Low-voltage prediction method and device for power distribution area

Abstract

The invention provides a low-voltage prediction method and device for a power distribution area. The method comprises the following steps: constructing a load prediction model and a load-voltage mapping model; acquiring current power load data of any one service node of the power distribution area; taking the current electrical load data as the input of a load data prediction model, and obtaining predicted electrical load data of the service node in a future preset duration; and taking the predicted electrical load data as the input of the load-voltage mapping model, and obtaining predicted voltage data of the service node in a future preset duration. According to the method, the node voltage prediction combination model taking the user load data of the service nodes as input is established, real-time and rapid low-voltage prediction of the power distribution area is realized, the prediction accuracy is high, full-coverage on-line prediction of the voltage of the power distribution area can be realized without information such as a power distribution area topological structure, line parameters and user voltage, and the power distribution area voltage prediction efficiency is improved. The training time is shorter, and the device is simpler, lighter and easy to implement.

Description

Technical Field

[0001] The present disclosure relates to the technical field of power grid data processing, and in particular to a method, device, storage medium and electronic device for predicting low voltage in a distribution station area. Background Art

[0002] The voltage qualification rate is an important indicator of electricity quality. Among them, the low voltage problem in the low-voltage distribution station area is an important cause of the reduction of the voltage qualification rate. Due to the characteristics of the low-voltage distribution network such as the long power supply radius, unreasonable wire line conditions, and dispersed power load, the voltage in the substation area is more likely to be low. In addition, there is currently a lack of effective online voltage monitoring methods, which often causes problems such as delayed perception of low voltage problems by users in the substation area, and passive and disorderly power grid repair work. Therefore, in order to facilitate timely treatment measures for low voltage problems in the substation area, simple and accurate low voltage prediction in the substation area is essential.

[0003] At present, the technology for voltage estimation mainly focuses on the calculation of power grid line loss. In terms of voltage estimation, traditional methods are mostly limited to using power flow calculation to achieve voltage estimation. Some proposed a quick judgment method for low voltage problems of feeders under various typical load distribution conditions based on the voltage loss coefficient table; some constructed a nonlinear estimation model of the maximum voltage loss of hybrid feeders based on the voltage loss analysis of overhead lines and cable lines, and used the Gauss-Raphson method to give a solution method for the parameters in the estimation model; some proposed a power demand model based on the Monte Carlo simulation method, and discussed the effectiveness of this method in analyzing voltage quality. However, these methods can only obtain the results of voltage drop through power flow calculation when the power grid topology, line parameters and voltage or power operation data are known, and the calculation process is relatively complicated, and it is impossible to achieve rapid voltage prediction in the distribution station area. Summary of the invention

[0004] The purpose of the embodiments of the present disclosure is to provide a method, device, storage medium and electronic device for predicting low voltage in a power distribution station area, so as to solve the problem of complicated voltage prediction process and low prediction efficiency in the power distribution station area in the prior art.

[0005] The embodiments of the present disclosure adopt the following technical solutions: a method for predicting low voltage in a distribution station area, comprising: constructing a load prediction model and a load-voltage mapping model, wherein the load prediction model is used to predict the power load data of the access node in the distribution station area, and the load-voltage mapping model is used to map the relationship between the power load data and the voltage of the access node; obtaining the current power load data of any access node in the distribution station area; using the current power load data as the input of the load data prediction model to obtain the predicted power load data of the access node within a preset time period in the future; using the predicted power load data as the input of the load-voltage mapping model to obtain the predicted voltage data of the access node within a preset time period in the future.

[0006] In some embodiments, the construction of the load prediction model and the load-voltage mapping model includes: obtaining historical power load data and historical voltage data of each user access node in the distribution station area; training a load data prediction model based on the historical power load data; and training a load-voltage mapping model based on the historical power load data and the historical voltage data.

[0007] In some embodiments, the load data prediction model is constructed based on a long short-term memory neural network.

[0008] In some embodiments, the long short-term memory neural network includes at least an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer; wherein each hidden layer includes at least a plurality of memory cells, and each memory cell includes at least an input gate, an output gate and a forget gate.

[0009] In some embodiments, the first hidden layer includes at least 288 memory cells, the second hidden layer includes at least 100 memory cells, and the third hidden layer includes at least 50 memory cells.

[0010] In some embodiments, the load-voltage mapping model is constructed based on a back-propagation neural network.

