A method for calculating reasonable interval of line loss in active area based on GRNN-PCA
Through the GRNN-PCA-based method, the active station area influencing factor is extracted, the GRNN network is trained for line loss calculation, and the high loss factor is analyzed using PCA to solve the problem of excessive data requirements in the existing technology, and the accurate calculation of reasonable ranges of line loss in the station area and the identification of high loss factors are achieved.
Patent Information
- Application Number
- CN202111107129.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-22
- Publication Date
- 2025-06-06
- Estimated Expiration
- 2041-09-22
AI Technical Summary
When the prior art calculates the line loss of a new active platform with distributed energy access, it is difficult to adapt to the random behavior of distributed energy, and the requirements for low-voltage platform network, distributed energy parameters and operating data are too high, resulting in technical bottlenecks in computing accuracy.
The active platform area line loss reasonable interval calculation method based on GRNN-PCA is adopted. The GRNN network is trained to calculate the line loss by extracting the platform area influence factor, and the high loss factor is analyzed in combination with PCA to identify and adjust the high loss factor.
Under the current technical conditions, the reasonable range of the station area line loss is more accurately calculated through existing feasible data and the factors affecting high losses are accurately identified, and the defects of traditional methods that require too high data requirements are overcome.
Smart Images
Figure CN114004273B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of application of artificial intelligence algorithms in power systems, and relates to a method for calculating a reasonable interval of line loss in an active transformer area, in particular to a method for calculating a reasonable interval of line loss in an active transformer area based on GRNN-PCA. Background Art
[0002] The line loss rate of the substation is a comprehensive technical indicator of the power company. As an important part of the line loss calculation, it involves various aspects such as distribution network planning, operation, maintenance, marketing electricity, metering, and reading and collection. It fully reflects the power company's calculation level of low-voltage substation equipment and users.
[0003] The reasonable value estimation of line loss in the substation is a non-physical dynamic line loss reference value introduced to guide the quantitative loss reduction target of the physical substation. If the real-time line loss is detected to be outside the reasonable range of theoretical calculation, the high-loss influencing factor identification method based on the line loss contribution should be used to study the interaction between high-loss influencing factors, and conduct high-loss factor investigation and control.
[0004] Traditional methods for estimating the reasonable value of line loss in low-voltage areas include the area loss rate method, voltage loss rate method, equivalent resistance method, and power flow method. However, for new active areas with distributed energy access, these methods have the following two major problems: First, considering the random behavior of distributed energy, the traditional line loss calculation method is no longer applicable; second, the accurate calculation of the existing improved algorithm requires too high requirements on the low-voltage area network, distributed energy parameters, and operating data. In the case of technical bottlenecks in the acquisition and calculation accuracy of low-voltage active area parameters, how to use more advanced algorithms through existing feasible data to realize the calculation of reasonable intervals of area line loss and the identification of high-loss influencing factors is the focus of active area line loss management.
[0005] After searching, no public documents of the prior art that are identical or similar to the present invention were found. Summary of the invention
[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a method for calculating the reasonable interval of line loss in active substations based on GRNN-PCA, which can realize the calculation of the reasonable interval of line loss in substations and the identification of high-loss influencing factors by using more advanced algorithms through existing feasible data.
[0007] The present invention solves the practical problem by adopting the following technical solutions:
[0008] A method for calculating a reasonable interval of line loss in an active area based on GRNN-PCA comprises the following steps:
[0009] Step 1, extract the impact factor of active area;
[0010] Step 2: Perform multi-level classification according to the capacity of the selected active area, and each type of active area includes the active area impact factor described in step 1;
[0011] Step 3: Train the GRNN network, perform GRNN line loss calculation on each level of the substation in step 2, and obtain the GRNN training results;
[0012] Step 4: Use the test set to calculate the relative error of the GRNN training result accuracy;
[0013] Step 5: If the relative error obtained in step 4 is less than the set value, the reasonable line loss interval is selected in combination with the line loss change interval calculated by GRNN. Otherwise, return to step 3 for retraining, and repeat steps 3 to 5 until the requirements are met.
