A real-time analysis method for low-voltage transformer area power supply stability

Abstract

The present application belongs to the field of electric power information technology, and particularly relates to a real-time analysis method for low-voltage power supply stability of a power supply area. The method is used for analyzing the current power supply area operation state according to the power monitoring data collected by a fusion terminal. The real-time analysis method comprises the following steps: S1: collecting the power information of each power consumption node in the area under different states. S2: performing normalization processing on the collected data to obtain a sample data set. S3: performing clustering on the sample data set to determine the cluster center and the number of categories. S4: constructing a BP neural network with a three-layer structure. S5: training the BP neural network and combining a classical genetic algorithm to optimize the threshold value of the BP neural network. S6: collecting real-time power information, performing normalization and clustering processing, and identifying the result by the trained BP neural network. The present application solves the problems of large amount of power information data in the area, high analysis difficulty, and difficulty in predicting the power supply stability of the area.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of electric power information technology, and particularly relates to a real-time analysis method for low-voltage power supply stability of a power supply area. BACKGROUND

[0002] An electric power system is a power production and consumption system composed of power plants, transmission and distribution lines, power supply and distribution stations, and power users. Its function is to convert primary energy in nature into electric energy through a power generation dynamic device, and then supply the electric energy to users through power transmission, power transformation and power distribution. In order to realize the coordination and control of each link, the electric power system also includes information and control systems applied to each link of power generation, power transmission, power distribution and marketing, and further measures, adjusts, controls, protects, communicates and dispatches the production and application process of electric energy to ensure that users obtain safe and high-quality electric energy.

[0003] The development goal of a modern electric power system is to realize electric power system automation. The fields of electric power system automation include automatic detection, adjustment and control of the production process, automatic safety protection of the system and components, automatic transmission of network information, automatic scheduling of system production, and automatic economic management of enterprises. The main goal of electric power system automation is to ensure the power quality (frequency and voltage) of power supply, ensure the safe and reliable operation of the system, and improve economic efficiency and management effectiveness.

[0004] In the power distribution and power consumption link, intelligent electric energy meters and fusion terminals and other devices can already realize remote collection of electric power information. Based on the collected electric power information big data, remote electric power monitoring and management can be realized. For example, in the existing electric power system, the fusion terminal and the intelligent electric energy meter can realize automatic meter reading in the power supply area, and realize functions such as automatic settlement of electricity charges in combination with an online payment system.

[0005] Electric power information big data can also be applied to electric power system operation state monitoring, etc., but in a single power distribution area, there are many electric power devices, and the load state of the power consumption nodes is complex. How to identify and predict the operation state of the electric power system through complex electric power information is still a technical problem to be solved. SUMMARY

[0006] In order to solve the problems of large amount of electric power information data in the power supply area, high analysis difficulty, and difficulty in predicting the power supply stability of the power supply area, the present application provides a real-time analysis method for low-voltage power supply stability of a power supply area.

[0007] The present application realizes the following technical scheme:

[0008] A real-time analysis method for low-voltage power supply stability of a power supply area, which is used to analyze the current power supply area operation state according to the electric power monitoring data collected by the fusion terminal. The real-time analysis method comprises the following steps:

[0009] S1: Collect the power information of each power consumption node in the substation area under different states as the sample data of the current node. The sample data includes: the power supply voltage V1, the power supply current I1, the power factor of the current node, the line loss rate ΔP, the device voltage V2, the device current I2, the real-time load P, the device temperature T of the substation transformer, and other related data related to the operation state of the power grid.

[0010] S2: According to the theoretical safety threshold in the running process, the large amount of sample data collected is normalized respectively, and then a sample data set containing all normalized sample data is obtained.

[0011] S3: Based on the adaptive fast search density peak value method, the sample data set of each node is clustered to determine the cluster center and the number of categories, and the node attribute data set after clustering is obtained.

[0012] S4: A BP neural network with three-layer structure is constructed, including input layer, hidden layer and output layer. In the BP neural network, the number of input layer nodes n is equal to the number of power consumption nodes in the current substation area; the number of output layer nodes is 1, and the number of hidden layer nodes is 2n+1.

[0013] S5: A large number of node attribute data sets under different operating states collected in advance are used as training samples, and the BP neural network constructed in the previous step is trained to update the weight value of the network model. At the same time, the threshold value of the BP neural network is optimized by combining the classical genetic algorithm in the training process. The trained BP neural network is used as the required substation operation state recognition model.

[0014] S6: By fusing the real-time power information of the substation transformer and all power consumption nodes in the substation collected by the terminal, the real-time power information is sequentially normalized and clustered to obtain the real-time node attribute data set. Then the real-time node attribute data set is input into the trained substation operation state recognition model, and the real-time operation state of the current substation is output by the model.

[0015] Among them, the real-time operation state of the substation is divided into normal and abnormal.

[0016] In the present application, the power supply voltage V1, the power supply current I1, the device voltage V2 and the device current I2 correspond to the voltage and current of each phase A, B and C respectively; specifically including: V 1A , V 1B , V 1C , I 1A , I 1B , I 1C , V 2A , V 2B , V2C , I 2A , I 2B , I 2C .

[0017] As a further improvement of the present application, in step S2, the normalization formula of the sample data is as follows:

[0018]

[0019] In the above formula, x represents the measured value of the current sample data; represents the normalized value of the current sample data; x max represents the upper limit of the theoretical safety threshold of the current sample data; x min represents the lower limit of the theoretical safety threshold of the current sample data; wherein, when there is no safety threshold lower limit for a certain sample data, then x min = 0.

