Voltage prediction method, device and equipment for photovoltaic power distribution network nodes and storage medium

Abstract

The invention discloses a voltage prediction method and device for photovoltaic power distribution network nodes, equipment and a storage medium. The method comprises the steps of obtaining real-time measurement data of each photovoltaic node in a power distribution network; inputting the real-time measurement data of the photovoltaic nodes into a deep residual network obtained by pre-training each non-photovoltaic node in the power distribution network; and inputting the real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node into a pre-trained voltage prediction model, and predicting the voltage amplitude of each node in the power distribution network at the future moment based on the voltage prediction model. The non-linear mapping relation between the real-time measurement data of the photovoltaic nodes and the voltage amplitude of the non-photovoltaic nodes is represented through the deep residual network, the problem that the real-time data of the non-photovoltaic nodes are difficult to obtain is solved, and therefore voltage prediction of all the nodes in the power distribution network is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of power distribution networks, and relates to a voltage prediction method, and in particular to a voltage prediction method, device, equipment and storage medium for photovoltaic power distribution network nodes. Background Art

[0002] With the development of photovoltaic power generation technology, large-scale photovoltaic power generation has been gradually connected to the power grid. However, photovoltaic power generation is volatile and random, which poses severe challenges to the stable operation of the power grid. For some distribution networks with high photovoltaic penetration rates, changes in photovoltaic output can easily lead to large fluctuations in the voltage of each node in the distribution network. Therefore, it is urgent to improve the real-time perception and short-term prediction capabilities of the power grid operation status to ensure the safe and reliable operation of the distribution network. The distribution network has many points and is limited by equipment costs and installation environments. It is difficult to deploy high-precision real-time measurement equipment at each node. It is difficult to obtain the real-time voltage and voltage development trend of nodes without real-time measurement.

[0003] In the related technologies, in response to the current situation of insufficient real-time measurement, pseudo-measurement data generation methods are often used. Traditional methods mostly use mathematical statistical models based on historical data, and do not effectively utilize some of the available real-time measurement data, resulting in large deviations between the generated data and the true value and low accuracy. Therefore, it is urgent to combine historical data with some of the available real-time measurement data, generate high-precision node voltage pseudo-measurement data in a data-driven way, and make short-term high-precision predictions of each node voltage based on real-time measurement and pseudo-measurement data, thereby improving the stability and reliability of power grid operation. Summary of the invention

[0004] In view of this, the present invention discloses a voltage prediction method, device, equipment and storage medium for photovoltaic distribution network nodes, which can solve the deficiencies existing in the related art.

[0005] To achieve the above purpose, the present invention discloses the following technical solutions:

[0006] According to a first aspect of the present invention, a voltage prediction method for a photovoltaic distribution network node is proposed, wherein the distribution network includes photovoltaic nodes and non-photovoltaic nodes, and the method includes:

[0007] Acquire real-time measurement data of each photovoltaic node in the distribution network, wherein the real-time measurement data includes the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node at the current moment;

[0008] Inputting the real-time measurement data of the photovoltaic node into a deep residual network pre-trained for each non-photovoltaic node in the distribution network, the deep residual network is used to generate the real-time voltage amplitude of the corresponding non-photovoltaic node at the current moment according to the real-time measurement data;

[0009] The real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node are input into a pre-trained voltage prediction model, and the voltage amplitude of each node in the distribution network at a future moment is predicted based on the voltage prediction model.

[0010] According to a second aspect of the present invention, a voltage prediction device for a photovoltaic distribution network node is proposed, wherein the distribution network includes photovoltaic nodes and non-photovoltaic nodes, and the method includes:

[0011] An acquisition unit: acquiring real-time measurement data of each photovoltaic node in the distribution network, wherein the real-time measurement data includes the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node at the current moment;

[0012] An input unit: inputs the real-time measurement data of the photovoltaic node into a deep residual network pre-trained for each non-photovoltaic node in the distribution network, and the deep residual network is used to generate the real-time voltage amplitude of the corresponding non-photovoltaic node at the current moment according to the real-time measurement data;

[0013] Prediction unit: input the real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node into a pre-trained voltage prediction model, and predict the voltage amplitude of each node in the distribution network at a future moment based on the voltage prediction model.