[0011] In some embodiments, the back-propagation neural network is trained based on the Levenberg-Marquardt method.

[0012] The disclosed embodiment provides a low voltage prediction device for a distribution station area, comprising: a model construction module, used to construct a load prediction model and a load-voltage mapping model, wherein the load prediction model is used to predict the power load data of the customer access node in the distribution station area, and the load-voltage mapping model is used to map the relationship between the power load data and the voltage of the customer access node; a data acquisition module, used to obtain the current power load data of any customer access node in the distribution station area; a first prediction module, used to use the current power load data as an input of the load data prediction model to obtain the predicted power load data of the customer access node within a preset time period in the future; and a second prediction module, used to use the predicted power load data as an input of the load-voltage mapping model to obtain the predicted voltage data of the customer access node within a preset time period in the future.

[0013] An embodiment of the present disclosure provides a storage medium storing a computer program, which implements the steps of the above-mentioned low voltage prediction method for a distribution station area when executed by a processor.

[0014] An embodiment of the present disclosure provides an electronic device, comprising at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned low voltage prediction method for a distribution station area when executing the computer program on the memory.

[0015] The beneficial effects of the embodiments of the present disclosure are as follows: by analyzing the voltage drop mechanism of the distribution station area, utilizing the correlation between the power load data and the voltage of the household nodes in the distribution station area, training the load prediction model and the load-voltage mapping model, and establishing a node voltage prediction combination model with the user load data of the household nodes as input, real-time and fast low voltage prediction of the distribution station area can be achieved with high prediction accuracy, and full coverage online prediction of the voltage in the distribution station area can be achieved without the need for information such as the substation topology structure, line parameters and user voltage, less training time, and more simple, lightweight and easy to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 A flowchart of a low voltage prediction method for a distribution station area according to a first embodiment of the present disclosure;

[0018] Figure 2A schematic diagram of a process for constructing a load prediction model and a load-voltage mapping model according to the first embodiment of the present disclosure;

[0019] Figure 3 This is a schematic diagram of the unit structure of the memory cell of the first embodiment of the present disclosure;

[0020] Figure 4 It is a schematic diagram of the architecture of the BP network of the first embodiment of the present disclosure;

[0021] Figure 5 A schematic diagram of load prediction results of a certain household node in a distribution station area according to the first embodiment of the present disclosure;

[0022] Figure 6 A schematic diagram showing a comparison result between a voltage prediction result and actual voltage data according to the first embodiment of the present disclosure;

[0023] Figure 7 It is a schematic diagram of the structure of a low voltage prediction device for a distribution station area according to the second embodiment of the present disclosure. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0025] The voltage qualification rate is an important indicator of electricity quality. Among them, the low voltage problem in the low-voltage distribution station area is an important cause of the reduction of the voltage qualification rate. Due to the characteristics of the low-voltage distribution network such as the long power supply radius, unreasonable wire line conditions, and dispersed power load, the voltage in the substation area is more likely to be low. In addition, there is currently a lack of effective online voltage monitoring methods, which often causes problems such as delayed perception of low voltage problems by users in the substation area, and passive and disorderly power grid repair work. Therefore, in order to facilitate timely treatment measures for low voltage problems in the substation area, simple and accurate low voltage prediction in the substation area is essential.

[0026] At present, the technology for voltage estimation mainly focuses on the calculation of power grid line loss. In terms of voltage estimation, traditional methods are mostly limited to using power flow calculation to achieve voltage estimation. Some proposed a quick judgment method for low voltage problems of feeders under various typical load distribution conditions based on the voltage loss coefficient table; some constructed a nonlinear estimation model of the maximum voltage loss of hybrid feeders based on the voltage loss analysis of overhead lines and cable lines, and used the Gauss-Raphson method to give a solution method for the parameters in the estimation model; some proposed a power demand model based on the Monte Carlo simulation method, and discussed the effectiveness of this method in analyzing voltage quality. However, these methods can only obtain the results of voltage drop through power flow calculation when the power grid topology, line parameters and voltage or power operation data are known, and the calculation process is relatively complicated, and it is impossible to achieve rapid voltage prediction in the distribution station area.

