Power distribution network line loss prediction method based on IMVO-SVM

Abstract

The invention relates to a power distribution network line loss prediction method based on an I MVO-SVM, and the method comprises the following steps: 1, analyzing a plurality of electrical indexes in a historical line loss rate record through a gray correlation analysis method, obtaining the correlation degree between different electrical indexes and the power distribution network line loss, and building a power distribution network electrical characteristic index system as training data; 2, setting an algorithm optimization strategy, and optimizing the multivariate universe optimization algorithm MVO by using the algorithm optimization strategy to obtain an improved multivariate universe optimization algorithm I MVO; step 3, constructing a support vector machine model SVM based on an improved multivariate universe optimization algorithm I MVO to obtain an I MVO-SVM model, and training the I MVO-SVM model by using the power distribution network electrical characteristic index system to obtain a power distribution network line loss prediction model; and step 4, predicting the power distribution network line loss based on the power distribution network line loss prediction model. Large-scale power grid operation data can be processed, real-time and accurate line loss prediction is provided, and technical support is provided for efficient management and operation and maintenance of a smart power grid.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a distribution network line loss prediction method based on IMVO-SVM. Background Art

[0002] The power loss of the distribution network (referred to as line loss) refers to the power loss in each link of transmission, transformation and distribution when the power is transmitted from the power grid to the user. The line loss rate is the ratio of power loss to total power supply in a certain period of time. It is an important indicator to measure the technical and economic feasibility of power grid enterprises. It involves planning, production, marketing, operation, finance and other fields, and comprehensively reflects the technical and economic level of power grid enterprises in planning, design, production, operation and management.

[0003] With the continuous development of smart grid technology, the monitoring, control and optimization capabilities of the power grid have been significantly improved, becoming more refined and automated. By monitoring the flow of power in real time, smart grids can efficiently collect and process large amounts of data, which provides a solid technical foundation for accurately predicting line losses. However, traditional line loss prediction methods may not be able to fully capture the complex nonlinear relationships in the power system, resulting in inaccurate prediction results. Today, the power grid has an increasing demand for real-time monitoring and rapid response, which places higher requirements on the accuracy and response speed of line loss prediction technology. Summary of the invention

[0004] The purpose of the present invention is to provide a distribution network line loss prediction method based on IMVO-SVM, which can solve the technical problems of slow training speed and low accuracy in distribution network line loss prediction of traditional prediction methods, and effectively improve the efficiency and accuracy of prediction.

[0005] The present invention is implemented by the following technical solution: a distribution network line loss prediction method based on IMVO-SVM, which comprises the following steps:

[0006] Step 1: Use grey correlation analysis to analyze multiple electrical indicators in the historical line loss rate records, obtain the correlation between different electrical indicators and distribution network line loss, and then establish a distribution network electrical characteristic indicator system as training data;

[0007] Step 2, setting an algorithm optimization strategy, and using the algorithm optimization strategy to optimize the multiverse optimization algorithm MVO to obtain an improved multiverse optimization algorithm IMVO;

[0008] Step 3, construct a support vector machine model SVM based on the improved multiverse optimization algorithm IMVO to obtain an IMVO-SVM model, and use the distribution network electrical characteristic index system to train the IMVO-SVM model to obtain a distribution network line loss prediction model;

[0009] Step 4: predict the distribution network line loss based on the distribution network line loss prediction model.

[0010] Compared with the previous technology, the beneficial effects of the present invention are:

[0011] 1. The support vector machine is optimized using the improved multiverse optimization algorithm, which can process large-scale power grid operation data, provide real-time and accurate line loss prediction, and provide technical support for the efficient management and operation and maintenance of smart grids.

