship of input-output data of a new energy transmission project provided by the present invention has the following beneficial effects: 1) A unified data processing framework is established, and different types of data are uniformly mapped to the same metric space through standardized processing, laying the foundation for subsequent analysis and modeling. Compared with the decentralized management mode in the prior art, the present invention realizes the system integration and unified processing of data, and improves the efficiency of data utilization.

[0059] 2) The innovative method of combining fast Fourier transform and principal component analysis is used to effectively extract the characteristics of equipment parameters. By converting equipment parameters from the time domain to the frequency domain, the dynamic characteristics of parameter changes are better captured; the principal component analysis is used to achieve feature dimensionality reduction, which retains the main information while reducing the computational complexity.

[0060] 3) Use equipment parameters as the central link of mapping. Equipment parameters have strong correlation and interactivity. For example, the impedance parameters of the transformer will affect the short-circuit current level of the system, which will affect the selection of the switchgear; the line parameters will affect the reactive balance of the system, which will affect the capacity configuration of the compensation device. Therefore, establishing a mapping relationship with equipment parameters as the center can effectively reflect the interaction between the various elements of the system and form a mapping relationship chain of input-equipment-electricity.

[0061] 4) A method for determining the weights of equipment parameters based on the interaction matrix was established to scientifically quantify the degree of influence of different parameters on system performance. By calculating the correlation coefficient and eigenvalue decomposition between parameters, the coupling relationship between parameters was accurately reflected, providing a theoretical basis for equipment selection and parameter optimization.

[0062] 5) An innovative power quality assessment method is proposed, and an improved neural network model is used to achieve a comprehensive assessment of indicators such as voltage, frequency, three-phase balance and harmonic content. Compared with traditional methods, this invention can more accurately reflect the impact of new energy access on system stability.

[0063] 6) A deep neural network model was constructed to realize the nonlinear mapping relationship between input and output data. Through the multi-layer network structure and appropriate activation function, the complex relationship between data can be effectively captured, and the prediction accuracy is significantly better than the traditional linear regression method.

[0064] In summary, the present invention solves the technical problem that the prior art mainly constructs input-output mapping based on the relationship between multiple costs and power generation, but fails to fully consider the relationship between different data. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flow chart of the method provided by the present invention; Figure 2 This is a comparison chart of the model accuracy between the method of the present invention and the traditional method; Figure 3 This is a comparison chart of the calculation time between the method of the present invention and the traditional mapping method; Figure 4 This is a graph showing changes in various indicators of power quality within 24 hours; Figure 5 This is a diagram of the training process of the deep neural network model; Figure 6 The figure is a comparison chart of resource consumption between the method of the present invention and the traditional mapping method. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0067] like Figure 1 As shown, it is a flow chart of a method for determining the mapping relationship of input-output data of a new energy transmission project provided by the present invention. The specific implementation methods of the steps of the present invention are described in detail below: The specific implementation method of step S10 is to provide basic data support for subsequent modeling and analysis through data collection and standardization. First, a professional power data collection system is used to obtain the historical operation data of the new energy transmission project, including construction investment data, equipment investment data, operation and maintenance investment data and output data. The collection of construction investment data requires the extraction of cost data such as infrastructure investment, design and survey investment, environmental impact assessment and water conservation investment, and construction investment from the project management system. These data are usually counted on a monthly basis. Equipment investment data requires the extraction of parameter information and cost information of various types of equipment from the equipment management system. The equipment parameter information mainly includes technical parameters of key equipment such as transformers, switchgear, transmission lines, reactive compensation, relay protection and secondary equipment. The equipment cost information includes procurement cost and installation and commissioning cost. Operation and maintenance investment data requires the extraction of relevant cost data such as daily maintenance, fault repair, technical transformation and emergency disposal from the operation and maintenance management system. For output data, it is necessary to extract output power data from the power dispatching automation system, including annual, monthly and daily power generation data, and extract power quality data such as voltage deviation, frequency deviation, three-phase imbalance and harmonic content from the power quality online monitoring system. The collected raw data is standardized, and the maximum and minimum value normalization method is used to uniformly map all types of data to the range of 0 to 1. The specific calculation formula is: ,in is the original data value, and are the minimum and maximum values ​​of this type of data, respectively. is the normalized data value. Standardization can eliminate the dimensional differences between different data and make all types of data comparable.