[0027] In order to solve the above problems, the first embodiment of the present disclosure provides a method for predicting low voltage in a distribution station area, and its flow chart is as follows: Figure 1 As shown, it mainly includes steps S10 to S40:

[0028] S10, constructing a load forecasting model and a load-voltage mapping model;

[0029] S20, obtaining current power load data of any access node in the distribution area;

[0030] S30, using the current power load data as the input of the load data prediction model to obtain the predicted power load data of the access node within a preset time period in the future;

[0031] S40, using the predicted power load data as an input of a load-voltage mapping model to obtain predicted voltage data of the access node within a preset time period in the future.

[0032] With the popularization of smart meters, the massive voltage and load data of users in the distribution area provide a realistic basis for data mining. Further, by analyzing the voltage drop mechanism in the distribution area and coordinating with intelligent data collection, the relationship between load data and voltage data can be mapped and represented by a neural network model. This embodiment constructs a load prediction model and a load-voltage mapping model, which are used to predict the change of power load data of the access node in the distribution area over time, and to represent the mapping relationship between power load data and voltage data; based on the trained load prediction model, the power load situation in the future preset time period can be predicted according to the current power load data of the access node, and then the load-voltage mapping model is combined with the predicted power load situation to determine the voltage situation of the access node in the future preset time period, and the prediction of the voltage of each access node in the distribution area is completed, so as to facilitate the early adoption of corresponding measures for low voltage problems.

[0033] It should be noted that there is no necessary execution order between the above steps S10 and S20. It is sufficient to ensure that the model building and training are completed before the actual prediction. Figure 1 The execution sequence shown in is only an example and does not limit the specific execution steps.

[0034] Specifically, when constructing the load forecasting model and the load-voltage mapping model, the following can be combined: Figure 2 The process shown in the figure is implemented, mainly including steps S11 to S13:

[0035] S11, obtaining historical power load data and historical voltage data of each user access node in the distribution station area;

[0036] S12, training a load data prediction model based on historical power load data;

[0037] S13, training a load-voltage mapping model according to historical power load data and historical voltage data.

[0038] When the model is actually trained, the historical power load data and historical voltage data of each access node can be obtained through the smart meters at each access node in the distribution station area and at each user. Specifically, it can be all power load data and historical voltage data within a certain period of time before the current moment. Due to the popular application of smart meters, all power consumption data at all times can be recorded as training samples for model training, and a part of them can be extracted as test samples.

[0039] The load data prediction model is mainly built by learning the changes of power load data over time series. In actual implementation, it can be constructed through a long short-term memory neural network (LSTM), that is, the LSTM network is used to fully learn the characteristics of different intervals and different levels of load time series data to realize the prediction of load data. The LSTM network includes at least an input layer, multiple hidden layers and an output layer, wherein each hidden layer includes at least multiple memory cells. When training, some parameters of the network need to be set and debugged first, such as the number of hidden layers, the number of memory cells, the dimensions of the input layer and the output layer, the number of hidden layers, and the time step.

[0040] Since load forecasting belongs to sequence forecasting, the input feature quantity is the historical power load data of the user group on all access nodes in the substation area, and the output feature quantity is the predicted power load data of the access node, so the input and output dimensions can be determined as one-dimensional. Considering that the number of hidden layers will greatly affect the performance of the neural network, this embodiment verifies the effect of the LSTM model with different hidden layers. The verification results are shown in Table 1. From the verification results, it can be seen that when the hidden layer is 3 layers, the model accuracy is the highest and the calculation time is short. Therefore, this embodiment sets a total of three hidden layers, namely the first hidden layer, the second hidden layer and the third hidden layer, and sets the first hidden layer to include at least 288 memory cells, the second hidden layer to include at least 100 memory cells, and the third hidden layer to include at least 50 memory cells.

[0041] Table 1

[0042] Number of hidden layers / layer Training time(s) Calculation time (ms) Accuracy (%) 1 59 0.18 93.6 2 149 0.29 94.2 3 388 4.39 98.9 4 891 8.44 96.5 5 173 12.11 95.8

[0043] In some embodiments, the Adam optimizer can be selected as the optimization algorithm of the LSTM network of this embodiment. The Adam optimizer has the advantages of fast convergence speed, simple implementation, and efficient calculation, and is widely used in neural network models. In addition, the gradient threshold of the LSTM network is set to 1, the initial learning rate is specified to be 0.005, and the number of training times is set to 250. After 125 rounds of training, the learning rate is reduced by multiplying by a factor of 0.2.

[0044] Figure 3 The schematic diagram of the unit structure of the memory cell is shown. Each memory cell includes at least an input gate, an output gate, and a forget gate. The input f of the forget gate is t From the hidden state h of the previous sequence t-1 With this sequence data x t It is calculated through the Sigmoid activation function, that is, it is used to determine whether the previous information has an impact on the unit memory at this time.

