Method, device and system for regulating and controlling voltage in transformer area based on multiple agents and chip

Abstract

The invention relates to the technical field of transformer area intelligent monitoring and management, in particular to a method, device and system for regulating and controlling voltage in a transformer area based on multiple agents and a chip. The system comprises a voltage regulation and control center, and a voltage regulation and control unit and an energy control node in a transformer area. And the voltage regulation and control center predicts future load and energy output of the transformer area by using the first intelligent agent, and generates a transformer area collaborative voltage regulation and control scheme in combination with the second intelligent agent. And the voltage regulation and control unit generates a specific voltage regulation and control scheme in the transformer area through a third agent according to the transformer area collaborative voltage regulation and control scheme and the real-time load information, and issues the specific voltage regulation and control scheme to the energy control node to execute voltage regulation and control. Through cooperative work of the intelligent agents, fine voltage regulation and control of the photovoltaic grid-connected point are effectively realized, voltage out-of-limit of the photovoltaic grid-connected point is avoided, voltage stability and electric energy quality of the power distribution network are improved, and powerful technical support is provided for access of renewable energy sources and intelligent operation of the power distribution network.

Description

Technical Field

[0001] The present disclosure relates to the technical field of intelligent monitoring and management of substations, and in particular to a method, device, system and chip for voltage regulation in a substation based on multiple intelligent agents. Background Art

[0002] With the transformation of the global energy structure and the rapid development of renewable energy, a large number of renewable distributed power sources (such as wind power generation and photovoltaic power generation equipment) have been connected to the distribution network. These distributed power sources have significant randomness and volatility because their output is affected by natural conditions (such as wind speed, light intensity, etc.). At the same time, the load in the distribution network is also becoming increasingly complex and uncertain. For example, the electricity consumption behaviors of industrial loads, commercial loads and residential loads vary greatly and are affected by multiple factors.

[0003] The random fluctuations in the output of distributed power sources and the uncertain fluctuations in loads have brought many challenges to the operation of distribution networks. Among them, large voltage fluctuations, voltage crossings, and increased network losses are particularly prominent, which seriously affect the quality of electric energy and the reliability of power supply. Traditional grid control methods are powerless in the face of these challenges. They are often unable to refine the load data within the sensing area (i.e., a specific area or range of power supply), resulting in low control accuracy. In addition, traditional control methods mostly rely on active control, that is, maintaining voltage stability by adjusting the output or load of the generator. This method not only has a slow response speed, but also may lead to an increase in electricity costs for users.

[0004] How to effectively achieve voltage regulation at photovoltaic grid-connected points, prevent voltage exceeding the limit at photovoltaic grid-connected points, improve voltage stability and power quality of distribution networks, and provide strong technical support for the access of renewable energy and intelligent operation of distribution networks is an urgent problem to be solved. Summary of the invention

[0005] In order to solve the problems in the related art, the embodiments of the present disclosure provide a method, device, system and chip for voltage control within a substation based on multi-agent.

[0006] In a first aspect, an embodiment of the present disclosure provides a multi-agent-based intra-area voltage control system, the voltage control system comprising: a voltage control center, a voltage control unit in each area in an area cluster, and an energy control node in each area, the voltage control center is respectively connected to the voltage control unit in each area, and the voltage control unit in any area in the area cluster is connected to the energy control node in any area, wherein:

[0007] The voltage control center is configured as follows:

[0008] Obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within a preset time in the future based on the historical load information and area characteristic information of each area and the meteorological information within a preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection;

[0009] According to the load information and energy output information of each substation within a preset future time, a substation coordinated voltage control scheme including a multi-point control curve is generated using a trained second agent; the substation coordinated voltage control scheme is sent to the voltage control unit in each substation; wherein the multi-point control curve includes a preset number of control points, and the control points represent the voltage control target corresponding to a specified time point within a preset future time, and the second agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism;

[0010] The voltage control unit in any substation in the substation cluster is configured as follows:

[0011] According to the coordinated voltage control scheme of the substation area and the real-time load information of any substation area, a trained third agent is used to generate a voltage control scheme within the substation area; the voltage control scheme within the substation area is sent to the energy control node in any substation area; wherein the third agent is trained by using the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism, using the historical load information, historical energy output information, substation area topology information and historical coordinated voltage control scheme of each substation area as input data;

[0012] Any energy control node in any of the substations is configured to perform voltage regulation based on the voltage regulation scheme in the substation.

[0013] According to an embodiment of the present disclosure, any energy control node in any of the substations is further configured as follows:

[0014] During the voltage regulation process based on the voltage regulation scheme in the substation, key parameters are monitored and recorded, the key parameters including: voltage level, power factor, active power and reactive power;

[0015] Obtaining a control effect evaluation index according to the key parameters, the control effect evaluation index includes: a voltage stability index and an energy efficiency index, and obtaining the control effect evaluation index according to the key parameters includes: determining the voltage stability index according to the voltage level, the voltage stability index includes: voltage fluctuation rate, voltage qualification rate and voltage average value and standard deviation; determining the energy efficiency index according to the power factor, the active power and the reactive power;

[0016] Compare the control effect evaluation index with the preset target value, analyze whether the control has achieved the expected effect, and obtain the evaluation result;

[0017] Forming feedback data based on the regulation effect evaluation index, the evaluation result and the key parameter, and sending the feedback data to the voltage regulation unit in any one of the substations;

[0018] The voltage control unit in any substation in the substation cluster is configured as follows:

[0019] Adding the feedback data to the training data set of the third agent, and retraining the third agent using the updated training data set;

[0020] The voltage control scheme within the substation is adjusted based on the retrained third intelligent agent.

[0021] According to an embodiment of the present disclosure, the residual connection-based CNN-GRU neural network model includes: an input module, a convolutional neural network CNN module, a gated recurrent unit GRU module and an output module, wherein a residual connection is included between the internal layers of the CNN module or between the internal layers of the GRU module or between the CNN module and the GRU module;

[0022] The method uses the trained first agent to predict the load information and energy output information of each substation within a preset time in the future based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future, including:

[0023] Based on the input module, the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are received, and the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are integrated into a multi-dimensional input tensor, wherein each dimension represents different time series data;

[0024] Extracting spatial features in the multi-dimensional input tensor through multi-layer convolution, or multi-layer convolution and residual connection based on the CNN module to form a feature map;

[0025] Processing the time series data in the feature graph through a multi-layer recurrent neural network or a multi-layer recurrent neural network and a residual connection based on the GRU module, thereby capturing the time dependency and generating a hidden state vector;

[0026] The hidden state vector is processed by the output module through a fully connected layer or a regression layer to generate a prediction result, and the prediction result includes: load information and energy output information of each substation within a preset time in the future.

[0027] According to an embodiment of the present disclosure, the substation characteristic information includes: substation type, installed capacity, grid connection mode and historical power generation efficiency; the substation type includes: photovoltaic power generation, wind power generation or a mixture of photovoltaic power generation and wind power generation;

[0028] When the area type is photovoltaic power generation, the meteorological information includes: temperature and light intensity;

[0029] When the substation type is wind power generation, the meteorological information includes: wind direction and wind speed;

[0030] When the substation type is a mixture of photovoltaic power generation and wind power generation, the meteorological information includes: temperature, light intensity, wind direction and wind speed.

[0031] According to an embodiment of the present disclosure, the Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism includes: an encoder, an attention module and a decoder, wherein the encoder and the decoder respectively include multiple layers of LSTM units connected in sequence;

[0032] The method of using the trained second agent to generate a coordinated voltage control scheme for each substation including a multi-point control curve based on the load information and energy output information of each substation within a preset time in the future includes:

[0033] The load information and energy output information of each substation within a preset future time are sorted into sequence data, wherein the sequence data includes a plurality of time steps, and each time step corresponds to a specific combination of load information and energy output information;

[0034] Processing each time step in the sequence data step by step based on the multiple LSTM units in the encoder, including: for each time step, the designated LSTM unit in the encoder generates a hidden state corresponding to the current time step according to the input data of the current time step and the hidden state of the previous time step;

[0035] After the multiple LSTM units in the encoder have processed each time step in the sequence data, based on the attention module, at each time step of the decoder, the attention weight is calculated according to the hidden state of the last layer of LSTM units in the encoder and the hidden state of all time steps of the encoder, and the hidden state of all time steps of the encoder is weighted and summed using the attention weight to obtain a weighted context vector, wherein the coordinated voltage control scheme of the substation area is characterized by an output sequence containing the preset number of control points, and the time step of the decoder corresponds to the time step relative to one control point in the output sequence;

[0036] The method comprises: for a current time step of the decoder, the LSTM unit in the decoder generates a control point of the current time step according to the weighted context vector and a control point of a previous time step;

[0037] After the decoder generates all the preset number of control points, the preset number of control points are combined into a coordinated voltage control scheme for an area including a multi-point control curve.

[0038] According to an embodiment of the present disclosure, the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism includes: an input module, an LSTM module, an attention mechanism module, an immediate feedback module and an output module; the voltage control scheme in the substation area is generated using the trained third agent according to the substation area collaborative voltage control scheme and the real-time load information of any substation area, including:

[0039] Preprocessing the coordinated voltage control scheme of the substation and the real-time load information of any substation based on the input module to generate corresponding time series format data;

[0040] Processing the time series format data based on the LSTM module includes: capturing long-term dependencies in the time series format data, obtaining time series features in the time series format data, and generating a corresponding hidden state vector group according to the long-term dependencies and the time series features;

[0041] Calculating the attention weight of each hidden state vector in the hidden state vector group based on the attention mechanism module, adding the results of multiplying each hidden state vector by its corresponding attention weight to obtain a weighted hidden state vector;

[0042] Based on the immediate feedback module, the weighted hidden state vector is compared with the immediate feedback information, an error signal is calculated, and an internal state parameter of the LSTM module or the attention mechanism module is adjusted according to the error signal, so that the LSTM module or the attention mechanism module regenerates the weighted hidden state vector according to the adjusted internal state parameter;

[0043] The weighted hidden state vector is decoded based on the output module to map the weighted hidden state vector to an output space corresponding to the voltage control scheme within the substation, a voltage control instruction corresponding to each energy control node is generated, and the voltage control scheme within the substation is generated based on the voltage control instructions of each energy control node.

[0044] According to an embodiment of the present disclosure, when the third intelligent agent is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism, the training weight of the station area topology structure information is gradually reduced according to a preset strategy until the input of the station area topology structure information is finally cancelled.

