Power grid equipment centralized control method, system and storage medium

Abstract

The present invention discloses a method, system, and storage medium for centralized control of power grid equipment. The method includes the following steps: S1: constructing a power grid situation assessment system based on macro- and micro-indicators of the power grid; S2: collecting expert semantic evaluation information of the macro-indicators of the power grid in a large number of time sections under the power grid situation assessment system, using fuzzy numbers to characterize the expert semantic evaluation information, and clarifying the fuzzy numbers; S3: determining the weights of each indicator and constructing a power grid situation monitoring matrix; S4: constructing a deep neural network model for each unit of the centralized control station, using the power grid situation of the sub-state after the control strategy is implemented obtained from the power grid situation monitoring matrix as a reward to train the deep neural network model; S5: inputting the real-time power grid situation of each unit of the centralized control station into the corresponding trained deep neural network to obtain the optimal control strategy under this state. The above method can realize optimized centralized control of the power grid under the coexistence of qualitative and quantitative indicators, and can make optimal centralized control decisions under different states.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system operation optimization, and in particular relates to a method, system and storage medium for centralized control of power grid equipment. Background Art

[0002] With the continuous development of smart grids, the inertial performance of existing power systems, which primarily rely on transmission motors, is becoming increasingly less significant. The integration of flexible loads and renewable energy into the grid has led to a significant increase in the number of factors that require monitoring and control within centralized control stations. This surge in factors has significantly increased the difficulty of prediction and control within centralized control stations. However, the three fundamental requirements of power system safety, reliability, and economy remain unchanged. Furthermore, to adapt to the complexities of today's times, predictive control requires the development of micro-indicators, including but not limited to substation operation monitoring, substation environmental monitoring, substation gas monitoring, and substation fire safety. Achieving centralized control across these numerous evaluation indicators requires the development of new control technologies to improve the current situation. Centralized control station technology based on reinforcement learning, as a relatively new type of power system control center, plays a crucial role in coordinating the various power supply units in modern power systems. Furthermore, for quantitative indicators, such as internal substation indicators, evaluation systems can provide specific values ​​for iterative learning. However, for qualitative indicators, such as economic efficiency and safety, evaluation systems struggle to provide specific values, and existing control methods are unable to incorporate these qualitative indicators into reinforcement learning. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to propose a method for centralized control of power grid equipment, which quantifies macro qualitative indicators using fuzzy numbers, and constructs a power grid status monitoring matrix for evaluating the power grid status of each unit of the centralized control station with quantitative micro indicators. The qualitative indicators are applied together with the quantitative indicators to the reinforcement learning process of the neural network, thereby realizing optimized centralized control of the power grid under the coexistence of qualitative and quantitative indicators, and making optimal centralized control decisions under different conditions.

[0004] Another object of the present invention is to provide a storage medium storing a computer program instantiating the above-mentioned centralized control method, and a power grid equipment centralized control system capable of implementing the above-mentioned centralized control method.

[0005] Technical solution: The centralized control method for power grid equipment described in the present invention specifically includes the following steps:

[0006] S1: Constructing macro- and micro-indicators of the power grid to form a power grid situation assessment system;

[0007] S2: Collect expert semantic evaluation information of power grid macro indicators in a large number of time sections under the power grid situation assessment system, use fuzzy numbers to characterize the expert semantic evaluation information, and clarify the fuzzy numbers;

[0008] S3: Determine the weight of each indicator and build a power grid situation monitoring matrix;

[0009] S4: Construct a deep neural network model for each unit of the centralized control station, and use the sub-state power grid status after the control strategy is implemented obtained from the power grid status monitoring matrix as a reward to train the deep neural network model;

[0010] S5: Input the real-time power grid status of each unit in the centralized control station into the corresponding trained deep neural network to obtain the optimal control strategy under this state.

[0011] Furthermore, in step S1, the macro indicators include economy, safety and reliability.

[0012] Furthermore, in step S1, the micro-indicators include the SOC of the energy storage unit, the grid-connected power of the photovoltaic unit, the grid-connected power of the wind turbine unit, and the grid-connected power of the traditional motor.

[0013] Furthermore, in step S2, trapezoidal fuzzy numbers are used to characterize the expert semantic evaluation information.

[0014] Furthermore, in step S2, the fuzzy number is clarified using the following formula:

[0015]

[0016] Where, is the brittle number after the trapezoidal fuzzy number is transformed, a, b, c, d are the trapezoidal fuzzy numbers respectively The corresponding four values.