[0045] Specifically, the input f of the forget gate t The expression is:

[0046] f t =σ(W f [h t-1 ,x t ]+b f ); (1)

[0047] Among them, W f and b fare the coefficients and bias coefficients of the linear relationship, and σ is the Sigmoid activation function. The input gate is used to calculate which information is saved to the state unit, including two parts of information, one is i t , which is used to represent how much information of the current input needs to be saved to the unit state, and the other part is It is used to represent the new information generated by the current input. The two parts generate a new memory state. The formula is:

[0048]

[0049] Therefore, the current cell state is the product of the forget gate input and the state of the previous moment plus the product of the two parts of the input gate, and the expression is:

[0050]

[0051] The state of the output gate is used to calculate the degree to which information is output at the current moment. The output gate is used to calculate the output value of the memory cell at time t. The formula is:

[0052]

[0053] After the model is built, the LSTM network can be trained based on the historical power load data of each household node in the current distribution station area. After the training round reaches the preset training round, the model training is considered complete. At this time, the prediction effect of the model can be evaluated in combination with the test data. Specifically, the loss function reflects the degree of fit of the model to the data, that is, the difference between the value predicted by the model and the actual value. The better the model fit effect, the smaller the loss function. In this embodiment, the root mean square error (MSE) can be used as the loss function of the model, and its calculation formula is:

[0054]

[0055] in, is the predicted power load data of the distribution station area, y t is the historical electricity load data, n is the number of samples, and t is the time. The root mean square error reflects the deviation between the predicted value and the true value, representing the overall reliability of the prediction. The smaller its value is, the smaller the prediction difference is, that is, the more reliable the prediction is overall. In actual implementation, data other than MSE can also be used to represent it, such as mean absolute error (MAE) or median absolute error (MedAE).

[0056] The load-voltage mapping model is mainly used to reflect the correlation between power load data and voltage data. It can be constructed using a back propagation neural network (BP), which uses the inherent logic between abstract learning data and nonlinear mathematical models to fit and establish the mapping relationship between data. In the framework of the BP network, it consists of three parts: the input layer, the hidden layer, and the output layer. Figure 4 As shown, where X 1 , X 2 ,……,X n Represents the input feature quantity, Y 1 , Y 2 ,……,Y m Representing the output feature quantity, combined with the neuron setting of the hidden layer, the mapping function of X and Y can be established when there is an objective logical relationship between X and Y.

[0057] In this embodiment, the power load data of the access node can be used as the input feature quantity, and the voltage data can be used as the output feature quantity. The BP network is used to establish a mapping relationship between the two to achieve voltage prediction in the distribution area. It should be noted that in the actual distribution area configuration, the topology generally does not change, so the feature quantity dimension of the BP network usually does not change.

[0058] In the actual implementation process, the hidden layer activation function of the BP network in this embodiment uses a rectified linear unit (ReLU) with strong nonlinear representation ability and robustness. For the output layer, this embodiment uses the Sigmoid function as the activation function, and its definition formula is:

[0059]

[0060] Its derivative with respect to x can be expressed by itself:

[0061]

[0062] Furthermore, in order to effectively avoid its shortcomings such as slow convergence speed, easy to fall into local optimum, poor stability, etc., this embodiment adopts the Levenberg-Marquardt method (Levenberg-Marquardt, LM) with the fastest propagation speed as the training algorithm.

[0063] After the BP network is constructed, it can be trained according to the historical power load data and historical voltage data, and the load-voltage mapping model can be obtained after training. In addition, the accuracy of the load-voltage mapping model can also be evaluated through test data, and the load-voltage mapping model can be further optimized based on the evaluation results.

[0064] This embodiment analyzes the voltage drop mechanism of the distribution station area, utilizes the relationship between the power load data and the voltage of the household nodes in the distribution station area, trains the load prediction model and the load-voltage mapping model, and establishes a node voltage prediction combination model with the user load data of the household nodes as input, so as to realize real-time and fast low voltage prediction of the distribution station area with high prediction accuracy. In addition, it can realize full coverage online prediction of the voltage in the distribution station area without the need for information such as the substation topology structure, line parameters and user voltage, and it takes less training time. It is simpler, lighter and easier to implement.