[0045] According to an embodiment of the present disclosure, the voltage control scheme in the substation includes the voltage control frequency and voltage control speed of each energy control node in the substation, and the active and reactive power adjustment sequence, wherein the voltage control speed includes: voltage down-regulation speed and voltage up-regulation speed, the voltage control frequency is related to the preset number of control points included in the multi-point control curve, the preset number of control points is 288 points, the multi-point control curve is a 288-point control curve, and the corresponding voltage control frequency is regulated once every 5 minutes. The voltage control based on the voltage control scheme in the substation includes:

[0046] Any energy control node in any one of the substations obtains the substation voltage control scheme corresponding to any one of the energy control nodes in the substation voltage control scheme;

[0047] The current voltage of any energy control node is regulated according to the voltage regulation frequency, voltage regulation speed and active and reactive power regulation sequence in the voltage regulation scheme in the substation of any energy control node, including: according to the voltage regulation frequency, the current voltage is monitored according to the corresponding regulation cycle, and in each regulation cycle, the current voltage data is obtained, and the current voltage data is compared with a preset maximum voltage threshold value, if the current voltage data is greater than the preset maximum voltage threshold value, the current voltage is lowered based on the voltage lowering speed in the voltage regulation speed, and if the current voltage data is less than the preset minimum voltage threshold value, the current voltage is increased based on the voltage increasing speed in the voltage regulation speed;

[0048] Wherein, when the current voltage of any energy control node is regulated, it is performed in accordance with the active and reactive power regulation sequence.

[0049] In a second aspect, a multi-agent-based voltage control method in a substation is provided in an embodiment of the present disclosure, which is applied to a voltage control center, wherein the voltage control center is respectively connected to a voltage control unit in each substation in a substation cluster, and a voltage control unit in any substation in the substation cluster is connected to an energy control node in any substation, and the voltage control method includes:

[0050] Obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within a preset time in the future based on the historical load information and area characteristic information of each area and the meteorological information within a preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection;

[0051] According to the load information and energy output information of each substation within a preset future time, a trained second intelligent agent is used to generate a substation coordinated voltage control plan including a multi-point control curve; the substation coordinated voltage control plan is sent to the voltage control unit in each substation; wherein the multi-point control curve contains a preset number of control points, and the control points represent the voltage control targets corresponding to the specified time points within the preset future time. The second intelligent agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism.

[0052] According to an embodiment of the present disclosure, the residual connection-based CNN-GRU neural network model includes: an input module, a convolutional neural network CNN module, a gated recurrent unit GRU module and an output module, wherein a residual connection is included between the internal layers of the CNN module or between the internal layers of the GRU module or between the CNN module and the GRU module;

[0053] The method uses the trained first agent to predict the load information and energy output information of each substation within a preset time in the future based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future, including:

[0054] Based on the input module, the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are received, and the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are integrated into a multi-dimensional input tensor, wherein each dimension represents different time series data;

[0055] Extracting spatial features in the multi-dimensional input tensor through multi-layer convolution, or multi-layer convolution and residual connection based on the CNN module to form a feature map;

[0056] Processing the time series data in the feature graph through a multi-layer recurrent neural network or a multi-layer recurrent neural network and a residual connection based on the GRU module, thereby capturing the time dependency and generating a hidden state vector;

[0057] The hidden state vector is processed by the output module through a fully connected layer or a regression layer to generate a prediction result, and the prediction result includes: load information and energy output information of each substation within a preset time in the future.

[0058] According to an embodiment of the present disclosure, the substation characteristic information includes: substation type, installed capacity, grid connection mode and historical power generation efficiency; the substation type includes: photovoltaic power generation, wind power generation or a mixture of photovoltaic power generation and wind power generation;

[0059] When the area type is photovoltaic power generation, the meteorological information includes: temperature and light intensity;

[0060] When the substation type is wind power generation, the meteorological information includes: wind direction and wind speed;

[0061] When the substation type is a mixture of photovoltaic power generation and wind power generation, the meteorological information includes: temperature, light intensity, wind direction and wind speed.

[0062] According to an embodiment of the present disclosure, the Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism includes: an encoder, an attention module and a decoder, wherein the encoder and the decoder respectively include multiple layers of LSTM units connected in sequence;

[0063] The method of using the trained second agent to generate a coordinated voltage control scheme for each substation including a multi-point control curve based on the load information and energy output information of each substation within a preset time in the future includes:

[0064] The load information and energy output information of each substation within a preset future time are sorted into sequence data, wherein the sequence data includes a plurality of time steps, and each time step corresponds to a specific combination of load information and energy output information;

[0065] Processing each time step in the sequence data step by step based on the multiple LSTM units in the encoder, including: for each time step, the designated LSTM unit in the encoder generates a hidden state corresponding to the current time step according to the input data of the current time step and the hidden state of the previous time step;

[0066] After the multiple LSTM units in the encoder have processed each time step in the sequence data, based on the attention module, at each time step of the decoder, the attention weight is calculated according to the hidden state of the last layer of LSTM units in the encoder and the hidden state of all time steps of the encoder, and the hidden state of all time steps of the encoder is weighted and summed using the attention weight to obtain a weighted context vector, wherein the coordinated voltage control scheme of the substation area is characterized by an output sequence containing the preset number of control points, and the time step of the decoder corresponds to the time step relative to one control point in the output sequence;

[0067] The method comprises: for a current time step of the decoder, the LSTM unit in the decoder generates a control point of the current time step according to the weighted context vector and a control point of a previous time step;

[0068] After the decoder generates all the preset number of control points, the preset number of control points are combined into a coordinated voltage control scheme for an area including a multi-point control curve.

[0069] In a third aspect, a multi-agent-based voltage control method in a substation is provided in an embodiment of the present disclosure, which is applied to a voltage control unit, wherein the voltage control unit is arranged in each substation in a substation cluster and is communicatively connected to a voltage control center, and a voltage control unit in any substation in the substation cluster is communicatively connected to an energy control node in any substation, and the voltage control method includes:

[0070] Receiving the coordinated voltage control plan of the substation area sent by the voltage control center;

[0071] According to the coordinated voltage control scheme of the substation and the real-time load information of any of the substations, a trained third intelligent agent is used to generate a voltage control scheme within the substation; the voltage control scheme within the substation is sent to the energy control node in any of the substations; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical coordinated voltage control schemes of each substation as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

[0072] According to an embodiment of the present disclosure, the voltage control method further includes:

[0073] Receiving feedback data sent by an energy control node in any of the substations;

[0074] Adding the feedback data to the training data set of the third agent, and retraining the third agent using the updated training data set;

[0075] The voltage control scheme within the substation is adjusted based on the retrained third intelligent agent.

[0076] According to an embodiment of the present disclosure, the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism includes: an input module, an LSTM module, an attention mechanism module, an immediate feedback module and an output module; the voltage control scheme in the substation area is generated using the trained third agent according to the substation area collaborative voltage control scheme and the real-time load information of any substation area, including:

[0077] Preprocessing the coordinated voltage control scheme of the substation and the real-time load information of any substation based on the input module to generate corresponding time series format data;

[0078] Processing the time series format data based on the LSTM module includes: capturing long-term dependencies in the time series format data, obtaining time series features in the time series format data, and generating a corresponding hidden state vector group according to the long-term dependencies and the time series features;

[0079] Calculating the attention weight of each hidden state vector in the hidden state vector group based on the attention mechanism module, adding the results of multiplying each hidden state vector by its corresponding attention weight to obtain a weighted hidden state vector;

[0080] Based on the immediate feedback module, the weighted hidden state vector is compared with the immediate feedback information, an error signal is calculated, and an internal state parameter of the LSTM module or the attention mechanism module is adjusted according to the error signal, so that the LSTM module or the attention mechanism module regenerates the weighted hidden state vector according to the adjusted internal state parameter;

[0081] The weighted hidden state vector is decoded based on the output module to map the weighted hidden state vector to an output space corresponding to the voltage control scheme within the substation, a voltage control instruction corresponding to each energy control node is generated, and the voltage control scheme within the substation is generated based on the voltage control instructions of each energy control node.

[0082] According to an embodiment of the present disclosure, when the third intelligent agent is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism, the training weight of the station area topology structure information is gradually reduced according to a preset strategy until the input of the station area topology structure information is finally cancelled.

[0083] In a fourth aspect, a multi-agent-based voltage control method in a substation is provided in an embodiment of the present disclosure, which is applied to an energy control node. The energy control node is arranged in any substation in a substation cluster and is communicatively connected to a voltage control unit in any substation. The voltage control units in each substation in the substation cluster are respectively communicatively connected to a voltage control center. The voltage control method includes:

[0084] Receiving the voltage control plan within the substation area sent by the voltage control unit;

[0085] Voltage regulation is performed based on the voltage regulation scheme within the substation.

[0086] According to an embodiment of the present disclosure, the voltage control method further includes:

[0087] During the voltage regulation process based on the voltage regulation scheme in the substation, key parameters are monitored and recorded, the key parameters including: voltage level, power factor, active power and reactive power;

[0088] Obtaining a control effect evaluation index according to the key parameters, the control effect evaluation index includes: a voltage stability index and an energy efficiency index, and obtaining the control effect evaluation index according to the key parameters includes: determining the voltage stability index according to the voltage level, the voltage stability index includes: voltage fluctuation rate, voltage qualification rate and voltage average value and standard deviation; determining the energy efficiency index according to the power factor, the active power and the reactive power;

[0089] Compare the control effect evaluation index with the preset target value, analyze whether the control has achieved the expected effect, and obtain the evaluation result;

[0090] Feedback data is formed based on the control effect evaluation index, the evaluation result and the key parameter, and the feedback data is sent to the voltage control unit in any one of the substations.

[0091] According to an embodiment of the present disclosure, the voltage control scheme in the substation includes the voltage control frequency and voltage control speed of each energy control node in the substation, and the active and reactive power adjustment sequence, wherein the voltage control speed includes: voltage down-regulation speed and voltage up-regulation speed, and the voltage control based on the voltage control scheme in the substation includes:

[0092] Any energy control node in any one of the substations obtains the substation voltage control scheme corresponding to any one of the energy control nodes in the substation voltage control scheme;

[0093] The current voltage of any energy control node is regulated according to the voltage regulation frequency, voltage regulation speed and active and reactive power regulation sequence in the voltage regulation scheme in the substation of any energy control node, including: according to the voltage regulation frequency, the current voltage is monitored according to the corresponding regulation cycle, and in each regulation cycle, the current voltage data is obtained, and the current voltage data is compared with a preset maximum voltage threshold value, if the current voltage data is greater than the preset maximum voltage threshold value, the current voltage is lowered based on the voltage lowering speed in the voltage regulation speed, and if the current voltage data is less than the preset minimum voltage threshold value, the current voltage is increased based on the voltage increasing speed in the voltage regulation speed;

[0094] Wherein, when the current voltage of any energy control node is regulated, it is performed in accordance with the active and reactive power regulation sequence.

[0095] In a fifth aspect, a multi-agent-based voltage control device in a substation is provided in an embodiment of the present disclosure, which is arranged in a voltage control center, wherein the voltage control center is respectively connected to the voltage control units in each substation in the substation cluster, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation, and the voltage control device includes:

[0096] The first agent prediction module is configured to: obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within a preset time in the future according to the historical load information and area characteristic information of each area and the meteorological information within a preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection;

[0097] The second intelligent agent prediction module is configured as follows: according to the load information and energy output information of each substation within a preset future time, use the trained second intelligent agent to generate a substation coordinated voltage control plan including a multi-point control curve; send the substation coordinated voltage control plan to the voltage control unit in each substation; wherein the multi-point control curve includes a preset number of control points, and the control points represent the voltage control targets at the specified time points within the preset future time. The second intelligent agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism.