[0014] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0015] processor;

[0016] a memory for storing processor-executable instructions;

[0017] The processor implements the steps of the method described in the first aspect by running the executable instructions.

[0018] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0019] It can be seen from the above technical solutions that the voltage prediction method for photovoltaic distribution network nodes disclosed in the present invention, on the one hand, uses a deep residual network to characterize the nonlinear mapping relationship between the real-time measurement data of photovoltaic nodes and the voltage amplitude of non-photovoltaic nodes, thereby solving the problem that real-time data of non-photovoltaic nodes is difficult to obtain, thereby realizing voltage prediction for all nodes in the distribution network; on the other hand, the voltage is predicted by a voltage prediction model obtained through historical data training, which combines real-time measurement data and historical data, improves the accuracy of voltage prediction, and provides efficient guidance for voltage estimation and prediction of distribution network nodes in actual projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of a voltage prediction method for a photovoltaic distribution network node provided by an exemplary embodiment.

[0021] Figure 2 It is a schematic diagram of a deep residual network structure provided by an exemplary embodiment.

[0022] Figure 3 It is a schematic diagram of a temporal convolution-long short-term memory network structure provided by an exemplary embodiment.

[0023] Figure 4 It is a schematic diagram of feature extraction in a voltage prediction model provided by an exemplary embodiment.

[0024] Figure 5 It is a schematic structural diagram of a device provided by an exemplary embodiment.

[0025] Figure 6 It is a block diagram of a voltage prediction device for a photovoltaic distribution network node provided by an exemplary embodiment. DETAILED DESCRIPTION

[0026] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of one or more embodiments of the present invention as detailed in the appended claims.

[0027] It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in the present invention. In some other embodiments, the steps included in the method may be more or less than those described in the present invention. In addition, a single step described in the present invention may be decomposed into multiple steps for description in other embodiments; and multiple steps described in the present invention may be combined into a single step for description in other embodiments.

[0028] To further illustrate the present invention, the following examples are provided:

[0029] With the development of photovoltaic power generation technology, large-scale photovoltaic power generation has been gradually connected to the power grid. However, photovoltaic power generation is volatile and random, which poses severe challenges to the stable operation of the power grid. For some distribution networks with high photovoltaic penetration rates, changes in photovoltaic output can easily lead to large fluctuations in the voltage of each node in the distribution network. Therefore, it is urgent to improve the real-time perception and short-term prediction capabilities of the power grid operation status to ensure the safe and reliable operation of the distribution network. The distribution network has many points and is limited by equipment costs and installation environments. It is difficult to deploy high-precision real-time measurement equipment at each node. It is difficult to obtain the real-time voltage and voltage development trend of nodes without real-time measurement.

[0030] In the related technologies, in response to the current situation of insufficient real-time measurement, pseudo-measurement data generation methods are often used. Traditional methods mostly use mathematical statistical models based on historical data, and do not effectively utilize some of the available real-time measurement data, resulting in large deviations between the generated data and the true value and low accuracy. Therefore, it is urgent to combine historical data with some of the available real-time measurement data, generate high-precision node voltage pseudo-measurement data in a data-driven way, and make short-term high-precision predictions of each node voltage based on real-time measurement and pseudo-measurement data, thereby improving the stability and reliability of power grid operation.

[0031] In order to solve the deficiencies existing in the related art, the present invention proposes a voltage prediction method for photovoltaic distribution network nodes.

[0032] Figure 1 The flowchart of a voltage prediction method for a photovoltaic distribution network node provided by an exemplary embodiment is as follows, wherein the distribution network includes photovoltaic nodes and non-photovoltaic nodes. Figure 1 As shown, the method may include the following steps:

[0033] Step 102: acquiring real-time measurement data of each photovoltaic node in the distribution network, wherein the real-time measurement data includes the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node at the current moment.