[0014] Moreover, the specific method of step 1 is:
[0015] For the calculation of theoretical line loss in active substations, basic line characteristic factors that can be directly detected by the substation master meter and have an impact on line loss are extracted, including power variance, three-phase imbalance, and load rate; and characteristic factors of photovoltaic distributed power supply operation are extracted, including photovoltaic daily power generation, photovoltaic grid-connected location, and number of photovoltaic users.
[0016] Moreover, the specific steps of step 3 include:
[0017] (1) Determine the radial basis function center of the hidden layer neuron
[0018] Assume that the training set sample input matrix is P and the output matrix is T
[0019]
[0020]
[0021] In the formula, p ij represents the i-th input of the j-th training sample; t ij represents the i-th input variable of the j-th training sample; R is the dimension of the input variable; S is the dimension of the output variable; Q is the number of samples in the training set;
[0022] Each neuron in the hidden layer corresponds to a training sample, that is, the radial basis function center corresponding to the Q hidden layer neurons is:
[0023] C=P' (8)
[0024] (2) Determine the hidden layer neuron threshold
[0025] The threshold corresponding to the Q hidden layer neurons is set as:
[0026] b 1 =[b11 ,b 12 ,…,b 1Q ]' (9)
[0027] In the formula, spread is the expansion speed of the radial basis function;
[0028] (3) Determine the weights between the hidden layer and the output layer
[0029] When the radial basis function center and threshold of the hidden layer neurons are determined, the output a of the hidden layer neurons i It can be calculated as follows:
[0030] a i =exp(-||Cp i || 2 b 1 ) (10)
[0031] In the formula, p i =[p i1 ,p i2 ,…,p iR ]' is the i-th training sample vector, and record
[0032] The connection weight W between the hidden layer and the input layer in GRNN is taken as the output matrix of the training set, that is,
[0033] W=T (11)
[0034] (4) Output layer neuron calculation
[0035] When the connection weights between the hidden layer and the output layer neurons are determined, the normalized dot product weight function normprod is used to calculate the output of the output layer neurons, namely:
[0036]
[0037] Among them, LW 2,1 is the output layer weight matrix.
[0038] is the predicted value of the line loss in the ith area, the activation function is selected as the linear transfer function purelin, and the GRNN line loss calculation of the area is completed, that is:
[0039]
[0040] Moreover, the specific method of step 4 is: select relative error as the evaluation index:
[0041]
[0042] In the formula, yi (i=1,2,…,n) is the true value of line loss in the ith substation.
[0043] Moreover, after step 5, the method further includes the following steps:
[0044] Step 6: Use the GRNN network trained in step 3 to predict the line loss of the area to be identified. If the predicted line loss is not within the reasonable range of step 5, the area is identified as a high-loss area.
[0045] Step 7, perform principal component analysis of influencing factors on the high-loss area identified in step 6;
[0046] Step 8: Conduct principal component analysis of the influencing factors of the high-damage areas obtained in step 7 and manage the high-damage areas.
[0047] Moreover, the specific method of step 7 is:
[0048] (1) Standardization of samples from high-loss areas
[0049] The characteristic matrix of the high-loss area is X:
[0050]
[0051] In the formula, x ij It represents the jth input feature of the i-th high-loss sample, U is the dimension of the input feature, and T is the number of high-loss areas.
[0052] Standardized transformation of high-loss area characteristics:
[0053]
[0054] in
[0055] (2) Calculate the correlation coefficient matrix for the above standardized matrix Z
[0056]
[0057] Where R is the correlation coefficient matrix, X U =[x 1 ,x 2 ,…,x U ],Z=[z ij ] T×U .
[0058] (3) Determine the principal components
[0059] Solve the characteristic equation of the sample correlation coefficient matrix R | R-λI U |=0 to get U characteristic roots, and determine the value of m according to the following formula so that the information utilization rate is above 85%
[0060]
[0061] For each λ j , j = 1, 2, ..., m, solve the equation system Rb = λ j b gets the unit eigenvector
[0062] (4) Convert standardized indicators into principal components
[0063]
[0064] V i =[v i1 ,v i2 ,...,v iU ] is the i-th principal component, i=1,2,…,U.
[0065] (5) Analyze the contribution of each factor to each principal component
[0066] Moreover, the specific steps of step 8 include:
[0067] (1) A correlation analysis is performed on the high loss factor that has a greater impact on the first principal component, and the high loss factor that is most convenient for the user to adjust is adjusted.