[0020] As a further improvement of the present application, in step S3, the adaptive fast search density peak method for clustering the sample data set of multiple nodes is as follows:

[0021] S31: Obtain the sample data set of any node, calculate the Euclidean distance d ij between any two sample data in the sample data set, and the calculation formula is as follows:

[0022]

[0023] In the above formula, A represents the sample data set, N represents the number of data points in the sample data set A; x i and x j represent two random data points in the sample data set A; dist() represents the Euclidean distance calculation function.

[0024] S32: According to the relationship between the Euclidean distance between the sample data and all other data points and the preset cutoff distance d0, calculate the local density p i of each data point in the sample data; the calculation formula is as follows:

[0025]

[0026] wherein, represents a self-defined classification function for distinguishing whether the Euclidean distance between the data point and the center point is less than the cutoff distance, and satisfies:

[0027]

[0028] S33: Based on the local density of each data point in the sample data set, calculate the distance q i, the calculation process is as follows:

[0029] determine whether the local density of the current data point is the maximum value in the sample data set:

[0030] (1) if yes, the current data point x i is the maximum distance from other data points in the sample data set as θ i , the calculation formula is: θ i = max j (d ij ), j = N.

[0031] (2) otherwise, the minimum distance between the data point x j with a local density greater than the current data point in the sample data set and the current data point x i is taken as θ i ; the calculation formula is: θ i = min(d ij ), x j : ρ j > ρ i .

[0032] S34: draw a decision graph according to ρ i and θ i of each sample data in the sample data set. The horizontal coordinate of each sample point in the decision graph is ρ i , and the vertical coordinate is θ i ; and then determine the cluster center and the number of categories according to the decision graph.

[0033] In the present application, the number of categories of the node attribute data set is 3; the category of each cluster center is determined by the corresponding power consumption node state. The category of each node in the node attribute data set is divided into "underload", "steady" or "overload"

[0034] As a further improvement of the present application, in the BP neural network constructed in step S4,

[0035] The hyperbolic tangent function Tanh is used as the activation function between the input layer and the hidden layer, and the Tanh activation function is as follows:

[0036]

[0037] The nonlinear transformation function Sigmoid is used as the activation function between the hidden layer and the output layer, sigmoid has a value range (0, 1) and is monotonous and continuous, and is differentiable everywhere, so it can realize the output of binary classification. The expression of the Sigmoid activation function is as follows:

[0038]

[0039] As a further improvement of the present application, in the BP neural network constructed in step S4,

[0040] The transfer formula from the input layer to the hidden layer is:

[0041]

[0042] In the above formula, x i is the input value of the i-th input layer, i = N, N represents the number of nodes of the input layer; H 1j is the output of the j-th node of the hidden layer, j = 2N + 1; f1 is a Tanh activation function, ω ij is the weight value between the i-th node of the input layer and the j-th node of the hidden layer, x i is the input value of the i-th node of the input layer, a j is the threshold value of the j-th node of the hidden layer.

[0043] The transfer formula from the hidden layer to the output layer is:

[0044]

[0045] In the above formula, y is the output value of the output layer, f2 is a Sigmoid activation function, ω j is the weight value between the j-th node of the hidden layer and the output layer, and b is the threshold value of the output layer.

[0046] As a further improvement of the present application, in step S5, the training process of the BP neural network is as follows:

[0047] S51: Set initial training parameters: including the number of iterations; target minimum error, learning rate η.

[0048] S52: Use the node attribute data set under different operating conditions collected in advance as the training sample, and perform forward propagation on the training sample by using the BP neural network.

[0049] S53: Calculate the sum of absolute values of relative errors E between the predicted output of the output layer and the expected output in each round of forward propagation process, and determine whether the target minimum error is met:

[0050] (1) Yes, the training process of the network model is completed.

[0051] (2) No, go to the back propagation process.

[0052] S54: Perform the back propagation process by using the gradient descent method, and dynamically update the weight values of the output layer, the weight values between the input layer and the hidden layer according to the weight value update formula from the hidden layer to the output layer and the weight value update formula between the input layer and the hidden layer.

[0053] S55: continue inputting the training sample into the BP neural network with updated weights, and re-perform the forward propagation of the next round.

[0054] S56: repeat steps S53-S54 until the preset iteration number is reached, or the sum of absolute values of relative error squares of the network model prediction output and the expected output meets the requirement of the target minimum error.

[0055] As a further improvement of the present application, in step S54, the weight update formula between the input layer and the hidden layer is as follows:

[0056]

[0057] In the above formula, ω i ′ j denotes the updated weight between the i-th node of the input layer and the j-th node of the hidden layer; denotes the expected output of the current input sample in the network model, y denotes the actual output of the current input sample in the network model, and E denotes the sum of squares of relative errors between the expected output and the actual output.

[0058] The weight update formula between the hidden layer and the output layer is as follows:

[0059]

[0060] In the above formula, ω j ′ denotes the updated weight between the j-th node of the hidden layer and the output layer.

[0061] As a further improvement of the present application, in step S5, the process of optimizing the threshold value of the BP neural network using the classical genetic algorithm is as follows:

[0062] S01: Convert the threshold value of the BP neural network into corresponding chromosome individuals using real number coding, and then randomly generate an initial population containing multiple chromosome individuals.

[0063] S02: Set the iteration termination condition of the classical genetic algorithm to be synchronous with the training phase of the BP neural network, and use the classical genetic algorithm to iteratively optimize the initial population, each iteration including the following contents:

[0064] S021: Calculate the fitness of each chromosome in the initial population using a preset fitness function.

[0065] S022: Perform selection operation on the initial population using the selection operator of the classical genetic algorithm.