[0012] 2. The present invention collects and analyzes the operation-related data of the smart grid in real time and provides line loss prediction information in a timely manner, supports rapid response to changes in the grid status, and enhances the real-time monitoring capability of the smart grid line loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a module block diagram of the present invention. DETAILED DESCRIPTION

[0014] The present invention is described in detail below in conjunction with the accompanying drawings, but the protection content of the present invention is not limited to the following:

[0015] like Figure 1 As shown, a distribution network line loss prediction method based on IMVO-SVM includes the following steps:

[0016] Step 1: Use grey correlation analysis to analyze multiple electrical indicators in the historical line loss rate records, obtain the correlation between different electrical indicators and distribution network line loss, and then establish a distribution network electrical characteristic indicator system as training data;

[0017] Step 2, setting an algorithm optimization strategy, and using the algorithm optimization strategy to optimize the multiverse optimization algorithm MVO to obtain an improved multiverse optimization algorithm IMVO;

[0018] Step 3, based on the improved multiverse optimization algorithm IMVO, a support vector machine model SVM is constructed to obtain an IMVO-SVM model, and the distribution network electrical characteristic index system is used to train the IMVO-SVM model to obtain a distribution network line loss prediction model;

[0019] Step 4: Based on the distribution network line loss prediction model, the distribution network line loss is predicted. The present invention optimizes the support vector machine using the improved multiverse optimization algorithm, which can process large-scale power grid operation data, provide real-time and accurate line loss prediction, and provide technical support for the efficient management and operation and maintenance of smart grids. It can also collect and analyze the operation-related data of smart grids in real time, provide line loss prediction information in a timely manner, support rapid response to changes in the power grid status, and enhance the real-time monitoring capability of smart grid line losses.

[0020] Furthermore, the historical line loss rate records electrical indicators that affect the line loss rate, and the electrical indicators include active power supply, reactive power supply, total transformer capacity, transformer active power, transformer reactive power, total line length, line equivalent cross-sectional area, trunk line cross-sectional area, total trunk line length, and total branch line length.

[0021] Furthermore, step 1 also includes the following steps:

[0022] Step 101, analyzing the historical line loss rate, and preliminarily selecting several electrical indicators that have the greatest impact on the line loss according to the change of the line loss rate;

[0023] Step 102: The theoretical line loss change of the distribution network is used as a reference sequence y, and the change of each electrical index is used as a different comparison sequence x. i ;

[0024] Step 103: Process the reference sequence y and each comparison sequence x using an extreme dimensionless processing method. i , and then calculate each comparison sequence x i Grey correlation coefficient with reference sequence y;

[0025] Step 104, calculate the grey correlation degree based on the grey correlation coefficient, then compare and sort the grey correlation degrees, and select several electrical indicators based on the grey correlation degree to establish a distribution network electrical characteristic indicator system as training data.

[0026] It should be noted that the calculation formula of the dimensionless processing method is:

[0027] Among them, x' i is the updated value of the dimensionless processing method, x i is the comparison sequence, max is the original value x i The maximum value and min value of x i The minimum value of

[0028] The calculation formula of the grey relational coefficient is:

[0029] Among them, x i (k) is the value at the kth position in the comparison sequence, y is the reference sequence, y(k) is the value at the kth position in the reference sequence, ξ i (k) is x i The grey correlation coefficient of (k) and y(k), ρ is the resolution coefficient and is generally taken as 0.5.

[0030] The calculation formula of the grey relational degree is: Here, n is the total number of sequences.

[0031] To verify the necessity, rationality and effectiveness of grey relational analysis, a distribution network in a certain area of Fujian Province was selected as the research object, and two 10 kV distribution networks, D1 and D2, were analyzed respectively. Distribution network D1 contains 400 10 kV lines, and the proportion of the monthly active power supply of special transformers in the monthly active power supply of 10 kV lines is 18.32%; Distribution network D2 contains 80 10 kV lines, and the proportion of the monthly active power supply of special transformers in the monthly active power supply of 10 kV lines is 55.87%.

[0032] Here, 10 electrical indexes reflecting the grid structure and operation state of the distribution network were selected as examples, and the sorting results of the grey relational degrees calculated by the grey relational analysis method are as follows in the table:

[0033] Serial number Electrical Specifications D1 Grey Relationship D2 Grey Relationship 1 Power supply 0.882 0.826 2 Power supply 0.809 0.782 3 Total capacity of transformer 0.786 0.732 4 Transformer active power 0.751 0.709 5 Transformer reactive power 0.738 0.692 6 Total line length 0.703 0.656 7 Equivalent cross-sectional area of ​​line 0.687 0.632 8 Main line cross-sectional area 0.662 0.621 9 Total length of main line 0.655 0.616 10 Total length of branch lines 0.643 0.609

[0034] It should be noted that in the grey relational analysis method, the comparison sequences are sorted according to the size of the corresponding grey relational degree h. If h1 < h2, the reference sequence y is more similar to the comparison sequence x2, that is, the electrical index h2 has a higher correlation with the distribution network line loss and can better reflect the distribution network line loss level.