[0034] The distribution network is an important part of the power system. It is responsible for reducing the voltage of high-voltage electricity sent from the transmission network and distributing it to various power users. Photovoltaic nodes refer to specific points in the distribution network that are connected to photovoltaic power generation systems. Photovoltaic power generation systems usually include solar panels, inverters and other necessary components, which convert solar energy into electrical energy and feed it into the distribution network through this node. Non-photovoltaic nodes refer to all other nodes in the distribution network that are not connected to the photovoltaic power generation system. These nodes may be connected to traditional power generation resources (such as thermal power generation, hydropower generation, etc.) or simply other connection points in the distribution network.

[0035] Since non-PV nodes are often equipped with traditional power systems, they may be relatively stable, with little change in voltage amplitude, and the frequency of uploading measurement data by the equipment used to detect non-PV nodes is relatively low. Therefore, compared with PV nodes, real-time measurement data of non-PV nodes is more difficult to obtain.

[0036] Step 104: input the real-time measurement data of the photovoltaic node into a deep residual network pre-trained for each non-photovoltaic node in the distribution network, and the deep residual network is used to generate the real-time voltage amplitude of the corresponding non-photovoltaic node at the current moment according to the real-time measurement data.

[0037] Step 106, inputting the real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node into a pre-trained voltage prediction model, and predicting the voltage amplitude of each node in the distribution network at a future time based on the voltage prediction model.

[0038] In this embodiment, on the one hand, a deep residual network is used to characterize the nonlinear mapping relationship from the real-time measurement data of the photovoltaic node to the voltage amplitude of the non-photovoltaic node, which solves the problem of difficulty in obtaining real-time data for non-photovoltaic nodes, thereby realizing voltage prediction for all nodes in the distribution network; on the other hand, the voltage is predicted by a voltage prediction model obtained through historical data training, which combines real-time measurement data and historical data to improve the accuracy of voltage prediction, thereby providing efficient guidance for distribution network node voltage estimation and prediction in actual projects.

[0039] In one embodiment, before performing voltage prediction on each node of the distribution network, the method further includes: obtaining historical meter data of all nodes in the distribution network, and performing power flow calculation on the historical meter data to obtain network training data; wherein the historical meter data includes historical power data of each node and network topology information of the distribution network, and the network training data includes: voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at different times, and voltage amplitude of each non-photovoltaic node.

[0040] Power flow calculation is a basic electrical calculation method in power system analysis. Its main purpose is to solve the voltage of each node when the power network parameters and the injection amount of each node are known. Specifically, the combined formula is as follows:

[0041]

[0042] Among them, P i is the active injection power of node i in the distribution network; Q i is the reactive power injected by node i in the distribution network; G ij is the sum of all conductances connected to node i; B ij is the sum of all susceptances connected to node i; V i is the voltage amplitude of node i in the distribution network; V j is the voltage amplitude of node j in the distribution network; θ ij is the electrical phase angle difference between node i and node j; n is the total number of nodes in the system.

[0043] The network training data obtained based on the above formula includes the voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at different times, and the voltage amplitude of each non-photovoltaic node. The data thus obtained will form the required data pair in is the voltage amplitude of each photovoltaic node in the distribution network; is the current amplitude of each photovoltaic node in the distribution network; is the injected active power of each photovoltaic node in the distribution network; is the injected reactive power of each photovoltaic node in the distribution network; V i It is the node voltage amplitude of each non-PV node in the distribution network.

[0044] In this embodiment, network training data is obtained by performing power flow calculation on historical electricity meter data, so as to prepare for the subsequent generation of a deep learning model or a training sample set for the network.

[0045] Furthermore, the method also includes: determining a deep residual network for each non-photovoltaic node, and generating a first training sample set based on the network training data, wherein each group of training samples in the first training sample set includes the voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at the same time, and the voltage amplitude of the corresponding non-photovoltaic node; training the deep residual network corresponding to each non-photovoltaic node according to the training sample set; wherein the trained deep residual network is used to characterize the nonlinear mapping relationship between the real-time measurement data of the photovoltaic node and the corresponding non-photovoltaic node.