[0065] The effectiveness of the low voltage prediction method of this embodiment is described below with reference to examples.

[0066] First, the power load data of the access nodes in a distribution substation for 24 hours a day and a total of one quarter are obtained. Then, a simulation test model is built using Simulink to obtain the corresponding voltage data. Finally, 2160 sets of power and voltage data of the substation are sorted out, and the first 90% of the data are used as training samples, and the remaining 10% of the data are used as test samples.

[0067] In this embodiment, the number of hidden layers of the load forecasting model is set to 3, and the number of memory cells in each layer is 288, 100, and 50 respectively. The Adam optimizer is selected as the optimization algorithm of the model, the gradient threshold is set to 1, the initial learning rate is specified to be 0.005, and the number of training times is set to 250. After 125 rounds of training, the learning rate is reduced by multiplying by a factor of 0.2. The specific calculation process is that the forget gate calculates the forget value through formula (1), the memory gate calculates the information to be remembered using formula (2), and then the output information and the memory information are added and transmitted to the next unit through formula (3) as the memory input of the memory gate of the next unit, and finally the output gate calculates the output value through formula (4).

[0068] For the load-voltage mapping model, the training error is set to 0.0001, the learning rate is set to 0.01, and the maximum number of iterations is set to 500. Since the neural network uses the 13 node powers of this substation as input and the 18 node voltages as output, based on experience and actual measurements, a 13-10-10-18 double hidden layer neural network can meet the requirements. The model training uses the Levenberg-Marquardt algorithm to automatically correct the model weights and biases, and finally establish the mapping between input features and output features.

[0069] The load forecast result of a randomly selected access node in the substation area is as follows: Figure 5 As shown in the figure, the first 90% of the node power data is used as training samples, and the LSTM load forecasting network model built in this paper is used to make short-term forecasts of the load in the substation area, and the predicted load data is obtained and compared with the remaining 10% of the load data. Figure 5It can be seen that the curve fluctuation trend of the node load prediction value is consistent with the actual value, and the degree of agreement is high. This is due to the fact that LSTM has the ability to maintain memory based on memory units, and has the ability to capture time series dependencies in time series data. In addition, since short-term load forecasting is not sensitive to weather factors, but has high requirements for the accuracy of historical data. Therefore, as long as the correctness of the data sample is guaranteed, the good accuracy of the LSTM-based load forecasting can be guaranteed. In addition, the root mean square error of the load forecasting result is calculated according to formula (5). The results show that the root mean square error of the prediction result of this method is approximately 0.1, and the prediction accuracy is high, indicating that the LSTM neural network model has a very impressive prediction effect on the node load in the distribution station area, which provides a guarantee for the reliable prediction of the station area voltage.

[0070] The prediction results of the above nodes are input into the BP network to obtain the corresponding voltage prediction results, which are compared with the actual voltage data. The comparison results are as follows: Figure 6 As shown. Figure 6 It can be seen that the estimated value and actual value of the node voltage are almost the same, indicating that the BP network has a good fitting effect on the node load and voltage in the distribution station area. At the same time, according to the simulation verification results, within the voltage error of 1V, the voltage prediction accuracy of this method is as high as 100%, which meets the actual engineering needs.

[0071] Based on the same inventive concept, the second embodiment of the present disclosure provides a low voltage prediction device for a distribution station area, and its structural schematic diagram is shown as follows: Figure 7 As shown, it mainly includes a model building module 10, a data acquisition module 20, a first prediction module 30 and a second prediction module 40, wherein the model building module 10 is used to build a load prediction model and a load-voltage mapping model, wherein the load prediction model is used to predict the power load data of the access node of the distribution station area, and the load-voltage mapping model is used to map the relationship between the power load data and the voltage of the access node; the data acquisition module 20 is used to obtain the current power load data of any access node in the distribution station area; the first prediction module 30 is used to use the current power load data as the input of the load data prediction model to obtain the predicted power load data of the access node within a preset time period in the future; the second prediction module 40 is used to use the predicted power load data as the input of the load-voltage mapping model to obtain the predicted voltage data of the access node within a preset time period in the future.

[0072] In some embodiments, the model building module 10 is specifically used to obtain historical power load data and historical voltage data of each access node in the distribution station area; train a load data prediction model based on historical power load data; and train a load-voltage mapping model based on historical power load data and historical voltage data.

[0073] In some embodiments, the load data prediction model is constructed based on a long short-term memory neural network.