[0035] Furthermore, according to the sorting of the grey relational degrees of the electrical indexes, different numbers of electrical index combinations are selected from top to bottom, and a distribution network line loss prediction model is established and trained based on the actual data. Among them, the electrical index combination with the smallest prediction error is used as the electrical characteristic index.

[0036] Furthermore, the dimensionless processing method is generally data normalization operation to eliminate the differences between different attributes in the data set, and then the training data is divided into a training data set and a test data set; The above comparison sequence x i and the reference sequence y are also values after data normalization processing.

[0037] Furthermore, step 2 also includes the following steps:

[0038] Step 201, set the algorithm optimization strategy, and the algorithm optimization strategy includes strategy A and strategy B;

[0039] Step 202, introduce the Logistic chaotic mapping function to solve the problem that the position update algorithm of the multi - universe optimization algorithm MVO is easy to fall into local extrema, and obtain the Logistic chaotic mapping universe position algorithm;

[0040] Step 203, combine the Logistic chaotic mapping universe position algorithm, strategy A and strategy B to obtain the improved multi - universe optimization algorithm IMVO.

[0041] Furthermore, the formula group of the MVO algorithm is:

[0042]

[0043] Among them, X best-universe(j) is the current optimal solution found in the jth dimension, TDR travel-distance is the travel distance rate, ub j is the upper bound of the j-th dimension of the search space, lb j is the lower bound of the j-th dimension of the search space, r2, r3, and r4 represent random values ​​in the interval [0,1], and WEP represents the probability of the existence of a wormhole;

[0044] In step 202, the Logistic chaotic mapping function is introduced to correct the generation of r4 in the algorithm, and the value of r4 is modified according to the number of iterations t to obtain r4 t The update formula is:

[0045] r4 t =a*r4 t-1 (1-r4 t-1 );

[0046] Where a is a constant and a∈[0,4], r4 t is the solution of r4 obtained by using the chaotic mapping function, r4 t-1 Represents the value of r4 in the previous iteration, and t is the number of iterations.

[0047] Furthermore, the strategy A is to adjust the wormhole existence probability WEP in a linear growth mode to the convergent wormhole existence probability CWEP in a logarithmic growth mode, and its calculation formula is:

[0048] Among them, t refers to the current number of iterations, T refers to the maximum number of iterations set, and WEP max Refers to the maximum probability of the existence of a wormhole, WEP min Refers to the minimum probability of the existence of a wormhole;

[0049] It should be noted that WEP max and WEP min It is usually a constant set by the algorithm, which is used to represent the upper and lower limits of exploration behavior;

[0050] The strategy B is to introduce a compression factor τ, which is calculated as follows:

[0051] Among them, t refers to the current number of iterations, T refers to the maximum number of iterations, and e is a natural constant (approximately equal to 2.718).

[0052] In summary, the improved multiverse optimization algorithm IMVO obtained by combining the Logistic chaos mapping universe position algorithm, strategy A and strategy B is expressed by the following function group:

[0053]

[0054] Where, X best-universe(j) is the current optimal solution found in the jth dimension, τ is the compression factor, TDR travel-distance is the travel distance rate, ub j is the upper bound of the j-th dimension of the search space, lb j is the lower bound of the jth dimension of the search space, COA(t) is the chaotic optimization algorithm function, r2, r3, r4 t It represents a random value in the interval [0,1], and CWEP is the probability of existence of a convergent wormhole;

[0055] Among them, r4 t =a*r4 t-1 (1-r4 t-1 ) is the update formula of r4 obtained after introducing the Logistic chaotic mapping function.

[0056] Furthermore, the travel distance rate TDR travel-distance The role of is to determine the distance the solution vector moves from the current position to the new position.

