CN114818455A

CN114818455A - Power system multi-agent transient state stability judging method and system for small amount of PMU sampling

Info

Publication number: CN114818455A
Application number: CN202111151150.9A
Authority: CN
Inventors: 王国政; 马士聪; 郭剑波; 卜广全; 王铁柱; 范士雄; 周子涵; 赵兵; 荆逸然; 徐浩田; 罗魁; 侯玮琳
Original assignee: State Grid Zhejiang Electric Power Co Ltd; China Electric Power Research Institute Co Ltd CEPRI
Current assignee: State Grid Zhejiang Electric Power Co Ltd; China Electric Power Research Institute Co Ltd CEPRI
Priority date: 2021-09-29
Filing date: 2021-09-29
Publication date: 2022-07-29
Anticipated expiration: 2041-09-29
Also published as: CN114818455B

Abstract

The invention discloses a power system multi-agent transient state stability judging method and system for sampling a small amount of PMUs. Wherein, the method comprises the following steps: the method comprises the steps of obtaining a sample data set, wherein the sample data in the sample data set are characteristic data collected by a plurality of PMU nodes in a simulated power system, and each sample data has a label representing whether the power system can keep transient stability; training a pre-training artificial intelligence network based on a sample data set, and extracting a key node set from a plurality of PMU nodes according to a training result, wherein the key nodes in the key node set are PMU nodes with a preset number selected from the plurality of PMU nodes; based on characteristic data corresponding to each key node in the key node set, training the artificial intelligent network to obtain a transient stability discrimination network; and inputting the electrical quantity data collected by PMU nodes consistent with the installation positions of all key nodes in the key node set in the actual power system into a transient stability judging network, and outputting the transient stability judging result of the actual power system.

Description

Power system multi-agent transient state stability judging method and system for small amount of PMU sampling

Technical Field

The present invention relates to the technical field of artificial intelligence in a power system, and more particularly, to a power system multi-agent transient state stability judging method and system with few PMU samples, a storage medium and an electronic device.

Background

With the development of smart power grids and the improvement of social power consumption, in order to ensure the economic and efficient operation of power grids, the transmission capacity of a power transmission network is gradually approaching the operation boundary of the power grids, and in recent years, with the access of high-permeability renewable energy sources and high-proportion power electronic devices to a power system, the operation mode of the power grids is changeable, the induction mechanism of the transient stability problem of the system is complex, and the processing speed of the traditional simulation analysis method is slower and slower. Although the stability problem does not occur frequently, once the transient stability of the system is damaged, the instability information is not obtained in time, and serious consequences such as generator set disconnection, power equipment damage, major power failure accidents and the like are possibly caused. Therefore, a reliable transient stability evaluation method is an important basis for ensuring safe and reliable operation of the power system.

The development of intelligent technology provides a new mode for judging the transient stability of the power system. The intelligent algorithm generally adopts an off-line training and on-line application mode, and the trained intelligent agent can meet the requirement of rapid real-time stability judgment. However, as the power system is larger and larger, the feature input dimension of the intelligent algorithm of the features is increased exponentially, and the training speed and the online application efficiency of the intelligent agent are seriously influenced. It has been shown that the computational power required for training complex neural networks has exceeded the speed of hardware development, and that the CO2 generated by the power consumed by the training process has reached an uneconomical state, so that it is important to characterize the data to simplify the neural network model.

Wide Area Monitoring Systems (WAMS) in power systems today collect data for a minimum unit based on PMU installation points. The existing feature extraction method (such as a self-encoder and a depth network) can reduce the dimension, but cannot extract the feature from the dimension of the installation position of a power system PMU. This is mainly because PMU location information is not typically part of the input data, and therefore the structure of a single agent is difficult to implement feature extraction with complex functions, requiring additional manual assistance.

In view of the technical problem that the conventional feature extraction method in the prior art is difficult to implement feature extraction of a power system with complex functions, an effective solution is not proposed at present.

Disclosure of Invention

The invention provides a multi-agent transient state stability judging method and a multi-agent transient state stability judging system for a power system with small amount of PMU sampling, aiming at the technical problem that the existing feature extraction method in the prior art is difficult to realize the feature extraction of the power system with complex functions.

According to one aspect of the invention, a power system multi-agent transient stability judging method for small-quantity PMU sampling is provided, which comprises the following steps:

the method comprises the steps of obtaining a sample data set, wherein the sample data in the sample data set are characteristic data collected by a plurality of PMU nodes in a simulated power system, and each sample data has a label representing whether the power system can keep transient stability;

training a pre-training artificial intelligence network based on a sample data set, and extracting a key node set from a plurality of PMU nodes according to a training result, wherein the key nodes in the key node set are PMU nodes with a preset number selected from the plurality of PMU nodes;

based on characteristic data corresponding to each key node in the key node set, training the artificial intelligent network to obtain a transient stability discrimination network;

acquiring electrical quantity data collected by PMU nodes consistent with the installation positions of all key nodes in the key node set in the actual power system, inputting the collected electrical quantity data into a transient stability judging network, and outputting the transient stability judging result of the actual power system.

Optionally, the obtaining the sample data set includes:

acquiring initial sample data with time domain information, which is acquired by a plurality of PMU nodes in a power system, by adopting a simulation mode;

selecting a voltage item of a PMU node from initial sample data as characteristic data; and

and generating labels which are corresponding to the characteristic data and represent whether the simulated power system can keep transient stability or not according to the maximum power angle difference delta max of the generator in the simulation process.

