Method for evaluating reactive voltage indexes of power distribution network under different operation conditions

Abstract

The invention discloses a power distribution network reactive voltage index evaluation method under different operation conditions, and relates to the field of voltage index evaluation, and the method comprises the following steps: S1, carrying out the division according to different operation states of a power distribution network, and formulating the corresponding operation conditions; s2, calculating six indexes including voltage deviation, a voltage fluctuation coefficient, a three-phase imbalance coefficient, a harmonic component ratio, frequency deviation and a power factor PF to construct a data set, and establishing and using a sparse encoder and a lasso regression model to determine a calculation mode of a voltage quality comprehensive index value; and S3, aiming at the operation condition of the power distribution network, considering the voltage quality comprehensive index value, and integrating four indexes including the line loss rate, the short circuit ratio and the voltage qualification rate to evaluate the reactive voltage of the power distribution network under the condition. According to the method, the operation health conditions of the power distribution network under various working conditions can be quickly identified, and a scientific basis is provided for optimization and adjustment; and an operation strategy of the power distribution network is guided in actual operation, and stable and efficient operation of the system is ensured.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of voltage index evaluation, and specifically to a method for evaluating reactive voltage index of a distribution network under different operating conditions. Background Art

[0002] The distribution network is an important part of the power system, and its operating status directly affects the power supply quality and the stability of the power system. During the operation of the distribution network, the reasonable control of reactive voltage is of great significance for improving the power factor of the system, reducing line losses, improving voltage quality, and increasing the utilization rate of power equipment.

[0003] Traditional reactive voltage control of distribution network mainly relies on manual experience, and is achieved by adjusting the reactive compensation equipment of substations and transformer taps of distribution lines. The reactive regulation speed of related methods is slow and the accuracy is low. In recent years, the concept of intelligent distribution network has been proposed. By introducing advanced information technology and intelligent control technology, the reactive voltage of distribution network can be managed automatically and intelligently. However, the evaluation index of reactive voltage of distribution network is not systematically established in related methods. Only a few indicators such as line loss and voltage qualification rate are often considered, and a comprehensive evaluation system has not been established. Summary of the invention

[0004] Based on this, the purpose of the present invention is to provide a method for evaluating reactive voltage indicators of a distribution network under different operating conditions to solve the technical problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for evaluating reactive voltage index of a distribution network under different operating conditions, the method comprising the following steps:

[0007] S1. Formulate the test conditions: divide the distribution network according to different operating conditions and formulate corresponding operating conditions;

[0008] S2. Establish reference values: Calculate voltage deviation (VD), voltage fluctuation coefficient (VFC), three-phase unbalance coefficient (VUF), harmonic component proportion (HCP), frequency deviation (FD), and power factor (PF) to construct a data set. Use the sparse autoencoder model to reduce the dimensions of the six indicators to 1, and then use the lasso regression model to determine the calculation method of the voltage quality comprehensive index value (CEI);

[0009] S3. Obtaining evaluation results: For each distribution network operating condition, the voltage quality comprehensive index value (CEI) is considered, and the line loss rate (LLR), short circuit ratio (SCR), and voltage qualification rate (VPR) are comprehensively considered to evaluate the reactive voltage of the distribution network under the condition and draw conclusions.

[0010] Preferably, the operating conditions described in S1, formulating the detection conditions, mainly include the following five types:

[0011] Peak load condition: the operating state when the demand for electricity is the greatest;

[0012] Off-peak load condition: the operating state when power demand is at its lowest;

[0013] New energy access conditions: operating status when new energy (such as photovoltaic and wind power) is accessed to the distribution network;

[0014] Fault condition: the operating status when some lines or equipment in the distribution network fail;

[0015] Standard operating conditions: the state in which the distribution network operates normally and stably.

[0016] Preferably, the step S2, establishing a reference value, further comprises the following sub-steps:

[0017] S21. Calculate the voltage deviation (VD), voltage fluctuation coefficient (VFC), three-phase unbalance coefficient (VUF), harmonic component proportion (HCP), frequency deviation (FD), and power factor (PF) for constructing a data set; and standardize the data;

[0018] S22, establishing and training a sparse autoencoder to reduce the dimensions of the six index values ​​to 1x1, and obtaining the output of the hidden layer as the voltage quality comprehensive index value (CEI) after dimension reduction;

[0019] S23. Use the lasso regression model to establish the linear relationship between the voltage quality comprehensive index value CEI and the six indicators, use the training set to fit the lasso regression model, and use the test set to verify it, and determine the calculation method of the voltage quality comprehensive index value (CEI).