[0066] S023: Perform crossover operation on the initial population using the crossover operator of the classical genetic algorithm.

[0067] S024: the mutation operator of the classical genetic algorithm is used to perform mutation operation on the initial population.

[0068] S03: in each population iteration process, the threshold value represented by the chromosome with the maximum fitness is output to the BP neural network as the threshold value of the BP neural network in the next training round.

[0069] As a further improvement of the application, in the classical genetic algorithm used in the application, the selection operator uses an elite reservation operator. The crossover operator calculates the similarity between any two individuals in the current population through a self-defined similarity function, and then performs double-point crossover on the two individuals with the lowest similarity. The mutation operator performs single-point mutation on the chromosomes with the highest and lowest fitness according to the preset mutation probability.

[0070] The technical solution provided by the application has the following beneficial effects:

[0071] The application designs a new identification method for the stability of power supply in a transformer area. The method first normalizes the power data of different nodes in the transformer area into standard data and then into a sample data set. Then, an adaptive fast search density peak value method is used to cluster the sample data set of each node. Finally, the clustered multi-node attribute data set is input into a model based on a BP neural network. The model identifies the power supply stability state of the transformer area according to the prediction sample and outputs the identification result.

[0072] The application trains the constructed BP neural network in combination with the collected sample data set, and applies the classical genetic algorithm to the update of the network model threshold value in the training process, thereby effectively shortening the iteration period of the training phase of the network model and improving the identification accuracy of the model.

[0073] Based on the identification method provided by the application, massive transformer area power information can be analyzed and processed, and the analysis result can be obtained quickly. The identification method provided by the application not only has high identification accuracy, but also improves the real-time performance of the algorithm. Moreover, the network model provided by the application can also learn autonomously according to the identification result in the application process, so that the identification accuracy of the method can be continuously improved. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 A step flowchart of a real-time analysis method for the stability of power supply in a low-voltage transformer area provided for embodiment 1 of the application.

[0075] Figure 2 A step flowchart of a real-time analysis method for the stability of power supply in a low-voltage transformer area provided for embodiment 1 of the application.

[0076] Figure 3The program flow chart of the training stage of the BP neural network constructed in Embodiment 1 of the present application.

[0077] Figure 4 The program flow chart of the optimization of the threshold value of the BP neural network by using the classical genetic algorithm in Embodiment 1 of the present application.

[0078] Figure 5 The system architecture diagram of the power supply stability monitoring system based on the fusion terminal provided in Embodiment 2 of the present application.

[0079] Figure 6 The topology diagram of the power supply stability monitoring system based on the fusion terminal.

[0080] Figure 7 The network architecture diagram of the BP neural network model constructed in Embodiment 2 of the present application. DETAILED DESCRIPTION

[0081] In order to make the objectives, technical solutions and advantages of the present application clearer and more comprehensible, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0082] Embodiment 1

[0083] The present embodiment provides a real-time analysis method for the power supply stability of a low-voltage transformer area, which is used to analyze the current operation state of the power supply transformer area according to the power monitoring data collected by the fusion terminal. As shown in the figure, the real-time analysis method specifically comprises the following steps: Figure 1

[0084] S1: Collect the power information of each power consumption node in the transformer area under different states as the sample data of the current node. The sample data includes: the power supply voltage V1, the power supply current I1, the power factor cosφ, and the line loss rate ΔP of the current node. The equipment voltage V2, the equipment current I2, the real-time load P, and the equipment temperature T of the transformer in the transformer area, and other related data related to the operation state of the power grid.

[0085] In addition to the abnormality caused by equipment failure or other accidents in the existing power grid, the most common reason affecting the stable operation of the power grid is the imbalance between supply and demand, and the impact of power demand fluctuations on the user side on the power grid. Therefore, the real-time analysis of the power supply stability in the present embodiment mainly analyzes the power supply and demand balance relationship through the power information of the power supply side and the user side in the power grid, and analyzes and predicts the possible abnormal state.

[0086] ​​Specifically, the power supply voltage V1, the power supply current I1, the device voltage V2, and the device current I2 collected in the embodiment correspond to the voltages and currents of phases A, B, and C, respectively. 1A 1B 1C 1A 1B 1C 2A 2B 2C 2A 2B 2C .

[0087] The power factor refers to the ratio of the active power to the apparent power in an alternating current circuit; it is a coefficient for measuring the efficiency of an electrical device. A low power factor indicates that the circuit has a large amount of reactive power for converting alternating magnetic fields, thereby reducing the utilization rate of the device and increasing the power supply loss. The power factor is related to the load properties of the power grid circuit; it is also an important technical indicator for evaluating the power system.

[0088] The line loss rate is the percentage of the power loss (line loss load) in the power grid to the power supplied to the power network (supply load). The line loss rate can evaluate the economy of the power system operation. The line loss rate is related to the load power factor of the power grid, and the voltage supplied by the power system to the load changes with the active and reactive power transmitted by the line. When the active power transmitted by the line and the initial voltage are constant, the more reactive power is transmitted, the greater the voltage loss of the line and the higher the line loss rate. Therefore, when the power factor is improved, the reactive power absorbed by the load from the system decreases, the voltage loss of the line also decreases, and the line loss rate decreases.

[0089] The device temperature is an indicator for evaluating the tolerance of the transformer device. The transformer needs to dissipate heat during operation. When the real-time load of the transformer increases, the device temperature also increases accordingly. If the environmental temperature and other conditions cause the transformer to be unable to dissipate heat effectively, the device temperature of the transformer will rapidly rise, which may cause the transformer to be unable to work at the rated load, thereby affecting the normal power supply in the transformer area. Therefore, the device temperature of the transformer needs to be considered when analyzing the operation state of the power grid.

