Distribution network engineering material demand prediction system based on gray level improvement GA-BP neural network

Abstract

The invention discloses a GA-BP neural network distribution network engineering material demand prediction system based on gray level improvement, and the system comprises a data importing module which is used for carrying out the format conversion of data extracted from source end data, and then importing the data; the data cleaning module is used for preprocessing the imported data; the grey correlation degree analysis module is used for calculating and analyzing the correlation degree between each influence factor of the distribution network material prediction and the distribution network material prediction by using grey correlation degree analysis; the GA algorithm optimization module is used for optimizing an initial weight and a threshold value of the BP neural network by utilizing the global optimization capability of the GA, so as to create a GA-BP neural network prediction model of network error reverse transmission based on GA optimization; the BP model training module is used for training by using the initial weight and the threshold optimized by the GA algorithm and utilizing a BP model to obtain a BP neural network model; and the data prediction module is used for performing prediction based on a BP neural network model. The prediction precision is improved.

Description

Technical Field

[0001] The invention relates to a power grid data processing technology, and in particular to a distribution network engineering material demand forecasting system based on a grayscale-improved GA-BP neural network. Background Art

[0002] With the continuous improvement of the modern intelligent supply chain system, the management requirements for the intensive level of power materials are getting higher and higher. Strengthening the comprehensive material planning management of "centralization, unification, leanness and efficiency" has become the core content of the supply chain management of power material companies. Compared with the main grid project, the distribution network project, as an important infrastructure facing end users and serving the people's livelihood, is mostly a short-term and fast project. Usually, the construction and renovation period is short, and the supply of equipment and materials is scattered, fragmented, low in certainty and high in timeliness. As a result, the related materials often have a wide variety, high supply response speed, high demand uncertainty, and large differences in demand characteristics. Therefore, how to select a reasonable forecasting and analysis model, combined with the characteristics of distribution network project construction, to carry out a reasonable distribution network material demand forecast has become one of the important tasks in the planning management of power material companies.

[0003] Power grid companies actively undertake the deployment, with the overall goal of building a high-quality, efficient, safe and lean first-class distribution network at home and abroad. Through the whole process, they improve the construction capacity of distribution network projects and implement supply chain reforms within power grid companies in accordance with the concept of "managing projects must also manage materials". Tracing back to the source, the inventory problem of distribution network projects is essentially a problem of supply chain source management, and ultimately returning to the starting point of material management, it is necessary to do a good job in material demand forecasting and fully guarantee the accuracy of data information on material demand plans for all categories of distribution network projects.

[0004] Scientific material demand forecasting is of great significance: first, it is conducive to power grid companies reducing inventory and solving the problem of absorbing inventory materials worth hundreds of millions of yuan from the root; second, it is conducive to power grid companies reducing costs. After the inventory decreases, the use of working capital will also decrease, and the capital cost will be reduced at the same time; third, it is conducive to power grid companies to respond and make decisions quickly. After the material constraints are reduced, the matching degree between the construction progress of the project and the material reserves will be improved, which is conducive to speeding up the progress of the project.

[0005] In summary, based on the characteristics of the distribution network materials projects, this paper introduces grey correlation analysis to rank the influence of factors affecting distribution network projects and material demand. The traditional BP neural network is improved. The GA-BP neural network distribution network project material demand forecasting technology based on grayscale improvement is a key technology to improve the accuracy and reliability of power grid material forecasting, which is of great significance to achieving the accuracy of distribution network material forecasting.

[0006] In the prior art, domestic scholars mainly use the exponential smoothing method to build models to predict procurement data for power grid material demand forecasting. Based on the multiple regression model, historical data training is used to obtain the influence coefficient of each factor on the emergency resource forecast, and the demand for various emergency materials after a disaster event occurs is quickly calculated. The multiple linear regression method can accurately measure the degree of correlation and fit between various factors, and improve the accuracy of emergency material demand forecasting, but this method is difficult to obtain historical data of explanatory variables.