[0020] Preferably, the voltage deviation (VD), voltage fluctuation coefficient (VFC), three-phase unbalance coefficient (VUF), harmonic component proportion (HCP), frequency deviation (FD), power factor (PF), and the definitions of each index are as shown in the following formula:

[0021]

[0022] Among them, V s is the actual voltage, V eis the rated voltage;

[0023]

[0024] Among them, V PP is the peak voltage fluctuation within a cycle, that is, V PP =V max -V min ;

[0025]

[0026] Among them, V2 is the effective value of negative sequence voltage, and V1 is the effective value of positive sequence voltage;

[0027]

[0028] Among them, U h is the effective value of the hth harmonic voltage, U1 is the effective value of the fundamental voltage, and U1 is h It is obtained by converting the time domain signal into the frequency domain signal by Fourier transform of the voltage signal;

[0029]

[0030] Among them, f2 is the actual frequency of the distribution network, and f1 is the standard frequency;

[0031]

[0032] Among them, P is the active power of the distribution network node, and Q is the reactive power of the distribution network node.

[0033] Preferably, in S21, the six indicators are calculated and the data is standardized by using a 0-1 normalization method to normalize the six indicator values, and the calculation method is as follows:

[0034]

[0035] in,

[0036] X is the original index value of one of the six index values; X min is the minimum value of all index values ​​in the sample of one of the six index values; max is the maximum value of all index values ​​in the sample of one of the six index values, X norm is the normalized index value, ranging from [0, 1].

[0037] Preferably, the S22 training sparse autoencoder specifically prevents overfitting during the training process, uses mean square error (MSE) as a loss function, first adds a regularization term to the loss function, and the loss function of the autoencoder is as follows:

[0038]

[0039] in,

[0040] x i : represents the input of the i-th neuron in the input layer,

[0041] y i : represents the output of the i-th neuron in the output layer,

[0042] λ: represents weight decay sparseness, n l Indicates the number of network layers, s l is the number of neurons in the lth layer;

[0043] Then by adding in the loss function of the autoencoder As a penalty term, the average activation of neurons in the hidden layer Take a value ρ close to 0;

[0044]

[0045] Get the corrected loss function:

[0046]

[0047] in,

[0048] s2 is the number of neurons in the hidden layer, β is the weight for controlling sparsity, and ρ is the sparsity parameter. is the average activation of the jth neuron.

[0049] Preferably, in the lasso regression model, in order to establish a linear relationship between the voltage quality comprehensive index value CEI and the six indicators, and use the data sets of the six indicators to fit the lasso regression model, it is assumed that there is a linear relationship between CEI and the six indicators, which can be expressed as:

[0050] CEI=ω1×VD+ω2×VFC+ω3×VUF+ω4×HCP+ω5×FD+ω6×PF

[0051] +b

[0052] Among them, ω1, ω2, ..., ω6 are regression coefficients, and b is the intercept.

[0053] Preferably, the steps of fitting the lasso regression model are as follows:

[0054] S231. Use Lasso regression to fit the model in the training set and estimate the regression coefficient and intercept. Specifically, by adding a penalty term of absolute value in the loss function, Lasso regression can compress some regression coefficients to zero, thereby realizing feature selection. Then, select the optimal regularization parameter λ through cross-validation, and use the training set to train the model, and estimate the regression coefficient and intercept by minimizing the loss function.

[0055] Its loss function is as follows:

[0056]

[0057] in,

[0058] n is the number of samples; y is i is the CEI value of the i-th sample; X i is the characteristic vector of the i-th sample, i.e., each voltage evaluation index; w is the regression coefficient; b is the intercept; λ is the regularization parameter, which controls the strength of the regularization term;

[0059] S232. Use the trained lasso regression model to predict the CEI value on the test set, calculate the model fitting error by calculating the R square and mean square error (MSE) indicators, and evaluate the model fitting effect; the mean square error (MSE) is used to measure the average error between the predicted value and the actual value, which is defined as:

[0060]

[0061] in,

[0062] m is the number of test set samples, y i is the CEI value obtained after SAE dimensionality reduction, is the predicted CEI value;

[0063] R squared (R 2 ) indicates the model's ability to explain the dependent variable. The closer its value is to 1, the better the model fit is.

[0064]

[0065] In the formula, is the average value of CEI obtained after SAE dimensionality reduction;

[0066] After judging the fitting effect, if the fitting effect is good, the calculation expression of CEI is determined, otherwise the regularization parameter λ value is reset, retraining and testing are performed, and finally the calculation expression of CEI is determined.