[0090] S2: According to the theoretical safety threshold of each data in the operation process, the collected large amount of sample data is normalized, and then a sample data set containing all the normalized sample data is obtained.

[0091] The normalization formula of the sample data is as follows:

[0092] ​​​​​​​​​​​

[0093] In the above formula, x represents the measured value of the current sample data; represents the normalized value of the current sample data; x max represents the upper limit of the theoretical safety threshold of the current sample data; x min represents the lower limit of the theoretical safety threshold of the current sample data; wherein, when there is no lower limit of the safety threshold for a certain sample data, x min = 0.

[0094] In this embodiment, the dimensions and units of each item of data used are different, which is not conducive to data analysis in the later stage. The embodiment eliminates the dimensional influence between different index items through normalization processing. After normalization processing, the embodiment also maps data of different dimensions to the interval (-1, 1), so that the data indicators are comparable. After data standardization processing of the original data, each index is in the same order of magnitude, which is suitable for comprehensive comparison and evaluation.

[0095] S3: Cluster the sample data set of each node based on the adaptive fast search density peak value method, determine the cluster center and the number of categories, and obtain the node attribute data set after clustering.

[0096] As shown in Figure 2 , the process of clustering the sample data set of multiple nodes by the adaptive fast search density peak value method is as follows:

[0097] S31: Obtain the sample data set of any node, calculate the Euclidean distance d ij between any two sample data in the sample data set, and the calculation formula is as follows:

[0098]

[0099] In the above formula, A represents the sample data set, N represents the number of data points in the sample data set A; x i and x j represent two random data points in the sample data set A; dist() represents the Euclidean distance calculation function.

[0100] S32: According to the relationship between the Euclidean distance between the sample data and all other data points and the preset cut-off distance d0, calculate the local density p i of each data point in the sample data. The calculation formula is as follows:

[0101]

[0102] wherein, represents a classification function for distinguishing whether the Euclidean distance between a data point and a center point is less than a cut-off distance, and satisfies:

[0103]

[0104] S33: Calculate the distance θ between the sample data and the density center based on the local density of each data point in the sample data set i The calculation process is as follows:

[0105] Determine whether the local density of the current data point is the maximum value in the sample data set:

[0106] (1) Yes, then the maximum distance between the current data point x i and other data points in the sample data set is taken as θ i , and the calculation formula is: θ i = max j (d ij ), j = N.

[0107] (2) Otherwise, the minimum distance between the data point x j with a local density greater than that of the current data point in the sample data set and the current data point x i is taken as θ i ; the calculation formula is: θ i = min(d ij ), x j : ρ j > ρ i .

[0108] S34: Draw a decision graph according to ρ i and θ i of each sample data in the sample data set. The horizontal coordinate of each sample point in the decision graph is ρ i , and the vertical coordinate is θ i ; and then determine the cluster center and the number of categories according to the decision graph.

[0109] Specifically, the category of each cluster center is determined by the corresponding node state. In this embodiment, the number of categories of the node attribute data set is 3; and the category of each node in the node attribute data set is divided into "underload", "steady" or "overload".

[0110] The adaptive fast search density peak value method provided in this embodiment determines the cluster center by analyzing the density of different data points, and the basis of the clustering is that the density of the cluster center is greater than the density of its surrounding neighboring points. This method can automatically obtain the number of categories from the collected data set without determining the initial cluster center, but quickly finds the density peak value of the data set of any shape by using the feature that the local density of the category center is always higher than the local density of its nearest neighbor point, effectively distributes the non-center sample points, and determines the cluster center by using the decision graph. The adaptive fast search density peak value method provided in this embodiment is suitable for multi-category clustering, and significantly improves the efficiency of clustering processing.

[0111] S4: Construct a BP neural network with three layers of structure, including input layer, hidden layer and output layer. In the BP neural network, the number of input layer nodes n is equal to the number of electricity nodes in the current district; the number of output layer nodes is 1, and the number of hidden layer nodes is 2n+1.

[0112] In the constructed BP neural network,

[0113] The hyperbolic tangent function Tanh is used as the activation function between the input layer and the hidden layer, and the Tanh activation function is as follows:

[0114]

[0115] The nonlinear transformation function Sigmoid is used as the activation function between the hidden layer and the output layer, sigmoid has a value range (0, 1), and is monotonic and continuous everywhere, and is differentiable everywhere, so it can realize the output of binary classification. The expression of the Sigmoid activation function is as follows:

[0116]

[0117] The transfer formula from the input layer to the hidden layer is:

[0118]

[0119] In the above formula, x i is the input value of the i-th input layer, i=N, N represents the number of nodes of the input layer; H 1j is the output of the j-th node of the hidden layer, j=2N+1; f1 is the Tanh activation function, ω ij is the weight value between the i-th node of the input layer and the j-th node of the hidden layer, x i is the input value of the i-th node of the input layer, a j is the threshold value of the j-th node of the hidden layer.

[0120] The transfer formula from the hidden layer to the output layer is:

[0121]

[0122] In the above formula, y is the output value of the output layer, f2 is the Sigmoid activation function, ω j is the weight value between the j-th node of the hidden layer and the output layer, and b is the threshold value of the output layer.

[0123] S5: A large number of node attribute data sets under different operating conditions collected in advance are used as training samples, and the BP neural network constructed in the previous step is trained by using the training samples to update the weight values of the network model. At the same time, the threshold value of the BP neural network is optimized by combining a classical genetic algorithm during the training process. The trained BP neural network is used as the required transformer area operating condition recognition model.

