A kind of auxiliary decision-making method and related device for static safety risk of power grid

Abstract

The present invention belongs to the technical field related to the power system, and discloses an auxiliary decision-making method and related devices for the static safety risk of the power grid; wherein the auxiliary decision-making method includes: obtaining the parameters and operation data of the power grid model, and using the trained intelligent agent to simulate the operation to obtain the auxiliary decision-making results; the intelligent agent training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy according to the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; uses the power grid regulation simulation function to verify the action strategy, calculates the adjusted equipment load rate, and ends the training when all the over-limits are eliminated. The present invention solves the problems existing in the prior art that the optimization problem is difficult to solve, the actual regulation experience is insufficient, and the decision results exceed the stability boundary.

Description

Technical Field

[0001] The present invention belongs to the technical field related to power systems, and in particular relates to an auxiliary decision-making method and related devices for static safety risks of power grids. Background Art

[0002] With the large-scale access of many source and load elements such as new energy power generation, distributed power sources, virtual power plants, electric vehicles, and energy storage to the power grid, the interactive influence of sources, networks, loads, and storage has expanded, the power grid trend has changed, the output and load intermittency and uncertainty have increased, and the risk of transmission line trend overload, bus voltage over-limit and other faults has become increasingly high. In severe cases, the power system may lose stability and cause major power outages. Overall, the difficulty of power grid regulation and operation continues to increase.

[0003] Grid control and auxiliary decision-making is an important means of supporting the safe and stable operation of the power system. Through reasonable dispatch optimization decisions, the power supply reliability, grid economy and power balance capacity of the power system can be improved. To further explain, grid control and auxiliary decision-making mainly includes real-time heavy overload adjustment, balance support and new energy consumption, voltage over-limit adjustment, fault handling, etc.

[0004] At present, the existing auxiliary decision-making methods for power grid regulation can be mainly divided into two categories: mechanism model-driven and data-driven. Among them, the auxiliary decision-making based on mechanism model-driven includes sensitivity analysis, linear programming, safety-constrained economic dispatch, etc. For example, the sensitivity analysis method regards the failure of power system equipment as a network power or parameter change disturbance in the normal operation of the system. According to the Taylor series expansion of the flow calculation, the sensitivity calculation equation is derived. The sensitivity analysis method has high calculation accuracy and speed, but when the optimization objectives and constraints are complex and changeable, it is difficult to obtain the optimal solution, and expert experience is not considered in the decision-making process, so it is difficult to guarantee the auxiliary decision results. In the safe and stable domain; in data-driven auxiliary decision-making, such as deep reinforcement learning, the optimal power system dispatch auxiliary decision-making strategy is obtained by simulating the decision-making process, avoiding the complete modeling of the power system. The reinforcement learning model is applied to power grid regulation auxiliary decision-making. The environment can be regarded as the power system itself, and the intelligent agent is the manager of the power system. Through the interaction between the intelligent agent and the environment, the intelligent agent can continuously learn the optimal decision-making strategy. The data-driven power grid regulation auxiliary decision has good performance when facing nonlinear and uncertainty problems, but it does not fully consider the constraints of the safe operation of the system, and the interpretability of the auxiliary decision is also lacking.

[0005] In summary, the existing mechanism model-driven or data-driven power grid control decision-making assistance methods still have some problems that need to be solved, mainly including:

[0006] 1) As the scale of grid computing grows, optimization problems become more complex and difficult to solve. Explanation: Due to the increasing scale of computing caused by the access of distributed new energy, energy storage, electric vehicles and other low-voltage equipment, the number of combinations of methods has exploded, and it is very difficult to speed up the calculation of existing traditional optimization algorithms. The decision-making objects of grid control have expanded from traditional units to millions of massive distributed resources. The decision-making objects are more multidimensional and the object characteristics are more complex. When optimizing with multiple constraints and multiple objectives, it is extremely difficult to solve the existing traditional optimization algorithms.