Optionally, the training of the pre-training artificial intelligence network based on the sample data set, and extracting a key node set from the plural PMU nodes according to a training result, include:

selecting any PMU node from the PMU nodes, and training a pre-training artificial intelligent network based on the characteristic data corresponding to the selected PMU node in the sample data set;

judging whether the accuracy of the pre-artificial intelligent network meets a preset accuracy threshold or not according to the training result of the pre-artificial intelligent network;

judging whether the number of the selected PMU nodes meets a preset minimum node number threshold value or not;

and when the accuracy of the artificial intelligence network before judgment meets a preset accuracy threshold and the number of the selected PMU nodes meets a preset minimum node number threshold, extracting the selected PMU nodes from the plurality of PMU nodes to obtain a key node set.

Optionally, the obtaining a transient stability discrimination network based on the feature data corresponding to each key node in the key node set and the trained artificial intelligent network includes:

after the electrical quantities corresponding to all key nodes in the key node set are input, the artificial intelligent network is pre-trained, and the accuracy of the pre-training is calculated;

when the accuracy of the post-artificial intelligent network is not greater than the precision threshold value, selecting a PMU node from the plurality of PMU nodes and adding the PMU node to the key node set to obtain a new key node set, wherein the selected PMU node belongs to other PMU nodes except the key node set;

and inputting the electric quantities corresponding to all the key nodes in the new key node set, and then pre-training the artificial intelligent network until the accuracy of the pre-training is greater than a precision threshold value, and stopping training to obtain a transient stability discrimination network.

Optionally, the pre-artificial intelligence network is a time-delay neural network and omits the feedback of the output, and the calculation formula of the pre-artificial intelligence network is as follows:

in the formula (I), the compound is shown in the specification, _ypre for the output of the prior artificial intelligence network, _yt for the output of the time-delay neural network at time t,

for time series, x is a vector and the superscript i is the number of the PMU node of the power system.

Optionally, the post-artificial intelligence network is composed of a bidirectional cyclic neural network and a BilSTM layer, and the obtained transient stability discriminant network comprises an input layer, the BilSTM layer, a Dropout layer, a full connection layer, a Softmax layer and a Classification layer.

Optionally, the method further comprises:

according to the maximum power angle difference delta of the generator _max And judging the label of each characteristic data, wherein the judgment formula is as follows:

wherein TSI is a parameter for determining whether the power angle of the simulated power system is stable, and y _Label Is a label for the characteristic data.

According to another aspect of the present invention, there is provided a power system multi-agent transient stability judging system with a small number of PMU samples, comprising:

the system comprises a sample data set acquisition module, a data processing module and a data processing module, wherein the sample data set acquisition module is used for acquiring a sample data set, the sample data in the sample data set are characteristic data acquired by a plurality of PMU nodes in a simulated power system, and each sample data has a label for representing whether the power system can keep transient stability;

a key node set determining module, configured to train a pre-training artificial intelligence network based on a sample data set, and extract a key node set from the plural PMU nodes according to a training result, where a key node in the key node set is a predetermined number of PMU nodes selected from the plural PMU nodes; and

the transient stability judging network determining module is used for acquiring a transient stability judging network by the artificial intelligent network after training based on the characteristic data corresponding to each key node in the key node set;

and the transient stability judging module is used for acquiring the electrical quantity data acquired by PMU nodes consistent with the installation positions of all the key nodes in the key node set in the actual power system, inputting the acquired electrical quantity data into the transient stability judging network, and outputting the transient stability judging result of the actual power system.

Optionally, the sample data set obtaining module is specifically configured to:

according to the maximum power angle difference delta of the generator in the simulation process _max And generating a label which is corresponding to each characteristic data and is used for representing whether the simulated power system can keep transient stability.

Optionally, the key node set determining module is specifically configured to:

Optionally, the transient stability discrimination network determining module is specifically configured to:

Optionally, the post-artificial intelligence network is composed of a bidirectional recurrent neural network and a BilStm layer, and the obtained transient stability discriminant network includes an input layer, a BilStm layer, a Dropout layer, a full link layer, a Softmax layer, and a Classification layer.

Optionally, the system further includes a tag determination module, specifically configured to: according to the maximum power angle difference delta of the generator _max And judging the label of each characteristic data, wherein the judgment formula is as follows:

According to a further aspect of the invention, there is provided a computer readable storage medium having stored thereon a computer program for executing the method of any of the above aspects of the invention.

According to still another aspect of the present invention, there is provided an electronic apparatus including: a processor; a memory for storing the processor-executable instructions; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method according to any one of the above aspects of the present invention.

Therefore, the invention uses the AI for AI multi-agent series connection matching mode, can automatically select key PMU nodes in the power grid, and realizes the integrated stability judgment flow considering the physical structure of the power grid. And key feature nodes can be effectively extracted through the Pre-AI model, the feature dimensionality of the data samples can be greatly reduced, the redundancy of an original data set is reduced, and the Post-AI model training efficiency is improved. And after the Post-AI model is trained, when the transient state stability judging method is applied, only the electric quantity on the key node set determined by the Pre-AI model in the actual power grid is collected and input into the Post-AI model for accurate stability/instability classification, and then the transient state stability judging result of the actual power grid can be output. In addition, the key characteristic node set extracted by the Pre-AI model corresponds to the spatial structure of the actual power grid, so that installation suggestions of PMU nodes can be provided for a system based on intelligent stability judgment in the future, and the installation cost is reduced. The mode for AI provides reference for the multi-agent to process complex problems in the power system.