[0068] The specific implementation method of step S20 is to extract device features by constructing a device parameter matrix and performing fast Fourier transform. First, the key parameters of all devices are arranged according to device type and parameter type to construct a device parameter matrix. Each row of the matrix represents a device and each column represents a key parameter. Considering the possible nonlinear relationship between device parameters, fast Fourier transform is used to process the parameter matrix and convert time domain information to frequency domain space. Fast Fourier transform is an algorithm for efficiently calculating discrete Fourier transform. Its core idea is to convert The point time domain series is decomposed into The Fourier transform of the point sequence reduces the computational complexity through the butterfly operation structure. During the transformation process, each row in the parameter matrix is ​​first regarded as a time domain sequence, and each sequence is fast Fourier transformed to obtain the corresponding frequency domain representation. Then the amplitude spectrum and phase spectrum information of the frequency domain signal are extracted. The amplitude spectrum reflects the intensity characteristics of the parameter change, and the phase spectrum reflects the phase relationship of the parameter change. The extracted amplitude spectrum and phase spectrum information are combined to form a device feature vector, which contains the main feature information of the device parameters.

[0069] The specific implementation method of step S30 is to use the principal component analysis method to reduce the dimension of the device eigenvector. Principal component analysis is a commonly used data dimensionality reduction method. Its core idea is to convert the original high-dimensional data into a set of new orthogonal bases through orthogonal transformation so that the projected data has the largest variance. First, calculate the covariance matrix of the device eigenvector, which reflects the correlation between each eigencomponent. Then perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalue and the corresponding eigenvector. The size of the eigenvalue indicates the degree of data change in the direction of the corresponding eigenvector. The larger the eigenvalue, the more significant the data change in this direction. Sort the eigenvectors according to the size of the eigenvalue, and select the top eigenvectors. The eigenvectors corresponding to the eigenvalues ​​constitute the dimension reduction transformation matrix, where The value of needs to ensure that the cumulative contribution rate reaches the preset threshold, usually 85%. Finally, the original device feature vector is projected onto the dimension reduction transformation matrix to obtain the reduced-dimensional core device feature vector. The dimension reduction process not only reduces the dimension of the data, but also retains the main feature information of the data.

[0070] The specific implementation method of step S40 is to determine the weight coefficients of the key parameters of each device by establishing an interaction influence matrix. First, an interaction influence matrix is ​​constructed based on the core device feature vector, and each element in the matrix represents the degree of correlation between different feature components. The Pearson correlation coefficient is used to calculate the correlation between feature components. The value range of the correlation coefficient is -1 to 1. The larger the absolute value, the stronger the correlation. For the interaction influence matrix, its eigenvalues ​​and eigenvectors are calculated. The eigenvalues ​​reflect the importance of each feature component in the overall feature space. By normalizing the eigenvalues, the weight coefficients of each feature component are obtained, and the sum of the weight coefficients is 1. The size of the weight coefficient reflects the degree of influence of the corresponding feature component on the overall performance of the system.

[0071] The specific implementation method of step S50 is to construct a mapping relationship between the comprehensive evaluation index of equipment parameters and other data. First, according to the weight coefficient obtained in step S40, the components of the core equipment feature vector are weighted and summed to obtain the comprehensive evaluation index of equipment parameters. This index comprehensively considers the importance of each characteristic component and can better reflect the overall characteristics of the equipment parameters. Then establish a mapping relationship between the comprehensive evaluation index and the construction investment data, operation and maintenance investment data, output power data and output power quality data. Taking into account the random disturbance factors that may exist in actual projects, a random error term is introduced in the mapping relationship. The error term obeys the normal distribution, the mean is 0, and the standard deviation is determined according to historical data statistics. The coefficients in the mapping relationship are estimated by the least squares method so that the mean square error between the predicted value and the actual observed value is minimized.

[0072] The specific implementation of step S60 is to construct and train a deep neural network model. A multilayer perceptron is selected as the network structure, including an input layer, multiple hidden layers and an output layer. The number of nodes in the input layer is equal to the dimension of the core equipment feature vector, and the number of nodes in the output layer is equal to the total dimension of the construction investment data, operation and maintenance investment data, output power data and output power quality data. The hidden layer adopts a multi-layer structure, and the number of nodes in each layer is determined by an empirical formula, generally taking the average value of the number of nodes in the input layer and the output layer. The hidden layer uses the ReLU activation function, which has the characteristics of simple calculation and stable gradient. The output layer selects a suitable activation function according to the characteristics of the output data, uses a linear activation function for continuous data, and uses a Sigmoid activation function for probabilistic data. The network is trained using the back propagation algorithm, and the network parameters are updated using the Adam optimizer, which combines the advantages of the momentum method and the adaptive learning rate method. The batch normalization technology is used during the training process to accelerate network convergence and prevent gradient disappearance. At the same time, the dropout technology is used to prevent the network from overfitting, and the dropout rate is set to 0.5.