[0067] Preferably, the line loss rate (LLR) is defined as the ratio of line loss power to total power supply, as shown in the following formula:

[0068]

[0069] in,

[0070] EQ loss EQ is the line loss, which refers to the total amount of energy loss caused by resistance within a certain period of time. sup The total power supply refers to the total amount of electric energy supplied by the power grid to users in the same period of time;

[0071] The short circuit ratio (SCR) is used to describe the strength of the grid connection point of the new energy power station, and its calculation method is shown in the following formula:

[0072]

[0073] in,

[0074] U cn is the grid connection point voltage of the power station, Z cn is the equivalent impedance of the power station grid connection point, S sc Installed capacity for power stations;

[0075] The voltage qualification rate (VPR) refers to the proportion of the time that the voltage in the power system remains within the specified range to the total time within a certain period of time; its definition is as shown in the following formula:

[0076]

[0077] in,

[0078] T p is the qualified voltage time, T is the total detection time, and the qualified voltage range is that the voltage deviation is within ±10% of the rated voltage.

[0079] In summary, the present invention mainly has the following beneficial effects:

[0080] The proposed indicator calculation method for the comprehensive indicator of distribution network voltage quality based on sparse autoencoder dimensionality reduction combined with lasso regression model can solve the problem that the weight allocation of each indicator in the traditional comprehensive evaluation index is too subjective and difficult to conform to the actual situation. The weight of each indicator is obtained through algorithm adaptively to construct a comprehensive evaluation index.

[0081] Then, on this basis, by making reasonable interval divisions for the comprehensive index value of voltage quality, line loss rate, voltage qualification rate and short-circuit ratio on the new energy access side, the operating status of the distribution network under different working conditions can be graded; such a division method helps to quickly identify the operating health status of the distribution network under various working conditions, and provide a scientific basis for subsequent optimization and adjustment; this grade division can also guide the operating strategy of the distribution network in actual operation to ensure the stable and efficient operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a flow chart of the steps of the reactive voltage index evaluation method of the distribution network of the present invention;

[0083] Figure 2 A neural network diagram of a three-layer structure of the automatic encoder of the present invention;

[0084] Figure 3 It is a table diagram of each operating state and index range of the peak load condition of the present invention;

[0085] Figure 4 It is a table diagram of various operating states and index ranges under low load conditions of the present invention;

[0086] Figure 5 It is a table diagram of various operating states and index ranges under the new energy access working condition of the present invention;

[0087] Figure 6 It is a table diagram of various operating states and index intervals under fault conditions of the present invention;

[0088] Figure 7 It is a table diagram of various operating states and index ranges of the conventional operating conditions of the present invention;

[0089] Figure 8 The flowchart of the voltage quality comprehensive evaluation index construction method of the present invention is shown in FIG. DETAILED DESCRIPTION

[0090] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0091] Example

[0092] like Figure 1 As shown, S1. Formulate the test conditions: divide according to the different operating states of the distribution network, and formulate the corresponding operating conditions. S1. Formulate the test conditions. The operating conditions mainly include the following five types:

[0093] Peak load condition: the operating state when the demand for electricity is the greatest;

[0094] Off-peak load condition: the operating state when power demand is at its lowest;

[0095] New energy access conditions: operating conditions when new energy such as photovoltaic and wind power are connected to the distribution network;

[0096] Fault condition: the operating status when some lines or equipment in the distribution network fail;

[0097] Standard operating conditions: the state in which the distribution network operates normally and stably.

[0098] It should be noted that, in this embodiment, the five working conditions mentioned are specifically as follows:

[0099] Peak load condition: the operating state when the power demand is the highest, and the reactive power demand and voltage fluctuations in the distribution network are large;

[0100] Off-peak load condition: the operating state when power demand is at its lowest, and the voltage is usually high;

[0101] Renewable energy access conditions: The operating state when new energy (such as photovoltaic and wind power) is accessed to the distribution network, resulting in voltage fluctuations and uncertainty in reactive power demand;

[0102] Fault condition: The operating state when some lines or equipment in the distribution network fail, which increases the difficulty of reactive power and voltage control;

[0103] Normal operating conditions: the state in which the distribution network operates normally and stably.

[0104] like Figure 1 , 2 , 8, S2, establish reference value: calculate voltage deviation VD, voltage fluctuation coefficient VFC, three-phase unbalance coefficient VUF, harmonic component proportion HCP, frequency deviation FD, power factor PF, a total of 6 indicators to build a data set, use the sparse autoencoder model to reduce the dimensions of the 6 indicators to 1, and then use the lasso regression model to determine the calculation method of the voltage quality comprehensive index value (CEI), S2, establish reference value, also includes the following sub-steps:

[0105] S21. Calculate the voltage deviation VD, voltage fluctuation coefficient VFC, three-phase unbalance coefficient VUF, harmonic component proportion HCP, frequency deviation FD, power factor PF, a total of 6 index values, which are used to construct a data set; and standardize the data;