Drawings

A more complete understanding of exemplary embodiments of the present invention may be had by reference to the following drawings in which:

FIG. 1 is a block diagram of an AI for AI model provided in an exemplary embodiment of the invention;

FIG. 2 is a flow chart of a power system multi-agent transient stability determination method with a small number of PMU samples according to an exemplary embodiment of the present invention;

FIG. 3 is a schematic diagram of a latency neural network according to an exemplary embodiment of the present invention;

FIG. 4 is a schematic overall flow chart of a power system multi-agent transient stability determination process with a small number of PMU samples according to an exemplary embodiment of the present invention;

FIG. 5 is a schematic diagram of a bi-directional recurrent neural network provided by an exemplary embodiment of the present invention;

FIG. 6 is a schematic diagram of the structure of an LSTM neuron provided in an exemplary embodiment of the present invention;

FIG. 7 is a schematic diagram of a transient stability discrimination network according to an exemplary embodiment of the present invention;

FIG. 8 is a schematic diagram of the accuracy η of the pre-artificial intelligence network versus the extracted set of key nodes provided by an exemplary embodiment of the present invention;

fig. 9 is a gamma diagram of an IEEE39 node system according to an exemplary embodiment of the present invention _Δt And the change relation of the accuracy rate along with the number of the key node set nodes;

fig. 10 is a gamma diagram of a 300-node system according to an exemplary embodiment of the present invention _Δt And the change relation of the accuracy rate along with the number of the key node set nodes;

FIG. 11 is a schematic structural diagram of a power system multi-agent transient stability determination system with a small number of PMU samples according to an exemplary embodiment of the present invention; and

fig. 12 is a structure of an electronic device according to an exemplary embodiment of the present invention.

Detailed Description

Hereinafter, example embodiments according to the present invention will be described in detail with reference to the accompanying drawings. It is to be understood that the described embodiments are merely a subset of embodiments of the invention and not all embodiments of the invention, with the understanding that the invention is not limited to the example embodiments described herein.

It should be noted that: the relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

It will be understood by those of skill in the art that the terms "first," "second," and the like in the embodiments of the present invention are used merely to distinguish one element, step, device, module, or the like from another element, and do not denote any particular technical or logical order therebetween.

It should also be understood that in embodiments of the present invention, "a plurality" may refer to two or more and "at least one" may refer to one, two or more.

It is also to be understood that any reference to any component, data, or structure in the embodiments of the invention may be generally understood as one or more, unless explicitly defined otherwise or stated to the contrary hereinafter.

In addition, the term "and/or" in the present invention is only one kind of association relationship describing the associated object, and means that there may be three kinds of relationships, for example, a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. In the present invention, the character "/" generally indicates that the preceding and following related objects are in an "or" relationship.

It should also be understood that the description of the embodiments of the present invention emphasizes the differences between the embodiments, and the same or similar parts may be referred to each other, so that the descriptions thereof are omitted for brevity.

Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail, but are intended to be part of the specification where appropriate.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

Embodiments of the invention are operational with numerous other general purpose or special purpose computing system environments or configurations, and with numerous other electronic devices, such as terminal devices, computer systems, servers, etc. Examples of well known terminal devices, computing systems, environments, and/or configurations that may be suitable for use with electronic devices, such as terminal devices, computer systems, servers, and the like, include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, hand-held or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, distributed cloud computing environments that include any of the above, and the like.

Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types. The computer system/server may be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.

Exemplary method

The invention firstly designs the system structure, and the specific implementation steps are as follows:

with the continuous development of power grids, the number of network nodes is increased, and if all the electric quantities collected by each node are input into a complex neural network for transient stability prediction, the model is inevitably complicated, more computing resources are consumed, and the computing time of online application is prolonged. The framework for building an AI for AI model by way of dual agent cooperation is illustrated herein in FIG. 1. The data full set represents the electrical characteristic quantity of each node acquired by the PMU, and the Pre-AI model firstly carries out rough classification on the original data to obtain the key characteristic set. The selected key feature set may not be optimal, but the transient stability judgment of the Post-AI model is guaranteed to be effective. After the Pre-AI model extracts the key feature set and inputs the key feature set as the feature of the Post-AI model, the training speed and the online operation speed of the Post-AI model are improved due to the reduction of the dimension of the feature set, and the effect of improving the larger electric network is more obvious. If the electric network is changed greatly or new technology and equipment are connected to the electric network, the Pre-AI model can be restarted to extract a new key feature set B, and at the moment, the Post-AI model can be trained from the feature set A to the feature set B in a transfer learning mode or a retraining mode to adapt to a new electric network structure.

Based on the AI for AI model, a power system multi-agent transient state stability judgment method (hereinafter referred to as 'TSA-SPS') based on a small amount of PMU sampling can be realized. In the TSA-SPS, the Pre-AI model has the main function of preprocessing the original data acquired by PMU nodes according to the accuracy of self-training and selecting the first nodes with the best accuracy. And then, PMU data corresponding to the selected nodes are input into a Post-AI model for accurate stability judgment and classification. Therefore, the TSA-SPS can effectively reduce the number of PMU nodes required for stability judgment, and can reduce the pressure of the PMU nodes in a data communication system.