[0068] (2) The GRNN trained in step 3 is used again to predict the line loss of the adjusted high-loss samples. If the line loss meets the reasonable range set in step 5, the high-loss area management is completed; otherwise, a correlation analysis is performed on the high-loss factor that has a greater impact on the next principal component, and the high-loss factor that is most convenient for user-side adjustment is adjusted until the GRNN prediction value of the high-loss area sample meets the reasonable range set in step 5. Usually, the third or fourth principal component can meet the requirements.
[0069] Advantages and beneficial effects of the present invention:
[0070] 1. The present invention provides a method for calculating the reasonable interval of active area line loss based on GRNN-PCA, predicts the reasonable interval of line loss based on historical data, detects the real-time line loss situation, finds the high loss moment and analyzes the high loss factors. The present invention overcomes the defects that the existing algorithms for calculating active area line loss have too high requirements on low-voltage area network, distributed energy parameters and network architecture data, and it is difficult to realize the reasonable interval calculation of area line loss and accurate identification of high loss influencing factors through existing feasible data when there are technical bottlenecks in the acquisition and calculation accuracy of low-voltage active area line parameters.
[0071] 2. The present invention fully considers the basic operation properties of the substation and the distributed energy grid-connected properties, and proposes a method for calculating the reasonable interval of line loss in the substation based on generalized neural network + principal component analysis (GRNN + PCA). In the early stage of the present invention, the line loss influencing factors of the active substation are extracted as GRNN feature input, training is performed, and the error analysis between the training results and the actual line loss is performed; in the mid-term, the trained GRNN is used to predict the reasonable interval of line loss, and the high loss is identified for the test samples containing high-loss substations; in the later stage, PCA is used to perform correlation analysis on the influencing factors of the high-loss substation, the most convenient high-loss factor for user-side adjustment is adjusted, and GRNN is used for post-adjustment inspection. Through experimental data analysis, the GRNN trained for the active substation of the present invention has good generalization ability and can realize the calculation of reasonable intervals of the substation. BRIEF DESCRIPTION OF THE DRAWINGS
[0072] Figure 1 It is a GRNN training structure diagram of the present invention;
[0073] Figure 2 It is a flow chart of algorithm implementation of the present invention;
[0074] Figure 3 This is a comparison result diagram of the GRNN verification of the present invention between the predicted value of the central line loss and the actual value;
[0075] Figure 4 It is a schematic diagram of a reasonable range of line loss in a substation area calculated by GRNN of the present invention;
[0076] Figure 5 This is a diagram of the treatment results of the high-loss area after PCA analysis of the present invention. DETAILED DESCRIPTION
[0077] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings:
[0078] A method for calculating the reasonable interval of line loss in active area based on GRNN-PCA, such as Figure 2 As shown, the following steps are included:
[0079] Step 1, extract the impact factor of active area;
[0080] The specific method of step 1 is:
[0081] For the calculation of theoretical line loss in active substations, we first extracted the basic line characteristic factors that can be directly detected by the substation master meter and have an impact on line loss, mainly including power variance, three-phase imbalance, and load rate. Secondly, we extracted the operating characteristic factors of photovoltaic distributed power sources, mainly including photovoltaic daily power generation, photovoltaic grid-connected location, and the number of photovoltaic users.
[0082] In this embodiment, 1) for the calculation of the theoretical line loss of the active substation, the basic line characteristic factors affecting the line loss are first extracted, mainly including: power variance, three-phase imbalance, and load rate.
[0083] (1) Power variance p 1 :The power fluctuation of the total meter in the area
[0084]
[0085] Where P i is the daily i-th detection power of the total meter in the area, is the average power detected by the total meter in the area every day, and n is the number of detections of the total meter in the area every day.
[0086] (2) Three-phase imbalance p 2 :Three-phase load balance in the substation area
[0087]
[0088] In the formula, I maxφ is the maximum phase load current at the detection point in the substation area, Refers to the average phase load current at the detection point in the substation area, and
[0089] (3) Load factor p 3 :Daily power supply situation in the substation area.