[0046] Deep Residual Network (ResNet) is a variant of traditional deep neural network. Unlike traditional neural network, which directly establishes the mapping relationship from input to output, i.e. H(x) = F(x), deep residual network establishes the mapping relationship from input to residual of expected output, i.e. H(x)-x = F(x). Through residual learning, when the number of network layers increases, the network learning efficiency can be higher, while reducing the impact of the gradient vanishing problem. Within a certain depth range, deep residual network has higher training efficiency and accuracy than traditional neural network.

[0047] For each non-photovoltaic node, a deep residual network is trained. The input of the deep residual network is the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node in the distribution network, and the output is the voltage amplitude of the non-photovoltaic node.

[0048] Specifically, a deep residual neural network is trained for each node. The deep residual network has an 18-layer structure, consisting of a fully connected layer and several residual blocks. Relu is used as the activation function in the residual block, and one-dimensional convolution is used to replace the original two-dimensional convolution layer. Dropout is used to prevent overfitting, and Gaussian noise is added to the residual block to enhance its learning and generalization capabilities. The network is trained in a regression manner, where the input features are the voltage amplitudes of each photovoltaic node in the distribution network. Current amplitude of each photovoltaic node in the distribution network Active power injected into each photovoltaic node in the distribution network Injected reactive power of each photovoltaic node in the distribution network The corresponding output is the node voltage amplitude V of the non-PV node in the distribution network i The target error function is composed of the weighted addition of the absolute mean error and the mean square error, and the Adam optimization algorithm is used for training.

[0049] like Figure 2 As shown in Figure 1, the deep residual network is composed of multiple residual blocks stacked together. Each basic residual block consists of a weight layer, an activation function layer, an identity mapping, and a normalization layer. H(x) is regarded as a basic mapping fitted by several stacked layers. x represents the input of the first layer in these layers, and the residual function is F(x) = H(x)-x. The main structural expression of each residual block is as follows:

[0050] x i+1 =x i +F(x i ,w i )

[0051] Among them, w i is the weight of the i-th residual block.

[0052] The deep residual network is composed of multiple residual blocks, and then the expression of the deep residual network can be obtained as:

[0053]

[0054] Among them, x0 is the input of the deep residual network, x L is the output of the deep residual network.

[0055] In this embodiment, a deep residual network is used to characterize the nonlinear mapping relationship from the real-time measurement data of the photovoltaic node to the voltage amplitude of the non-photovoltaic node. On the one hand, it solves the problem of difficulty in obtaining real-time data of non-photovoltaic nodes, thereby realizing voltage prediction of all nodes in the distribution network; on the other hand, since the deep residual network has higher training efficiency and accuracy than the traditional neural network, and the deep residual network adopts a residual structure, the residual structure can be quickly and flexibly adjusted according to actual conditions, and can efficiently handle multi-node and multi-layer network training, and can provide guidance for voltage estimation and prediction of larger-scale distribution network systems.

[0056] In one embodiment, the method further includes: generating a second training sample set based on the network training data, each group of training samples in the second training sample set including: the voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at the same time, the voltage amplitude of each non-photovoltaic node at the same time, and the actual voltage amplitude of each node in the distribution network at the next moment; inputting the second training sample set into a model to be trained constructed based on a temporal convolution-long short-term memory network, so that the model to be trained outputs the predicted voltage amplitude of each node at the next moment; comparing the actual voltage amplitude with the predicted voltage amplitude, and optimizing the model to be trained according to the comparison result until the prediction effect of the model to be trained meets expectations.

[0057] Temporal convolutional neural network (TCN) is a new network structure based on convolutional neural network (CNN), which is suitable for processing time series data. The main structure of TCN includes causal convolution and dilated convolution. This structure makes the network layers have causal relationships, and each layer of the network has stronger perception ability, which is conducive to the network to extract time series features. In addition, residual links are added to solve the problem of gradient disappearance or gradient explosion. Long short-term memory neural network (LSTM) is a variant of recurrent neural network RNN. It introduces the gate structure of input gate, output gate and forget gate, and introduces the cell state function. It is good at processing long-term and short-term information in time series. The temporal convolution-long short-term memory (TCN-LSTM) network uses TCN based on the attention mechanism to extract high-dimensional time series features of long history data, and assigns different weights to the extracted feature variables to increase the effect of important features. It uses LSTM's powerful nonlinear fitting ability for time series data to capture the evolution trend of high-dimensional time series features, and further realizes the prediction of node voltage amplitude changes.