[0106] S22, establishing and training a sparse autoencoder to reduce the dimensions of the six index values ​​to 1x1, and obtaining the output of the hidden layer as a reference value of the voltage quality comprehensive index value CEI evaluation result;

[0107] S23. Use the lasso regression model to establish the linear relationship between the voltage quality comprehensive index value CEI and the six indicators. Use the training set to fit the lasso regression model and use the test set to verify it. Determine the calculation method of the voltage quality comprehensive index value CEI, voltage deviation VD, voltage fluctuation coefficient VFC, three-phase unbalance coefficient VUF, harmonic component proportion HCP, frequency deviation FD, power factor PF. The definitions of each indicator are as shown in the following formula:

[0108]

[0109] Among them, V s is the actual voltage, V e is the rated voltage;

[0110]

[0111] Among them, V PP is the peak voltage fluctuation within a cycle, that is, V PP =V max -V min ;

[0112]

[0113] Among them, V2 is the effective value of negative sequence voltage, and V1 is the effective value of positive sequence voltage;

[0114]

[0115] Among them, U h is the effective value of the hth harmonic voltage, U1 is the effective value of the fundamental voltage, and U1 is h It is obtained by converting the time domain signal into the frequency domain signal by Fourier transform of the voltage signal;

[0116]

[0117] Among them, f2 is the actual frequency of the distribution network, and f1 is the standard frequency;

[0118]

[0119] Among them, P is the active power of the distribution network node, Q is the reactive power of the distribution network node, and in S21, six indicators are calculated and the data is standardized. The specific method is to use the 0-1 normalization method to normalize the six indicator values. The calculation method is as follows:

[0120]

[0121] in,

[0122] X is the original index value of one of the six index values; X min is the minimum value of all index values ​​in the sample of one of the six index values; max is the maximum value of all index values ​​in the sample of one of the six index values, X norm is the normalized index value, ranging from [0, 1];

[0123] S22 training sparse autoencoder specifically to prevent overfitting during training, using mean square error MSE as the loss function, first add a regularization term to the loss function, the loss function of the autoencoder is as follows:

[0124]

[0125] in,

[0126] x i : represents the input of the i-th neuron in the input layer,

[0127] y i : represents the output of the i-th neuron in the output layer,

[0128] λ: represents weight decay sparseness, n l Indicates the number of network layers, s l is the number of neurons in the lth layer;

[0129] Then by adding in the loss function of the autoencoder As a penalty term, the average activation of neurons in the hidden layer Take a value ρ close to 0;

[0130]

[0131] Get the corrected loss function:

[0132]

[0133] in,

[0134] s2 is the number of neurons in the hidden layer, β is the weight for controlling sparsity, and ρ is the sparsity parameter. is the average activation of the jth neuron. In the lasso regression model, in order to establish the linear relationship between the voltage quality comprehensive index CEI and the six indicators, and use the data sets of the six indicators to fit the lasso regression model, it is assumed that there is a linear relationship between CEI and the six indicators, which can be expressed as:

[0135] CEI=ω1×VD+ω2×VFC+ω3×VUF+ω4×HCP+ω5×FD+ω6×PF

[0136] +b

[0137] Among them, ω1, ω2, ..., ω6 are regression coefficients, b is the intercept, and Lasso regression is used to fit the model in the training set to estimate the regression coefficients and intercept. Specifically, by adding a penalty term of absolute value in the loss function, Lasso regression can compress some regression coefficients to zero, thereby realizing feature selection; then, the optimal regularization parameter λ is selected through cross-validation, and the training set is used for model training, and the regression coefficients and intercept are estimated by minimizing the loss function;

[0138] Its loss function is as follows:

[0139]

[0140] in,

[0141] n is the number of samples; y is i is the CEI value of the i-th sample; X i is the characteristic vector of the i-th sample, i.e., each voltage evaluation index; w is the regression coefficient; b is the intercept; λ is the regularization parameter, which controls the strength of the regularization term;

[0142] S232. Use the trained lasso regression model to predict the CEI value on the test set, calculate the model fitting error by calculating the R square and mean square error MSE indicators, and evaluate the model fitting effect; the mean square error (MSE) is used to measure the average error between the predicted value and the actual value, which is defined as:

[0143]

[0144] in,

[0145] m is the number of test set samples, y i is the CEI value obtained after SAE dimensionality reduction, is the predicted CEI value;

[0146] R squared 2 It indicates the explanatory power of the model on the dependent variable. The closer its value is to 1, the better the model fit is.

[0147]

[0148] In the formula, is the average value of CEI obtained after SAE dimensionality reduction;

[0149] After judging the fitting effect, if the fitting effect is good, the calculation expression of CEI is determined, otherwise the regularization parameter λ value is reset, retraining and testing are performed, and finally the calculation expression of CEI is determined.