[0090] p 3 =W / 24S 变N (3)
[0091] Where W is the daily power supply in the substation, S 变N The rated capacity of the transformer in the substation
[0092] 2) Secondly, the operating characteristic factors of photovoltaic distributed power sources were extracted, mainly including: photovoltaic daily power generation, photovoltaic grid-connected location, and number of photovoltaic users.
[0093] (1) Daily photovoltaic power generation p 4 : Total daily power generation of photovoltaic distributed power sources in the area
[0094]
[0095] Where P PVij is the photovoltaic power of the i-th photovoltaic power source at the jth second, and m is the number of photovoltaic power sources in the area.
[0096] (2) Percentage of photovoltaic users p 5 :The proportion of photovoltaic users in the area
[0097] p 5 =NPV / N (5)
[0098] Where N PV is the number of users who directly consume photovoltaic power in the substation area, and N is the total number of users in the substation area;
[0099] (3) Photovoltaic grid-connected position p 6 : The grid-connected position indicates the distance between the photovoltaic grid-connected point and the busbar.
[0100] Step 2: Perform multi-level classification according to the selected active area capacity, and each type of active area includes the active area influencing factors described in step 1, namely: line basic characteristic factors and photovoltaic distributed power generation operation characteristic factors.
[0101] In this embodiment, the active area classification in step 2 is classified into four levels according to the selected area capacity, as shown in Table 1:
[0102] Table 1 Transformer capacity classification in the substation area
[0103]
[0104]
[0105] Step 3: Train the GRNN network, perform GRNN line loss calculation on each level of the substation in step 2, and obtain the GRNN training results;
[0106] The GRNN structure is as follows Figure 1 As shown, the specific steps of step 3 include:
[0107] (1) Determine the radial basis function center of the hidden layer neuron
[0108] Assume that the training set sample input matrix is P and the output matrix is T
[0109]
[0110]
[0111] In the formula, p ij represents the i-th input of the j-th training sample; t ij represents the i-th input variable of the j-th training sample; R is the dimension of the input variable; S is the dimension of the output variable; Q is the number of samples in the training set;
[0112] Each neuron in the hidden layer corresponds to a training sample, that is, the radial basis function center corresponding to the Q hidden layer neurons is:
[0113] C=P' (8)
[0114] (2) Determine the hidden layer neuron threshold
[0115] The threshold corresponding to the Q hidden layer neurons is set as:
[0116] b 1 =[b 11 ,b 12 ,…,b 1Q ]' (9)
[0117] In the formula, spread is the expansion speed of the radial basis function;
[0118] (3) Determine the weights between the hidden layer and the output layer
[0119] When the radial basis function center and threshold of the hidden layer neurons are determined, the output a of the hidden layer neurons i It can be calculated as follows:
[0120] a i =exp(-||Cp i || 2 b 1 ) (10)
[0121] In the formula, p i =[p i1 ,p i2 ,…,p iR ]' is the i-th training sample vector, and record
[0122] The connection weight W between the hidden layer and the input layer in GRNN is taken as the output matrix of the training set, that is,
[0123] W=T (11)
[0124] (4) Output layer neuron calculation
[0125] When the connection weights between the hidden layer and the output layer neurons are determined, the normalized dot product weight function normprod is used to calculate the output of the output layer neurons, namely:
[0126]
[0127] Among them, LW 2,1 is the output layer weight matrix.
[0128] is the predicted value of the line loss in the ith area, the activation function is selected as the linear transfer function purelin, and the GRNN line loss calculation of the area is completed, that is:
[0129]
[0130] Step 4: Use the test set to evaluate the accuracy of the GRNN training results, and obtain the evaluation results. The relative error is used as the evaluation indicator:
[0131]
[0132] In the formula, y i (i=1,2,…,n) is the true value of line loss in the ith substation.
[0133] Step 5: If the relative error obtained in step 4 is less than the set value (20%), the reasonable line loss interval is selected in combination with the line loss change interval calculated by GRNN. Otherwise, return to step 3 for retraining, and repeat steps 3 to 5 until the requirements are met.
[0134] Step 6: Use the GRNN network trained in step 3 to predict the line loss of the area to be identified. If the predicted line loss is not within the reasonable range of step 5, the area is identified as a high-loss area.