Fig. 2 is a flowchart illustrating a power system multi-agent transient stability determination method with a small number of PMU samples according to an exemplary embodiment of the present invention. The present embodiment can be applied to an electronic device, and as shown in fig. 2, the method 200 for determining the transient state of a multi-agent in a power system includes the following steps:

step 201, a sample data set is obtained, wherein sample data in the sample data set is characteristic data acquired by a plurality of PMU nodes in a simulated power system, and each sample data has a label representing whether the power system can keep transient stability.

As an embodiment, obtaining a sample data set includes: acquiring initial sample data with time domain information, which is acquired by a plurality of PMU nodes in a power system, by adopting a simulation mode; selecting a voltage item of a PMU node from initial sample data as characteristic data; and according to the maximum power angle difference delta of the generator in the simulation process _max And generating a label which is corresponding to each characteristic data and is used for representing whether the simulated power system can keep transient stability.

Generally, considering that the probability of the power grid in actual operation is extremely small, the data collected by the PMU nodes is highly biased data concentrated in a normal state, which is not beneficial to the effect of the intelligent algorithm researched by us. Therefore, the invention adopts a simulation mode to obtain the data with time domain information, and the sampling frequency of the simulated PMU node is 50 Hz. For transient stability judgment, only voltage terms (including voltage amplitude and phase angle) of PMU nodes are selected as characteristic quantities of interest. The initial sample data collected is as follows:

wherein M represents the total node number of the power transmission network under study, T represents the length of the time sequence, and the total characteristic dimension of the network with one M nodes is 2M.

As an example, the maximum power angle difference delta according to the generator _max And judging the label of each characteristic data, wherein the judgment formula is as follows:

Step 202, training the pre-training artificial intelligence network based on the sample data set, and extracting a key node set from the plural PMU nodes according to a training result, wherein the key nodes in the key node set are PMU nodes with a predetermined number selected from the plural PMU nodes.

As an embodiment, the pre-artificial intelligence network is a time-delay neural network and omits feedback of output, and the calculation formula of the pre-artificial intelligence network is as follows:

Generally, a time-evolution dynamic process is performed after a power system is disturbed, and compared with a time-delay neural network, the time-delay neural network can better conform to a physical process of the power system after the power system is disturbed than a time-delay-free neural network, so that a bidirectional long-time and short-time memory network (BiLSTM) is adopted in a transient stability judging network (corresponding to a Post-AI model in fig. 1) for finally judging the transient stability of a power grid. The following two factors are therefore mainly considered in the process of designing the Pre-artificial intelligent network (corresponding to the Pre-AI model in fig. 1):

1) the Pre-AI model and the Post-AI model have similar intelligent structures so as to ensure that the key features extracted by the Pre-AI model can be utilized by the Post-AI model;

2) the Pre-AI model should have a faster running speed to ensure the efficiency of feature selection.

Based on the above two pointsThe Pre-AI model is selected as a simple time-delayed neural network (TDNN) and output feedback is omitted. The working principle is shown in FIG. 3, where y _t Representing the output of TDNN at time t, the input of the model being a time series

To

x is a vector and the superscript i is the number of the power system node. The formula for TDNN is as follows:

Compared to a normal fully-connected ANN network, if the time-domain evolution characteristics are to be considered, the TDNNs can share the weight matrix w in the time dimension, and thus can perform more efficiently on the time series.

As an embodiment, training a pre-training artificial intelligence network based on a sample data set, and extracting a key node set from a plurality of PMU nodes according to a training result, the method comprises the following steps: selecting any PMU node from the PMU nodes, and training a pre-training artificial intelligent network based on the characteristic data corresponding to the selected PMU node in the sample data set; judging whether the accuracy of the pre-artificial intelligent network meets a preset accuracy threshold or not according to the training result of the pre-artificial intelligent network; judging whether the number of the selected PMU nodes meets a preset minimum node number threshold value or not; and when the accuracy of the artificial intelligence network before judgment meets a preset accuracy threshold and the number of the selected PMU nodes meets a preset minimum node number threshold, extracting the selected PMU nodes from the plurality of PMU nodes to obtain a key node set.

Generally, the raw data collected by PMU nodes generally has high redundancy, and for an electrical network including M nodes, only data collected by several PMU nodes may play a major role in final transient state judgment, and other nodes only play a weak role, and if too much data with weak roles is available, a large amount of computing resources are consumed in the neural network training process, and the final effect is not good. The goal of the Pre-AI model is to select the nodes that play a major role in the spatial dimension, and only use the information of these nodes as input to the Post-AI model. The overall flow diagram of the power system multi-agent transient state stability judgment method for sampling by using a small number of PMUs is shown in fig. 4.

Specifically, firstly, the Pre-AI model selects any PMU node from a plurality of PMU nodes and sets the PMU node as a key node S _k ¹ And is used only in the key node S _k ¹ Training the Pre-AI model (i.e., the Pre-artificial intelligent network) under the corresponding feature data set if the accuracy of the Pre-AI model on the test set does not meet the set minimum accuracy eta _min And a minimum number of nodes n _min And if required, changing the node. If no nodes which are not traversed exist, selecting the nodes with the best effect and adding another node until a key node set S meeting the requirement is found _k ⁿ 。

And 203, training the artificial intelligent network based on the characteristic data corresponding to each key node in the key node set to obtain a transient stability judging network.