[0017] Furthermore, in step S3, the weight value of each indicator is determined by the BWM method.

[0018] Furthermore, in step S3, a power grid situation monitoring matrix is ​​constructed using the MARCOS method.

[0019] Furthermore, step S3 includes the following steps:

[0020] S3.1: Determine the most important and least important indicators, and generate relationship matrices between the most important indicator and other indicators, and between other indicators and the least important indicator. and

[0021]

[0022]

[0023] in Indicates the importance of the most important indicator relative to the nth indicator among other indicators, Indicates the importance of the nth indicator relative to the least important indicator among other indicators, and Both are represented by trapezoidal fuzzy numbers;

[0024] S3.2: Use the following model to determine the weight of each indicator:

[0025] minξ *

[0026] Need to meet

[0027] Where, ξ * is the objective function, represented by trapezoidal fuzzy numbers, The trapezoidal fuzzy number representing the weight of the j-th indicator, Trapezoidal fuzzy numbers representing the weights of the most important indicators, Trapezoidal fuzzy number representing the weight of the least important indicator, (a Bj ,b Bj ,c Bj ,d Bj )for The jth value in (a jW ,b jW ,c jW ,d jW )for The jth value in .

[0028] Furthermore, step S3 further includes the following steps:

[0029] S3.3: Use the MARCOS method to construct the following power grid situation monitoring matrix:

[0030]

[0031] Among them C i Represents the i-th indicator, A i represents the power grid situation at the i-th moment, AAI represents the set of the worst power grid situations at all moments, i.e., r AAIi is r 1i ~r mi The worst value in , AI represents the best state set in the power grid at all times, i.e. r AIi is r 1i ~r mi The optimal value in ;

[0032] S3.4: Initialize the power grid status monitoring matrix using the following formula to generate the final monitoring matrix:

[0033] If j∈Cost

[0034] If j∈Benefit

[0035] v ij =n ij ×R(w j )

[0036] Among them, Cost represents an indicator where the smaller the better, and Benefit represents an indicator where the larger the better;

[0037] S3.5: Use the following formula to obtain the final monitoring results of the power grid status:

[0038]

[0039] in,

[0040] Furthermore, the step S4 includes:

[0041] S4.1: Determine the state space, action space, and state transition matrix of each unit in the centralized control system, construct a deep neural network, and assign infinite reward parameters to the optimal state among a large number of time slices of the power grid state.

[0042] S4.2: From high to low weights, select the indicators corresponding to the optimal situation whose error is greater than or equal to the set threshold e for optimization control;

[0043] S4.3: Determine exploration or exploitation based on the greedy strategy. If exploration is selected, randomly generate the action change value of the corresponding unit from the action space as the control action; if exploitation is selected, use the main network mainNET to generate the control action of the corresponding unit.

[0044] S4.4: Each optimized control state, control action, sub-state, and reward are stored as samples in the experience replay pool, where the reward is the power grid status of the sub-state;

[0045] S4.5: At every certain step, the parameters of the main network mainNET are passed to the target network targetNET;

[0046] S4.6: Randomly extract a certain number of samples from the experience replay pool, calculate the loss function, and update the neural network parameters of the main network mainNET through gradient descent;

[0047] S4.7: Repeat steps S4.2 to S4.6 until the error between the predicted Q value and the actual Q value between the main network mainNET and the target network targetNET is less than the set threshold.

[0048] The power grid equipment centralized control system described in the present invention is used to implement the above-mentioned power grid centralized control method, including: a power grid situation monitoring module, which is established based on the power grid situation evaluation system formed by the macro-indicators and micro-indicators of the power grid, and is used to evaluate the power grid situation of each unit in the centralized control station; a deep neural network model, which is used to obtain the optimal control strategy for each unit based on the real-time power grid situation of each unit evaluated by the power grid situation monitoring module.

[0049] The storage medium of the present invention stores a computer program, and the computer program is configured to implement the above-mentioned method for centralized control of power grid equipment when running.