[0098] In a sixth aspect, a multi-agent-based voltage control device in a substation is provided in an embodiment of the present disclosure, which is arranged in a voltage control unit, and the voltage control unit is arranged in each substation in a substation cluster, and is communicatively connected to a voltage control center, and the voltage control unit in any substation in the substation cluster is communicatively connected to an energy control node in any substation, and the voltage control device includes:

[0099] The area coordinated voltage control scheme receiving module is configured to: receive the area coordinated voltage control scheme sent by the voltage control center;

[0100] The third intelligent agent prediction module is configured to: generate a voltage control plan within the substation using the trained third intelligent agent according to the substation coordinated voltage control plan and the real-time load information of any substation; send the voltage control plan within the substation to the energy control node in any substation; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical substation coordinated voltage control plan as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

[0101] In a seventh aspect, a multi-agent-based voltage control device in a substation is provided in an embodiment of the present disclosure, which is arranged in an energy control node, and the energy control node is arranged in any substation in a substation cluster, and is communicatively connected to a voltage control unit in any substation, and the voltage control units in each substation in the substation cluster are respectively communicatively connected to a voltage control center, and the voltage control device includes:

[0102] The intra-area voltage control scheme receiving module is configured to: receive the intra-area voltage control scheme sent by the voltage control unit;

[0103] The execution model of the voltage control scheme in the substation is configured to perform voltage control based on the voltage control scheme in the substation.

[0104] In the eighth aspect, a chip is provided in an embodiment of the present disclosure, comprising the voltage control device described in any one of aspects 5 to 7; or, comprising a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the voltage control method described in any one of aspects 2 to 4 of the claims.

[0105] In the ninth aspect, an electronic device is provided in an embodiment of the present disclosure, comprising a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the voltage control method described in any one of the second to fourth aspects.

[0106] In the tenth aspect, a computer-readable storage medium is provided in an embodiment of the present disclosure, on which computer instructions are stored. When the computer instructions are executed by a processor, the voltage control method described in any one of the second to fourth aspects is implemented.

[0107] In an eleventh aspect, a computer program product is provided in an embodiment of the present disclosure, including a computer program, and when the computer program is executed by a processor, the voltage control method described in any one of the second to fourth aspects is implemented.

[0108] According to the technical solution provided by the embodiment of the present disclosure, a multi-agent-based voltage control system in the substation is provided, which includes a voltage control center and voltage control units and energy control nodes in the substation. The voltage control center uses the first agent (based on the CNN-GRU neural network) to predict the future load and energy output of the substation, and then combines the second agent (based on the Seq2Seq neural network of LSTM and attention mechanism) to generate a coordinated voltage control plan for the substation. The multi-point control curve in the plan clarifies the voltage control target at each future time point. The voltage control unit in the substation generates a specific voltage control plan in the substation through the third agent (based on the LSTM neural network of instant feedback and attention mechanism) according to the coordinated voltage control plan of the substation and real-time load information, and sends it to the energy control node to perform voltage control. The present invention expands a single reinforcement learning problem into a multi-agent collaborative control problem through an on-site voltage coordination control algorithm based on multi-agent reinforcement learning, thereby meeting the needs of multi-regional voltage collaborative control. At the same time, through the collaborative work of various agents, fine voltage regulation of photovoltaic grid-connected points is effectively achieved, voltage exceeding the limit of photovoltaic grid-connected points is avoided, and the voltage stability and power quality of the distribution network are improved, providing strong technical support for the access of renewable energy and the intelligent operation of the distribution network.

[0109] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] Other features, objectives and advantages of the present disclosure will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings:

[0111] Figure 1 A schematic diagram of the structure of a multi-agent-based voltage control system in a substation according to an embodiment of the present disclosure is shown;

[0112] Figure 2 A schematic diagram showing prediction using a CNN-GRU neural network model based on residual connection according to an embodiment of the present disclosure;

[0113] Figure 3 A schematic diagram showing a method of generating a coordinated voltage regulation scheme for a substation using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism according to an embodiment of the present disclosure;

[0114] Figure 4 A schematic diagram showing a method of generating a voltage control scheme within a substation using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism according to an embodiment of the present disclosure;

[0115] Figure 5 A flow chart of a voltage control method in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0116] Figure 6 A flow chart of another method for voltage control in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0117] Figure 7 A flow chart of another method for voltage control in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0118] Figure 8 A flow chart of another method for voltage control in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0119] Fig. 9 A flow chart of a voltage control method in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0120] Fig.10 A flow chart of another method for voltage control in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0121] Fig.11A structural block diagram of a voltage control device in a substation based on multi-agents according to an embodiment of the present disclosure is shown;

[0122] Fig.12 A structural block diagram of another multi-agent-based voltage control device in a substation according to an embodiment of the present disclosure is shown;

[0123] Fig.13 A structural block diagram of another multi-agent-based voltage control device in a substation according to an embodiment of the present disclosure is shown;

[0124] Fig.14 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0125] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0126] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or a combination thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or a combination thereof exist or are added.

[0127] It should also be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0128] As mentioned above, the random fluctuations in the output of distributed power sources and the uncertain fluctuations in loads in the prior art have brought many challenges to the operation of the distribution network. Among them, the problems of large voltage fluctuations, voltage crossing the line, and increased network losses are particularly prominent, which seriously affect the quality of electric energy and the reliability of power supply. Traditional power grid control methods are powerless in the face of these challenges. They are often unable to refine the load data within the sensing area (i.e., a specific area or range of power supply), resulting in low control accuracy. In addition, traditional control methods mostly rely on active control, that is, maintaining voltage stability by adjusting the output or load of the generator. This method not only has a slow response speed, but also may lead to an increase in electricity costs for users.

[0129] In order to solve the above problems in the prior art, the present invention focuses on building an efficient, intelligent and adaptive voltage control system in the substation, adopts a multi-agent architecture, and realizes accurate prediction, coordinated control and real-time response of the voltage in the substation.

[0130] Figure 1FIG. 2 shows a schematic diagram of a voltage control system in a substation based on a multi-agent system according to an embodiment of the present disclosure. Figure 1 As shown, the voltage control system includes: a voltage control center, a cluster ( Figure 1 The voltage control unit in each area and the energy control node set in each area are respectively connected to the voltage control unit in each area ( Figure 1 In the example of connecting two voltage control units), the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation ( Figure 1 In the example above, connecting two energy control nodes is used for explanation).

[0131] Among them, the substation cluster in the present disclosure refers to a collection of multiple geographically adjacent or interrelated substations (i.e., power supply areas or distribution transformer service areas) in terms of power supply. These substations usually have similar load characteristics, energy output potential, or face similar voltage regulation challenges in the power grid. By managing and regulating these substations as a whole, more efficient energy utilization and more optimized voltage control can be achieved. Suppose a city is divided into multiple power supply areas, each of which is powered by one or more distribution transformers. These areas (substations) may be classified into the same substation cluster due to their geographical proximity, similar user types (such as residential areas, commercial areas, industrial areas, etc.), or similar grid structures.

[0132] The voltage control center in the present disclosure is the core of the entire voltage control system, located in the control center of the power company, and connected to the voltage control units in each substation through a high-speed communication network. The voltage control center achieves efficient data processing and decision-making through advanced intelligent agent technology (such as the first intelligent agent, the second intelligent agent) and neural network models (such as CNN-GRU, LSTM-Seq2Seq).

[0133] The voltage control unit in the present disclosure is arranged in each substation, and is a bridge between the voltage control center and the energy control node. For example, in a residential substation, the voltage control unit can be installed near the distribution transformer or integrated in the smart meter system.

[0134] The energy control nodes in the present disclosure are specific execution units of the voltage control scheme in the substation area, and can be distributed energy devices, such as photovoltaic power stations, wind power stations, etc. They perform voltage control operations according to the received voltage control scheme in the substation area to achieve voltage stability and optimization.

[0135] According to an embodiment of the present disclosure, the voltage control center is configured to: obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within the preset time in the future based on the historical load information and area characteristic information of each area, as well as the meteorological information within the preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection; according to the load information and energy output information of each area within the preset time in the future, use the trained second agent to generate an area coordinated voltage control scheme including a multi-point control curve; send the area coordinated voltage control scheme to the voltage control unit in each area; wherein the multi-point control curve includes a preset number of control points, and the control point represents the voltage control target corresponding to a specified time point within the preset time in the future, and the second agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each area as input data, and is trained using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism.

[0136] According to an embodiment of the present disclosure, the voltage control unit in any substation in the substation cluster is configured to: generate a voltage control plan within the substation using a trained third intelligent agent based on the substation collaborative voltage control plan and the real-time load information of any substation; send the voltage control plan within the substation to the energy control node in any substation; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical substation collaborative voltage control plan of each substation as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

[0137] According to an embodiment of the present disclosure, any energy control node in any substation is configured to perform voltage control based on the voltage control scheme in the substation.

[0138] The following is a detailed description of the specific functions of the voltage control center, voltage control unit and energy control node configured in the voltage control system.

[0139] The voltage control center is usually a central control system, which is mainly responsible for collecting, analyzing and processing historical load information, substation characteristic information and meteorological information from each substation to predict future load and energy output, and formulate coordinated voltage control plans for the substations accordingly.

[0140] The load information includes but is not limited to: the power load data of the substation, which refers to the total power demanded by the power users for the power system at a certain moment. The historical load information refers to the past power load situation of the substation, and the substation characteristic information includes but is not limited to: substation type, installed capacity, grid connection method and historical power generation efficiency. The substation type includes but is not limited to: photovoltaic power generation, wind power generation or a combination of photovoltaic power generation and wind power generation.

[0141] Among them, when the voltage control center predicts the load information and energy output information of each area within the preset time in the future based on the first intelligent agent, the meteorological information used depends on the type of area used. Specifically, when the area type is photovoltaic power generation, that is, the energy control node in the area is a photovoltaic power station. Since for photovoltaic power stations, light intensity is a key factor affecting photovoltaic power generation, and temperature will affect the efficiency of photovoltaic cells, it is also a meteorological parameter that photovoltaic power stations need to pay attention to. Therefore, the meteorological information may include: temperature and light intensity; similarly, when the area type is wind power generation, that is, the energy control node in the area is a wind power station, since wind speed is the main factor determining wind power generation, and wind direction will affect the wind receiving area and efficiency of wind turbines, therefore, the meteorological information may include: wind direction and wind speed; and when the area type is a mixture of photovoltaic power generation and wind power generation, that is, there are both wind power stations and photovoltaic power stations in the area, therefore, the meteorological information may include: temperature, light intensity, wind direction and wind speed.