As one embodiment, the post-artificial intelligence network is composed of a bidirectional cyclic neural network and a BilSTM layer, and the obtained transient stability discriminant network comprises an input layer, the BilSTM layer, a Dropout layer, a full connection layer, a Softmax layer and a Classication layer.

Generally, signals collected by PMU nodes have time sequence characteristics, so a dynamic process of grid instability can be better reflected by adopting a Recurrent Neural Network (RNN). The conventional uni-directional RNN can only acquire information of a positive sequence time series when processing the time series. However, the feature extraction of a time sequence from the time negative sequence direction can also acquire some key information so as to enhance the accuracy and generalization performance of the whole neural network. Therefore, the bidirectional recurrent neural network has better performance, and can acquire more information by using less data, so that the Pre-AI model can compress the number of PMU nodes to a greater extent. The structure of the bidirectional recurrent neural network is shown in fig. 5, which includes not only the forward layer but also the reverse layer. Taking time t as an example, the calculation formula of the recurrent neural network is as follows:

wherein h is _→ And h _← Representing the outputs of the forward and reverse layers, respectively, b → and b ← representing the biases on the forward and reverse layers, respectively, b _y Offset, x, representing the final output _t And yt respectively represent input and output corresponding to the time t, and w is a weight matrix. If only the result of the last moment is of interest, only the value y of the last moment can be output _tmax 。

Because the traditional neuron has the phenomenon of short-time memory, the LSTM is used as the variant of the traditional RNN, so that the memory capacity is prolonged, and the defect that a standard RNN unit is easy to have gradient diffusion is overcome. The network structure of LSTM neurons is shown in fig. 6.

Substituting LSTM neurons into the bi-directional circulation network shown in FIG. 5 can build a BiLSTM layer, and building a Post-AI model (corresponding to the transient stability discriminant network) using the BiLSTM layer, as shown in FIG. 7. In order to prevent overfitting during training, a Dropout layer is added after the BilSTM layer, the probability of the Dropout layer is set to be 0.5, 128 hidden units are contained in the BilSTM layer, and the output size of the fully-connected layer is 2.

As an embodiment, the obtaining of the transient stability discrimination network based on the feature data corresponding to each key node in the key node set and the trained artificial intelligent network includes: after the electrical quantities corresponding to all key nodes in the key node set are input, the artificial intelligent network is pre-trained, and the accuracy of the pre-training is calculated; when the accuracy of the post-artificial intelligent network is not greater than the precision threshold value, selecting a PMU node from the plurality of PMU nodes and adding the PMU node to the key node set to obtain a new key node set, wherein the selected PMU node belongs to other PMU nodes except the key node set; and inputting the electric quantities corresponding to all the key nodes in the new key node set, and then pre-training the artificial intelligent network until the accuracy of the pre-training is greater than a precision threshold value, and stopping training to obtain a transient stability discrimination network.

Set key nodes

Inputting the electric quantity corresponding to each key node into a Post artificial intelligent network (corresponding to a Post-AI model) for pre-training, and calculating the accuracy gamma of the pre-training, if the gamma is less than the precision requirement gamma min of the Post-AI model, then performing pre-training on the electric quantity corresponding to each key node

And adding a node until the trained Post-AI model network is output after the requirement is met, thereby obtaining the transient stability discrimination network shown in FIG. 7. Wherein, the loss function of the Post-AI model training is expressed by cross entropy error as follows:

wherein N represents the number of training set samples,

representing the predicted value of the neural network.

In addition, the neural network parameters of the Post-AI model are optimized through a minimized loss function L in the training process by using an adma optimization algorithm, the learning rate is reduced by 0.4 time after 60 iterations, and the mini-branch size is 512.

Thus, the present invention introduces a hyper-parameter η in the AI for AI mode _min And n _min To correlate the fit between the Pre-AI model and the Post-AI model. Wherein eta is _min The main function of (a) is to ensure the accuracy and stability of the Pre-AI model training result, n _min The main function of (1) is to ensure the stability of Post-AI model training, n _min The larger the value of (A), the higher the stability of the Post-AI model results. Due to the complex structure and long training time of the Post-AI model, the occurrence of gamma < gamma should be avoided as much as possible _min By adjusting η _min And n _min The value of (n) can ensure that the accuracy rate gamma of the Post-AI model is greater than a threshold value, and the invention obtains eta in an IEEE39 node system _min ＝86％，n _min ＝5。

And 204, acquiring electrical quantity data acquired by PMU nodes consistent with the installation positions of all key nodes in the key node set in the actual power system, inputting the acquired electrical quantity data into a transient stability judgment network, and outputting a transient stability judgment result of the actual power system.

In the embodiment of the invention, the transient stability judging network can be obtained after the post artificial intelligent network training is finished, and when the transient stability judging network is applied, only the electric quantity on the key node set determined by the pre artificial intelligent network in the actual power grid is required to be collected and input into the transient stability judging network for accurate stability/instability classification, so that the transient stability judging result of the actual power grid is output.