[0007] Traditional forecasting methods such as time series and regression models are difficult to cope with the nonlinearity and dynamics of material demand, especially in the big data environment. Scholars have begun to study intelligent technologies, including neural networks and optimization algorithms, to improve material demand forecasting. Based on the type of power material projects, the influencing factors of power materials are selected based on grey correlation analysis, and finally a BP (GA-BP) neural network forecasting model optimized by genetic algorithm based on grey correlation analysis is constructed.

[0008] The shortcomings of existing technologies are mainly reflected in the low reliability of material demand forecasts. The annual feasibility study estimate (investment pre-plan) of the distribution network project is very different from the actual investment amount. One is that the feasibility study estimate of the current year is far less than the investment of the next year, and the other is that the feasibility study estimate of the current year is far greater than the investment of the next year. When the feasibility study estimate of the current year is far less than the investment of the next year, the material demand forecast will be too small, resulting in a shortage of materials at the start of construction, which will inevitably affect the progress of the project; when the feasibility study estimate of the current year is far greater than the investment of the next year, the material demand forecast will be too large, resulting in excess materials at the start of construction, which will inevitably cause inventory backlogs. Carrying out the first batch of material demand forecasts based on the feasibility study (investment pre-plan) will cause the demand forecast to be too large or too small (because the feasibility study and actual investment are very random and have a large difference). Therefore, the reliability of the traditional forecasting model is very low. When the demand changes randomly, this forecasting model cannot adapt well to the randomness of material demand.

[0009] The grayscale-improved GA-BP neural network distribution network engineering material demand forecasting system provided by the present invention is proposed to address these deficiencies in the prior art. Summary of the invention

[0010] The purpose of the present invention is to provide a distribution network engineering material demand forecasting system based on grayscale improvement of GA-BP neural network, which can solve the technical problems mentioned in the background technology.

[0011] A distribution network engineering material demand forecasting system based on grayscale improved GA-BP neural network, the system comprising:

[0012] The data import module is used to convert the data extracted from the source into a new format and import it;

[0013] The data cleaning module is used to pre-process the imported data, including data cleaning, data integration, data transformation, data standardization, abnormal data deletion, and divide the pre-processed data into training sets and test sets;

[0014] Grey correlation analysis module, which uses grey correlation analysis to calculate and analyze the correlation between various influencing factors of distribution network material forecast and distribution network material forecast;

[0015] GA algorithm optimization module, which uses GA's global optimization ability to optimize the initial weights and thresholds of the BP neural network, thereby creating a GA-BP neural network prediction model based on the reverse transmission of the network error optimized by GA;

[0016] The BP model training module uses the initial weights and thresholds optimized by the GA algorithm to train the BP model and obtain the BP neural network model;

[0017] Data prediction module, which makes predictions based on BP neural network model.

[0018] As a further technical solution of the present invention, the grey relational analysis module comprises:

[0019] Dimensionless processing: Since the source data is usually different from other series in scope and unit, it is often dimensionless for easy comparison, using averaging or initialization.

[0020] Solve the grey correlation coefficient;

[0021] Solve the grey relational degree;

[0022] Grey relational sorting.

[0023] As a further technical solution of the present invention, the GA algorithm optimization module includes:

[0024] Initialize basic parameters: set the size of the genetic algorithm population and the number of iterations, assign probabilities to the selection, crossover, and mutation of core operations, and determine the fitness function;

[0025] Create an initial population: encode the network weights and thresholds of the BP model training module;

[0026] Fitness value calculation: Calculate the fitness value according to the fitness function;

[0027] Population evaluation: Sort the calculated fitness values ​​and record the best individuals through selection, crossover, and mutation operations.

[0028] As a further technical solution of the present invention, selection, crossover and mutation: firstly, a roulette wheel selection operator is adopted so that the probability of each individual in the population being selected is proportional to its fitness value; secondly, two-point crossover is used as a crossover operator, two crossover points are randomly set in the individual chromosome, and some genes are exchanged between the crossover points to increase the diversity of the population, in order to produce better individuals in the subsequent process.