[0150] It should be noted that, in this embodiment, by collecting historical data and real-time monitoring data related to voltage quality, and according to corresponding calculation formulas, a total of 6 index values ​​are calculated, including voltage deviation VD, voltage fluctuation coefficient VFC, three-phase unbalance coefficient VUF, harmonic component proportion HCP, frequency deviation FD, and power factor PF.

[0151] Among them, since different voltage quality indicators have different dimensions and ranges, we use the 0-1 normalization method to normalize the six indicator values.

[0152] An autoencoder (AE) is a neural network with a three-layer structure, such as Figure 2 As shown, it includes an input layer, an intermediate hidden layer, and an output layer. The autoencoder is trained to make the input values ​​as equal as possible, and then the input layer and the hidden layer are combined to obtain the autoencoder.

[0153] Among them, the activation value of the neurons in the middle hidden layer can represent the value of the neurons in the input layer. When the number of neurons in the hidden layer is less than the number of neurons in the input layer, the autoencoder can extract data features and reduce the dimension of the input data.

[0154] Based on this, the autoencoder uses the back propagation algorithm to find the optimal network parameters to minimize the error between the input layer and the output layer. Therefore, the mean square error (MSE) is used as the loss function, and in order to prevent overfitting during training,

[0155] The number of input layer nodes of the sparse autoencoder is set to 6, corresponding to the 6 voltage quality indicators contained in each sample, the number of hidden layer nodes is set to 1, and the number of output layer nodes is also 6, which are used to reconstruct the input data.

[0156] After the training is completed, the output of the hidden layer is obtained as the voltage quality comprehensive index value CEI obtained after dimensionality reduction. This index value contains the main information of 6 index values.

[0157] On the basis of the comprehensive voltage quality index CEI, short circuit ratio SCR, line loss rate LLR and voltage qualification rate VPR are supplemented as evaluation indicators.

[0158] like Figure 1 , 3 , 4, 5, 6, 7, as shown, S3, evaluation results acquisition: for each distribution network operating condition, consider the voltage quality comprehensive index value CEI, and comprehensively consider the line loss rate LLR, short circuit ratio SCR, voltage qualification rate VPR, a total of four indicators, evaluate the reactive voltage of the distribution network under this condition, and conclude that the line loss rate LLR is defined by the ratio of line loss power to total power supply, as shown in the following formula:

[0159]

[0160] in,

[0161] EQ loss EQ is the line loss, which refers to the total amount of energy loss caused by resistance within a certain period of time. sup The total power supply refers to the total amount of electric energy supplied by the power grid to users in the same period of time;

[0162] The short circuit ratio SCR is used to describe the strength of the grid connection point of a new energy power station. Its calculation method is shown in the following formula:

[0163]

[0164] in,

[0165] U cn is the grid connection point voltage of the power station, Z cn is the equivalent impedance of the power station grid connection point, S sc Installed capacity for power stations;

[0166] The voltage qualification rate VPR refers to the proportion of the time that the voltage in the power system remains within the specified range to the total time within a certain period of time; its definition is shown in the following formula:

[0167]

[0168] in,

[0169] T p is the qualified voltage time, T is the total detection time, and the qualified voltage range is that the voltage deviation is within ±10% of the rated voltage.

[0170] It should be noted that, in this embodiment, after the calculation of the index is completed, the voltage quality of the distribution network under different operating conditions is evaluated.

[0171] For peak load conditions, new energy access is not considered for the time being. In this case, the distribution network has various operating states and index ranges. Level 1 indicates that the line loss is low, the distribution network voltage fluctuation is small, and the reactive power demand is well controlled; Level 2 indicates that the distribution network voltage fluctuation and line loss are slightly large, but overall they are still within the acceptable range; Level 3 indicates that the distribution network voltage fluctuation and line loss are large, and the voltage may be unqualified in some areas; Level 4 indicates that the voltage fluctuation is extremely large, the system line loss is too high, the reactive power voltage control is difficult, and adjustment measures are urgently needed;

[0172] For low-load conditions, the access of new energy is not considered for the time being. In this case, the various operating states and index intervals of the distribution network, under this condition, the various evaluation levels of the distribution network voltage are the same as those under peak load conditions;

[0173] Under the condition of renewable energy access, four index values ​​are calculated. In this case, the distribution network has various operating states and index ranges. Level 1 indicates that renewable energy access is stable and the system has strong adaptability to renewable energy fluctuations, and voltage fluctuations and reactive power demand are effectively controlled; Level 2 indicates that the system has good adaptability to renewable energy, with occasional slight voltage fluctuations and high overall voltage quality; Level 3 indicates that the system has average adaptability, with large fluctuations in renewable energy access and obvious voltage fluctuations, and reactive power compensation needs to be further optimized; Level 4 indicates that the system has poor adaptability to renewable energy, severe voltage fluctuations, and difficulty in system regulation, and it is recommended to reduce the renewable energy grid-connected capacity;