[0135] Step 7, perform principal component analysis of influencing factors on the high-loss area identified in step 6;
[0136] The specific method of step 7 is:
[0137] (1) Standardization of samples from high-loss areas
[0138] The characteristic matrix of the high-loss area is X:
[0139]
[0140] In the formula, x ij It represents the jth input feature of the i-th high-loss sample, U is the dimension of the input feature, and T is the number of high-loss areas.
[0141] Standardized transformation of high-loss area characteristics:
[0142]
[0143] in
[0144] (2) Calculate the correlation coefficient matrix for the above standardized matrix Z
[0145]
[0146] Where R is the correlation coefficient matrix, X U =[x 1 ,x 2 ,…,x U ],Z=[z ij ] T×U .
[0147] (3) Determine the principal components
[0148] Solve the characteristic equation of the sample correlation coefficient matrix R | R-λI U |=0 to get U characteristic roots, and determine the value of m according to the following formula so that the information utilization rate is above 85%
[0149]
[0150] For each λ j , j = 1, 2, ..., m, solve the equation system Rb = λ j b gets the unit eigenvector
[0151] (4) Convert standardized indicators into principal components
[0152]
[0153] V i =[v i1 ,v i2 ,...,v iU ] is the i-th principal component, i=1,2,…,U.
[0154] (5) Analyze the contribution of each factor to each principal component
[0155] Step 8: Conduct principal component analysis of the influencing factors of the high-damage areas obtained in step 7, and manage the high-damage areas;
[0156] The specific steps of step 8 include:
[0157] (1) A correlation analysis is performed on the high loss factor that has a greater impact on the first principal component, and the high loss factor that is most convenient for the user to adjust is adjusted.
[0158] (2) Reuse the GRNN trained in step 3 to predict line loss for the adjusted high-loss samples. If the line loss meets the reasonable range set in step 5, the high-loss area management is completed. Otherwise, perform correlation analysis on the high-loss factors that have a greater impact on the next principal component, and adjust the high-loss factors that are most convenient for user-side adjustment until the GRNN prediction value of the high-loss area sample meets the reasonable range set in step 5, usually the third or fourth principal component can meet the requirements.
[0159] The GRNN-PCA-based active area reasonable interval calculation method of the present invention is used to predict the reasonable interval of GRNN area line loss and manage the high line loss of PCA in a photovoltaic active area (level 3 area) in Tianjin, my country. 100 test samples and 100 test samples in the experimental results are selected as the results. The GRNN area line loss prediction result is compared with the true value. Figure 3 As shown, the reasonable interval selection is as follows Figure 4 As shown, the treatment results after high loss factor PCA analysis are as follows Figure 5 shown.
[0160] Figure 3 It can be seen that the calculation results basically match the true value well. Through the relative error analysis in step 4, the relative error of the GRNN prediction of the demonstration sample is 21.3%, indicating that the invention has good line loss calculation ability. Figure 4 It shows that the high-loss area is obviously outside the reasonable line loss range, and the GRNN recognition value is consistent with the true value, which shows that the invention has good high-loss recognition ability. Figure 5 It shows that after PCA high-loss management, the predicted loss value of the original high-loss area line has fallen within a reasonable range, indicating that the invention has good high-loss management capabilities.
[0161] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0162] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0163] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0164] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
Claims
1. A method for calculating the reasonable interval of line loss in active area based on GRNN-PCA. The following steps are involved: Step 1, extract the active area impact factor; Step 2: Perform multi-level classification according to the capacity of the selected active area, and each type of active area includes the active area impact factor described in step 1; Step 3: Train the GRNN network, perform GRNN line loss calculation on each level of the substation in step 2, and obtain the GRNN training results; Step 4: Use the test set to calculate the relative error of the GRNN training result accuracy; Step 5: If the relative error obtained in step 4 is less than the set value, the reasonable line loss interval is selected in combination with the line loss change interval calculated by GRNN, otherwise return to step 3 for retraining, and repeat steps 3 to 5 until the requirements are met; After step 5, the following steps are also included: Step 6: Use the GRNN network trained in step 3 to predict the line loss of the area to be identified. If the predicted line loss is not within the reasonable range of step 5, the area is identified as a high-loss area. Step 7, perform principal component analysis of influencing factors on the high-loss area identified in step 6; Step 8: Conduct principal component analysis of the influencing factors of the high-damage areas obtained in step 7, and manage the high-damage areas; The specific steps of step 8 include: (1) Conduct correlation analysis on the high loss factor that has a high impact on the first principal component, and adjust the high loss factor that is most convenient for the user to adjust; (2) The GRNN trained in step 3 is used again to predict the line loss of the adjusted high-loss samples. If the line loss meets the reasonable range set in step 5, the high-loss area management is completed; otherwise, a correlation analysis is performed on the high-loss factor with the highest impact on the next principal component, and the high-loss factor that is most convenient for user-side adjustment is adjusted until the GRNN prediction value of the high-loss area sample meets the reasonable range set in step 5. Usually, the third or fourth principal component can meet the requirements.