[0029] As a further technical solution of the present invention, the BP model training module includes:

[0030] Establish BP model:

[0031] Assume that the nodes of the input layer are , the number of nodes in the hidden layer is , the final output of the output layer is , and are the activation functions of the hidden layer and the output layer, respectively. is the weight from the input layer to the hidden layer, is the weight from the hidden layer to the output layer, the algorithm flow of the BP neural network is as follows:

[0032] In the forward propagation process of information, the information is first transmitted from the input layer to the hidden layer, and the output value of the hidden layer for:

[0033] ;

[0034] Where: ; is the number of nodes in the input layer;

[0035] Information continues to be transmitted from the hidden layer to the output layer, and the output value of the output layer for:

[0036] ;

[0037] Where: ; is the number of hidden layer nodes;

[0038] The first forward pass of the neural network has been completed, since the actual output and actual value There are differences, resulting in errors for:

[0039] ;

[0040] Where: ; is the number of nodes in the output layer;

[0041] The error is transmitted backwards, and the BP neural network adjusts the weights based on the gradient descent algorithm. The weight update formula is:

[0042] ;

[0043] In the formula is the learning rate;

[0044] The error is adjusted by continuously looping the above process until the algorithm converges, then the training process ends and the final result is output;

[0045] Model training: During the model training process, the GA algorithm optimization module is used to optimize the initial weights and thresholds of the BP neural network;

[0046] Use the data imported by the data import module to train the BP neural network;

[0047] Observe the training effect of the BP neural network, and after the BP neural network training is completed, adjust the parameters of the BP neural network according to the training results until the model reaches the optimal value, and obtain an optimal BP neural network model;

[0048] Model verification: Use the test set to evaluate the model performance and calculate indicators such as mean square error (MSE) and mean absolute error (MAE). Plot the prediction results to compare the difference between the actual value and the predicted value to determine the model effect.

[0049] Beneficial effects achieved by the present invention:

[0050] 1. Improved prediction accuracy

[0051] Introduction of gray model: Gray system theory is good at dealing with small samples and uncertainty problems. By combining the gray prediction model with the BP neural network, it can improve the trend extraction and processing capabilities of the original data and improve the prediction accuracy of material demand.

[0052] Genetic algorithm optimization: GA can globally optimize the initial weights and thresholds of the BP neural network, avoiding the traditional BP neural network from falling into the local optimal problem, thereby further improving the prediction accuracy.

[0053] 2. Adaptability to small sample data

[0054] The material demand data of distribution network projects are often small samples, nonlinear or lack regularity. The introduction of grayscale model significantly enhances the robustness of the model in dealing with these characteristics, making up for the shortcomings of single BP neural network in its insufficient processing ability for small samples.

[0055] 3. Resource allocation optimization

[0056] Improved inventory management: Accurately forecasting material needs can reduce the backlog of excess inventory and avoid project delays caused by insufficient demand.

[0057] Improved logistics efficiency: Logistics arrangements can be made based on more accurate demand forecasts, reducing resource waste and transportation costs.

[0058] 4. Cost reduction

[0059] Reduce losses caused by forecast errors: Reducing forecast errors directly reduces additional expenses caused by shortages or surpluses of materials.

[0060] Optimize procurement plans: Accurate demand forecasts support more scientific procurement decisions, reducing over-purchasing and the resulting financial costs.

[0061] 5. Enhanced dynamic adaptability

[0062] With the dynamic changes in distribution network engineering demand (such as seasonal or temporary demand), the improved GA-BP model can quickly adjust the prediction results and provide real-time demand response support.

[0063] 6. Improvement of intelligent management capabilities

[0064] Integrate grayscale theory, genetic algorithms and neural network technology into distribution network material demand management to enhance the digitalization and intelligence level of enterprises and promote the transformation to modern management. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A schematic diagram of the system operation flow provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The technical solution of the present invention is described in detail below in conjunction with the specific drawings.