Analysis by calculation example:

the invention adopts PSD power grid commercial simulation calculation software released by China electric academy of sciences, and utilizes an IEEE39 node system to generate an example, wherein a generator and load change respectively take 0.8, 1 and 1.2 of base values, the system is subjected to N-1 fault scanning, the fault time is set to be 0-0.4s, and 9000 samples are generated in total.

1) Firstly, selecting a key node set by utilizing a Pre-AI model:

because the Pre-AI model is set to extract the rough characteristics of the key nodes and does not need to be trained on the whole sample set, only 500 samples are extracted to be used as the sample set of the Pre-AI model, 85% of the sample set is taken as a training set, 15% of the sample set is taken as a testing set, and the time length T is taken to be 3, so that the high-efficiency running speed is ensured. Extracting key characteristic nodes of the IEEE39 node system by utilizing a Pre-AI model, and training to obtain different characteristicsThe training effect under the node set is shown in fig. 8. The abscissa represents the adding sequence of the selected node set, the accuracy rate can be rapidly improved for the addition of the first nodes, and when the number of extracted key nodes is 3, eta is met _min >86% but to ensure stability of Post-AI model training and to prevent gamma < gamma for Post-AI model _min In case (1), therefore, n is taken _min 5. As the number of nodes increases, nodes which have weak effects on training results are merged into a node set, which may result in accuracy drop back, and this is also consistent with the training experience of machine learning, and it is not so good that more input features are, so feature screening is generally performed at the early stage of training. Finally, as the number of nodes is further increased, the accuracy rate can keep a relatively stable value.

It should be noted that, for each selection of the node set, the number of training iterations is ensured to make the neural network stable enough. Before the iteration number is 60, the training accuracy curve has obvious oscillation, which may cause the result of training the extracted node set not to converge, and the maximum iteration number taken by the Pre-AI model is set as 100. Table 1 lists node sets selected by the Pre-AI model for 5 times, and it can be seen that the similarity of the key node sets selected by the Pre-AI model is very high, especially the first three nodes, which also indicates that the convergence of the operation result of the Pre-AI model is good, and a stable data set can be provided for the Post-AI model.

TABLE 1 η _min Key node set selected by Pre-AI model under 86% condition

2) Determining the training duration and the network performance of the Post-AI model:

the CPU used for training is Intel i 56500 and 3.2GHz, and is used for calculating the time required by training. Since the training time difference is large by using different CPUs or GPUs, the relative reduction gamma of the calculation time is adopted here _Δt Representing the difference in demand for computing resources, as shown in the following equation:

wherein, t ₀ Indicates the duration of time it takes for training not to preprocess through the Pre-AI model, t _pre Represents the duration, t, taken by the Pre-AI model to extract the key nodes _post Representing the length of time it takes for Post-AI model training.

The calculated relative reduction of training time under different key node training sets is shown in fig. 9. It can be seen that the key feature nodes are preprocessed through the Pre-AI model, the training time of the Post-AI model can be effectively reduced, when the number of feature node sets is less than 20, more than one fifth of calculated amount can be saved, and even if 25 feature nodes are extracted, the calculated amount can be reduced by 14%. Fig. 9 shows the amount of computation saved by the Post-AI model in 1 computation, and in general, the Post-AI model will train to select the result with the best training effect for multiple times, and the amount of computation saved is considerable.

As shown in fig. 10, as can be seen from the training result of the neural network, compared with the mode of randomly selecting a bus node set, the calculation amount can be greatly reduced by using the training mode of the AI for AI model, the performance of the Post-AI model can still maintain relatively high accuracy, and the accuracy of the test set is maintained above 98%, which indicates that the key feature nodes extracted by the method are effective.

For further explaining the advantages of the TSA-SPS, an AC/DC series-parallel power grid 300 node is established according to an actual AC/DC power supply form and by combining related operation data. The 300-node standard example was analyzed using the AI for AI model of TSA-SPS, and the results are shown in FIG. 10.

Calculating gamma using the formula _Δt As shown in FIG. 10, when the number of key node sets is below 50, the Post-AI model can save more than 85% of calculation amount once, the relative saving amount of calculation time is much higher than that of 39 nodes, and the model accuracy of TSA-SPS is also kept above 97.8%. The method also shows that the TSA-SPS model can be used in a large power grid with a large number of nodesThe training efficiency of the intelligent model is effectively improved.

In addition, as shown in fig. 8 and 10, through model verification, it is found that when the power grid is provided with 39 nodes, the number of extracted key nodes is 5, and the model stability and accuracy of the TSA-SPS are optimal. When the power grid is provided with 300 nodes, and the number of the extracted key nodes is 50, the model stability and accuracy of the TSA-SPS are optimal. Therefore, the transient stability judgment method can realize the transient stability judgment of the power system by only using data collected by PMUs on fewer nodes.

Therefore, the invention explores the cooperation of multiple agents in the transient state stability judgment process of the power system, and provides a power system multi-agent transient state stability judgment method (TSA-SPS) with a small amount of PMU sampling in the mode of AI for AI model, and the method has the following beneficial effects:

1) in the TSA-SPS, a Pre-AI model can extract a key node set in a power grid according to the positions of PMU nodes, and transient stability judgment can be realized by only using data of PMUs on fewer nodes.

2) Because the number of required nodes is reduced, the calculation resources required by Post-AI model training can be obviously reduced, the training efficiency of the neural network is improved, and the effect of improving the larger network is obvious.