[0124] In this embodiment, as shown in Figure 3 the training process of the BP neural network is as follows:

[0125] S51: Set initial training parameters, including the number of iterations, the target minimum error, and the learning rate η.

[0126] S52: Use the node attribute data sets under different operating conditions collected in advance as training samples, and perform forward propagation of the training samples by using the BP neural network.

[0127] S53: Calculate the sum of the absolute values of the relative errors between the predicted output and the expected output of the output layer during each round of forward propagation E, and determine whether the target minimum error is met:

[0128] (1) Yes, the training process of the network model is completed.

[0129] (2) No, go to the back propagation process.

[0130] S54: Perform the back propagation process by using the gradient descent method, and dynamically update the weight values of the output layer, the input layer, and the hidden layer according to the weight value update formula from the hidden layer to the output layer and the weight value update formula between the input layer and the hidden layer.

[0131] Specifically, the weight value update formula between the input layer and the hidden layer is as follows:

[0132]

[0133] In the above formula, ω i ′ j represents the updated weight value between the i-th node of the input layer and the j-th node of the hidden layer; represents the expected output of the current input sample in the network model, y represents the actual output of the current input sample in the network model, and E represents the sum of the squares of the relative errors between the expected output and the actual output.

[0134] The weight value update formula between the hidden layer and the output layer is as follows:

[0135]

[0136] In the above formula, ω jrepresents the weight value between the updated jth node of the hidden layer and the output layer.

[0137] S55: Continue to input the training sample into the BP neural network with updated weight values, and re-perform the next round of forward propagation.

[0138] S56: Repeat steps S53-S54 until the preset number of iterations is reached, or the sum of the absolute values of the relative errors of the network model prediction output and the expected output meets the target minimum error requirement.

[0139] S6: Obtain the real-time node attribute data set by fusing the real-time power information of the transformer in the terminal area and all power consumption nodes in the terminal area, and sequentially performing normalization and clustering processing on the real-time power information. Then input the real-time node attribute data set into the terminal area operation state recognition model that has been trained, and output the real-time operation state of the current terminal area from the model. The real-time operation state of the terminal area is divided into normal and abnormal.

[0140] In particular, in the present embodiment, as shown in Figure 4 The process of optimizing the threshold value of the BP neural network using the classical genetic algorithm is as follows:

[0141] S01: Convert the threshold value of the BP neural network into corresponding chromosome individuals using real number coding, and then randomly generate an initial population containing multiple chromosome individuals.

[0142] S02: Set the iteration termination condition of the classical genetic algorithm to be synchronized with the training phase of the BP neural network, and use the classical genetic algorithm to iteratively optimize the initial population. Each iteration includes the following contents:

[0143] S021: Calculate the fitness of each chromosome in the initial population using a predetermined fitness function.

[0144] S022: Perform selection operation on the initial population using the selection operator of the classical genetic algorithm.

[0145] S023: Perform crossover operation on the initial population using the crossover operator of the classical genetic algorithm.

[0146] S024: Perform mutation operation on the initial population using the mutation operator of the classical genetic algorithm.

[0147] S03: In each population iteration process, output the threshold value represented by the chromosome with the maximum fitness to the BP neural network as the threshold value of the BP neural network in the next training round.

[0148] In the classical genetic algorithm used in this embodiment, the selection operator uses an elite reservation operator. The crossover operator calculates the similarity between any two individuals in the current population through a self-defined similarity function, and then performs double-point crossover on the two individuals with the lowest similarity. The mutation operator performs single-point mutation on the chromosomes with the highest and lowest fitness according to the preset mutation probability.

[0149] In the final stage of the technical solution, the BP neural network is used to complete the prediction task of the power supply stability state of the power grid according to the node attribute data set of different nodes. The BP neural network uses a step-by-step approximation method. In the continuous algorithm training process, the error analysis is performed on the obtained result and the expected result, and then the weight and threshold are modified, and the model that can output the consistent result with the expected result is obtained step by step. In particular, in order to improve the training effect of the network model, the genetic algorithm is also introduced into the threshold updating process of the network model, thereby significantly improving the convergence rate of the BP neural network model. The network model trained in this embodiment has a high response speed, and thus can dynamically evaluate the power supply state of the transformer area according to the real-time collected data.

[0150] Embodiment 2

[0151] On the basis of embodiment 1, this embodiment further provides a transformer area power supply stability monitoring system based on a fusion terminal. The monitoring is used to collect power information of a transformer and each power consumption node in a transformer area, and then the collected power information is used to perform online analysis on the operation state of the transformer area through the real-time analysis method of the low-voltage transformer area power supply stability in embodiment 1, and to predict the power supply stability of the transformer area.

[0152] As shown in Figure 5 , the monitoring system provided by this embodiment includes a data acquisition device and a master station server. The data acquisition device is installed on the site of the power distribution transformer area, and the master station server is set in the cloud. The data acquisition device and the master station server are in communication connection.

[0153] Among them, the data acquisition device includes: a smart electric energy meter, a concentrator and a fusion terminal. The smart electric energy meter is installed at each power consumption node of the power user, and is used to collect the power information on the user side. The concentrator is used to obtain the collection data of the smart electric energy meter at each power consumption node. The concentrator and the fusion terminal are in communication connection, and the concentrator sends the data obtained from different power consumption nodes to the fusion terminal. The fusion terminal is also electrically connected with the transformer of the low-voltage power grid, and thus obtains the power information on the power grid side. The fusion terminal and the master station server are in communication connection, and the fusion terminal sends the synchronously collected power information on the user side and the power information on the power grid side to the master station server.