[0073] The specific implementation method of step S70 is to verify and tune the trained deep neural network model. Randomly select a part of the data from the actual engineering data as the verification set, and the number of verification sets generally accounts for 20% of the total data volume. Input the verification set data into the trained network to obtain the prediction result. Calculate the root mean square error between the prediction result and the actual value, which reflects the prediction accuracy of the model. If the prediction accuracy does not meet the preset requirements, the network structure and parameters need to be adjusted. The adjustment of the network structure includes changing the number of hidden layers and the number of nodes in each layer. Generally, the grid search method is used to determine the optimal network structure. The adjustment of network parameters includes modifying hyperparameters such as learning rate, batch size, dropout rate, etc., and the Bayesian optimization method can be used to automatically search for the optimal hyperparameter combination. Through repeated adjustment and verification, until the prediction accuracy of the model reaches the preset requirements, a unified input-output mapping model is finally output. The preset accuracy requirements vary depending on the specific application scenario. Generally, a root mean square error less than 0.1 can be considered to have good prediction performance.

[0074] The specific implementation methods of the above steps fully consider the characteristics and actual needs of the new energy transmission project, and adopt a variety of advanced data processing and machine learning methods to achieve accurate modeling of the mapping relationship between input and output data. This method has strong versatility and practicality, and can provide effective decision support for the planning, design and operation management of the new energy transmission project.

[0075] Specifically, the principle of the present invention is: the principle of constructing a mapping relationship centered on device parameters is based on the following scientific understanding: First of all, equipment parameters are the most basic determinants of the performance of new energy transmission projects. The rated parameters and operating parameters of the equipment directly affect the transmission capacity, loss level and reliability of the system. Through the systematic analysis of equipment parameters, the technical characteristics and economic performance of the project can be accurately grasped.

[0076] Secondly, equipment parameters are highly correlated and interactive. For example, the impedance parameters of the transformer will affect the short-circuit current level of the system, which in turn affects the selection of the switchgear; the line parameters will affect the reactive balance of the system, which in turn affects the capacity configuration of the compensation device. Therefore, establishing a mapping relationship centered on equipment parameters can effectively reflect the interaction between the various elements of the system.

[0077] In terms of data processing, the principle of fast Fourier transform adopted in this invention is based on the periodic characteristics of equipment parameter changes. The intermittent and fluctuating nature of renewable energy generation will lead to periodic changes in the load of the transmission system, which will be reflected in the dynamic characteristics of the equipment parameters. Through frequency domain analysis, this dynamic characteristic can be better captured.

[0078] The principle of principal component analysis for dimensionality reduction is based on the correlation between device parameters. By projecting the original parameter space into a new feature space through orthogonal transformation, the main information can be expressed with fewer feature components. This method not only reduces the computational complexity, but also eliminates the problem of collinearity between parameters.

[0079] When establishing the mapping relationship, the principle of using a deep neural network is based on the nonlinear characteristics of the input-output relationship. The multi-layer structure of the network can extract features layer by layer and realize complex nonlinear mapping through different activation functions. This method is more suitable for describing the complex relationships in the new energy transmission project than the traditional linear model.

[0080] The logic of the scheme of the present invention is reflected in the progressive relationship between data processing and model construction: starting from basic parameter collection, unified expression of data is achieved through standardization processing; dynamic characteristics of parameters are extracted using fast Fourier transform; dimensionality reduction optimization of characteristics is achieved through principal component analysis; parameter weights are determined based on the interaction influence matrix; and finally a deep neural network is constructed to achieve modeling of mapping relationships. Each step has a clear goal and theoretical support, forming a complete technical chain.

[0081] In summary, the technical solution of the present invention centered on equipment parameters not only conforms to the physical characteristics of the new energy transmission project, but also meets the needs of engineering practice, and can better guide the planning, design, operation and maintenance of the new energy transmission project.