[0007] 2) The construction of mechanism-based auxiliary decision-making functions is not yet complete, and the business support capabilities need to be improved; although the existing traditional mechanism-based model methods have achieved certain practical effects in some application scenarios, such as the sensitivity analysis method to achieve active power control and voltage over-limit adjustment, in other scenarios such as fault handling and emergency plan preparation, they still rely heavily on stability regulations, operating procedures and other regulations, and the technical support capabilities of the auxiliary decision-making function still need to be strengthened;

[0008] 3) How to combine the auxiliary decision-making function with the control experience and expert knowledge to ensure that the decision results are completely within the stability boundary requires further in-depth exploration and research; Explanation-wise, auxiliary decisions in scenarios such as fault handling and over-limit adjustment rely heavily on expert experience. When faced with complex problems such as cascading power grid failures, auxiliary decisions generated by existing traditional methods are difficult to obtain the optimal solution. In actual applications, dispatching and operating personnel are still required to make adjustments based on their own experience, and the accuracy and adaptability of decisions need to be improved. Summary of the invention

[0009] The purpose of the present invention is to provide a method and related device for assisting decision-making for static safety risks of power grids, so as to solve one or more of the above-mentioned technical problems. The technical solution disclosed in the present invention is specifically an auxiliary decision-making solution for static safety risks of power grids that integrates mechanism, data and knowledge, which solves the problems existing in the prior art, such as the difficulty in solving optimization problems, insufficient consideration of actual control experience and decision results exceeding the stability boundary.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] In a first aspect, the present invention provides an auxiliary decision-making method for static security risks of a power grid, comprising:

[0012] Obtaining grid model parameters and operation data of the power grid;

[0013] Based on the acquired power grid model parameters and operation data, the trained intelligent agent is used to perform simulation operation to obtain auxiliary decision-making results of the power grid;

[0014] The training steps of the agent include:

[0015] Acquire a training sample data set and a strategy space of a power grid; wherein the training sample data set includes power grid model data and a safety and stability boundary;

[0016] The training sample data set and strategy space of the power grid are input into the intelligent agent built based on the deep reinforcement learning model and trained to obtain a trained intelligent agent; the training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy according to the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; the action strategy is verified by using the power grid regulation simulation function, and the adjusted equipment load rate is calculated. The training ends when all over-limits are eliminated.

[0017] A further improvement of the present invention is that

[0018] In the step of inputting the training sample data set and strategy space of the power grid into the intelligent agent constructed based on the deep reinforcement learning model and training,

[0019] The state space of the agent It is expressed as:

[0020] ;

[0021] In the formula, for t The active output of hydro-thermal nuclear power units at all times; for t The new energy units are always producing power; for t Always balance the active output of the units; for t Load power at all times; for t Branch load rate at any moment; for t The node voltage load rate at the moment;

[0022] The action space of the agent It is expressed as:

[0023] ;

[0024] In the formula, for t Active output adjustment value of hydro-thermal nuclear power units at any given moment; for t Active output adjustment value of pumped storage unit at any moment; for t Active output adjustment value of new energy units at all times; for tAlways balance the active output adjustment value of the unit; for t The DC power adjustment value at the moment; for t The load power adjustment value can be adjusted at any time;

[0025] The agent's reward function It is expressed as:

[0026] ;

[0027] In the formula, , , , They are t Normalized values ​​of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item at the moment; , , , They are the weights of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item.

[0028] A further improvement of the present invention is that

[0029] The agent's reward function In the calculation process,

[0030] The limit-crossing judgment condition is expressed as:

[0031] ;

[0032] In the formula, F is the weighted crossing function; is the ground state crossing subfunction; It is the N-1 scanning limit-crossing subfunction; Scan the out-of-limit subfunction for the fault set; , , They are the weighting coefficients of base state over-limit, N-1 scan over-limit, and fault set scan over-limit respectively;

[0033] Among them, the weighted limit function F Each type of over-limit sub-function in the above is composed of AC line over-limit and the voltage of the computing node exceeds the limit It consists of two items;

[0034] ;

[0035] ;

[0036] In the formula, , are the number of AC lines and the number of computing nodes respectively; , Respectively i The current value of one AC line, i 2 voltage values ​​of the calculation nodes; , Respectively i 1 AC line limit, i Limit of 2 computing nodes; A very small correction.

[0037] A further improvement of the present invention is that

[0038] The agent's reward function In the calculation process,

[0039] t The calculation expression of the line crossing penalty item at time is:

[0040] ;in, ;

[0041] t The calculation expression of the node voltage over-limit penalty term at time is:

[0042] ;

[0043] t The calculation expression of the over-limit penalty item processed by the balancing machine at the moment is:

[0044] ;

[0045] In the formula, To balance the number of units; for t Moment i Active output of a balancing unit; for t Moment i The upper limit of active output of a balancing unit; for t Moment i The lower limit of active output of a balancing unit;

[0046] t The calculation expression of the control experience reward item at the moment is:

[0047] ;

[0048] In the formula, fort Action at all times j Belongs to the original strategy space The number of for t Total number of actions at a time.