[0067] See also Figure 1 The embodiment of the present invention provides a distribution network engineering material demand forecasting system based on grayscale improvement of GA-BP neural network, which calculates and analyzes the correlation between various influencing factors of distribution network material forecasting and distribution network material forecasting, and predicts distribution network materials with the help of data extraction module, data import module, gray correlation analysis module, GA algorithm optimization module, BP model training module, and data prediction module. The system includes:

[0068] The data import module is used to convert the data extracted from the source into a new format and import it;

[0069] The data cleaning module is used to pre-process the imported data, including data cleaning, data integration, data transformation, data standardization, abnormal data deletion, and divide the pre-processed data into training sets and test sets;

[0070] Grey correlation analysis module, which uses grey correlation analysis to calculate and analyze the correlation between various influencing factors of distribution network material forecast and distribution network material forecast;

[0071] GA algorithm optimization module. Since the BP neural network is prone to fall into local minima during the training process and the prediction error may be too large, the present invention uses the global optimization ability of GA to optimize the initial weights and thresholds of the BP neural network, thereby creating a GA-BP neural network prediction model based on the reverse transfer of the network error optimized by GA;

[0072] The BP model training module uses the initial weights and thresholds optimized by the GA algorithm to train the BP model and obtain the BP neural network model;

[0073] Data prediction module, which makes predictions based on BP neural network model.

[0074] In this embodiment, the grey relational analysis module includes:

[0075] Dimensionless processing: Since the source data is usually different from other series in scope and unit, it is often dimensionless for easy comparison, using averaging or initialization.

[0076] Solve the grey correlation coefficient;

[0077] Solve the grey relational degree;

[0078] Grey relational sorting.

[0079] In this embodiment, the GA algorithm optimization module includes:

[0080] Initialize basic parameters: set the size of the genetic algorithm population and the number of iterations, assign probabilities to the selection, crossover, and mutation of core operations, and determine the fitness function;

[0081] Create an initial population: encode the network weights and thresholds of the BP model training module;

[0082] Fitness value calculation: Calculate the fitness value according to the fitness function;

[0083] Population evaluation: Sort the calculated fitness values ​​and record the best individuals through selection, crossover, and mutation operations.

[0084] In this embodiment, selection, crossover and mutation: firstly, a roulette wheel selection operator is used so that the probability of each individual in the population being selected is proportional to its fitness value; secondly, two-point crossover is used as a crossover operator, two crossover points are randomly set in the individual chromosome, and some genes are exchanged between the crossover points to increase the diversity of the population, in order to produce better individuals in the subsequent process.

[0085] In this embodiment, the BP model training module includes:

[0086] Establish BP model:

[0087] Assume that the nodes of the input layer are , the number of nodes in the hidden layer is , the final output of the output layer is , and are the activation functions of the hidden layer and the output layer, respectively. is the weight from the input layer to the hidden layer, is the weight from the hidden layer to the output layer, the algorithm flow of the BP neural network is as follows:

[0088] In the forward propagation process of information, the information is first transmitted from the input layer to the hidden layer, and the output value of the hidden layer for:

[0089] ;

[0090] Where: ; is the number of nodes in the input layer;

[0091] Information continues to be transmitted from the hidden layer to the output layer, and the output value of the output layer for:

[0092] ;

[0093] Where: ; is the number of hidden layer nodes;

[0094] The first forward pass of the neural network has been completed, since the actual output and actual value There are differences, resulting in errors for:

[0095] ;

[0096] Where: ; is the number of nodes in the output layer;

[0097] The error is transmitted backwards, and the BP neural network adjusts the weights based on the gradient descent algorithm. The weight update formula is:

[0098] ;

[0099] In the formula is the learning rate;

[0100] The error is adjusted by continuously looping the above process until the algorithm converges, then the training process ends and the final result is output;

[0101] Model training: During the model training process, the GA algorithm optimization module is used to optimize the initial weights and thresholds of the BP neural network;

[0102] Use the data imported by the data import module to train the BP neural network;

[0103] Observe the training effect of the BP neural network, and after the BP neural network training is completed, adjust the parameters of the BP neural network according to the training results until the model reaches the optimal value, and obtain an optimal BP neural network model;

[0104] Model verification: Use the test set to evaluate the model performance and calculate indicators such as mean square error (MSE) and mean absolute error (MAE). Plot the prediction results to compare the difference between the actual value and the predicted value to determine the model effect.