[0082] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: This embodiment takes a certain wind power transmission project as an example to illustrate the method for determining the input-output data mapping relationship of the new energy transmission project. The total installed capacity of the wind power transmission project is 1 million kilowatts, the construction period is 24 months, and it has been put into operation for 12 months. The project includes a 220kV booster station and an 80-kilometer-long 220kV transmission line. The specific implementation process of the method of the present invention is described in detail below.

[0083] First, step S10 is performed to collect historical operation data of the wind power transmission project. Construction input data is extracted from the project management system, and the specific data is shown in Table 1 below.

[0084] Table 1 Construction investment table

[0085] Equipment investment data is extracted from the equipment management system, mainly including parameters and cost information of transformers, switchgear, transmission lines, reactive power compensation, relay protection and secondary equipment. The specific data of equipment parameters are shown in Table 2 below.

[0086] Table 2 Equipment parameters

[0087] The equipment cost data is shown in Table 3 below.

[0088] Table 3 Equipment cost table

[0089] The operation and maintenance input data is extracted from the operation and maintenance management system. The specific data is shown in Table 4 below.

[0090] Table 4 Operation and maintenance investment table

[0091] The output power data in the output data is extracted from the power dispatching automation system. The specific data is shown in Table 5 below.

[0092] Table 5 Output Electric Energy Table

[0093] The power quality data is extracted from the power quality online monitoring system, the sampling time is the last 24 hours, and the number of sampling points is 8640. All collected data are standardized, and the maximum and minimum value normalization method is used to map the data to the range of 0 to 1.

[0094] Next, step S20 is executed to construct a device parameter matrix and perform feature extraction. The 18 key parameters of the six types of devices are constructed into a 6×18 device parameter matrix, and the matrix is ​​fast Fourier transformed to obtain a complex feature matrix. The complex feature matrix is ​​converted to the real domain, and the amplitude spectrum and phase spectrum information are extracted to form a device feature vector. The dimension of the device feature vector is 36, including 18 amplitude components and 18 phase components.

[0095] Then execute step S30, and use the principal component analysis method to reduce the dimension of the 36-dimensional device feature vector. Calculate the covariance matrix of the feature vector and perform eigenvalue decomposition. Sort by the size of the eigenvalues, select the eigenvectors corresponding to the first 8 eigenvalues ​​required to achieve a cumulative contribution rate of 85%, and construct a dimensionality reduction transformation matrix. Project the original eigenvector onto the dimensionality reduction transformation matrix to obtain an 8-dimensional core device feature vector. Traditional methods usually use simple linear mapping or statistical-based regression methods to establish input-output relationships, such as multivariate linear regression and stepwise regression. These methods are simple to calculate but have low accuracy, and cannot effectively handle nonlinear relationships and high-dimensional data. In terms of dimensionality reduction, traditional methods mainly use linear methods such as PCA, which cannot fully extract nonlinear features in the data. In terms of mapping relationship modeling, traditional methods often use a single mathematical model, lack the ability to comprehensively process multi-source heterogeneous data, and do not consider the complex interactions between data. Such as Figure 2As shown in the figure, the model accuracy of the method of the present invention and the traditional PCA method under different feature dimensions is compared. The horizontal axis represents the feature dimension after dimensionality reduction, and the vertical axis represents the accuracy of the model. It can be seen that the method of the present invention is superior to the traditional PCA method in all dimensions, especially in low dimensions. Figure 3 As shown in the figure, the calculation time comparison between the method of the present invention and the traditional mapping method under different data amounts is shown. The horizontal axis represents the data amount, and the vertical axis represents the calculation time (logarithmic coordinate). The results show that the calculation time of the method of the present invention is much lower than that of the traditional method, and the advantage becomes more obvious as the data amount increases.

[0096] Step S40 is further executed to construct an 8×8 interaction matrix and calculate the correlation coefficients between the components of the core eigenvector. The interaction matrix is ​​subjected to eigenvalue decomposition and the weight coefficients of each component are obtained by eigenvalue normalization. The weight coefficients of the eight eigencomponents are 0.22, 0.18, 0.15, 0.12, 0.10, 0.09, 0.08, and 0.06, respectively. Figure 4 As shown in Figure 2, the changes in various power quality indicators within 24 hours, including voltage deviation, frequency deviation and total harmonic distortion rate, are shown. These curves reflect the dynamic change characteristics of power quality indicators.