[0142] When making predictions, the voltage control center considers the type of substation and the corresponding meteorological information, because the characteristic information of the substation can reflect the power generation capacity and historical performance of the substation. Using this information as input data allows the first intelligent agent to more accurately understand the specific situation of the substation, thereby improving the accuracy of predictions of future loads and energy output. At the same time, it also enables the voltage control system to better adapt to different types of substations, whether photovoltaic power generation, wind power generation or mixed types, can be targeted for prediction and processing. In addition, by using corresponding meteorological information for different substation types (such as photovoltaic power generation considering temperature and light intensity, wind power generation considering wind direction and wind speed), it can more accurately reflect the impact of meteorological conditions on the power generation capacity of the substation, further improving the accuracy of the prediction.

[0143] For the first agent, the main task is to predict the load information and energy output information of each substation within a preset time in the future. This requires the model to be able to extract spatial features (such as substation feature information) and capture the time dependencies in time series data (such as historical load information). The present disclosure adopts a CNN-GRU neural network model based on residual connection to specifically implement the first agent. It is a deep learning model that combines convolutional neural network (CNN) and gated recurrent unit (GRU), and introduces residual connection. This structure can use CNN to extract spatial features and GRU to capture the time dependencies in time series data. At the same time, residual connection can further improve the training effect and prediction accuracy of the model.

[0144] According to an embodiment of the present disclosure, the CNN-GRU neural network model based on residual connection includes: an input module, a convolutional neural network CNN module, a gated recurrent unit GRU module and an output module, wherein the CNN module may include multiple layers of CNN, the GRU module may include multiple layers of GRU, and residual connections are included between the internal layers of the CNN module or between the internal layers of the GRU module or between the CNN module and the GRU module.

[0145] Figure 2 A schematic diagram showing prediction using a CNN-GRU neural network model based on residual connection according to an embodiment of the present disclosure. Figure 2 The following example is taken to illustrate that the internal layers of the GRU module contain residual connections. Figure 2 As shown, when the trained first agent is used to predict the load information and energy output information of each substation within a preset time in the future based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future:

[0146] First, based on the input module, the historical load information and the characteristic information of each substation, as well as the meteorological information within a preset time in the future are received, and the historical load information and the characteristic information of each substation, as well as the meteorological information within a preset time in the future are integrated into a multi-dimensional input tensor, wherein each dimension represents different time series data; then, based on the CNN module, the spatial features in the multi-dimensional input tensor are extracted through multi-layer convolution, or multi-layer convolution and residual connection to form a feature map; then, based on the GRU module, the time series data in the feature map is processed through a multi-layer recurrent neural network, or a multi-layer recurrent neural network and residual connection to capture the time dependency and generate a hidden state vector; finally, based on the output module, the hidden state vector is processed through a fully connected layer or a regression layer to generate a prediction result, and the prediction result includes: the load information and energy output information of each substation within the preset time in the future.

[0147] For the second agent, the main task is to generate a coordinated voltage control scheme for the substation containing a multi-point control curve based on the load information and energy output information of each substation within a preset time in the future. This disclosure takes into account the complexity of the coordinated voltage control scheme for the substation and the need for fine control, requiring the model to be able to dynamically pay attention to key information and the ability to process complex data. Specifically, a Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism is used to implement the second agent. The Seq2Seq neural network model based on LSTM and the attention mechanism can learn the complex correlations between each substation and generate accurate coordinated voltage control schemes for the substation or other related prediction results.

[0148] According to an embodiment of the present disclosure, the Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism includes: an encoder, an attention module and a decoder, and the encoder and the decoder respectively include multiple layers of LSTM units connected in sequence.

[0149] Figure 3 A schematic diagram showing a method for generating a coordinated voltage regulation scheme for a substation using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism according to an embodiment of the present disclosure. Figure 3 As shown, when the trained second agent is used to generate a coordinated voltage control scheme for each substation including a multi-point control curve based on the load information and energy output information of each substation within a preset time in the future:

[0150] First, the load information and energy output information of each substation within a preset future time are sorted into sequence data, wherein the sequence data includes multiple time steps, and each time step corresponds to a specific combination of load information and energy output information; then, each time step in the sequence data is processed step by step based on multiple LSTM units in the encoder, including: for each time step, the designated LSTM unit in the encoder generates a hidden state corresponding to the current time step according to the input data of the current time step and the hidden state of the previous time step; thereafter, after the multiple LSTM units in the encoder have processed each time step in the sequence data, based on the attention module, at each time step of the decoder, the hidden state of the last layer of LSTM units in the encoder and all time steps of the encoder are generated; The attention weight is calculated based on the hidden state of the encoder, and the attention weight is used to perform weighted summation on the hidden states of all time steps of the encoder to obtain a weighted context vector, wherein the substation coordinated voltage control scheme is characterized by an output sequence containing the preset number of control points, and the time step of the decoder corresponds to the time step relative to a control point in the output sequence; then, each control point in the output sequence is gradually generated based on multiple LSTM units in the decoder, including: for the current time step of the decoder, the LSTM unit in the decoder generates the control point of the current time step according to the weighted context vector and the control point of the previous time step; finally, when the decoder has generated all the preset number of control points, the preset number of control points are combined into a substation coordinated voltage control scheme containing a multi-point control curve.

[0151] When training the second agent, reverse overload information and voltage over-limit information are used as part of the input data, wherein reverse overload information mainly refers to data and information related to transformer reverse overload fault. Transformer reverse overload refers to a fault in which the direction of the transformer's electric force is opposite to the rated value and its magnitude exceeds the rated load. This fault may cause damage to the transformer or even cause serious consequences such as fire. Reverse overload information includes: current direction information (indicating whether the current flows in the reverse direction), current magnitude information, duration information (recording the duration of the reverse overload fault), temperature information, etc.; voltage over-limit information refers to the situation in which the voltage in the power system exceeds the rated value. It may cause damage to the equipment or even cause equipment failure or failure. Voltage over-limit information includes: voltage value information, over-limit type information (indicating whether the voltage is too high or too low to cause the over-limit), duration information, and occurrence location information, etc.

[0152] The present disclosure uses reverse heavy overload information and voltage over-limit information as training data to train the second agent, which has the following important significance:

[0153] 1. Improve prediction accuracy: This information contains abnormal conditions that transformers and power systems may encounter in actual operation. Through training, the second agent can learn the characteristics and patterns of these abnormal conditions, thereby improving the prediction accuracy of future load information and energy output information.

[0154] 2. Enhance model robustness: Since reverse overload and voltage overlimit are common fault types in power systems, using relevant information as training data can make the second agent more familiar and adaptable to these abnormal situations. In this way, in actual applications, even when encountering similar fault conditions, the model can maintain good performance and stability.

[0155] 3. Optimize voltage control scheme: By considering reverse heavy overload and voltage over-limit information, the second agent can more comprehensively consider the actual operation status and possible risk factors of the power system when generating the coordinated voltage control scheme of the substation area. This helps to generate a more reasonable and effective voltage control scheme and improve the operation efficiency and stability of the power system.

[0156] 4. Improve system safety: Timely identification and response to abnormal situations such as reverse overload and voltage over-limit are crucial to ensure the safe operation of the power system. By training the second agent, rapid response and effective handling of these abnormal situations can be achieved, thereby reducing the probability and impact of failures and improving the safety of the entire power system.

[0157] As for the voltage control unit, it is a local control system set up in each substation, which can be composed of distribution terminals (such as data transmission units DTU, feeder terminal units FTU, etc.) or transformers with intelligent control functions, etc., and is mainly responsible for using the third intelligent body to generate a voltage control plan in the substation. In the specific implementation of the present disclosure, a substation fusion terminal equipped with a Falcon SAC 300 chip is used as the voltage control unit.

[0158] For the third agent used by the voltage control unit, the main task is to generate a voltage control plan within the substation according to the substation collaborative voltage control plan and the real-time load information of any substation. The present disclosure adopts an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism to specifically implement the third agent. By combining the attention mechanism with LSTM, the ability of the Seq2Seq model to process sequence data can be further enhanced. LSTM is responsible for capturing long-term dependencies in the sequence, while the attention mechanism helps the model pay more attention to key information in the input sequence when generating outputs. In addition, the voltage control task requires the model to be real-time and dynamic, and to be able to quickly adjust the voltage control plan according to real-time load information and other changes. The Seq2Seq model combines the attention mechanism and LSTM to improve the response speed and adaptability of the model while maintaining prediction accuracy.

[0159] According to an embodiment of the present disclosure, the LSTM neural network model based on the instant feedback mechanism and the attention mechanism includes: an input module, an LSTM module, an attention mechanism module, an instant feedback module and an output module.

[0160] Figure 4 A schematic diagram showing a voltage regulation scheme within a substation using an LSTM neural network model based on an instant feedback mechanism and an attention mechanism according to an embodiment of the present disclosure. Figure 4 As shown, when the voltage control scheme in the substation is generated using the trained third agent according to the substation coordinated voltage control scheme and the real-time load information of any substation:

[0161] First, based on the input module, the coordinated voltage control scheme of the substation and the real-time load information of any substation are preprocessed to generate corresponding time series format data; then, based on the LSTM module, the time series format data is processed, including: capturing the long-term dependency in the time series format data, obtaining the timing characteristics in the time series format data, and generating a corresponding hidden state vector group according to the long-term dependency and the timing characteristics; then, based on the attention mechanism module, the attention weight of each hidden state vector in the hidden state vector group is calculated, and the results of multiplying each hidden state vector by its corresponding attention weight are added to obtain a weighted hidden state vector. quantity; then, based on the instant feedback module, the weighted hidden state vector is compared with the instant feedback information, an error signal is calculated, and the internal state parameters of the LSTM module or the attention mechanism module are adjusted according to the error signal, so that the LSTM module or the attention mechanism module regenerates the weighted hidden state vector according to the adjusted internal state parameters; finally, based on the output module, the weighted hidden state vector is decoded to map the weighted hidden state vector to an output space corresponding to the voltage control scheme in the substation, a voltage control instruction corresponding to each energy control node is generated, and the voltage control scheme in the substation is generated based on the voltage control instructions of each energy control node.

[0162] Among them, when the third agent is trained using the LSTM neural network model based on the instant feedback mechanism and the attention mechanism, the training weight of the station area topological structure information is gradually reduced according to the preset strategy until the input of the station area topological structure information is finally cancelled, thereby reducing the algorithm's dependence on the topological structure. This solution has the following beneficial technical effects:

[0163] 1. Enhanced the generalization ability of the model: The substation topology information usually has specific regional and grid layout characteristics. Gradually reducing its training weight and eventually canceling the input can prompt the model to pay more attention to universal characteristics such as load changes and energy output, thereby enhancing the generalization ability of the model in different substations and different grid structures.

[0164] 2. Improved forecasting flexibility: When the topology of the substation changes (such as adding new lines, equipment modification, etc.), if the model relies too much on topology information, it may lead to deviations in the forecast results. By gradually reducing the weight of topology information, the model can adapt more flexibly to changes in the grid structure and maintain a high forecasting accuracy.

[0165] 3. Reduced data dependence: Substation topology information may require detailed grid drawings and on-site surveys to obtain, which increases the difficulty and cost of data collection. Gradually reducing the weight of topology information can reduce the model's dependence on such data, allowing the model to maintain good performance even when the data is incomplete or difficult to obtain.