[0049] A further improvement of the present invention is that

[0050] The agent's reward function In the calculation process,

[0051] Line over-limit penalty The value range is (-∞,0), and the node voltage over-limit penalty term The value range is (-∞,1), and the balancing machine handles the over-limit penalty item The value range is (-1,0), and the control experience reward item The value range of is [0,1];

[0052] right , , , The normalized expressions are:

[0053] ;

[0054] , , , The value range of is in [0,1].

[0055] A further improvement of the present invention is that

[0056] The steps of obtaining the training sample data set of the power grid include: raw data extraction, data cleaning, data integration, data screening and feature selection; among them,

[0057] The original data extraction includes: obtaining the power grid model file, state estimation result file, and online static safety analysis calculation result file from the control cloud big data platform;

[0058] Data cleaning includes: parsing the power grid model file and state estimation result file according to the Protocol Buffers protocol, parsing the online static safety analysis calculation result file according to the JSON format, and performing timestamp matching verification on the parsed results to exclude the moments of data missing, and obtaining three types of data after data cleaning: power grid model, state estimation, and online static safety analysis calculation result;

[0059] Data integration includes: arranging three types of data based on the cleaned power grid model, state estimation, and online static safety analysis calculation results according to the order of data time to obtain time series data;

[0060] Data screening includes: based on time series data, targeting the static safety risks of the power grid, first screening out the base state heavy overload equipment, N-1 over-limit equipment, and related equipment with a close electrical distance to the heavy overload equipment, and then screening out the model parameters and power flow data of all equipment;

[0061] Feature selection includes: based on the model parameters and flow data obtained by screening, selecting section time, base state over-limit equipment flow value, N-1 over-limit equipment flow value, associated equipment flow value, over-limit equipment stable limit value, and network topology parameters to construct a training sample data set.

[0062] In a second aspect, the present invention provides an auxiliary decision-making system for static security risks of a power grid, comprising:

[0063] A data acquisition module, used to acquire grid model parameters and operation data of the grid;

[0064] The simulation operation module is used to perform simulation operation based on the acquired power grid model parameters and operation data using the trained intelligent agent to obtain auxiliary decision-making results of the power grid;

[0065] The training steps of the agent include:

[0066] Acquire a training sample data set and a strategy space of a power grid; wherein the training sample data set includes power grid model data and a safety and stability boundary;

[0067] The training sample data set and strategy space of the power grid are input into the intelligent agent built based on the deep reinforcement learning model and trained to obtain a trained intelligent agent; the training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy according to the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; the action strategy is verified by using the power grid regulation simulation function, and the adjusted equipment load rate is calculated. The training ends when all over-limits are eliminated.

[0068] A further improvement of the present invention is that

[0069] In the step of inputting the training sample data set and strategy space of the power grid into the intelligent agent constructed based on the deep reinforcement learning model and training,

[0070] The state space of the agent It is expressed as:

[0071] ;

[0072] In the formula, for t The active output of hydro-thermal nuclear power units at all times; for t The new energy units are always producing power; for t Always balance the active output of the units; for t Load power at all times; for t Branch load rate at any moment; for t The node voltage load rate at the moment;

[0073] The action space of the agent It is expressed as:

[0074] ;

[0075] In the formula, for t Active output adjustment value of hydro-thermal nuclear power units at any given moment; for t Active output adjustment value of pumped storage unit at any moment; for t Active output adjustment value of new energy units at all times; for t Always balance the active output adjustment value of the unit; for t The DC power adjustment value at the moment; for t The load power adjustment value can be adjusted at any time;

[0076] The agent's reward function It is expressed as:

[0077] ;

[0078] In the formula, , , , They are t Normalized values ​​of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item at the moment; , , , They are the weights of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item.

[0079] A further improvement of the present invention is that

[0080] The agent's reward function In the calculation process,

[0081] The limit-crossing judgment condition is expressed as:

[0082] ;

[0083] In the formula, F is the weighted crossing function; is the ground state crossing subfunction; It is the N-1 scanning limit-crossing subfunction; Scan the out-of-limit subfunction for the fault set; , , They are the weighting coefficients of base state over-limit, N-1 scan over-limit, and fault set scan over-limit respectively;

[0084] Among them, the weighted limit function F Each type of over-limit sub-function in the above is composed of AC line over-limit and the voltage of the computing node exceeds the limit It consists of two items;

[0085] ;

[0086] ;

[0087] In the formula, , are the number of AC lines and the number of computing nodes respectively; , Respectively i The current value of one AC line, i 2 voltage values ​​of the calculation nodes; , Respectively i 1 AC line limit, i Limit of 2 computing nodes; A very small correction.