[0097] Execute step S50 to construct a comprehensive evaluation index of equipment parameters according to the weight coefficient. Introduce a random error term with a mean of 0 and a standard deviation of 0.05, and establish a mapping relationship between the comprehensive evaluation index and the construction investment data, operation and maintenance investment data, output power data, and output power quality data. Use the least squares method to estimate the coefficients in the mapping relationship.

[0098] Execute step S60 to build a deep neural network model. The number of input layer nodes is 8, corresponding to the dimension of the core equipment feature vector. The number of output layer nodes is 12, corresponding to 4 types of construction investment data, 4 types of operation and maintenance investment data, 3 types of output power data and 1 type of comprehensive power quality data. Design 3 hidden layers with 10, 10, and 10 nodes respectively. The hidden layer uses the ReLU activation function, and the output layer uses the linear activation function. Use the Adam optimizer, the initial learning rate is set to 0.001, the batch size is 32, and the number of training rounds is 1000. Figure 5 As shown in the figure, the training process of the deep neural network model is shown, including the changing curves of training loss and validation loss. It can be seen that the loss function gradually converges during the model training process, and there is no obvious overfitting phenomenon.

[0099] Finally, step S70 is performed to randomly select 20% of the 12-month historical data as the validation set, that is, about 72 days of data. The validation set is input into the trained network model to calculate the root mean square error between the predicted result and the actual value. The root mean square error of the first round of verification is 0.15, which does not meet the preset accuracy requirement of 0.1. The network structure is adjusted by grid search, and the number of hidden layer nodes is adjusted to 12, 10, and 8. The hyperparameters are adjusted using the Bayesian optimization method, and the learning rate is adjusted to 0.0008, the batch size is adjusted to 64, and the dropout rate is adjusted to 0.4. The root mean square error of the adjusted model on the validation set is reduced to 0.08, meeting the preset accuracy requirements.

[0100] The final input-output mapping model performs well in practical applications. Taking the real-time data of a certain day as an example, the core equipment feature vector of that day was input, and the model predicted a daily power generation of 7.285 million kWh, with a relative error of only 0.21% from the actual value of 7.3 million kWh. The predicted voltage deviation was 3.2%, the frequency deviation was 0.12Hz, the three-phase imbalance was 1.1%, and the total harmonic distortion was 2.8%, all of which were within a reasonable range. This model can provide effective decision-making support for the operation and management of wind power transmission projects, helping operation and maintenance personnel to promptly discover and deal with potential problems. Figure 6 As shown, the comparison of resource consumption between the method of the present invention and the traditional mapping method is shown, including two indicators: memory usage and CPU utilization. The results show that the method of the present invention has obvious advantages in resource utilization efficiency.

[0101] It can be seen from this embodiment that the method of the present invention has a good application effect in actual engineering. The method comprehensively considers factors such as equipment parameters, input costs and operating conditions, and adopts advanced data processing and machine learning methods to achieve accurate modeling of the mapping relationship between input and output data. The implementation process of the method is clear and definite, and the settings of various parameters and indicators are reasonable, which can provide reference and reference for similar projects.

[0102] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for determining the mapping relationship of input-output data of a new energy transmission project, characterized in that: The following steps are involved: S10, obtaining historical input data and output data of the new energy transmission project, and performing standardization on the historical input data and output data, wherein the historical input data includes construction input data, equipment input data, and operation and maintenance input data, and the output data includes output power data and output power quality data; S20, constructing a device parameter matrix for key parameters of all types of devices, where each row represents a type of device and each column represents a key parameter, performing a fast Fourier transform on the device parameter matrix to obtain a complex feature matrix of device parameters, and converting the complex feature matrix into a real domain feature vector to obtain a device feature vector; S30, using a principal component analysis method to perform dimensionality reduction processing on the device feature vector, extracting the main feature components, and constructing a core device feature vector after dimensionality reduction; S40, establishing an interaction matrix between components within the core device feature vector, calculating the eigenvalue of the interaction matrix, and determining the weight coefficient of each key parameter of the device; S50, constructing a comprehensive evaluation index of equipment parameters according to the weight coefficient, and establishing an integrated mapping relationship between the comprehensive evaluation index of equipment parameters and the construction investment data, the operation and maintenance investment data, the output power data, and the output power quality data; S60, constructing a multi-input and multi-output deep neural network structure, taking the core equipment feature vector as the input layer, taking the construction investment data, the operation and maintenance investment data, the output power data and the output power quality data as the output layer, and using the integrated mapping relationship to perform network training; S70. Use a verification data set in actual engineering data to verify the prediction accuracy of the deep neural network, adjust the parameters of the deep neural network according to the prediction accuracy, until the prediction accuracy reaches a preset accuracy requirement, and output a unified input-output mapping model.