[0105] The data prediction module comprises:

[0106] Using the distribution network material collection data of the data import module and the model trained by the BP model training module, select the material category to predict the distribution network materials.

[0107] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0108] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A distribution network engineering material demand forecasting system based on grayscale improved GA-BP neural network, characterized in that: The system comprises: The data import module is used to convert the data extracted from the source into a new format and import it; The data cleaning module is used to pre-process the imported data, including data cleaning, data integration, data transformation, data standardization, abnormal data deletion, and divide the pre-processed data into training sets and test sets; Grey correlation analysis module, which uses grey correlation analysis to calculate and analyze the correlation between various influencing factors of distribution network material forecast and distribution network material forecast; GA algorithm optimization module, which uses GA's global optimization ability to optimize the initial weights and thresholds of the BP neural network, thereby creating a GA-BP neural network prediction model based on the reverse transmission of the network error optimized by GA; The BP model training module uses the initial weights and thresholds optimized by the GA algorithm to train the BP model and obtain the BP neural network model; Data prediction module, which makes predictions based on BP neural network model.

2. According to claim 1, a GA-BP neural network distribution network engineering material demand forecasting system based on grayscale improvement is characterized in that: The grey relational analysis module comprises: Dimensionless processing: Since the source data is usually different from other series in scope and unit, it is dimensionless, using averaging or initialization. Solve the grey correlation coefficient; Solve the grey relational degree; Grey relational sorting.

3. The distribution network engineering material demand forecasting system based on grayscale improvement GA-BP neural network according to claim 1 is characterized in that: The GA algorithm optimization module includes: Initialize basic parameters: set the size of the genetic algorithm population and the number of iterations, assign probabilities to the selection, crossover, and mutation of core operations, and determine the fitness function; Create an initial population: encode the network weights and thresholds of the BP model training module; Fitness value calculation: Calculate the fitness value according to the fitness function; Population evaluation: Sort the calculated fitness values ​​and record the best individuals through selection, crossover, and mutation operations.

4. The distribution network engineering material demand forecasting system based on grayscale improvement GA-BP neural network according to claim 3 is characterized in that: Selection, crossover and mutation: First, the roulette wheel selection operator is used to make the probability of each individual in the population being selected proportional to its fitness value; secondly, two-point crossover is used as the crossover operator, two crossover points are randomly set in the individual chromosome, and some genes are exchanged between the crossover points to increase the diversity of the population, in order to produce better individuals in the subsequent process.

5. The distribution network engineering material demand forecasting system based on grayscale improvement GA-BP neural network according to claim 1 is characterized in that: The BP model training module includes: Establish BP model: Assume that the nodes of the input layer are , the number of nodes in the hidden layer is , the final output of the output layer is , and are the activation functions of the hidden layer and the output layer, respectively. is the weight from the input layer to the hidden layer, is the weight from the hidden layer to the output layer, the algorithm flow of the BP neural network is as follows: In the forward propagation process of information, the information is first transmitted from the input layer to the hidden layer, and the output value of the hidden layer for: ； Where: ; is the number of nodes in the input layer; Information continues to be transmitted from the hidden layer to the output layer, and the output value of the output layer for: ； Where: ; is the number of hidden layer nodes; The first forward pass of the neural network has been completed, since the actual output and actual value There are differences, resulting in errors for: ； Where: ; is the number of nodes in the output layer; The error is transmitted backwards, and the BP neural network adjusts the weights based on the gradient descent algorithm. The weight update formula is: ； In the formula is the learning rate; The error is adjusted in a continuous cycle until the algorithm converges, then the training process ends and the final result is output; Model training: During the model training process, the GA algorithm optimization module is used to optimize the initial weights and thresholds of the BP neural network; Use the data imported by the data import module to train the BP neural network; Observe the training effect of the BP neural network, and after the BP neural network training is completed, adjust the parameters of the BP neural network according to the training results until the model reaches the optimal value, and obtain an optimal BP neural network model; Model validation: Use the test set to evaluate the model performance, calculate the mean square error and mean absolute error, plot the predicted results and compare the difference between the actual value and the predicted value to judge the model effect.

Images