[0050] Beneficial effects: Compared with the existing technology, the present invention has the following advantages: macro qualitative indicators can be applied to the reinforcement learning of neural networks, and the obtained deep neural network can realize the optimized centralized control of the power grid under the coexistence of qualitative and quantitative indicators, and make the optimal centralized control decision under different conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a method for centralized control of power grid equipment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0053] Reference Figure 1 According to an embodiment of the present invention, a method for centralized control of power grid equipment includes the following steps:

[0054] Step 1: Construct macro- and micro-indicators of the power grid to form a power grid situation assessment system;

[0055] Step 2: Collect expert semantic evaluation information of power grid macro indicators in a large number of time sections under the power grid situation assessment system, use fuzzy numbers to characterize the expert semantic evaluation information, and clarify the fuzzy numbers;

[0056] Step 3: Determine the weight of each indicator and build a power grid situation monitoring matrix;

[0057] Step 4: Construct a deep neural network model for each unit of the centralized control station, and use the sub-state power grid status after the control strategy is implemented obtained from the power grid status monitoring matrix as a reward to train the deep neural network model;

[0058] Step 5: Input the real-time power grid status of each unit in the centralized control station into the corresponding trained deep neural network to obtain the optimal control strategy under this state.

[0059] By adopting the above method, qualitative macro-indicators are quantified through fuzzy numbers, so that they can be used to construct a power grid situation monitoring matrix with quantitative micro-indicators, realizing the evaluation of the power grid situation under the coexistence of qualitative and quantitative indicators. The quantified power grid situation under a large number of time sections obtained by monitoring and evaluating the power grid situation monitoring matrix is ​​put into the reinforcement learning process of the deep neural network, and the deep neural network obtained by training can realize the optimization and centralized control of the power grid under the coexistence of qualitative and quantitative indicators. The power grid equipment centralized control method described in the present invention can use experts to evaluate qualitative indicators, and the optimal control strategy obtained can make both qualitative and quantitative indicators of the power grid better.

[0060] It is understandable that in step 3, the comprehensive scoring method, comparative analysis method, nominal group method, Delphi method or compromise solution method (MARCOS) can be used to obtain the power grid status under multiple indicators and construct a power grid status monitoring matrix.

[0061] In this embodiment, macro indicators include economy (A1), safety (A2), and reliability (A3), while micro indicators include energy storage unit SOC (A4), photovoltaic unit grid power (A5), wind turbine unit grid power (A6), and traditional motor grid power (A7). Macro indicators can be supplemented with other qualitative indicators based on actual control needs, while micro indicators are formulated based on the grid structure. The semantic descriptions of macro indicators include "very good," "good," "average," "low," and "very low," and are all characterized using trapezoidal fuzzy numerical values, as shown in Table 1.

[0062] Table 1 Fuzzy numerical table of macro indicators

[0063]

[0064] And use the following formula to clarify the trapezoidal fuzzy number:

[0065]

[0066] where a, b, c, and d are trapezoidal fuzzy numbers The corresponding four values ​​are is the brittle number after the trapezoidal fuzzy number is converted.

[0067] In step 3, the best-worst (BWM) method is used to determine the weight of each indicator in the indicator system. First, the most important and least important indicators need to be selected, and the relationship matrix between the most important indicator and other indicators, and other indicators and the least important indicator needs to be established respectively:

[0068]

[0069]

[0070] in Indicates the importance of the most important indicator relative to the nth indicator among other indicators, Indicates the importance of the nth indicator relative to the least important indicator among other indicators. Represents the relationship matrix between the most important indicator and other indicators, The relationship matrix between the worst indicator and other indicators is represented. The importance of the most important indicator relative to other indicators and the least important indicator relative to other indicators are evaluated by experts. They can be evaluated using semantic evaluations such as "very important", "generally important", "equally important", "generally unimportant", and "very unimportant", and are also represented by trapezoidal fuzzy numbers.

[0071] The weight of each indicator is determined by the following model:

[0072] minξ * (4)

[0073] Need to meet

[0074] where ξ * is the objective function of the weighted solution model, which is a set unknown variable and is represented by trapezoidal fuzzy numbers. Trapezoidal fuzzy numbers representing the weights of the most important indicators, The trapezoidal fuzzy number representing the weight of the j-th indicator, Trapezoidal fuzzy number representing the weight of the least important indicator. (a Bj ,b Bj ,c Bj ,d Bj )for The jth quantity in (a jW ,b jW ,c jW ,d jW )for The jth quantity in .

[0075]

[0076] in, The trapezoidal fuzzy number representing the weight of index j, is the brittle number after the trapezoidal fuzzy number is converted. is the weight of each indicator.

[0077] In this embodiment, the MARCOS method is used to construct a power grid situation monitoring matrix:

[0078]

[0079] Among them, C i Represents the i-th indicator, A iRepresents the power grid status at the i-th moment. AAI represents the set of the worst power grid status at all moments, i.e., r AAIi is r 1i ~r mi The worst value in AI is the opposite of AAI, representing the best set of situations in the power grid at all times.