[0154] As shown in Figure 6As shown, a data normalization module, a clustering model based on an adaptive fast search density peak value algorithm, and a transformer area operation state recognition model based on a BP neural network are running in the master station server. The data normalization module is used for normalizing the collected power information, and then obtaining a sample data set containing normalized data of all nodes. The clustering model is used for clustering the sample data set of each node, determining the number of categories and the clustering center, and obtaining a node attribute data set containing state information of all nodes. The transformer area operation state recognition model is used for predicting the current power supply stability state of the transformer area according to the node attribute data set; the clustering center of the clustering model is three, corresponding to the “underload”, “stable” or “overload” state of the node respectively. The prediction result of the transformer area operation state recognition model is divided into two states: normal and abnormal.

[0155] The concentrator is in communication connection with the smart meter through an RS485 serial bus interface. The fusion terminal is in communication connection with the concentrator and the transformer area voltage regulator in the form of Ethernet or power carrier communication. The fusion terminal is in communication connection with the master station server in the form of 4G, 5G mobile communication or Ethernet communication.

[0156] In this embodiment, the power information of the user side collected by the fusion terminal includes: power supply voltage V1, power supply current I1, power factor Line loss rate ΔP. The power information of the power grid side includes: device voltage V2 of the transformer area transformer, device current I2, real-time load P, device temperature T.

[0157] The normalization module is used for mapping data of different dimensions to the interval (-1, 1) by using a normalization formula, and the normalization formula is as follows: Wherein, x represents the measured value of the current sample data; x represents the normalized value of the current sample data; x max x represents the upper limit of the theoretical safety threshold of the current sample data; x min x represents the lower limit of the theoretical safety threshold of the current sample data. When there is no safety threshold lower limit for a certain sample data, x min = 0.

[0158] In the monitoring system, the clustering model adopts a clustering model based on an adaptive fast search density peak value algorithm to cluster the sample data set, and the clustering process of the model is as follows:

[0159] First, the sample data set of any node is obtained, the Euclidean distance d ij between any two sample data in the sample data set is calculated, and the calculation formula is as follows:

[0160]

[0161] In the above formula, A represents a sample data set, N represents the number of data points in the sample data set A; x i and x j represent two random data points in the sample data set A; and dist() represents an Euclidean distance calculation function.

[0162] Secondly, according to the relationship between the Euclidean distance between the sample data and all other data points and the preset cut-off distance d0, the local density p i of each data point in the sample data is calculated; the calculation formula is as follows:

[0163]

[0164] wherein, represents a self-defined classification function for distinguishing whether the Euclidean distance between a data point and a center point is less than a cut-off distance, and satisfies:

[0165]

[0166] Then, based on the local density of each data point in the sample data set, the distance q i between the sample data and the density center is calculated, and the calculation process is as follows: it is judged whether the local density of the current data point is the maximum value in the sample data set: if yes, the maximum distance between the current data point x i and other data points in the sample data set is taken as q i , and the calculation formula is: q i = max j (d ij ), j = N. Otherwise, the minimum distance between the data point x j with a local density greater than the current data point in the sample data set and the current data point x i is taken as q i ; and the calculation formula is: q i = min(d ij ), x j : p j > p i .

[0167] Finally, the decision graph is drawn according to p i and q i of each sample data in the sample data set. In the decision graph, the horizontal coordinate of each sample point is p i , and the vertical coordinate is q i ; and then the cluster center and the number of categories are determined according to the decision graph.

[0168] The district operation state recognition model constructed in this embodiment is a BP neural network with a three-layer structure. As Figure 7As shown, the BP neural network includes an input layer, a hidden layer and an output layer. The number of input layer nodes is equal to the number of power consumption nodes in the current district, the number of output layer nodes is 1; the number of hidden layer nodes = 2*input layer node number + 1.

[0169] In the BP neural network, the hyperbolic tangent function Tanh is used as the activation function between the input layer and the hidden layer, and the Tanh activation function is as follows:

[0170]

[0171] The nonlinear transformation function Sigmoid is used as the activation function between the hidden layer and the output layer; the expression of the Sigmoid activation function is as follows:

[0172]

[0173] In the BP neural network constructed in this embodiment, the transfer formula from the input layer to the hidden layer is as follows:

[0174]

[0175] In the above formula, x i is the input value of the i-th input layer, i = N, N represents the number of nodes of the input layer. H 1j is the output of the j-th node of the hidden layer, j = 2N + 1. f1 is the Tanh activation function, ω ij is the weight value between the i-th node of the input layer and the j-th node of the hidden layer, x i is the input value of the i-th node of the input layer, a j is the threshold value of the j-th node of the hidden layer.

[0176] The transfer formula from the hidden layer to the output layer is as follows:

[0177]

[0178] In the above formula, y is the output value of the output layer, f2 is the Sigmoid activation function, ω j is the weight value between the j-th node of the hidden layer and the output layer, and b is the threshold value of the output layer.

[0179] In the district operation state recognition model provided in this embodiment, the training process of the BP neural network is as follows:

[0180] S1: Set initial training parameters: including the number of iterations; target minimum error, learning rate η.

[0181] S2: Use the node attribute data set collected in advance under different operation states as training samples, and use the BP neural network to perform forward propagation on the training samples.

[0182] S3: Calculate the absolute value of the relative error square sum of the predicted output of the output layer and the expected output in each forward propagation process, and determine whether the target minimum error is met:

[0183] (1) Yes, the training process of the network model is completed.

[0184] (2) No, enter the back propagation process.

[0185] S4: The gradient descent method is used for the back propagation process, and the output layer weight, the weight between the input layer and the hidden layer are dynamically updated according to the weight update formula between the hidden layer and the output layer, and the weight update formula between the input layer and the hidden layer.