[0166] 4. Accelerated model training: The topological structure information of the substation usually contains a large number of nodes and connection relationships, and processing this information requires additional computing resources. Gradually reducing the weight of the topological structure information can simplify the input of the model, reduce the amount of calculation, and thus accelerate the model training process.

[0167] 5. Improved model robustness: By gradually reducing and eventually eliminating the input of the station area topology information, the model is forced to learn to extract useful information from other more common features, which helps to improve the model's robustness to noise and abnormal data, allowing the model to maintain stable prediction performance in the face of uncertainty and interference.

[0168] 6. Easy model update and maintenance: As the power grid develops and changes, the substation topology information may change. By reducing the model's reliance on such information, the model can be updated and maintained more easily without the need to frequently adjust the model structure or re-collect data.

[0169] Energy control nodes usually refer to devices located at the end of the distribution network or the user end in the power system, such as reactive power compensation devices (such as SVC, SVG), distributed energy access devices (such as photovoltaic inverters, wind power converters), etc. These devices can perform corresponding operations according to the received control instructions or strategies to achieve the regulation and control of power load, voltage, current and other parameters. In this disclosure, it is mainly responsible for executing the voltage control plan within the substation.

[0170] According to an embodiment of the present disclosure, the voltage control scheme within the substation includes the voltage control frequency and voltage control speed of each energy control node in the substation, as well as the active and reactive power adjustment sequence, wherein the voltage control speed includes: voltage reduction speed and voltage increase speed, and the voltage control frequency is related to a preset number of control points contained in the multi-point control curve.

[0171] The present disclosure combines the communication transmission stability factor and the real-time factor when designing the voltage control frequency, that is, when determining the specific number of points of the control curve used. Through repeated experiments, it is found that when 1 minute is used as the time node for data transmission, the data transmission produces repeated and unstable phenomena, which indicates that the shorter time interval may cause conflicts or network congestion during the data transmission process, thereby affecting the accuracy and reliability of the data. Comparing the data transmission results and voltage change trends of 5 minutes, 10 minutes, and 15 minutes as time nodes, it is found that the data transmission curve is the smoothest at 5 minutes, and the transmitted data does not have a repetitive phenomenon, which indicates that appropriately extending the time interval of data transmission helps to reduce interference and errors in the data transmission process and improve the stability and reliability of the data. Since the stability and real-time performance of data transmission directly affect the generation of voltage control strategy and the accuracy of voltage control, the present disclosure chooses to use 5 minutes as the time node for voltage data collection and voltage control, that is, within 24 hours, it is regulated every 5 minutes. In this way, while ensuring a certain real-time performance, the stability of data transmission and the accuracy of voltage control are optimized, and a good balance between real-time performance and stability is achieved. Corresponding to the control curve, a 288-point control curve needs to be generated.

[0172] Then, when the preset number of control points is 288 points, the multi-point control curve is a 288-point control curve, corresponding to the voltage control frequency of once every 5 minutes; when the preset number of control points is 96 points, the multi-point control curve is a 96-point control curve, corresponding to the voltage control frequency of once every 15 minutes.

[0173] According to an embodiment of the present disclosure, the voltage regulation based on the voltage regulation scheme in the substation includes:

[0174] Any energy control node in any of the substations obtains the substation voltage control scheme corresponding to any of the energy control nodes in the substation voltage control scheme; and controls the current voltage of any of the energy control nodes according to the voltage control frequency, voltage control speed and active and reactive power adjustment sequence in the substation voltage control scheme of any of the energy control nodes, including: monitoring the current voltage according to the corresponding control cycle according to the voltage control frequency, obtaining the current voltage data in each control cycle, comparing the current voltage data with a preset maximum voltage threshold, if the current voltage data is greater than the preset maximum voltage threshold, then reducing the current voltage based on the voltage reduction speed in the voltage control speed, if the current voltage data is less than the preset minimum voltage threshold, then increasing the current voltage based on the voltage increase speed in the voltage control speed; wherein, when controlling the current voltage of any of the energy control nodes, it is performed in accordance with the active and reactive power adjustment sequence.

[0175] When formulating the voltage control plan within the substation, the present disclosure is based on the substation coordinated voltage control plan formulated by the voltage control unit. Since the substation coordinated voltage control plan is based on the perspective of the entire substation cluster, it comprehensively considers the load conditions, energy output and future meteorological information of each substation, etc., and provides global guidance for the voltage control plan within the substation, which helps to avoid local optimization problems that may arise when regulating a single substation, and achieve voltage stability and optimization in the global range. Therefore, through the input of the substation coordinated voltage control plan, the voltage control plan within the substation can be better coordinated with other substations to ensure that the voltage control actions between the substations are consistent and avoid mutual interference and conflict, which helps to improve the voltage control efficiency and effect of the entire substation cluster.

[0176] In addition, the coordinated voltage control scheme of the substation is predicted based on the load information and energy output information within the preset time in the future, which provides forward-looking guidance for the voltage control scheme in the substation. The energy control node can understand the future voltage control needs in advance, make corresponding preparations and adjustments, so as to better cope with possible load changes and voltage fluctuations. In addition, the coordinated voltage control scheme of the substation uses trained intelligent agents for prediction, which has high accuracy. This helps to improve the accuracy of the voltage control scheme in the substation, reduce the control error and unnecessary adjustment times, and thus reduce the impact on the power grid and users.

[0177] In addition, through the input of the coordinated voltage control plan of the substation, the voltage control plan within the substation can better understand the voltage control goals and strategies of the entire substation cluster, so as to formulate a more scientific and reasonable control plan. This helps to optimize the voltage control strategy and improve the efficiency and effect of control. In addition, the voltage control plan within the substation can quickly respond and adjust the voltage output according to the requirements of the coordinated voltage control plan of the substation. This helps to improve the response speed of the entire substation cluster to load changes and voltage fluctuations, and ensure the stability and reliability of voltage.

[0178] In addition, any energy control node in any of the above-mentioned areas is further configured as:

[0179] During the voltage control process based on the voltage control scheme within the substation, key parameters are monitored and recorded, and the key parameters include: voltage level, power factor, active power and reactive power; wherein, the voltage level includes but is not limited to: real-time voltage value (including the voltage size before and after control), voltage fluctuation range (i.e., the maximum and minimum values ​​of the voltage during the control process).

[0180] The control effect evaluation index is obtained according to the key parameters, and the control effect evaluation index includes: voltage stability index and energy efficiency index. The control effect evaluation index is obtained according to the key parameters, including: determining the voltage stability index according to the voltage level, and the voltage stability index includes: voltage fluctuation rate (obtained by calculating the ratio of the voltage fluctuation range after regulation to the rated voltage), voltage qualification rate (achieved by counting the proportion of time when the voltage is within the allowable range) and voltage average and standard deviation, wherein the voltage average and standard deviation reflect the overall level and dispersion of the voltage, and these indicators are used to evaluate whether the voltage is stable within the specified range, and the impact of regulation on voltage stability; determining the energy efficiency index according to the power factor, the active power and the reactive power, and the energy efficiency index is used to evaluate the effect of regulation on improving the efficiency of electric energy utilization; wherein the power factor can be directly used as one of the energy efficiency indicators to reflect the efficiency of electric energy utilization. In addition to the power factor, the energy efficiency index also includes the energy efficiency ratio, which is obtained by calculating the ratio of the active power after regulation to the total energy consumption, reflecting the improvement of the energy efficiency of regulation. The regulation effect evaluation index is compared with the preset target value to analyze whether the regulation has achieved the expected effect and obtain the evaluation result; feedback data is formed based on the regulation effect evaluation index, the evaluation result and the key parameters, and the feedback data is sent to the voltage regulation unit in any substation, so that the voltage regulation unit in any substation in the substation cluster adds the feedback data to the training data set of the third intelligent agent, retrains the third intelligent agent with the updated training data set, and adjusts the voltage regulation scheme in the substation based on the retrained third intelligent agent.

[0181] The energy control node in the present disclosure can accurately evaluate the actual effect of voltage regulation by real-time monitoring and recording key parameters; then these key parameters are converted into regulation effect evaluation indicators (such as voltage stability indicators and energy efficiency indicators), and compared with the preset target values, so that deviations and deficiencies in the regulation process can be discovered in a timely manner. Finally, the third agent is retrained based on the feedback data, so that it can predict and regulate voltage more accurately, thereby improving the accuracy and efficiency of voltage regulation.

[0182] Therefore, the present invention significantly improves the accuracy and efficiency of voltage regulation by obtaining the regulation results and feeding them back to the third intelligent agent for retraining, and adjusting the voltage regulation plan in the substation based on the retrained third intelligent agent, enhances the system's adaptability and learning ability, optimizes the voltage regulation strategy and reduces energy consumption, and improves the reliability and stability of the system.

[0183] Figure 5A flow chart of a multi-agent-based voltage control method in a substation according to an embodiment of the present disclosure is shown. The voltage control method is applied to a voltage control center, and the voltage control center is respectively connected to the voltage control units in each substation in the substation cluster, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation, specifically as follows Figure 5 As shown, the voltage control method includes the following steps S510-S520:

[0184] In step S510, the historical load information and substation characteristic information of each substation are obtained through the voltage control unit in each substation in the substation cluster; based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future, the trained first intelligent agent is used to predict the load information and energy output information of each substation within a preset time in the future; wherein the first intelligent agent is trained using a CNN-GRU neural network model based on residual connection.

[0185] In step S520, based on the load information and energy output information of each substation within a preset future time, a trained second intelligent agent is used to generate a substation coordinated voltage control plan including a multi-point control curve; the substation coordinated voltage control plan is sent to the voltage control units in each substation; wherein the multi-point control curve includes a preset number of control points, and the control points represent voltage control targets corresponding to specified time points within a preset future time. The second intelligent agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism.

[0186] According to an embodiment of the present disclosure, the residual connection-based CNN-GRU neural network model includes: an input module, a convolutional neural network CNN module, a gated recurrent unit GRU module and an output module, wherein a residual connection is included between the internal layers of the CNN module or between the internal layers of the GRU module or between the CNN module and the GRU module;

[0187] The method uses the trained first agent to predict the load information and energy output information of each substation within a preset time in the future based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future, including:

[0188] Based on the input module, the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are received, and the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are integrated into a multi-dimensional input tensor, wherein each dimension represents different time series data;

[0189] Extracting spatial features in the multi-dimensional input tensor through multi-layer convolution, or multi-layer convolution and residual connection based on the CNN module to form a feature map;

[0190] Processing the time series data in the feature graph through a multi-layer recurrent neural network or a multi-layer recurrent neural network and a residual connection based on the GRU module, thereby capturing the time dependency and generating a hidden state vector;

[0191] The hidden state vector is processed by the output module through a fully connected layer or a regression layer to generate a prediction result, and the prediction result includes: load information and energy output information of each substation within a preset time in the future.