[0057] Furthermore, constructing a support vector machine model SVM based on the improved multiverse optimization algorithm IMVO, and obtaining the IMVO-SVM model also includes the following steps:

[0058] Step S301, load the support vector machine SVM (hereinafter referred to as SVM for short), first introduce the Lagrange multiplier α into the SVM i Transformed into the optimization problem of Lagrangian function, the SVM objective function is obtained:

[0059] Then, the Lagrange multiplier α is introduced into the equality constraint of the dual problem. i , and get the formula group of the objective function:

[0060] In the formula group:

[0061] α i is a Lagrange multiplier and 0≤α i ≤C,α i It can also represent the contribution of training sample i to the decision boundary by solving α i The optimal decision boundary of SVM can be found;

[0062] y is the label of sample i, usually taking a value of ±1, indicating the category to which the sample belongs;

[0063] x is the feature vector of sample i, representing the input data of the sample;

[0064] m is the number of training samples;

[0065] C is a penalty factor used to control the degree of relaxation of the soft interval, which balances the training error rate and the complexity of the decision boundary;

[0066] Step S302, change the inner product x i ·x j , define x i and x j The mapping φ is used as the transformation relationship from low-dimensional sample space to high-dimensional space, which transforms the nonlinear problem into a linear problem. After the transformation, the optimization objective function becomes:

[0067] Step S303, introducing a kernel function K(x,z) in the SVM to map the linearly inseparable data in the low-dimensional space to the high-dimensional space to make it linearly separable, and the calculation formula is: K(x,z)=φ(x)·φ(z); wherein x and z are both feature vectors;

[0068] Optionally, the kernel function is a Gaussian kernel function K1(x,z), which is calculated as follows:

[0069] K1(x,z)=exp(-y||xz||2), where γ is a hyperparameter of the Gaussian kernel function, which is used to control the width of the Gaussian kernel function. The larger the γ value, the smaller the influence range of the Gaussian kernel function K1(x,z); the smaller the γ value, the larger the influence range of the Gaussian kernel function K1(x,z);

[0070] Step S304, loading the improved multiverse optimization algorithm IMVO (hereinafter referred to as IMVO), setting the maximum number of iterations of IMVO, the number of universes, the penalty factor of SVM and the value range of the kernel parameter, etc.; then initializing the universe position, in IMVO, each universe and the individual in the universe represents a group (C, g), which is randomly initialized according to the parameter limit range, where g is the kernel parameter and C is the penalty factor. The penalty factor is an important hyperparameter of SVM and is used to control overfitting and underfitting;

[0071] Step S305, calculate the expansion rate of the current universe and sort it, and then select a white hole through the roulette principle (i.e., roulette algorithm); update the convergent wormhole existence probability CWEP and travel distance rate TDR in IMVO travel-distance , and perform boundary checking;

[0072] Step 306, update the expansion rate of the universe. If the expansion rate of the universe is better than the current expansion rate of the universe, update the current universe to save the expansion rate. Otherwise, keep the current universe. Then update the position of the universe and find the optimal individual in the optimal universe.

[0073] Step 307, check whether the convergence condition is met, if not, continue iteration and update the universe position; if the convergence condition is met, output the corresponding result to obtain the support vector machine model optimized based on the improved multiverse optimization algorithm, that is, the IMVO-SVM model;

[0074] Furthermore, the convergence condition is to find a good enough universe or reach a maximum number of iterations;

[0075] It should be noted that most of the mechanisms of MVO are inherited in the IMVO-SVM model, including white holes, black holes, wormholes, and the expansion rate of the universe. Since these mechanisms are existing technologies, only the following concepts are briefly introduced here:

[0076] The white hole mechanism is used to generate new solutions in the universe;

[0077] The black hole mechanism is used to discard bad features in the current solution of the universe;

[0078] The wormhole mechanism is used to increase diversity and randomly change the current solution to help escape from the local optimum.

[0079] The expansion rate, whose changes affect the generation of white holes, black holes, and wormholes in the IMVO algorithm, is of great significance to the update of the individual positions of the universe and the balance and stability of the entire multiverse space; and the expansion rate of each universe is positively correlated with the related fitness function; the relationship between the expansion rate and white holes, black holes, and wormholes is as follows:

[0080] ①The faster the universe expands, the greater the probability of white holes appearing;

[0081] ②The higher the expansion rate of the universe, the lower the probability of black holes appearing;

[0082] ③ When the universe has a high expansion rate, it will cause objects in the universe to move in white holes;

[0083] ④ When the universe has a low expansion rate, it will cause objects in the universe to be received from black holes;

[0084] ⑤ Ignoring the expansion rate, all objects can move to the optimal universe through wormholes.