[0186] The weight update formula between the input layer and the hidden layer is as follows:

[0187]

[0188] In the above formula, ω i ′ j represents the updated weight between the i th node of the input layer and the j th node of the hidden layer, represents the expected output of the current input sample in the network model, y represents the actual output of the current input sample in the network model, and E represents the relative error square sum of the expected output and the actual output.

[0189] The weight update formula between the hidden layer and the output layer is as follows:

[0190]

[0191] In the above formula, ω j ′ represents the updated weight between the j th node of the hidden layer and the output layer.

[0192] S5: Continue to input the training sample into the BP neural network with updated weights, and perform the next round of forward propagation.

[0193] S6: Loop steps S3-S4 until the preset iteration number is reached, or the relative error absolute value square sum of the network model prediction output and the expected output meets the target minimum error requirement.

[0194] In this embodiment, the iteration number and the error less than the preset value are used as the iteration termination conditions at the same time, and when any one of the two conditions is reached, the recursive process of the planting algorithm training is terminated.

[0195] In this embodiment, in order to improve the training effect and convergence rate of the algorithm model, a genetic algorithm is also used for threshold optimization in the BP neural network training process, and the optimization process is as follows: In the foregoing embodiment, it has been described in detail, and will not be repeated here.

[0196] Embodiment 3

[0197] The embodiment provides a real-time analysis device for low-voltage power supply stability of a power supply area. The analysis device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the real-time analysis method for low-voltage power supply stability of a power supply area in Embodiment 1 are implemented, and the prediction result of the power supply stability of the power supply area is analyzed according to the collected power information in the power supply area.

[0198] The computer device can be a smartphone, a tablet computer, a notebook computer, a desktop computer, a rack server, a blade server, a tower server, or a cabinet server (including a single server or a server cluster composed of multiple servers), which can execute programs. The computer device in the embodiment at least includes, but is not limited to, a memory and a processor which can be connected to each other through a system bus.

[0199] In the embodiment, the memory (i.e., a readable storage medium) includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory can be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. of the computer device. Of course, the memory can include both the internal storage unit and the external storage device of the computer device. In the embodiment, the memory is usually used to store an operating system and various application software installed on the computer device, etc. In addition, the memory can also be used to temporarily store various data that have been output or will be output.

[0200] The processor in some embodiments can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor is usually used to control the overall operation of the computer device. In the embodiment, the processor is used to run the program code or process data stored in the memory.

[0201] The above merely describes preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A real-time analysis method for low-voltage power supply stability of a power supply area, which is used to analyze the current power supply area operation state according to the power monitoring data collected by a fusion terminal; characterized in that, The real-time analysis method comprises the following steps: S1: collect the power information of each power consumption node in the transformer area under different states as the sample data of the current node; the sample data includes: the power supply voltage V1, the power supply current I1, the power factor of the current node, the line loss rate ΔP; the equipment voltage V2, the equipment current I2, the real-time load P, the equipment temperature T of the transformer in the transformer area, and other related data related to the operation state of the power grid; S2: According to the theoretical safety threshold of each data in the running process, the collected large amount of sample data is normalized respectively, and then a sample data set containing all the normalized sample data is obtained; S3: Based on the adaptive fast search density peak value method, the sample data set of each node is clustered to determine the cluster center and the number of categories, and a node attribute data set after clustering is obtained; S4: A BP neural network with a three-layer structure is constructed; the BP neural network comprises an input layer, a hidden layer and an output layer; the number of input layer nodes n is equal to the number of power consumption nodes in the current transformer area, the number of output layer nodes is 1, and the number of hidden layer nodes is 2n+1; S5: A large number of node attribute data sets under different running states collected in advance are used as training samples, the BP neural network constructed in the above step is trained by using the training samples, and the weights of the network model are updated; and in the training process, the threshold value of the BP neural network is optimized by combining the classical genetic algorithm; the trained BP neural network is used as the required transformer area running state recognition model; S6: The real-time power information of the transformer area transformer and all power consumption nodes in the transformer area is collected by a fusion terminal, the real-time power information is sequentially normalized and clustered to obtain real-time node attribute data sets, and then the real-time node attribute data sets are input into the trained transformer area running state recognition model, and the real-time running state of the current transformer area is output by the model; wherein the real-time running state of the transformer area is divided into normal and abnormal.

2. The real-time analysis method for low-voltage transformer area power supply stability according to claim 1, characterized in that: The power supply voltage V1, the power supply current I1, the equipment voltage V2 and the equipment current I2 correspond to the voltages and currents of phases A, B and C respectively; Including: V 1A , V 1B , V 1C , I 1A , I 1B , I 1C , V 2A , V 2B , V 2C , I 2A , I 2B , I 2C .

3. The real-time analysis method for low-voltage area power supply stability according to claim 2, characterized in that: In step S2, the normalization processing formula of the sample data is as follows: In the above formula, x represents the measured value of the current sample data; represents the normalized value of the current sample data; x max represents the upper limit of the theoretical safety threshold of the current sample data; x min represents the lower limit of the theoretical safety threshold of the current sample data; wherein, when there is no lower limit of the safety threshold for a certain sample data, then x min = 0.