3) The key node set extracted by the Pre-AI model can provide guidance for selecting the position of the PMU of the power system, and the number of the needed PMUs is reduced, so that the communication pressure in the WAMS system is reduced, and the installation cost of the PMU and the construction cost of the communication system are reduced.

4) The AI for AI model established by the invention is not limited to transient state stability judgment in the power grid, can be expanded and applied to other power grid applications, and can avoid the problem of dimension disaster caused by directly using a single neural network to process high-dimensional data. Meanwhile, the working mode of the AI for AI model also provides reference and reference for the cooperation and application of multi-agent technology in the power system or other industrial systems.

Therefore, the invention can automatically select key PMU nodes in the power grid by using an AI for AI multi-agent series matching mode, realizes an integrated stability judging process considering the physical structure of the power grid, can effectively extract key feature nodes and greatly reduce the feature dimensions of data samples through the Pre-AI model, reduces the redundancy of an original data set and improves the training efficiency of a Post-AI model. And after the Post-AI model is trained, when the transient state stability judging method is applied, only the electric quantity on the key node set determined by the Pre-AI model in the actual power grid is collected and input into the Post-AI model for accurate stability/instability classification, and then the transient state stability judging result of the actual power grid can be output. In addition, the key node set extracted by the Pre-AI model corresponds to the spatial structure of the actual power grid, so that PMU installation suggestions can be provided for a system based on intelligent stability judgment in the future, and the installation cost is reduced. The mode for AI provides reference for the multi-agent to process complex problems in the power system.

Exemplary System

Fig. 11 is a schematic structural diagram of a power system multi-agent transient stability judging system with a small number of PMU samples according to an exemplary embodiment of the present invention. As shown in fig. 11, system 1100 includes: a sample data set obtaining module 1101, configured to obtain a sample data set, where sample data in the sample data set is feature data acquired by multiple PMU nodes in a simulated power system, and each sample data has a label indicating whether the power system can maintain transient stability; a key node set determining module 1102, configured to train a pre-training artificial intelligence network based on a sample data set, and extract a key node set from the plural PMU nodes according to a training result, where a key node in the key node set is a predetermined number of PMU nodes selected from the plural PMU nodes; the transient stability judging network determining module 1103 is used for acquiring a transient stability judging network by the trained artificial intelligent network based on the feature data corresponding to each key node in the key node set; and the transient stability judging module 1104 is configured to acquire electrical quantity data acquired by PMU nodes in the actual power system, which are consistent with the installation positions of the key nodes in the key node set, input the acquired electrical quantity data to the transient stability judging network, and output a transient stability judging result of the actual power system.

In an embodiment, the sample data set obtaining module 1101 is specifically configured to: by means of simulationTaking initial sample data with time domain information collected by a plurality of PMU nodes in a power system; selecting a voltage item of a PMU node from initial sample data as characteristic data; and according to the maximum power angle difference delta of the generator in the simulation process _max And generating a label which is corresponding to each characteristic data and is used for representing whether the simulated power system can keep transient stability.

In an embodiment, the key node set determining module 1102 is specifically configured to: selecting any PMU node from the plurality of PMU nodes, and training a pre-training artificial intelligent network on the basis of the characteristic data corresponding to the selected PMU node in the sample data set; judging whether the accuracy of the pre-artificial intelligent network meets a preset accuracy threshold or not according to the training result of the pre-artificial intelligent network; judging whether the number of the selected PMU nodes meets a preset minimum node number threshold value or not; and when the accuracy of the artificial intelligence network before judgment meets a preset accuracy threshold and the number of the selected PMU nodes meets a preset minimum node number threshold, extracting the selected PMU nodes from the plurality of PMU nodes to obtain a key node set.

In one embodiment, the transient stability determination network determining module 1103 is specifically configured to: after the electrical quantities corresponding to all key nodes in the key node set are input, the artificial intelligent network is pre-trained, and the accuracy of the pre-training is calculated; when the accuracy of the post-artificial intelligent network is not greater than the precision threshold value, selecting a PMU node from the plurality of PMU nodes and adding the PMU node to the key node set to obtain a new key node set, wherein the selected PMU node belongs to other PMU nodes except the key node set; and inputting the electric quantities corresponding to all the key nodes in the new key node set, and then pre-training the artificial intelligent network until the accuracy of the pre-training is greater than a precision threshold value, and stopping training to obtain a transient stability discrimination network.

In one embodiment, the pre-artificial intelligence network is a time-delay neural network and omits the feedback of the output, and the calculation formula of the pre-artificial intelligence network is as follows:

In one embodiment, the post-artificial intelligence network is composed of a bidirectional recurrent neural network and a BilSTM layer, and the obtained transient stability discriminant network comprises an input layer, the BilSTM layer, a Dropout layer, a full connectivity layer, a Softmax layer and a Classication layer.

In one embodiment, the system 1100 further comprises a tag determination module, specifically configured to: according to the maximum power angle difference delta of the generator _max And judging the label of each characteristic data, wherein the judgment formula is as follows:

The power system multi-agent transient state stability judging system 1100 for small amount of PMU sampling according to the embodiment of the present invention corresponds to the power system multi-agent transient state stability judging method 200 for small amount of PMU sampling according to another embodiment of the present invention, and is not described herein again.