[0074] In some embodiments, the long short-term memory neural network includes at least an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer; wherein each hidden layer includes at least a plurality of memory cells, and each memory cell includes at least an input gate, an output gate and a forget gate.

[0075] In some embodiments, the first hidden layer includes at least 288 memory cells, the second hidden layer includes at least 100 memory cells, and the third hidden layer includes at least 50 memory cells.

[0076] In some embodiments, the load-voltage mapping model is constructed based on a back-propagation neural network.

[0077] In some embodiments, the back-propagation neural network is trained based on the Levenberg-Marquardt method.

[0078] This embodiment analyzes the voltage drop mechanism of the distribution station area, utilizes the relationship between the power load data and the voltage of the household nodes in the distribution station area, trains the load prediction model and the load-voltage mapping model, and establishes a node voltage prediction combination model with the user load data of the household nodes as input, so as to realize real-time and fast low voltage prediction of the distribution station area with high prediction accuracy. In addition, it can realize full coverage online prediction of the voltage in the distribution station area without the need for information such as the substation topology structure, line parameters and user voltage, and it takes less training time. It is simpler, lighter and easier to implement.

[0079] Based on the same inventive concept, a third embodiment of the present disclosure provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the low voltage prediction method for a distribution station area provided by the first embodiment of the present disclosure.

[0080] Based on the same inventive concept, the fourth embodiment of the present disclosure provides an electronic device, comprising at least a memory and a processor, wherein a computer program is stored on the memory, and when the processor executes the computer program on the memory, the steps of the low voltage prediction method for the distribution station area provided in the first embodiment of the present disclosure are implemented.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A method for predicting low voltage in a distribution station area. It is characterized in that include: Constructing a load prediction model and a load-voltage mapping model, wherein the load prediction model is used to predict the power load data of the access node of the distribution station area, and the load-voltage mapping model is used to map the relationship between the power load data and the voltage of the access node; Obtain the current power load data of any access node in the distribution area; Using the current power load data as input to a load data prediction model, the predicted power load data of the access node within a preset time period in the future is obtained; The predicted power load data is used as the input of the load-voltage mapping model to obtain the predicted voltage data of the user access node within a future preset time period.

2. The low voltage prediction method according to claim 1, It is characterized in that The construction of the load prediction model and the load-voltage mapping model includes: Obtain the historical power load data and historical voltage data of each user access node in the distribution station area; Training a load data prediction model based on the historical power load data; A load-voltage mapping model is trained according to the historical power load data and the historical voltage data.

3. The low voltage prediction method according to claim 2, It is characterized in that The load data prediction model is constructed based on a long short-term memory neural network.

4. The low voltage prediction method according to claim 3, It is characterized in that The long short-term memory neural network comprises at least an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer; wherein each hidden layer comprises at least a plurality of memory cells, and each memory cell comprises at least an input gate, an output gate and a forget gate.

5. The low voltage prediction method according to claim 4, It is characterized in that The first hidden layer includes at least 288 memory cells, the second hidden layer includes at least 100 memory cells, and the third hidden layer includes at least 50 memory cells.

6. The low voltage prediction method according to claim 1, It is characterized in that The load-voltage mapping model is constructed based on a back-propagation neural network.

7. The low voltage prediction method according to claim 6, It is characterized in that The back propagation neural network is trained based on the Levenberg-Marquardt method.

8. A low voltage prediction device for a distribution station area, It is characterized in that include: A model building module, used to build a load prediction model and a load-voltage mapping model, wherein the load prediction model is used to predict the power load data of the user access node in the distribution station area, and the load-voltage mapping model is used to map the relationship between the power load data and the voltage of the user access node; A data acquisition module is used to obtain the current power load data of any access node in the distribution area; A first prediction module, configured to use the current power load data as an input of a load data prediction model to obtain predicted power load data of the access node within a preset time period in the future; The second prediction module is used to use the predicted power load data as the input of the load-voltage mapping model to obtain the predicted voltage data of the user node within a preset time period in the future.

9. A storage medium storing a computer program, It is characterized in that When the computer program is executed by a processor, the steps of the method for predicting low voltage in a distribution station area according to any one of claims 1 to 7 are implemented.

10. An electronic device comprising at least a memory and a processor, wherein the memory stores a computer program. It is characterized in that The processor implements the steps of the low voltage prediction method for the distribution station area according to any one of claims 1 to 7 when executing the computer program on the memory.

Images