[0080] And initialize it through the following formula to obtain the final monitoring matrix:

[0081] v ij =n ij ×R(w j ) (8)

[0082] in:

[0083]

[0084] Among them, Cost means the smaller the better, that is, the cost indicator; Benefit means the larger the better, that is, the benefit indicator.

[0085] By obtaining the evaluation results of experts on a large number of macro indicators of time sections, and characterizing the evaluation semantics through fuzzy numbers, and finally clarifying them using formula (1), the above monitoring matrix is ​​constructed and initialized with the corresponding micro indicators.

[0086] Finally, the final power grid situation monitoring result is obtained through the following formula:

[0087]

[0088] in,

[0089] Then, the state space, action space and state transfer matrix of each unit in the centralized control station are determined, a deep neural network is constructed, and the sub-state after the action is used as a reward for training the neural network, where the reward parameter is infinite for the optimal situation assessment result.

[0090] The quality of learning in each round is reflected by the reward and punishment value. The cumulative reward in each round is defined as:

[0091]

[0092] Where rt and rt+1 are the reward and punishment values ​​at the current and next moments respectively; the discount rate γ k ∈[0,1], determines the importance of the reward or punishment value of k at the future moment to the present.

[0093] The indicator with the largest weight among energy storage, photovoltaics, wind turbines, and traditional motors is selected in turn to determine whether the error between the indicator and the indicator corresponding to the optimal situation is less than e; if the result is yes, the indicator with the second largest weight is selected, and so on; if the result is no, the indicator is optimized and controlled.

[0094] According to the ε-greedy strategy, the exploration and utilization judgment is made; if the result is exploration, the action of the corresponding indicator is a random value; if it is utilization, the current learning result of the deep neural network mainNET is used to generate the change value of the corresponding unit output as the action, as shown in the following formula:

[0095]

[0096] Where ε is a fixed constant in the interval [0, 1]; β is randomly generated by a computer in the interval [0, 1]. When β < ε, the agent randomly selects an action in the action space; otherwise, it selects the action with the greatest value in the current state.

[0097] Iterates and continuously learns through the Bellman equation, and stores samples such as Reward, State, Action, Nextstate, and terminate after each Action in the Experience replay pool. Every 100 steps, the parameters of the mainNET are passed to the targetNET.

[0098] The learning cycle continues until the error between the predicted Q value and the actual Q value between mainNET and targetNET is small enough to generate a deep final neural network. The loss function of the neural network is defined as:

[0099]

[0100] Where ω is a parameter in the neural network.

[0101] During the training process, a portion of samples are randomly drawn from the experience pool in batches each time, the loss function is calculated, and the parameter ω is updated through stochastic gradient descent to complete the training of the neural network.

[0102] According to an embodiment of the present invention, a centralized control system for power grid equipment can implement the above-mentioned centralized control method for power grid equipment, including a power grid status monitoring module and a deep neural network model. The power grid status monitoring module is established based on a power grid status assessment system formed by macro-indicators and micro-indicators of the power grid, and is used to assess the power grid status of each unit in the centralized control station; the deep neural network model is used to obtain the optimal control strategy for each unit based on the real-time power grid status of each unit assessed by the power grid status monitoring module. According to an embodiment of the present invention, a storage medium stores a computer program that instantiates the above-mentioned centralized control method for power grid equipment.

[0103] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] The present invention is described with reference to flowcharts and / or block diagrams of methods, systems, and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0105] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0106] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art may still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

Claims

1. A method for centralized control of power grid equipment, characterized in that: The steps include: S1: Constructing macro- and micro-indicators of the power grid to form a power grid situation assessment system; S2: Collect expert semantic evaluation information of macro indicators of the power grid in a large number of time sections under the power grid situation assessment system, use fuzzy numbers to characterize the expert semantic evaluation information, and clarify the fuzzy numbers; S3: Determine the weight of each indicator and build a power grid situation monitoring matrix; S4: Build a deep neural network model for each unit of the centralized control station, and use the sub-state power grid status after the control strategy is implemented obtained from the power grid status monitoring matrix as a reward to train the deep neural network model; S4.1: Determine the state space, action space, and state transition matrix of each unit in the centralized control system, construct a deep neural network, and assign infinite reward parameters to the optimal state among a large number of time slices of the power grid state. S4.2: From high to low weights, select the indicators corresponding to the optimal situation whose error is greater than or equal to the set threshold e for optimization control; S4.3: Determine exploration or exploitation based on the greedy strategy. If exploration is selected, randomly generate the action change value of the corresponding unit from the action space as the control action; if exploitation is selected, use the main network mainNET to generate the control action of the corresponding unit. S4.4: Each optimized control state, control action, sub-state, and reward are stored as samples in the experience replay pool, where the reward is the power grid status of the sub-state; S4.5: At every certain step, the parameters of the main network mainNET are passed to the target network targetNET; S4.6: Randomly extract a certain number of samples from the experience replay pool, calculate the loss function, and update the neural network parameters of the main network mainNET through gradient descent; S4.7: Repeat steps S4.2 to S4.6 until the error between the predicted Q value and the actual Q value of the main network mainNET and the target network targetNET is less than the set threshold; S5: Input the real-time power grid status of each unit in the centralized control station into the corresponding trained deep neural network to obtain the optimal control strategy under this state.