[0174] Under fault conditions, new energy access is not considered for the time being. In this case, the distribution network has various operating states and index ranges. Level 1 means that the fault has little impact, the system can quickly recover and stabilize, and the voltage level is still good under the influence of the fault; Level 2 means that the system can recover in a relatively short time, the voltage is controlled within a reasonable range, and there is occasional slight instability; Level 3 means that the system recovers slowly after the fault, the stability is general, the voltage level is relatively unstable, and the fault recovery capability needs to be strengthened; Level 4 means that the system has poor stability, fault recovery is difficult, the voltage fluctuates severely, and a large range of unqualified systems may occur;

[0175] Conventional operating conditions, when the distribution network is operating normally and stably, various operating states and index ranges. Under this condition, the various evaluation levels of the distribution network voltage are the same as those under peak load conditions.

[0176] The working principle of the present invention is:

[0177] In this method, the five working conditions mentioned are:

[0178] Peak load condition: the operating state when the power demand is the highest, and the reactive power demand and voltage fluctuations in the distribution network are large;

[0179] Off-peak load condition: the operating state when power demand is at its lowest, and the voltage is usually high;

[0180] Renewable energy access conditions: The operating state when new energy (such as photovoltaic and wind power) is accessed to the distribution network, resulting in voltage fluctuations and uncertainty in reactive power demand;

[0181] Fault condition: The operating state when some lines or equipment in the distribution network fail, which increases the difficulty of reactive power and voltage control;

[0182] Normal operation condition: the state in which the distribution network operates normally and stably;

[0183] By collecting historical data and real-time monitoring data related to voltage quality, and according to the corresponding calculation formula, a total of 6 index values ​​are calculated, including voltage deviation VD, voltage fluctuation coefficient VFC, three-phase unbalance coefficient VUF, harmonic component proportion HCP, frequency deviation FD, and power factor PF.

[0184] Among them, since different voltage quality indicators have different dimensions and ranges, we use the 0-1 normalization method to normalize the six indicator values.

[0185] An autoencoder (AE) is a neural network with a three-layer structure, such as Figure 2 As shown, it includes an input layer, an intermediate hidden layer, and an output layer. The autoencoder is trained to make the input values ​​as equal as possible, and then the input layer and the hidden layer are combined to obtain the autoencoder.

[0186] Among them, the activation value of the neurons in the middle hidden layer can represent the value of the neurons in the input layer. When the number of neurons in the hidden layer is less than the number of neurons in the input layer, the autoencoder can extract data features and reduce the dimension of the input data.

[0187] Based on this, the autoencoder uses the back propagation algorithm to find the optimal network parameters to minimize the error between the input layer and the output layer. Therefore, the mean square error (MSE) is used as the loss function, and in order to prevent overfitting during training,

[0188] The number of input layer nodes of the sparse autoencoder is set to 6, corresponding to the 6 voltage quality indicators contained in each sample, the number of hidden layer nodes is set to 1, and the number of output layer nodes is also 6, which are used to reconstruct the input data.

[0189] After the training is completed, the output of the hidden layer is obtained as the voltage quality comprehensive index value CEI obtained after dimensionality reduction. This index value contains the main information of 6 index values.

[0190] On the basis of the voltage quality comprehensive index value CEI, short circuit ratio SCR, line loss rate LLR, and voltage qualification rate VPR are added as evaluation indicators;

[0191] After completing the calculation of the indicators, the voltage quality of the distribution network under different operating conditions is evaluated.

[0192] 1. For peak load conditions, the access of new energy is not considered for the time being. In this case, the operating states and index ranges of the distribution network are as follows: Figure 3As shown in the figure, Level 1 means that the line loss is low, the distribution network voltage fluctuation is small, and the reactive power demand is well controlled; Level 2 means that the distribution network voltage fluctuation and line loss are slightly large, but generally still within the acceptable range; Level 3 means that the distribution network voltage fluctuation and line loss are large, and the voltage may be unqualified in some areas; Level 4 means that the voltage fluctuation is extremely large, the system line loss is too high, the reactive power voltage control is difficult, and adjustment measures are urgently needed;

[0193] 2. For low-load conditions, the integration of new energy sources is not considered for the time being. In this case, the operating states and index ranges of the distribution network are as follows: Figure 4 As shown, under this condition, the various evaluation levels of the distribution network voltage are the same as those under the peak load condition;