2. The method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 1, characterized in that: The construction investment data include infrastructure investment, design and investigation investment, environmental assessment and water conservation investment, and construction investment; The equipment investment data includes equipment parameters and equipment costs of all equipment in the project, and the equipment costs include procurement costs, installation and commissioning costs; The operation and maintenance investment data include daily maintenance costs, fault repair costs, technical transformation costs, and emergency response costs; The output power data includes annual power generation, monthly power generation, and daily power generation; The output power quality data includes voltage deviation, frequency deviation, three-phase unbalance, and harmonic content.

3. The method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 2 is characterized in that: The device parameters include at least: Transformer key parameters: rated capacity, impedance percentage, no-load loss rate, load loss rate; Key parameters of switchgear: rated voltage, rated current, breaking capacity; Key parameters of transmission lines: line length, conductor cross section, tower load; Key parameters of reactive power compensation: compensation capacity, response time, and adjustment range; Key parameters of relay protection: action time, return coefficient, reliability coefficient; Key parameters of secondary equipment: measurement accuracy, sampling frequency, and communication bandwidth.

4. The method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 3 is characterized in that: The device parameter matrix construction and feature extraction process are specifically expressed as follows: 1) Construction of original matrix of device parameters: ; In the formula, For the The first The normalized values ​​of the key parameters are , ; is the total number of device types; is the total number of key parameters for each device; 2) Fast Fourier Transform Process: ; In the formula, is the complex domain representation after Fourier transform; is the frequency component number; is an imaginary unit; is the time series index; 3) The complex feature matrix is ​​converted to the real number domain: ; ; In the formula, is the amplitude spectrum; is the phase spectrum; and are the real and imaginary parts of the complex characteristic, respectively; 4) Device feature vector construction: ; In the formula, For the The feature vector of the device.

5. The method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 4 is characterized in that: The process of constructing the interaction influence matrix is ​​as follows: 1) Construct the interaction influence matrix: ; In the formula, For the The characteristic component and The correlation coefficient of the feature components is calculated by Pearson correlation coefficient: ； 2) Calculate the weight coefficient: ; In the formula, The interaction matrix No. eigenvalues; For the The weight coefficient of the feature component.

6. A method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 5, characterized in that: The process of building the integrated mapping relationship is as follows: 1) Calculation of comprehensive evaluation index of equipment parameters: ; In the formula, It is a comprehensive evaluation indicator; is a random error term, which follows a normal distribution ; 2) Mapping relationship construction: ; In the formula, Input vector for construction; Provide vectors for operation and maintenance; is the output electric energy vector; is the power quality vector; are unknown coefficients, obtained by least squares fitting.

7. A method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 6, characterized in that: The multi-input multi-output deep neural network adopts an encoder-decoder architecture. The input layer receives the core device feature vector after dimensionality reduction processing. The encoder is composed of 3 layers of bidirectional LSTM. The hidden layer dimension of each LSTM layer is 256, and a Dropout layer is added between the LSTM layers to prevent overfitting. An 8-head self-attention mechanism is set after the encoder output to learn the correlation between different features; The decoder adopts a multi-branch structure, which includes four independent decoding branches: construction investment, operation and maintenance investment, output power and power quality. Each branch independently learns the corresponding mapping relationship.

8. The method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 7, characterized in that: The calculation steps of the power quality data include: Step 1: Collect real-time operation data of the power system of the new energy transmission project; Step 2: construct matrix groups respectively, including the power quality equation matrix group when new energy is not connected and the power quality equation matrix group after new energy is connected; Step 3: Perform singular value decomposition on each matrix in the constructed matrix group to obtain a stable equation matrix and a variable equation matrix; Step 4: Calculate the power quality vector based on the transformation equation matrix.

9. A method for determining the mapping relationship of input-output data of a new energy transmission project according to claim 8, characterized in that: The improved BP neural network is used to calculate the power quality vector based on the change equation matrix.

10. A method for determining a mapping relationship of input-output data of a new energy transmission project according to claim 9, characterized in that: The power quality vector specifically includes: voltage deviation, frequency deviation, three-phase unbalance, and harmonic content.

Images