[0085] Furthermore, the process of using the distribution network electrical characteristic index system to train the IMVO-SVM model to obtain the distribution network line loss prediction model also includes the following steps:

[0086] Step 308, initializing the IMVO-SVM model parameters, including the number of iterations, universe positions, and the number of universes, and then loading the training data set into the IMVO-SVM model;

[0087] Step 309, the IMVO-SVM model continuously updates the expansion rate of the universe, the convergent wormhole existence probability CWEP, and the travel distance rate TDR based on the number of iterations. travel-distance And parameters such as compression factor τ;

[0088] Step 3010, the IMVO-SVM model selects the minimum value of the fitness value as the optimal solution, repeats step 3010 until the number of iterations ends or a sufficiently good universe is found, saves the parameters of the IMVO-SVM under the universe and outputs the distribution network line loss prediction model;

[0089] Step 3011, testing the precision and accuracy of the distribution network line loss prediction model based on the test data set, and evaluating the distribution network line loss prediction model by calculating the mean square error MSE and the determination coefficient R2:

[0090] The calculation formula of mean square error MSE is:

[0091] The calculation formula of the determination coefficient R2 is:

[0092] Among them, y i is the actual value of line loss, is the line loss prediction value, m is the number of wiring, is the average value of line loss; the comparison results with BP neural network are shown in the following table:

[0093]

[0094] Based on the comparison of the mean square error and determination coefficient data of the prediction values ​​of the support vector machine optimized by the improved multiverse optimization algorithm (IMVO-SVM) and the BP neural network, the results on both data sets show that the line loss prediction model of the IMVO-SVM has a smaller mean square error MSE and a larger determination coefficient R2, indicating that the model proposed in the present invention has higher prediction accuracy.

[0095] Although the present invention is illustrated and described by specific embodiments and their alternatives, it should be understood that various changes and modifications within the spirit and scope of the present invention can be implemented. Therefore, it should be understood that the present invention is not limited in any sense except by the limitations of the attached claims and their equivalents.

Claims

1. A distribution network line loss prediction method based on IMVO-SVM, characterized in that: It includes the following steps: Step 1: Use grey correlation analysis to analyze multiple electrical indicators in the historical line loss rate records, obtain the correlation between different electrical indicators and distribution network line loss, and then establish a distribution network electrical characteristic indicator system as training data; Step 2, setting an algorithm optimization strategy, and using the algorithm optimization strategy to optimize the multiverse optimization algorithm MVO to obtain an improved multiverse optimization algorithm IMVO; Step 3, construct a support vector machine model SVM based on the improved multiverse optimization algorithm IMVO to obtain an IMVO-SVM model, and use the distribution network electrical characteristic index system to train the IMVO-SVM model to obtain a distribution network line loss prediction model; Step 4: predict the distribution network line loss based on the distribution network line loss prediction model.

2. The distribution network line loss prediction method based on IMVO-SVM according to claim 1 is characterized in that: The historical line loss rate records electrical indicators that affect the line loss rate, and the electrical indicators include active power supply, reactive power supply, total transformer capacity, transformer active power, transformer reactive power, total line length, line equivalent cross-sectional area, trunk line cross-sectional area, total trunk line length and total branch line length.

3. A distribution network line loss prediction method based on IMVO-SVM according to claim 2, characterized in that: Step 1 is divided into the following steps: Step 101, analyzing the historical line loss rate, and preliminarily selecting several electrical indicators that have the greatest impact on the line loss according to the change of the line loss rate; Step 102, taking the theoretical line loss change of the distribution network as a reference sequence, and taking the change of each electrical index as a different comparison sequence; Step 103, using an extreme dimensionless processing method to process the reference sequence and each comparison sequence, and then calculating the grey correlation coefficient between each comparison sequence and the reference sequence; Step 104, calculate the grey correlation degree based on the grey correlation coefficient, then compare and sort the grey correlation degrees, and select several electrical indicators based on the grey correlation degree to establish a distribution network electrical characteristic indicator system as training data.