[0194] 3. Calculate the values ​​of four indicators under the condition of new energy access. In this case, the operating states and indicator ranges of the distribution network are as follows: Figure 5 As shown, Level 1 means that the access to renewable energy is stable and the system is highly adaptable to the fluctuation of renewable energy, and voltage fluctuation and reactive power demand are effectively controlled; Level 2 means that the system has good adaptability to renewable energy, with occasional slight voltage fluctuations and high voltage quality overall; Level 3 means that the system has average adaptability, with large fluctuations in the access to renewable energy and obvious voltage fluctuations, and reactive power compensation needs to be further optimized; Level 4 means that the system has poor adaptability to renewable energy, severe voltage fluctuations, and difficulty in system regulation, and it is recommended to reduce the grid-connected capacity of renewable energy;

[0195] 4. Under fault conditions, the access of new energy is not considered for the time being. In this case, the operating states and index ranges of the distribution network are as follows: Figure 6 As shown, Level 1 means that the fault has little impact, the system can quickly recover and stabilize, and the voltage level is still good under the influence of the fault; Level 2 means that the system can recover in a relatively short time, the voltage is controlled within a reasonable range, and there is occasional slight instability; Level 3 means that the system recovers slowly after the fault, the stability is average, the voltage level is relatively unstable, and the fault recovery capability needs to be strengthened; Level 4 means that the system has poor stability, fault recovery is difficult, the voltage fluctuates severely, and a large range of unqualified may occur;

[0196] 5. Normal operation conditions: When the distribution network is operating normally and stably, the operating conditions and index ranges are as follows: Figure 7 As shown in the figure, under this condition, the various evaluation levels of the distribution network voltage are the same as those under the peak load condition.

[0197] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for evaluating reactive power voltage index of distribution network under different operating conditions, characterized in that ,The distribution network reactive voltage index evaluation method comprises the following steps: S1. Formulate the test conditions: divide the distribution network according to different operating conditions and formulate corresponding operating conditions; S2. Establish reference values: Calculate voltage deviation (VD), voltage fluctuation coefficient (VFC), three-phase unbalance coefficient (VUF), harmonic component proportion (HCP), frequency deviation (FD), and power factor (PF) to construct a data set. Use the sparse autoencoder model to reduce the dimensions of the six indicators to 1, and then use the lasso regression model to determine the calculation method of the voltage quality comprehensive index value (CEI); S3. Obtaining evaluation results: For each distribution network operating condition, the voltage quality comprehensive index value (CEI) is considered, and the line loss rate (LLR), short circuit ratio (SCR), and voltage qualification rate (VPR) are comprehensively considered to evaluate the reactive voltage of the distribution network under the condition and draw conclusions.

2. The method for evaluating reactive voltage index of a distribution network under different operating conditions according to claim 1 is characterized in that: The operating conditions described in S1, formulating the test conditions, mainly include the following five types: Peak load condition: the operating state when the demand for electricity is the greatest; Off-peak load condition: the operating state when power demand is at its lowest; New energy access conditions: operating status when new energy (such as photovoltaic and wind power) is accessed to the distribution network; Fault condition: the operating status when some lines or equipment in the distribution network fail; Standard operating conditions: the state in which the distribution network operates normally and stably.

3. The method for evaluating reactive voltage index of a distribution network under different operating conditions according to claim 1 is characterized in that: The step S2, establishing a reference value, further comprises the following sub-steps: S21. Calculate the voltage deviation (VD), voltage fluctuation coefficient (VFC), three-phase unbalance coefficient (VUF), harmonic component proportion (HCP), frequency deviation (FD), and power factor (PF) for constructing a data set; and standardize the data; S22, establishing and training a sparse autoencoder to reduce the dimensions of the six indicator values ​​to 1x1, and obtaining the output of the hidden layer as the voltage quality comprehensive index value (CEI) after dimension reduction; S23. Use the lasso regression model to establish the linear relationship between the voltage quality comprehensive index value CEI and the six indicators, use the training set to fit the lasso regression model, and use the test set to verify it, and determine the calculation method of the voltage quality comprehensive index value (CEI).

4. The method for evaluating reactive voltage index of a distribution network under different operating conditions according to claim 3 is characterized in that: The voltage deviation (VD), voltage fluctuation coefficient (VFC), three-phase unbalance coefficient (VUF), harmonic component proportion (HCP), frequency deviation (FD), power factor (PF), and the definitions of each index are shown in the following formula: Among them, V s is the actual voltage, V e is the rated voltage; Among them, V PP is the peak voltage fluctuation within a cycle, that is, V PP =V max -V min ; Among them, V2 is the effective value of negative sequence voltage, and V1 is the effective value of positive sequence voltage; Among them, U h is the effective value of the hth harmonic voltage, U1 is the effective value of the fundamental voltage, and U1 is h It is obtained by converting the time domain signal into the frequency domain signal by Fourier transform of the voltage signal; Among them, f2 is the actual frequency of the distribution network, and f1 is the standard frequency; Among them, P is the active power of the distribution network node, and Q is the reactive power of the distribution network node.