[0058] The voltage prediction model may include a feature extraction layer, an attention layer, a long short-term memory network layer, and a fully connected layer. The feature extraction layer is used to extract the time series feature vector of the data input into the voltage prediction model according to the dilated causal convolution of the time series convolutional network, the attention layer is used to weight the time series feature vector, the long short-term memory network layer is used to transfer the weighted time series feature vector and the voltage fluctuation trend according to the hidden state of the long short-term memory network, and the fully connected layer is used to output the voltage amplitude of each node in the distribution network at the next moment.

[0059] For each node in the distribution network, based on the node voltage data, a temporal convolution-long short-term memory network with attention mechanism is trained. Figure 3 As shown in Figure 1, the temporal convolution-long short-term memory network with attention mechanism is mainly composed of TCN layer, attention layer and LSTM layer. Figure 4 As shown in Figure 1, after the time series convolution network extracts the features of the voltage sequence of each node, it weights the feature vector through the attention layer. The initial voltage data is expanded by the time series convolution and causal convolution to obtain a high-dimensional time series feature vector H i,t =[h 1,t ,h 2,t ,…,g i,t ], using the attention mechanism to add weight A to the high-dimensional time series feature vector i,t =[a 1,t ,a 2,t ,…,a i,t ], the weight calculation formula is as follows:

[0060]

[0061] Among them, s(h i ,q) is the weight coefficient, and its calculation formula is as follows:

[0062] s(h i ,q)=v T tanh(Wh i +Uq)

[0063] Among them, W, U and v are the network's learnable parameter matrices and bias vectors, and h i is the input high-dimensional time series feature vector, and q is the query vector.

[0064] This method introduces learnable parameters W, U and v to transform the query vector q and the original input high-dimensional time series feature vector h i Mapped to different vector spaces for weight calculation, the calculated weight A is further i,t =[a 1,t ,a 2,t ,…,a i,t ] and the original eigenvector H i,t =[h 1,t ,h 2,t ,…,h i,t ] and multiply them to get the weighted characteristic voltage vector S i,t =[s 1,t ,s 2,t ,…,s i,t ] as the input of the subsequent long short-term memory network. The weighted feature voltage vector S extracted by the temporal convolutional network feature i,t The long short-term memory network is input in sequence. The long short-term memory network transmits the weighted feature vector through the hidden state and learns the voltage fluctuation trend. Furthermore, the processed feature vector is input into the fully connected layer, and the voltage amplitude at the specified future moment is output.

[0065] In one embodiment, the distribution network is a photovoltaic high penetration distribution network. A photovoltaic high penetration distribution network refers to a distribution network in which photovoltaic power generation accounts for a high proportion of the distribution network and has a significant impact on the operation of the distribution network.

[0066] Figure 5 is a schematic structural diagram of a device provided by an exemplary embodiment. Figure 5At the hardware level, the device includes a processor 502, an internal bus 504, a network interface 506, a memory 509, and a non-volatile memory 510, and may also include hardware required for other functions. One or more embodiments of the present invention may be implemented based on software, such as the processor 502 reading the corresponding computer program from the non-volatile memory 510 into the memory 508 and then running it. Of course, in addition to the software implementation, one or more embodiments of the present invention do not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but may also be hardware or logic devices.

[0067] Please refer to Figure 6 , a voltage prediction device for photovoltaic distribution network nodes can be applied to Figure 6 In the device shown, to implement the technical solution of the present invention, the distribution network includes photovoltaic nodes and non-photovoltaic nodes, and the device may include:

[0068] An acquisition unit 602 is used to acquire real-time measurement data of each photovoltaic node in the distribution network, wherein the real-time measurement data includes the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node at the current moment;

[0069] A first input unit 604 is used to input the real-time measurement data of the photovoltaic node into a deep residual network pre-trained for each non-photovoltaic node in the distribution network, and the deep residual network is used to generate a real-time voltage amplitude of the corresponding non-photovoltaic node at the current moment according to the real-time measurement data;

[0070] The prediction unit 606 is used to input the real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node into a pre-trained voltage prediction model, and predict the voltage amplitude of each node in the distribution network at a future time based on the voltage prediction model.