4. The real-time analysis method for low-voltage area power supply stability according to claim 1, characterized in that: In step S3, the process of clustering the sample data set of multiple nodes by the adaptive fast search density peak value method is as follows: S31: Obtain a sample data set of any node, and calculate the Euclidean distance d between any two sample data in the sample data set ij The calculation formula is as follows: In the above formula, A represents the sample data set, N represents the number of data points in the sample data set A; x i and x j represent two random data points in the sample data set A; dist() represents the Euclidean distance calculation function; S32: Calculate the local density p of each data point in the sample data according to the relationship between the Euclidean distance between the sample data and all other data points and the preset cut-off distance d0 i ; the calculation formula is as follows: wherein, represents a custom classification function for distinguishing whether the Euclidean distance of a data point from a center point is less than a cutoff distance, and satisfies: S33: Calculate the distance θ of the sample data from the density center based on the local density at each data point of the sample data set i , the calculation process is as follows: judging whether the local density of the current data point is the maximum value in the sample data set; if yes, taking the maximum distance between the current data point x i and other data points in the sample data set as θ i ; the calculation formula is: θ i = max j (d ij ), j = N; otherwise, taking the minimum distance between the data point x j with local density greater than the current data point in the sample data set and the current data point x i as θ i ; the calculation formula is: θ i = min(d ij ), x j : p j > p i ; S34: according to the ρ of each sample data in the sample data set i and θ i Draw a decision graph, in which the horizontal coordinate of each sample point is ρ i , and the vertical coordinate is θ i ; and then determine the cluster center and the number of categories according to the decision graph; the number of categories is 3; the category of each cluster center is determined by the corresponding power consumption node state; the category of each node in the node attribute data set is divided into "underload", "steady" or "overload".

5. The real-time analysis method for low-voltage area power supply stability according to claim 1, characterized in that: In the BP neural network constructed in step S4, The hyperbolic tangent function Tanh is used as an activation function between the input layer and the hidden layer, and the Tanh activation function is as follows: A nonlinear transformation function Sigmoid is used as an activation function between the hidden layer and the output layer; the expression of the Sigmoid activation function is as follows:

6. The real-time analysis method for low-voltage area power supply stability according to claim 5, characterized in that: In the BP neural network constructed in step S4, The transfer formula from the input layer to the hidden layer is as follows: In the above formula, x i is the input value of the i-th input layer, i = N, N represents the number of nodes of the input layer; H 1j is the output of the j-th node of the hidden layer, j = 2N + 1; f1 is a Tanh activation function, ω ij is the weight value between the i-th node of the input layer and the j-th node of the hidden layer, x i is the input value of the i-th node of the input layer, a j is the threshold value of the j-th node of the hidden layer; The transfer formula from the hidden layer to the output layer is as follows: In the above formula, y is the output value of the output layer, f2 is a Sigmoid activation function, ω j is the weight between the jth node of the hidden layer and the output layer, and b is the threshold value of the output layer.

7. The real-time analysis method for low-voltage area power supply stability according to claim 6, characterized in that: In step S5, the training process of the BP neural network is as follows: S51: Set the initial training parameters, including the number of iterations, the target minimum error and the learning rate η; S52: The node attribute data sets collected in advance under different running states are used as training samples, and the training samples are propagated forward by using the BP neural network; S53: Calculate the sum of the absolute values of the relative errors between the predicted output and the expected output of the output layer in each round of forward propagation, and determine whether the target minimum error is met; if yes, the training process of the network model is completed, otherwise, the reverse propagation process is entered; S54: the back propagation process is performed by using the gradient descent method, and the output layer weight values and the weight values between the input layer and the hidden layer are dynamically updated according to the weight value updating formula from the hidden layer to the output layer and the weight value updating formula between the input layer and the hidden layer; S55: the training sample is continuously input into the BP neural network after the weight values are updated, and the next round of forward propagation is performed again; S56: steps S53-S54 are cycled until a preset iteration number is reached, or the relative error absolute value square sum of the network model prediction output and the expected output meets the target minimum error requirement.

8. The real-time analysis method for low-voltage area power supply stability according to claim 7, characterized in that: In step S54, the weight value updating formula between the input layer and the hidden layer is as follows: In the above formula, ω′ ij represents the weight value between the updated input layer i-th node and the hidden layer j-th node; represents the expected output of the current input sample in the network model, y represents the actual output of the current input sample in the network model, and E represents the relative error sum of squares of the expected output and the actual output. The weight value updating formula between the hidden layer and the output layer is as follows: In the above equation, ω'j j denotes the updated weight between the jth node of the hidden layer and the output layer.

9. The real-time analysis method for low-voltage area power supply stability according to claim 8, characterized in that: In step S5, The process of optimizing the threshold value of the BP neural network by using the classical genetic algorithm is as follows: S01: the threshold value of the BP neural network is converted into a corresponding chromosome individual by using real number coding, and an initial population containing multiple chromosome individuals is randomly generated; S02: the iteration termination condition of the classical genetic algorithm is set to be synchronous with the training stage of the BP neural network, and the initial population is iteratively optimized by using the classical genetic algorithm, and each iteration includes the following contents; S021: the fitness of each chromosome in the initial population is calculated by using a preset fitness function; S022: the selection operator of the classical genetic algorithm is used to perform selection operation on the initial population; S023: the crossover operator of the classical genetic algorithm is used to perform crossover operation on the initial population; S024: the mutation operator of the classical genetic algorithm is used to perform mutation operation on the initial population; S03: in each population iteration process, the threshold value represented by the chromosome with the highest fitness is output to the BP neural network as the threshold value of the BP neural network in the next training round.

10. The real-time analysis method for low-voltage area power supply stability according to claim 9, characterized in that: The selection operator uses an elite reservation operator; the crossover operator calculates the similarity between any two individuals in the current population by using a self-defined similarity function, and then performs double-point crossover on the two individuals with the lowest similarity; the mutation operator performs single-point mutation on the chromosome with the highest fitness and the chromosome with the lowest fitness according to a preset mutation probability.

Images