5. The method for evaluating reactive voltage index of distribution network under different operating conditions according to claim 3 is characterized in that: In S21, the six indicators are calculated and the data is standardized. Specifically, the six indicator values ​​are normalized using a 0-1 normalization method. The calculation method is as follows: in, X is the original index value of one of the six index values; X min is the minimum value of all index values ​​in the sample of one of the six index values; max is the maximum value of all index values ​​in the sample of one of the six index values, X norm is the normalized index value, ranging from [0, 1].

6. The method for evaluating reactive power voltage index of a distribution network under different operating conditions according to claim 3 is characterized in that: The S22 training sparse autoencoder is specifically to prevent overfitting during the training process, using mean square error (MSE) as the loss function, first adding a regularization term to the loss function, the loss function of the autoencoder is as follows: in, x i : represents the input of the i-th neuron in the input layer, y i : represents the output of the i-th neuron in the output layer, λ: represents weight decay sparseness, n1 represents the number of network layers, and s1 represents the number of neurons in the lth layer; Then by adding in the loss function of the autoencoder As a penalty term, the average activation of neurons in the hidden layer Take a value ρ close to 0; Get the corrected loss function: in, s2 is the number of neurons in the hidden layer, β is the weight for controlling sparsity, and ρ is the sparsity parameter. is the average activation of the jth neuron.

7. The method for evaluating reactive power voltage index of a distribution network under different operating conditions according to claim 3 is characterized in that: In the lasso regression model, in order to establish a linear relationship between the voltage quality comprehensive index value CEI and the six indicators, and use the data sets of the six indicators to fit the lasso regression model, it is assumed that there is a linear relationship between CEI and the six indicators, which can be expressed as: CEI=ω1×VD+ω2×VFC+ω3×VUF+ω4×HCP+ω5×FD+ω6×PF+b Among them, ω1, ω2, ..., ω6 are regression coefficients, and b is the intercept.

8. The method for evaluating reactive power voltage index of a distribution network under different operating conditions according to claim 7 is characterized in that: The steps for fitting the lasso regression model are as follows: S231. Use Lasso regression to fit the model in the training set and estimate the regression coefficient and intercept. Specifically, by adding a penalty term of absolute value in the loss function, Lasso regression can compress some regression coefficients to zero, thereby realizing feature selection. Then, select the optimal regularization parameter λ through cross-validation, and use the training set to train the model, and estimate the regression coefficient and intercept by minimizing the loss function. Its loss function is as follows: in, n is the number of samples; y is i is the CEI value of the i-th sample; X i is the characteristic vector of the i-th sample, i.e., each voltage evaluation index; w is the regression coefficient; b is the intercept; λ is the regularization parameter, which controls the strength of the regularization term; S232. Use the trained lasso regression model to predict the CEI value on the test set, calculate the model fitting error by calculating the R square and mean square error (MSE) indicators, and evaluate the model fitting effect; the mean square error (MSE) is used to measure the average error between the predicted value and the actual value, which is defined as: in, m is the number of test set samples, y i is the CEI value obtained after SAE dimensionality reduction, is the predicted CEI value; R squared (R 2 ) indicates the model's ability to explain the dependent variable. The closer its value is to 1, the better the model fit is. In the formula, is the average value of CEI obtained after SAE dimensionality reduction; After judging the fitting effect, if the fitting effect is good, the calculation expression of CEI is determined, otherwise the regularization parameter λ value is reset, retraining and testing are performed, and finally the calculation expression of CEI is determined.

9. The method for evaluating reactive power voltage index of a distribution network under different operating conditions according to claim 1, characterized in that: The line loss rate (LLR) is defined as the ratio of line loss power to total power supply, as shown in the following formula: in, EQ loss EQ is the line loss, which refers to the total amount of energy loss caused by resistance within a certain period of time. sup The total power supply refers to the total amount of electric energy supplied by the power grid to users in the same period of time; The short circuit ratio (SCR) is used to describe the strength of the grid connection point of the new energy power station, and its calculation method is shown in the following formula: in, U cn is the grid connection point voltage of the power station, Z cn is the equivalent impedance of the power station grid connection point, S sc Installed capacity for power stations; The voltage qualification rate (VPR) refers to the proportion of the time that the voltage in the power system remains within the specified range to the total time within a certain period of time; its definition is as shown in the following formula: in, T p is the qualified voltage time, T is the total detection time, and the qualified voltage range is that the voltage deviation is within ±10% of the rated voltage.

Images