[0071] Optionally, before performing voltage prediction on each node of the distribution network, the method further includes:

[0072] The calculation unit 608 is used to obtain the historical meter data of all nodes in the distribution network, and perform power flow calculation on the historical meter data to obtain network training data; wherein the historical meter data includes the historical power data of each node and the network topology information of the distribution network, and the network training data includes: the voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at different times, and the voltage amplitude of each non-photovoltaic node.

[0073] Optionally, the method further includes:

[0074] A first generating unit 610 is used to determine a deep residual network for each non-photovoltaic node, and generate a first training sample set according to the network training data, wherein each group of training samples in the first training sample set includes the voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at the same time, and the voltage amplitude of the corresponding non-photovoltaic node;

[0075] The training unit 612 is used to train the deep residual network corresponding to each non-photovoltaic node according to the training sample set; wherein the trained deep residual network is used to characterize the nonlinear mapping relationship between the real-time measurement data of the photovoltaic node and the corresponding non-photovoltaic node.

[0076] Optionally, the method further includes:

[0077] A second generating unit 614 is used to generate a second training sample set according to the network training data, wherein each group of training samples in the second training sample set includes: the voltage amplitude, current amplitude, active power, and reactive power of each photovoltaic node at the same time, the voltage amplitude of each non-photovoltaic node at the same time, and the real voltage amplitude of each node in the distribution network at the next time;

[0078] A second input unit 616 is used to input the second training sample set into a to-be-trained model constructed based on a temporal convolution-long short-term memory network, so that the to-be-trained model outputs a predicted voltage amplitude of each node at the next moment;

[0079] The optimization unit 618 is used to compare the actual voltage amplitude with the predicted voltage amplitude, and optimize the model to be trained according to the comparison result until the prediction effect of the model to be trained meets expectations.

[0080] Optionally, the voltage prediction model includes a feature extraction layer, an attention layer, a long short-term memory network layer, and a fully connected layer. The feature extraction layer is used to extract the time series feature vector of the data input into the voltage prediction model according to the dilated causal convolution of the time series convolutional network, the attention layer is used to weight the time series feature vector, the long short-term memory network layer is used to transfer the weighted time series feature vector and the voltage fluctuation trend according to the hidden state of the long short-term memory network, and the fully connected layer is used to output the voltage amplitude of each node in the distribution network at the next moment.

[0081] Optionally, the deep residual network includes multiple residual blocks, each residual block includes a weight layer, an activation function layer, an identity mapping and a normalization layer, and the output of each residual block is the output of the next residual block.

[0082] Optionally, the distribution network is a photovoltaic high penetration distribution network.

[0083] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, a game console, a tablet computer, a wearable device or a combination of any of these devices.

[0084] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0085] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0086] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0087] With respect to the computer-readable medium (or computer-readable storage medium) as described above or in any other form, computer instructions may be stored thereon, and when the instructions are executed by a processor, one or more of the above-mentioned embodiments are implemented, thereby realizing the technical solution of the present invention.

[0088] The present invention also proposes a computer program, which, when executed by a processor, implements one or more of the above embodiments, thereby realizing the technical solution of the present invention. The computer program may be specifically recorded in the above or any other form of computer-readable medium, and the present invention is not limited thereto.

[0089] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0090] The above describes specific embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] The terms used in one or more embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of the present invention. The singular forms of "a", "said" and "the" used in one or more embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0092] It should be understood that although the terms first, second, third, etc. may be used to describe various information in one or more embodiments of the present invention, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0093] The above description is merely a preferred embodiment of one or more embodiments of the present invention and is not intended to limit one or more embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present invention shall be included in the scope of protection of one or more embodiments of the present invention.