2. According to the method for calculating the reasonable interval of active area line loss based on GRNN-PCA according to claim 1, The following steps are involved: The specific method of step 1 is: For the calculation of theoretical line loss in active substations, basic line characteristic factors that can be directly detected by the substation master meter and have an impact on line loss are extracted, including power variance, three-phase imbalance, and load rate; and characteristic factors of photovoltaic distributed power generation operation are extracted, including photovoltaic daily power generation, photovoltaic grid-connected location, and number of photovoltaic users.
3. According to the GRNN-PCA-based method for calculating the reasonable interval of active area line loss according to claim 1, The following steps are involved: The specific steps of step 3 include: (1) Determine the radial basis function center of the hidden layer neuron Assume that the training set sample input matrix is P and the output matrix is T In the formula, p ij represents the i-th input of the j-th training sample; t ij represents the i-th output variable of the j-th training sample; R is the dimension of the input variable; S is the dimension of the output variable; Q is the number of samples in the training set; Each neuron in the hidden layer corresponds to a training sample, that is, the radial basis function center corresponding to the Q hidden layer neurons is: C=P' (8) (2) Determine the hidden layer neuron threshold The threshold corresponding to the Q hidden layer neurons is set as: b 1 =[b 11 ,b 12 ,…,b 1Q ]' (9) In the formula, spread is the expansion speed of the radial basis function; (3) Determine the weights between the hidden layer and the output layer When the radial basis function center and threshold of the hidden layer neurons are determined, the output a of the hidden layer neurons i It can be calculated as follows: a i =exp(-||C-p i || 2 b 1 ) (10) In the formula, p i =[p i1 ,p i2 ,…,p iR ]' is the i-th training sample vector, and record The connection weight W between the hidden layer and the input layer in GRNN is taken as the output matrix of the training set, that is, W=T (11) (4) Output layer neuron calculation When the connection weights between the hidden layer and the output layer neurons are determined, the normalized dot product weight function normprod is used to calculate the output of the output layer neurons, namely: Among them, LW 2,1 is the output layer weight matrix; is the predicted value of the line loss in the ith area, the activation function is selected as the linear transfer function purelin, and the GRNN line loss calculation of the area is completed, that is:
4. According to claim 1, a method for calculating the reasonable interval of active area line loss based on GRNN-PCA, The following steps are involved: The specific method of step 4 is: select relative error as the evaluation index: In the formula, y i (i=1,2,…,n) is the true value of line loss in the ith substation.
5. According to the method for calculating the reasonable interval of active area line loss based on GRNN-PCA according to claim 1, The following steps are involved: The specific method of step 7 is: (1) Standardization of samples from high-loss areas The characteristic matrix of the high-loss area is X: In the formula, x ij represents the jth input feature quantity of the i-th high-loss sample, U is the dimension of the input feature, and T is the number of high-loss areas; Standardized transformation of high-loss area characteristics: in (2) Calculate the correlation coefficient matrix for the standardized matrix Z Where R is the correlation coefficient matrix, X U =[x 1 ,x 2 ,…,x U ],Z=[z ij ] T×U ; (3) Determine the principal components Solve the characteristic equation of the sample correlation coefficient matrix R | R-λI U |=0 to get U characteristic roots, and determine the value of m according to the following formula so that the information utilization rate is above 85% For each λ j , j = 1, 2, ..., m, solve the equation system Rb = λ j b gets the unit eigenvector (4) Convert standardized indicators into principal components V i =[v i1 ,v i2 ,...,v iU ] is the i-th principal component, i = 1, 2, …, U; (5) Analyze the contribution of each factor to each principal component.