4. A distribution network line loss prediction method based on IMVO-SVM according to claim 1, characterized in that , in step 2, the following steps are also included: Step 201, setting an algorithm optimization strategy, the algorithm optimization strategy includes strategy A and strategy B; Step 202, introducing the Logistic chaos mapping function to solve the problem that the universe position update algorithm in the multiverse optimization algorithm MVO is prone to fall into the local extreme value, and obtaining the Logistic chaos mapping universe position algorithm; Step 203, combining the Logistic chaos mapping universe position algorithm, strategy A and strategy B to obtain an improved multiverse optimization algorithm IMVO.

5. A distribution network line loss prediction method based on IMVO-SVM according to claim 4, characterized in that ,The strategy A is to adjust the wormhole existence probability WEP in a linear growth mode to the convergent wormhole existence probability CWEP in a logarithmic growth mode. The calculation formula of the convergent wormhole existence probability CWEP is: Among them, t refers to the current number of iterations, T refers to the maximum number of iterations set, and WEP max Refers to the maximum probability of the existence of a wormhole, WEP min Refers to the minimum probability of the existence of a wormhole; WEP max and WEP min It is usually a constant set by the algorithm, which is used to represent the upper and lower limits of exploration behavior; The strategy B is to introduce a compression factor τ, and the calculation formula of the compression factor τ is: Among them, t refers to the current number of iterations, T refers to the maximum number of iterations, and e is a natural constant.

6. A distribution network line loss prediction method based on IMVO-SVM according to claim 5, characterized in that: The improved multiverse optimization algorithm IMVO obtained by combining the Logistic chaos mapping universe position algorithm, strategy A and strategy B is expressed by the following function group: Where, X best-universe(j) is the current optimal solution found in the jth dimension, τ is the compression factor, TDR travel-distance is the travel distance rate, ub j is the upper bound of the j-th dimension of the search space, lb j is the lower bound of the jth dimension of the search space, COA(t) is the chaotic optimization algorithm function, r2, r3, r4 t It represents a random value in the interval [0,1], and CWEP is the probability of existence of a convergent wormhole; Among them, r4 t =a*r4 t-1 (1-r4 t-1 ) is the update formula of r4 obtained after introducing the Logistic chaotic mapping function.

7. A distribution network line loss prediction method based on IMVO-SVM according to claim 1, characterized in that: Based on the improved multiverse optimization algorithm IMVO, a support vector machine model SVM is constructed, and the IMVO-SVM model also includes the following steps: Step 301, load the support vector machine SVM and introduce the Lagrange multiplier α i Get the formula group of the objective function: Step 302, change the inner product x i ·x j , define x i and x j The mapping φ is used as the transformation relationship from low-dimensional sample space to high-dimensional space; Step 303, introducing a kernel function K(x,z) into the SVM to map the linearly inseparable data in the low-dimensional space to the high-dimensional space to make it linearly separable; Step 304, loading the improved multiverse optimization algorithm IMVO, setting the maximum number of iterations of IMVO, the number of universes, the penalty factor of SVM and the value range of the kernel parameter; Step 305, calculate the expansion rate of the current universe and sort it, and then select a white hole through the roulette principle; update the convergent wormhole existence probability CWEP and travel distance rate TDR in IMVO travel-distance , and perform boundary checking; Step 306, update the universe expansion rate. If the universe expansion rate is better than the current universe expansion rate, update the current universe to save the expansion rate. Otherwise, keep the current universe. Then update the universe position and find the optimal solution X in the optimal universe. best-universe(j) ; Step 307, check whether the convergence condition is met. If not, continue iterating and update the universe position. If the convergence condition is met, output the corresponding result to obtain the support vector machine model optimized based on the improved multiverse optimization algorithm, that is, the IMVO-SVM model.

8. The method for predicting line loss in a distribution network based on IMVO-SVM according to claim 7 is characterized in that: The distribution network line loss prediction model is evaluated by calculating the mean square error MSE and the determination coefficient R2. The calculation formula of the mean square error MSE is: The calculation formula of the determination coefficient R2 is: Among them, y i is the actual value of line loss, is the line loss prediction value, m is the number of wiring, is the average line loss value.

Images