Claims

1. A voltage prediction method for a photovoltaic distribution network node, characterized in that: The distribution network includes photovoltaic nodes and non-photovoltaic nodes, and the method includes: Acquire real-time measurement data of each photovoltaic node in the distribution network, wherein the real-time measurement data includes the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node at the current moment; Inputting the real-time measurement data of the photovoltaic node into a deep residual network pre-trained for each non-photovoltaic node in the distribution network, the deep residual network is used to generate the real-time voltage amplitude of the corresponding non-photovoltaic node at the current moment according to the real-time measurement data; The real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node are input into a pre-trained voltage prediction model, and the voltage amplitude of each node in the distribution network at a future moment is predicted based on the voltage prediction model.

2. The method according to claim 1, characterized in that Before performing voltage prediction on each node of the distribution network, the method further includes: Obtain historical meter data of all nodes in the distribution network, and perform power flow calculation on the historical meter data to obtain network training data; wherein the historical meter data includes historical power data of each node and network topology information of the distribution network, and the network training data includes: voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at different times, and voltage amplitude of each non-photovoltaic node.

3. The method according to claim 2, characterized in that The method further comprises: Determine a deep residual network for each non-photovoltaic node, and generate a first training sample set based on the network training data, wherein each group of training samples in the first training sample set includes the voltage amplitude, current amplitude, active power, reactive power of each photovoltaic node at the same time, and the voltage amplitude of the corresponding non-photovoltaic node; The deep residual network corresponding to each non-photovoltaic node is trained according to the training sample set; wherein the trained deep residual network is used to characterize the nonlinear mapping relationship between the real-time measurement data of the photovoltaic node and the corresponding non-photovoltaic node.

4. The method according to claim 2, characterized in that: The method further comprises: Generate a second training sample set according to the network training data, wherein each group of training samples in the second training sample set includes: the voltage amplitude, current amplitude, active power, and reactive power of each photovoltaic node at the same time, the voltage amplitude of each non-photovoltaic node at the same time, and the real voltage amplitude of each node in the distribution network at the next time; Inputting the second training sample set into a to-be-trained model constructed based on a temporal convolution-long short-term memory network, so that the to-be-trained model outputs a predicted voltage amplitude of each node at the next moment; The actual voltage amplitude and the predicted voltage amplitude are compared, and the model to be trained is optimized according to the comparison result until the prediction effect of the model to be trained meets expectations.

5. The method according to claim 1, characterized in that: The voltage prediction model includes a feature extraction layer, an attention layer, a long short-term memory network layer, and a fully connected layer. The feature extraction layer is used to extract the time series feature vector of the data input into the voltage prediction model according to the dilated causal convolution of the time series convolutional network, the attention layer is used to weight the time series feature vector, the long short-term memory network layer is used to transfer the weighted time series feature vector and the voltage fluctuation trend according to the hidden state of the long short-term memory network, and the fully connected layer is used to output the voltage amplitude of each node in the distribution network at the next moment.

6. The method according to claim 1, characterized in that The deep residual network includes multiple residual blocks, each residual block includes a weight layer, an activation function layer, an identity mapping and a normalization layer, and the output of each residual block is the output of the next residual block.

7. The method according to claim 1, characterized in that The distribution network is a photovoltaic high penetration distribution network.

8. A voltage prediction device for a photovoltaic distribution network node, characterized in that: The distribution network includes photovoltaic nodes and non-photovoltaic nodes, and the method includes: An acquisition unit: acquiring real-time measurement data of each photovoltaic node in the distribution network, wherein the real-time measurement data includes the voltage amplitude, current amplitude, active power and reactive power of each photovoltaic node at the current moment; An input unit: inputs the real-time measurement data of the photovoltaic node into a deep residual network pre-trained for each non-photovoltaic node in the distribution network, and the deep residual network is used to generate the real-time voltage amplitude of the corresponding non-photovoltaic node at the current moment according to the real-time measurement data; Prediction unit: input the real-time voltage amplitude corresponding to each photovoltaic node and the real-time voltage amplitude corresponding to each non-photovoltaic node into a pre-trained voltage prediction model, and predict the voltage amplitude of each node in the distribution network at a future moment based on the voltage prediction model.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor implements the steps of the method according to any one of claims 1 to 7 by running the executable instructions.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images