[0192] According to an embodiment of the present disclosure, the substation characteristic information includes but is not limited to: substation type, installed capacity, grid connection mode and historical power generation efficiency; the substation type includes: photovoltaic power generation, wind power generation or a combination of photovoltaic power generation and wind power generation;

[0193] When the area type is photovoltaic power generation, the meteorological information includes: temperature and light intensity;

[0194] When the substation type is wind power generation, the meteorological information includes: wind direction and wind speed;

[0195] When the substation type is a mixture of photovoltaic power generation and wind power generation, the meteorological information includes: temperature, light intensity, wind direction and wind speed.

[0196] According to an embodiment of the present disclosure, the Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism includes: an encoder, an attention module and a decoder, wherein the encoder and the decoder respectively include multiple layers of LSTM units connected in sequence;

[0197] According to an embodiment of the present disclosure, the method of using the trained second agent to generate a coordinated voltage control scheme for each substation including a multi-point control curve based on the load information and energy output information of each substation within a preset future time includes:

[0198] The load information and energy output information of each substation within a preset future time are sorted into sequence data, wherein the sequence data includes a plurality of time steps, and each time step corresponds to a specific combination of load information and energy output information;

[0199] Processing each time step in the sequence data step by step based on the multiple LSTM units in the encoder, including: for each time step, the designated LSTM unit in the encoder generates a hidden state corresponding to the current time step according to the input data of the current time step and the hidden state of the previous time step;

[0200] After the multiple LSTM units in the encoder have processed each time step in the sequence data, based on the attention module, at each time step of the decoder, the attention weight is calculated according to the hidden state of the last layer of LSTM units in the encoder and the hidden state of all time steps of the encoder, and the hidden state of all time steps of the encoder is weighted and summed using the attention weight to obtain a weighted context vector, wherein the coordinated voltage control scheme of the substation area is characterized by an output sequence containing the preset number of control points, and the time step of the decoder corresponds to the time step relative to one control point in the output sequence;

[0201] The method comprises: for a current time step of the decoder, the LSTM unit in the decoder generates a control point of the current time step according to the weighted context vector and a control point of a previous time step;

[0202] After the decoder generates all the preset number of control points, the preset number of control points are combined into a coordinated voltage control scheme for an area including a multi-point control curve.

[0203] Figure 6 Another flow chart of a multi-agent-based voltage control method in a substation according to an embodiment of the present disclosure is shown. The voltage control method is applied to a voltage control unit, which is arranged in each substation in a substation cluster and is connected to a voltage control center in communication. The voltage control unit in any substation in the substation cluster is connected to an energy control node in any substation in communication. Figure 6 As shown, the voltage control method includes the following steps S610-S620:

[0204] In step S610, a coordinated voltage control scheme for substations sent by the voltage control center is received.

[0205] In step S620, based on the coordinated voltage control scheme of the substation and the real-time load information of any of the substations, a trained third intelligent agent is used to generate a voltage control scheme within the substation; the voltage control scheme within the substation is sent to the energy control node in any of the substations; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical coordinated voltage control schemes of each substation as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

[0206] Figure 7 A flow chart of another method for voltage regulation in a substation based on multi-agents according to an embodiment of the present disclosure is shown. The voltage regulation method further includes the following steps S630-S640:

[0207] In step S630, feedback data sent by the energy control node in any of the substations is received; the feedback data is added to the training data set of the third agent, and the third agent is retrained using the updated training data set.

[0208] In step S640, the voltage control scheme in the substation is adjusted based on the retrained third agent.

[0209] According to an embodiment of the present disclosure, the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism includes: an input module, an LSTM module, an attention mechanism module, an immediate feedback module and an output module; the voltage control scheme in the substation area is generated using the trained third agent according to the substation area collaborative voltage control scheme and the real-time load information of any substation area, including:

[0210] Preprocessing the coordinated voltage control scheme of the substation and the real-time load information of any substation based on the input module to generate corresponding time series format data;

[0211] Processing the time series format data based on the LSTM module includes: capturing long-term dependencies in the time series format data, obtaining time series features in the time series format data, and generating a corresponding hidden state vector group according to the long-term dependencies and the time series features;

[0212] Calculating the attention weight of each hidden state vector in the hidden state vector group based on the attention mechanism module, adding the results of multiplying each hidden state vector by its corresponding attention weight to obtain a weighted hidden state vector;

[0213] Based on the immediate feedback module, the weighted hidden state vector is compared with the immediate feedback information, an error signal is calculated, and an internal state parameter of the LSTM module or the attention mechanism module is adjusted according to the error signal, so that the LSTM module or the attention mechanism module regenerates the weighted hidden state vector according to the adjusted internal state parameter;

[0214] The weighted hidden state vector is decoded based on the output module to map the weighted hidden state vector to an output space corresponding to the voltage control scheme within the substation, a voltage control instruction corresponding to each energy control node is generated, and the voltage control scheme within the substation is generated based on the voltage control instructions of each energy control node.

[0215] Among them, when the third intelligent agent is trained using the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism, the training weight of the station area topological structure information is gradually reduced according to the preset strategy until the input of the station area topological structure information is finally cancelled.

[0216] Figure 8 A flow chart of another method for voltage control in a substation based on multi-agents according to an embodiment of the present disclosure is shown. The voltage control method is applied to an energy control node, which is arranged in any substation in a substation cluster and is connected to a voltage control unit in any substation in a communication connection. The voltage control units in each substation in the substation cluster are connected to a voltage control center in communication connection. Figure 8 As shown, the voltage control method includes the following steps S810-S820:

[0217] In step S810, a voltage control scheme within a substation area sent by the voltage control unit is received.

[0218] In step S820, voltage control is performed based on the intra-station voltage control scheme.

[0219] The voltage control method further includes:

[0220] Fig. 9 FIG. 2 shows another flow chart of a method for voltage control in a substation based on multi-agents according to an embodiment of the present disclosure. In addition to the above steps S810 to S820, the specific steps are as follows: Fig. 9 As shown, the voltage control method further includes the following steps S830-S850:

[0221] In step S830, during the voltage regulation process based on the voltage regulation scheme within the substation, key parameters are monitored and recorded, and the key parameters include: voltage level, power factor, active power and reactive power.

[0222] In step S840, a control effect evaluation index is obtained according to the key parameters, and the control effect evaluation index includes: a voltage stability index and an energy efficiency index. The control effect evaluation index is obtained according to the key parameters, including: determining the voltage stability index according to the voltage level, and the voltage stability index includes: voltage fluctuation rate, voltage qualification rate and voltage average and standard deviation; determining the energy efficiency index according to the power factor, the active power and the reactive power; comparing the control effect evaluation index with the preset target value, analyzing whether the control achieves the expected effect, and obtaining the evaluation result.

[0223] In step S850, feedback data is formed based on the control effect evaluation index, the evaluation result and the key parameter, and the feedback data is sent to the voltage control unit in any one of the substations.

[0224] According to an embodiment of the present disclosure, the voltage control scheme in the substation includes the voltage control frequency and voltage control speed of each energy control node in the substation, and the active and reactive power adjustment sequence, wherein the voltage control speed includes: voltage down-regulation speed and voltage up-regulation speed, and the voltage control based on the voltage control scheme in the substation includes:

[0225] Any energy control node in any one of the substations obtains the substation voltage control scheme corresponding to the any energy control node in the substation voltage control scheme.

[0226] The current voltage of any energy control node is regulated according to the voltage regulation frequency, voltage regulation speed and active and reactive power regulation sequence in the voltage regulation scheme in the substation of any energy control node, including: according to the voltage regulation frequency, the current voltage is monitored according to the corresponding regulation cycle, in each regulation cycle, the current voltage data is obtained, and the current voltage data is compared with the preset maximum voltage threshold, if the current voltage data is greater than the preset maximum voltage threshold, the current voltage is lowered based on the voltage reduction speed in the voltage regulation speed, if the current voltage data is less than the preset minimum voltage threshold, the current voltage is increased based on the voltage increase speed in the voltage regulation speed. Wherein, when the current voltage of any energy control node is regulated, it is carried out according to the active and reactive power regulation sequence.

[0227] Based on the functions performed by the voltage control center, voltage control unit and energy control node in the voltage control system in the substation area as mentioned above, Fig.10 FIG. 2 shows a flow chart of another method for voltage regulation in a substation based on multi-agent according to an embodiment of the present disclosure. Fig.10As shown, it includes the above steps S510-520, S610-S620 and S810-S820.

[0228] Fig.11 The structure block diagram of a multi-agent-based voltage control device in a substation according to an embodiment of the present disclosure is shown. The voltage control device is arranged in a voltage control center, and the voltage control center is respectively connected to the voltage control units in each substation in the substation cluster, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation, such as Fig.11 As shown, the voltage control device 1100 includes: a first agent prediction module, which is configured to: obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within the future preset time according to the historical load information and area characteristic information of each area, as well as the meteorological information within the future preset time; wherein the first agent is trained using the CNN-GRU neural network model based on residual connection; the second agent prediction module is configured to: predict the load information and energy output information of each area within the future preset time according to the historical load information and area characteristic information of each area, as well as the meteorological information within the future preset time The load information and energy output information in the area are used, and a coordinated voltage control plan for the area including a multi-point control curve is generated using the trained second intelligent agent; the coordinated voltage control plan for the area is sent to the voltage control units in each area; wherein the multi-point control curve includes a preset number of control points, and the control points represent the voltage control targets corresponding to the specified time points within the preset future time. The second intelligent agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each area as input data, and is trained using a Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism.

[0229] Fig.12 The structural block diagram of another multi-agent-based voltage control device in a substation according to an embodiment of the present disclosure is shown. The voltage control device is arranged in a voltage control unit, and the voltage control unit is arranged in each substation in the substation cluster and is connected to the voltage control center in communication. The voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation in communication. Fig.12As shown, the voltage control device 1200 includes: a substation collaborative voltage control scheme receiving module, which is configured to: receive the substation collaborative voltage control scheme sent by the voltage control center; a third intelligent agent prediction module, which is configured to: generate a voltage control scheme within the substation using a trained third intelligent agent based on the substation collaborative voltage control scheme and the real-time load information of any substation; and send the voltage control scheme within the substation to the energy control node in any substation; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical substation collaborative voltage control schemes of each substation as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

[0230] Fig.13 A structural block diagram of another multi-agent-based voltage control device in a substation according to an embodiment of the present disclosure is shown. The voltage control device is arranged at an energy control node, and the energy control node is arranged in any substation in a substation cluster, and is connected to a voltage control unit in any substation in a communication connection, and the voltage control unit in each substation in the substation cluster is connected to a voltage control center in communication connection, such as Fig.13 As shown, the voltage control device 1300 includes: an intra-substation voltage control scheme receiving module, which is configured to: receive the intra-substation voltage control scheme sent by the voltage control unit; and an intra-substation voltage control scheme execution model, which is configured to: perform voltage control based on the intra-substation voltage control scheme.

[0231] The present disclosure also provides a chip, comprising a voltage control device as described in any one of the above device embodiments; or, comprising a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the voltage control method as described in any one of the above method embodiments.