2. The centralized control method for power grid equipment according to claim 1, characterized in that: In step S1, the macro indicators include economy, safety and reliability.

3. The centralized control method for power grid equipment according to claim 2, characterized in that: In step S1, the micro-indicators include the SOC of the energy storage unit, the grid-connected power of the photovoltaic unit, the grid-connected power of the wind turbine unit, and the grid-connected power of the traditional motor.

4. The centralized control method for power grid equipment according to claim 1, characterized in that: In step S2, trapezoidal fuzzy numbers are used to characterize the expert semantic evaluation information.

5. The centralized control method for power grid equipment according to claim 4, characterized in that: In step S2, the fuzzy number is clarified using the following formula: Where, is the brittle number after the trapezoidal fuzzy number is transformed, a, b, c, d are the trapezoidal fuzzy numbers respectively The corresponding four values.

6. The method for centralized control of power grid equipment according to claim 1, characterized in that: In step S3, the weight value of each indicator is determined by the BWM method.

7. The centralized control method for power grid equipment according to claim 1, characterized in that: In step S3, a power grid situation monitoring matrix is ​​constructed using the MARCOS method.

8. The centralized control method for power grid equipment according to claim 6, characterized in that: The step S3 comprises the following steps: S3.1: Determine the most important and least important indicators, and generate relationship matrices between the most important indicator and other indicators, and between other indicators and the least important indicator. and : in Indicates the importance of the most important indicator relative to the nth indicator among other indicators, Indicates the importance of the nth indicator relative to the least important indicator among other indicators, and Both are represented by trapezoidal fuzzy numbers; S3.2: Use the following model to determine the weight of each indicator: minξ * Where, ξ * is the objective function, represented by trapezoidal fuzzy numbers, The trapezoidal fuzzy number representing the weight of the j-th indicator, Trapezoidal fuzzy numbers representing the weights of the most important indicators, Trapezoidal fuzzy number representing the weight of the least important indicator, (a Bj ,b Bj ,c Bj ,d Bj )for The jth value in (a jW ,b jW ,c jW ,d jW )for The jth value in .

9. The method for centralized control of power grid equipment according to claim 8, characterized in that: Described step S3 also comprises the following steps: S3.3: Use the MARCOS method to construct the following power grid situation monitoring matrix: Among them C i Represents the i-th indicator, A i represents the power grid situation at the i-th moment, AAI represents the set of the worst power grid situations at all moments, i.e., r AAIi is r 1i ~r mi The worst value in , AI represents the best state set in the power grid at all times, i.e. r AIi is r 1i ~r mi The optimal value in ; S3.4: Initialize the power grid status monitoring matrix using the following formula to generate the final monitoring matrix: If j∈Cost If j∈Benefit v ij =n ij ×R(w j ) Among them, Cost represents an indicator where the smaller the better, and Benefit represents an indicator where the larger the better; S3.5: Use the following formula to obtain the final monitoring results of the power grid status: in, 10. A power grid equipment centralized control system according to the power grid equipment centralized control method according to any one of claims 1 to 9, characterized in that: include: The power grid situation monitoring module is established based on the power grid situation assessment system formed by the macro-indicators and micro-indicators of the power grid, and is used to evaluate the power grid situation of each unit in the centralized control station; A deep neural network model is used to obtain the optimal control strategy for each unit based on the real-time grid status of each unit evaluated by the grid status monitoring module.

11. A storage medium storing a computer program, characterized in that: The computer program is configured to implement the method for centralized control of power grid equipment according to any one of claims 1 to 9 when running.

Images