Exemplary electronic device

Fig. 12 is a structure of an electronic device according to an exemplary embodiment of the present invention. The electronic device may be either or both of the first device and the second device, or a stand-alone device separate from them, which stand-alone device may communicate with the first device and the second device to receive the acquired input signals therefrom. FIG. 12 illustrates a block diagram of an electronic device in accordance with an embodiment of the present invention. As shown in fig. 12, the electronic device 120 includes one or more processors 121 and a memory 122.

The processor 121 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

Memory 122 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium and executed by the processor 121 to implement the method for information mining on historical change records of the software program of the various embodiments of the present invention described above and/or other desired functions. In one example, the electronic device may further include: an input system 123 and an output system 124, which are interconnected by a bus system and/or other form of connection mechanism (not shown).

The input system 123 may also include, for example, a keyboard, a mouse, and the like.

The output system 124 may output various information to the outside. The output devices 124 may include, for example, a display, speakers, a printer, and a communication network and remote output devices connected thereto, among others.

Of course, for the sake of simplicity, only some of the components of the electronic device relevant to the present invention are shown in fig. 12, and components such as buses, input/output interfaces, and the like are omitted. In addition, the electronic device may include any other suitable components, depending on the particular application.

Exemplary meterComputer program product and computer-readable storage medium

In addition to the above-described methods and apparatus, embodiments of the present invention may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the method of information mining of historical change records according to various embodiments of the present invention described in the "exemplary methods" section above of this specification.

The computer program product may write program code for carrying out operations for embodiments of the present invention in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.

Furthermore, embodiments of the present invention may also be a computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the steps in the method of information mining of historical change records according to various embodiments of the present invention described in the "exemplary methods" section above of this specification.

The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The basic principles of the present invention have been described above with reference to specific embodiments, but it should be noted that the advantages, effects, etc. mentioned in the present invention are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present invention. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the invention is not limited to the specific details described above.

In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts in the embodiments are referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The block diagrams of devices, systems, apparatuses, and systems involved in the present invention are merely illustrative examples and are not intended to require or imply that the devices, systems, apparatuses, and systems must be connected, arranged, or configured in the manner shown in the block diagrams. These devices, systems, apparatuses, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

The method and system of the present invention may be implemented in a number of ways. For example, the methods and systems of the present invention may be implemented in software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustrative purposes only, and the steps of the method of the present invention are not limited to the order specifically described above unless specifically indicated otherwise. Furthermore, in some embodiments, the present invention may also be embodied as a program recorded in a recording medium, the program including machine-readable instructions for implementing a method according to the present invention. Thus, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

It should also be noted that in the systems, apparatus and methods of the present invention, the various components or steps may be broken down and/or re-combined. These decompositions and/or recombinations are to be regarded as equivalents of the present invention. The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the invention. Thus, the present invention is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the invention to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. A power system multi-agent transient state stability judging method for small amount of PMU sampling is characterized by comprising the following steps:

2. The method of claim 1, wherein said obtaining a sample data set comprises:

3. The method of claim 1, wherein training the pre-training artificial intelligence network based on the sample data set and extracting a set of key nodes from the plurality of PMU nodes according to a training result comprises:

4. The method of claim 1, wherein the training of the artificial intelligent network based on the feature data corresponding to each key node in the set of key nodes to obtain the transient stability discrimination network comprises:

5. The method of claim 1, wherein the pre-artificial intelligence network is a time-delay neural network and omits feedback of the output, and the pre-artificial intelligence network has a calculation formula as follows:

in the formula, y _pre For the output of the pre-artificial intelligence network, y _t For the output of the time-delay neural network at time t,

6. The method of claim 1, wherein the post-artificial intelligence network is composed of a bi-directional recurrent neural network and a BilsTM layer, and the resulting transient stability discriminant network comprises an input layer, the BilsTM layer, a Dropout layer, a fully connected layer, a Softmax layer, and a Classification layer.

7. The method of claim 2, further comprising:

8. A steady system is judged to electric power system multi-agent transient state of a small amount of PMU sampling, its characterized in that includes:

9. The system of claim 8, wherein the sample data set acquisition module is specifically configured to:

10. The system of claim 8, wherein the key node set determining module is specifically configured to:

11. The system of claim 8, wherein the transient stability discrimination network determination module is specifically configured to:

when the accuracy of the post-artificial intelligent network is not larger than a precision threshold value, selecting a PMU node from the plurality of PMU nodes to add to the key node set to obtain a new key node set, wherein the selected PMU node belongs to other PMU nodes except the key node set;

12. The system of claim 8, wherein the pre-artificial intelligence network is a time-delay neural network and omits feedback of the output, and the pre-artificial intelligence network has a calculation formula as follows:

13. The system of claim 8, wherein the post-artificial intelligence network is comprised of a bi-directional recurrent neural network and a BilsTM layer, and the resulting transient stability discriminant network comprises an input layer, a BilsTM layer, a Dropout layer, a fully connected layer, a Softmax layer, and a Classification layer.

14. The system of claim 9, further comprising a tag determination module, specifically configured to: according toMaximum power angle difference delta of generator _max And judging the label of each characteristic data, wherein the judgment formula is as follows:

15. A computer-readable storage medium, characterized in that the storage medium stores a computer program for performing the method of any of the preceding claims 1-7.

16. An electronic device, characterized in that the electronic device comprises:

a processor;

a memory for storing the processor-executable instructions;

the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method of any one of claims 1 to 7.