[0232] Fig.14 1 shows a structural block diagram of an electronic device according to an embodiment of the present disclosure. Fig.14 As shown, the electronic device includes a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the voltage control method as described in any one of the above method embodiments.

[0233] The present disclosure also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or computer system in the above embodiment; or a computer-readable storage medium that exists independently and is not assembled into a device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the voltage regulation method described in the present disclosure.

[0234] The present disclosure further provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the voltage control method described in any one of the present disclosure is implemented.

[0235] According to the technical solution provided by the embodiment of the present disclosure, a multi-agent-based voltage control system in the substation is provided, which includes a voltage control center and a voltage control unit and an energy control node in the substation. The voltage control center uses the first agent (based on CNN-GRU neural network) to predict the future load and energy output of the substation, and then combines the second agent (Seq2Seq neural network based on LSTM and attention mechanism) to generate a coordinated voltage control scheme for the substation. The multi-point control curve in the scheme clarifies the voltage control target at each time point in the future. The voltage control unit in the substation generates a specific voltage control scheme in the substation according to the coordinated voltage control scheme of the substation and real-time load information through the third agent (LSTM neural network based on instant feedback and attention mechanism), and sends it to the energy control node to perform voltage control. Through the collaborative work of each agent, the fine voltage control of the photovoltaic grid-connected point is effectively realized, the voltage limit of the photovoltaic grid-connected point is avoided, the voltage stability and power quality of the distribution network are improved, and a strong technical support is provided for the access of renewable energy and the intelligent operation of the distribution network.

[0236] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other.

Claims

1. A voltage control system in a substation based on multi-agent, characterized in that: The voltage control system includes: a voltage control center, a voltage control unit in each substation in the substation cluster, and an energy control node in each substation. The voltage control center is respectively connected to the voltage control unit in each substation, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation, wherein: The voltage control center is configured as follows: Obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within a preset time in the future based on the historical load information and area characteristic information of each area and the meteorological information within a preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection; According to the load information and energy output information of each substation within a preset future time, a substation coordinated voltage control scheme including a multi-point control curve is generated using a trained second agent; the substation coordinated voltage control scheme is sent to the voltage control unit in each substation; wherein the multi-point control curve includes a preset number of control points, and the control points represent the voltage control target corresponding to a specified time point within a preset future time, and the second agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism; The voltage control unit in any substation in the substation cluster is configured as follows: According to the coordinated voltage control scheme of the substation area and the real-time load information of any substation area, a trained third agent is used to generate a voltage control scheme within the substation area; the voltage control scheme within the substation area is sent to the energy control node in any substation area; wherein the third agent is trained by using the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism, using the historical load information, historical energy output information, substation area topology information and historical coordinated voltage control scheme of each substation area as input data; Any energy control node in any of the substations is configured to perform voltage regulation based on the voltage regulation scheme in the substation.

2. The voltage control system according to claim 1, characterized in that: Any energy control node in any of the above-mentioned areas is further configured as: During the voltage regulation process based on the voltage regulation scheme in the substation, key parameters are monitored and recorded, the key parameters including: voltage level, power factor, active power and reactive power; Obtaining a control effect evaluation index according to the key parameters, the control effect evaluation index includes: a voltage stability index and an energy efficiency index, and obtaining the control effect evaluation index according to the key parameters includes: determining the voltage stability index according to the voltage level, the voltage stability index includes: voltage fluctuation rate, voltage qualification rate and voltage average value and standard deviation; determining the energy efficiency index according to the power factor, the active power and the reactive power; Compare the control effect evaluation index with the preset target value, analyze whether the control has achieved the expected effect, and obtain the evaluation result; Forming feedback data based on the regulation effect evaluation index, the evaluation result and the key parameter, and sending the feedback data to the voltage regulation unit in any one of the substations; The voltage control unit in any substation in the substation cluster is configured as follows: Adding the feedback data to the training data set of the third agent, and retraining the third agent using the updated training data set; The voltage control scheme within the substation is adjusted based on the retrained third intelligent agent.

3. The voltage control system according to claim 1, characterized in that: The CNN-GRU neural network model based on residual connection includes: an input module, a convolutional neural network CNN module, a gated recurrent unit GRU module and an output module, wherein residual connections are included between internal layers of the CNN module or between internal layers of the GRU module or between the CNN module and the GRU module; The method uses the trained first agent to predict the load information and energy output information of each substation within a preset time in the future based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future, including: Based on the input module, the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are received, and the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are integrated into a multi-dimensional input tensor, wherein each dimension represents different time series data; Extracting spatial features in the multi-dimensional input tensor through multi-layer convolution, or multi-layer convolution and residual connection based on the CNN module to form a feature map; Processing the time series data in the feature graph through a multi-layer recurrent neural network or a multi-layer recurrent neural network and a residual connection based on the GRU module, thereby capturing the time dependency and generating a hidden state vector; The hidden state vector is processed by the output module through a fully connected layer or a regression layer to generate a prediction result, and the prediction result includes: load information and energy output information of each substation within a preset time in the future.

4. The voltage control system according to claim 1, characterized in that: in, The area characteristic information includes: area type, installed capacity, grid connection mode and historical power generation efficiency; the area type includes: photovoltaic power generation, wind power generation or a combination of photovoltaic power generation and wind power generation; When the area type is photovoltaic power generation, the meteorological information includes: temperature and light intensity; When the substation type is wind power generation, the meteorological information includes: wind direction and wind speed; When the substation type is a mixture of photovoltaic power generation and wind power generation, the meteorological information includes: temperature, light intensity, wind direction and wind speed.

5. The voltage control system according to claim 1, characterized in that: The Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism includes: an encoder, an attention module and a decoder, wherein the encoder and the decoder respectively include multiple layers of LSTM units connected in sequence; The method of using the trained second agent to generate a coordinated voltage control scheme for each substation including a multi-point control curve based on the load information and energy output information of each substation within a preset time in the future includes: Arrange the load information and energy output information of each substation within a preset future time into sequence data, wherein the sequence data includes multiple time steps, and each time step corresponds to a specific combination of load information and energy output information; Processing each time step in the sequence data step by step based on the multiple LSTM units in the encoder, including: for each time step, the designated LSTM unit in the encoder generates a hidden state corresponding to the current time step according to the input data of the current time step and the hidden state of the previous time step; After the multiple LSTM units in the encoder have processed each time step in the sequence data, based on the attention module, at each time step of the decoder, the attention weight is calculated according to the hidden state of the last layer of LSTM units in the encoder and the hidden state of all time steps of the encoder, and the hidden state of all time steps of the encoder is weighted and summed using the attention weight to obtain a weighted context vector, wherein the coordinated voltage control scheme of the substation area is characterized by an output sequence containing the preset number of control points, and the time step of the decoder corresponds to the time step relative to one control point in the output sequence; The method comprises: for a current time step of the decoder, the LSTM unit in the decoder generates a control point of the current time step according to the weighted context vector and a control point of a previous time step; After the decoder generates all the preset number of control points, the preset number of control points are combined into a coordinated voltage control scheme for an area including a multi-point control curve.

6. The voltage control system according to claim 1, characterized in that: The LSTM neural network model based on the instant feedback mechanism and the attention mechanism includes: an input module, an LSTM module, an attention mechanism module, an instant feedback module and an output module; the voltage control scheme in the substation area is generated by using the trained third agent according to the substation area coordinated voltage control scheme and the real-time load information of any substation area, including: Preprocessing the coordinated voltage control scheme of the substation and the real-time load information of any substation based on the input module to generate corresponding time series format data; Processing the time series format data based on the LSTM module includes: capturing long-term dependencies in the time series format data, obtaining time series features in the time series format data, and generating a corresponding hidden state vector group according to the long-term dependencies and the time series features; Calculating the attention weight of each hidden state vector in the hidden state vector group based on the attention mechanism module, adding the results of multiplying each hidden state vector by its corresponding attention weight to obtain a weighted hidden state vector; Based on the immediate feedback module, the weighted hidden state vector is compared with the immediate feedback information, an error signal is calculated, and an internal state parameter of the LSTM module or the attention mechanism module is adjusted according to the error signal, so that the LSTM module or the attention mechanism module regenerates the weighted hidden state vector according to the adjusted internal state parameter; The weighted hidden state vector is decoded based on the output module to map the weighted hidden state vector to an output space corresponding to the voltage control scheme within the substation, a voltage control instruction corresponding to each energy control node is generated, and the voltage control scheme within the substation is generated based on the voltage control instructions of each energy control node.

7. The voltage control system according to claim 1, characterized in that: in, When the third agent is trained using the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism, the training weight of the station area topological structure information is gradually reduced according to a preset strategy until the input of the station area topological structure information is finally cancelled.

8. The voltage control system according to claim 1, characterized in that: The voltage control scheme in the substation includes the voltage control frequency and voltage control speed of each energy control node in the substation, and the active and reactive power adjustment sequence, wherein the voltage control speed includes: voltage down-regulation speed and voltage up-regulation speed, the voltage control frequency is related to the preset number of control points included in the multi-point control curve, the preset number of control points is 288 points, the multi-point control curve is a 288-point control curve, the corresponding voltage control frequency is regulated once every 5 minutes, and the voltage control based on the voltage control scheme in the substation includes: Any energy control node in any one of the substations obtains the substation voltage control scheme corresponding to any one of the energy control nodes in the substation voltage control scheme; The current voltage of any energy control node is regulated according to the voltage regulation frequency, voltage regulation speed and active and reactive power regulation sequence in the voltage regulation scheme in the substation of any energy control node, including: according to the voltage regulation frequency, the current voltage is monitored according to the corresponding regulation cycle, and in each regulation cycle, the current voltage data is obtained, and the current voltage data is compared with a preset maximum voltage threshold value, if the current voltage data is greater than the preset maximum voltage threshold value, the current voltage is lowered based on the voltage lowering speed in the voltage regulation speed, and if the current voltage data is less than the preset minimum voltage threshold value, the current voltage is increased based on the voltage increasing speed in the voltage regulation speed; Wherein, when the current voltage of any energy control node is regulated, it is performed in accordance with the active and reactive power regulation sequence.

9. A voltage control method within a substation based on multi-agent, characterized in that: Applied to a voltage control center, the voltage control center is respectively connected to the voltage control units in each substation in the substation cluster, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation. The voltage control method includes: Obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within a preset time in the future based on the historical load information and area characteristic information of each area and the meteorological information within a preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection; According to the load information and energy output information of each substation within a preset future time, a trained second intelligent agent is used to generate a substation coordinated voltage control plan including a multi-point control curve; the substation coordinated voltage control plan is sent to the voltage control unit in each substation; wherein the multi-point control curve contains a preset number of control points, and the control points represent the voltage control targets corresponding to the specified time points within the preset future time. The second intelligent agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism.

10. The voltage control method according to claim 9, characterized in that: The CNN-GRU neural network model based on residual connection includes: an input module, a convolutional neural network CNN module, a gated recurrent unit GRU module and an output module, wherein residual connections are included between internal layers of the CNN module or between internal layers of the GRU module or between the CNN module and the GRU module; The method uses the trained first agent to predict the load information and energy output information of each substation within a preset time in the future based on the historical load information and substation characteristic information of each substation, as well as the meteorological information within a preset time in the future, including: Based on the input module, the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are received, and the historical load information and the characteristic information of each substation area, as well as the meteorological information within a preset time in the future are integrated into a multi-dimensional input tensor, wherein each dimension represents different time series data; Extracting spatial features in the multi-dimensional input tensor through multi-layer convolution, or multi-layer convolution and residual connection based on the CNN module to form a feature map; Processing the time series data in the feature graph through a multi-layer recurrent neural network or a multi-layer recurrent neural network and a residual connection based on the GRU module, thereby capturing the time dependency and generating a hidden state vector; The hidden state vector is processed by the output module through a fully connected layer or a regression layer to generate a prediction result, and the prediction result includes: load information and energy output information of each substation within a preset time in the future.

11. The voltage control method according to claim 9, characterized in that: in, The area characteristic information includes: area type, installed capacity, grid connection mode and historical power generation efficiency; the area type includes: photovoltaic power generation, wind power generation or a combination of photovoltaic power generation and wind power generation; When the area type is photovoltaic power generation, the meteorological information includes: temperature and light intensity; When the substation type is wind power generation, the meteorological information includes: wind direction and wind speed; When the substation type is a mixture of photovoltaic power generation and wind power generation, the meteorological information includes: temperature, light intensity, wind direction and wind speed.

12. The voltage control method according to claim 9, characterized in that: The Seq2Seq neural network model based on the long short-term memory network LSTM and the attention mechanism includes: an encoder, an attention module and a decoder, wherein the encoder and the decoder respectively include multiple layers of LSTM units connected in sequence; The method of using the trained second agent to generate a coordinated voltage control scheme for each substation including a multi-point control curve based on the load information and energy output information of each substation within a preset time in the future includes: Arrange the load information and energy output information of each substation within a preset future time into sequence data, wherein the sequence data includes multiple time steps, and each time step corresponds to a specific combination of load information and energy output information; Processing each time step in the sequence data step by step based on the multiple LSTM units in the encoder, including: for each time step, the designated LSTM unit in the encoder generates a hidden state corresponding to the current time step according to the input data of the current time step and the hidden state of the previous time step; After the multiple LSTM units in the encoder have processed each time step in the sequence data, based on the attention module, at each time step of the decoder, the attention weight is calculated according to the hidden state of the last layer of LSTM units in the encoder and the hidden state of all time steps of the encoder, and the hidden state of all time steps of the encoder is weighted and summed using the attention weight to obtain a weighted context vector, wherein the coordinated voltage control scheme of the substation area is characterized by an output sequence containing the preset number of control points, and the time step of the decoder corresponds to the time step relative to one control point in the output sequence; The method comprises: for a current time step of the decoder, the LSTM unit in the decoder generates a control point of the current time step according to the weighted context vector and a control point of a previous time step; After the decoder generates all the preset number of control points, the preset number of control points are combined into a coordinated voltage control scheme for an area including a multi-point control curve.

13. A voltage control method in a substation based on multi-agent, characterized in that: Applied to a voltage control unit, the voltage control unit is arranged in each substation in the substation cluster, and is connected to the voltage control center in communication, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation in communication, and the voltage control method includes: Receiving the coordinated voltage control plan of the substation area sent by the voltage control center; According to the coordinated voltage control scheme of the substation and the real-time load information of any of the substations, a trained third intelligent agent is used to generate a voltage control scheme within the substation; the voltage control scheme within the substation is sent to the energy control node in any of the substations; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical coordinated voltage control schemes of each substation as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

14. The voltage control method according to claim 13, characterized in that: The voltage control method further includes: Receiving feedback data sent by an energy control node in any of the substations; Adding the feedback data to the training data set of the third agent, and retraining the third agent using the updated training data set; The voltage control scheme within the substation is adjusted based on the retrained third intelligent agent.

15. The voltage control method according to claim 13, characterized in that: The LSTM neural network model based on the instant feedback mechanism and the attention mechanism includes: an input module, an LSTM module, an attention mechanism module, an instant feedback module and an output module; the voltage control scheme in the substation area is generated by using the trained third agent according to the substation area coordinated voltage control scheme and the real-time load information of any substation area, including: Preprocessing the coordinated voltage control scheme of the substation and the real-time load information of any substation based on the input module to generate corresponding time series format data; Processing the time series format data based on the LSTM module includes: capturing long-term dependencies in the time series format data, obtaining time series features in the time series format data, and generating a corresponding hidden state vector group according to the long-term dependencies and the time series features; Calculating the attention weight of each hidden state vector in the hidden state vector group based on the attention mechanism module, adding the results of multiplying each hidden state vector by its corresponding attention weight to obtain a weighted hidden state vector; Based on the immediate feedback module, the weighted hidden state vector is compared with the immediate feedback information, an error signal is calculated, and an internal state parameter of the LSTM module or the attention mechanism module is adjusted according to the error signal, so that the LSTM module or the attention mechanism module regenerates the weighted hidden state vector according to the adjusted internal state parameter; The weighted hidden state vector is decoded based on the output module to map the weighted hidden state vector to an output space corresponding to the voltage control scheme within the substation, a voltage control instruction corresponding to each energy control node is generated, and the voltage control scheme within the substation is generated based on the voltage control instructions of each energy control node.

16. The voltage control method according to claim 13, characterized in that: in, When the third agent is trained using the LSTM neural network model based on the immediate feedback mechanism and the attention mechanism, the training weight of the station area topological structure information is gradually reduced according to a preset strategy until the input of the station area topological structure information is finally cancelled.

17. A voltage control method in a substation based on multi-agent, characterized in that: Applied to an energy control node, the energy control node is arranged in any substation in a substation cluster, and is communicatively connected to a voltage control unit in any substation, and the voltage control units in each substation in the substation cluster are respectively communicatively connected to a voltage control center, and the voltage control method includes: Receiving the voltage control plan within the substation area sent by the voltage control unit; Voltage regulation is performed based on the voltage regulation scheme within the substation.

18. The voltage control method according to claim 17, characterized in that: The voltage control method further includes: During the voltage regulation process based on the voltage regulation scheme in the substation, key parameters are monitored and recorded, the key parameters including: voltage level, power factor, active power and reactive power; Obtaining a control effect evaluation index according to the key parameters, the control effect evaluation index includes: a voltage stability index and an energy efficiency index, and obtaining the control effect evaluation index according to the key parameters includes: determining the voltage stability index according to the voltage level, the voltage stability index includes: voltage fluctuation rate, voltage qualification rate and voltage average value and standard deviation; determining the energy efficiency index according to the power factor, the active power and the reactive power; Compare the control effect evaluation index with the preset target value, analyze whether the control has achieved the expected effect, and obtain the evaluation result; Feedback data is formed based on the control effect evaluation index, the evaluation result and the key parameter, and the feedback data is sent to the voltage control unit in any one of the substations.

19. The voltage control method according to claim 17, characterized in that: The voltage control scheme in the substation area includes the voltage control frequency and voltage control speed of each energy control node in the substation area, and the active and reactive power adjustment sequence, wherein the voltage control speed includes: voltage down-regulation speed and voltage up-regulation speed, and the voltage control based on the voltage control scheme in the substation area includes: Any energy control node in any one of the substations obtains the substation voltage control scheme corresponding to any one of the energy control nodes in the substation voltage control scheme; The current voltage of any energy control node is regulated according to the voltage regulation frequency, voltage regulation speed and active and reactive power regulation sequence in the voltage regulation scheme in the substation of any energy control node, including: according to the voltage regulation frequency, the current voltage is monitored according to the corresponding regulation cycle, and in each regulation cycle, the current voltage data is obtained, and the current voltage data is compared with a preset maximum voltage threshold value, if the current voltage data is greater than the preset maximum voltage threshold value, the current voltage is lowered based on the voltage lowering speed in the voltage regulation speed, and if the current voltage data is less than the preset minimum voltage threshold value, the current voltage is increased based on the voltage increasing speed in the voltage regulation speed; Wherein, when the current voltage of any energy control node is regulated, it is performed in accordance with the active and reactive power regulation sequence.

20. A voltage control device in a substation based on multi-agent, characterized in that: The voltage control center is provided, and the voltage control center is respectively connected to the voltage control units in each substation in the substation cluster, and the voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation. The voltage control device includes: The first agent prediction module is configured to: obtain historical load information and area characteristic information of each area through the voltage control unit in each area in the area cluster; use the trained first agent to predict the load information and energy output information of each area within a preset time in the future according to the historical load information and area characteristic information of each area and the meteorological information within a preset time in the future; wherein the first agent is trained using a CNN-GRU neural network model based on residual connection; The second intelligent agent prediction module is configured as follows: according to the load information and energy output information of each substation within a preset future time, use the trained second intelligent agent to generate a substation coordinated voltage control plan including a multi-point control curve; send the substation coordinated voltage control plan to the voltage control unit in each substation; wherein the multi-point control curve includes a preset number of control points, and the control points represent the voltage control targets at the specified time points within the preset future time. The second intelligent agent uses the historical load information, historical energy output information, reverse heavy overload information and voltage over-limit information of each substation as input data, and is trained using a Seq2Seq neural network model based on a long short-term memory network LSTM and an attention mechanism.

21. A voltage control device in a substation based on multi-agent, characterized in that: The voltage control unit is arranged in each substation in the substation cluster, and is connected to the voltage control center in communication. The voltage control unit in any substation in the substation cluster is connected to the energy control node in any substation. The voltage control device includes: The area coordinated voltage control scheme receiving module is configured to: receive the area coordinated voltage control scheme sent by the voltage control center; The third intelligent agent prediction module is configured to: generate a voltage control plan within the substation using the trained third intelligent agent according to the substation coordinated voltage control plan and the real-time load information of any substation; send the voltage control plan within the substation to the energy control node in any substation; wherein the third intelligent agent uses the historical load information, historical energy output information, substation topology information and historical substation coordinated voltage control plan as input data, and is trained using an LSTM neural network model based on an immediate feedback mechanism and an attention mechanism.

22. A voltage control device in a substation based on multi-agent, characterized in that: The energy control node is arranged in any substation in the substation cluster, and is connected to the voltage control unit in any substation. The voltage control unit in each substation in the substation cluster is connected to the voltage control center. The voltage control device includes: The intra-area voltage control scheme receiving module is configured to: receive the intra-area voltage control scheme sent by the voltage control unit; The execution model of the voltage control scheme in the substation is configured to perform voltage control based on the voltage control scheme in the substation.

23. A chip, characterized in that: It comprises the voltage control device as described in any one of claims 20 to 22; or, it comprises a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the voltage control method as described in any one of claims 9 to 19.

24. An electronic device, characterized in that: It comprises a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the voltage control method according to any one of claims 9 to 19.

25. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the voltage control method according to any one of claims 9 to 19 is implemented.

26. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the voltage control method according to any one of claims 9 to 19 is implemented.

Images