[0088] A further improvement of the present invention is that

[0089] The agent's reward function In the calculation process,

[0090] t The calculation expression of the line crossing penalty item at time is:

[0091] ;in, ;

[0092] t The calculation expression of the node voltage over-limit penalty term at time is:

[0093] ;

[0094] t The calculation expression of the over-limit penalty item processed by the balancing machine at the moment is:

[0095] ;

[0096] In the formula, To balance the number of units; for t Moment i Active output of a balancing unit; for t Moment i The upper limit of active output of a balancing unit; for t Moment i The lower limit of active output of a balancing unit;

[0097] t The calculation expression of the control experience reward item at the moment is:

[0098] ;

[0099] In the formula, for t Action at all times j Belongs to the original strategy space The number of for t Total number of actions at a time.

[0100] A further improvement of the present invention is that

[0101] The agent's reward function In the calculation process,

[0102] Line over-limit penalty The value range is (-∞,0), and the node voltage over-limit penalty term The value range is (-∞,1), and the balancing machine handles the over-limit penalty item The value range is (-1,0), and the control experience reward item The value range of is [0,1];

[0103] right , , , The normalized expressions are:

[0104] ;

[0105] , , , The value range of is in [0,1].

[0106] A further improvement of the present invention is that

[0107] The steps of obtaining the training sample data set of the power grid include: raw data extraction, data cleaning, data integration, data screening and feature selection; among them,

[0108] The original data extraction includes: obtaining the power grid model file, state estimation result file, and online static safety analysis calculation result file from the control cloud big data platform;

[0109] Data cleaning includes: parsing the power grid model file and state estimation result file according to the Protocol Buffers protocol, parsing the online static safety analysis calculation result file according to the JSON format, and performing timestamp matching verification on the parsed results to exclude the moments of data missing, and obtaining three types of data after data cleaning: power grid model, state estimation, and online static safety analysis calculation result;

[0110] Data integration includes: arranging three types of data based on the cleaned power grid model, state estimation, and online static safety analysis calculation results according to the order of data time to obtain time series data;

[0111] Data screening includes: based on time series data, targeting the static safety risks of the power grid, first screening out the base state heavy overload equipment, N-1 over-limit equipment, and related equipment with a close electrical distance to the heavy overload equipment, and then screening out the model parameters and power flow data of all equipment;

[0112] Feature selection includes: based on the model parameters and flow data obtained by screening, selecting section time, base state over-limit equipment flow value, N-1 over-limit equipment flow value, associated equipment flow value, over-limit equipment stable limit value, and network topology parameters to construct a training sample data set.

[0113] According to a third aspect of the present invention, there is provided an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an auxiliary decision-making method for static safety risks of a power grid as described in any one of the first aspect of the present invention is implemented.

[0114] According to a fourth aspect of the present invention, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, an auxiliary decision-making method for static safety risks of a power grid as described in any one of the first aspect of the present invention is implemented.

[0115] Compared with the prior art, the present invention has the following beneficial effects:

[0116] The present invention specifically discloses a static safety risk auxiliary decision-making scheme for power grids that integrates mechanisms, data, and knowledge. It is suitable for static safety risk auxiliary decision-making for scenarios such as base state over-limit, N-1 scanning, and fault set scanning of power grid equipment such as AC lines, transformers, and bus voltages, and solves the problems of difficulty in solving optimization problems, insufficient consideration of actual control experience, and decision results exceeding the stability boundary in the prior art. Specifically, in view of the problem that the optimization solution is difficult due to the growth of power grid computing scale, the present invention constructs an intelligent agent model based on a data-driven method, and trains the intelligent agent offline using massive power grid operation model data, so that it has the ability to automatically generate auxiliary decision-making online; especially for large-scale power grids, multiple constraints, and multi-target adjustment scenarios, there is no need to solve complex mathematical equations. According to the network topology relationship learning, strategy retrieval, and safe and stable state judgment capabilities of the intelligent agent offline training, auxiliary decisions are directly generated, avoiding the problem of long time and difficulty in solving traditional methods. In summary, the method of the present invention can adapt to the problems of the continuous growth of the computing scale of new power systems, the explosion of the number of mode combinations, and the difficulty of solving traditional optimization algorithms, and has improved the calculation speed and scope of application of auxiliary decision-making.

[0117] In view of the problem that traditional decision-making support methods are heavily dependent on control experience such as stability regulations and operating procedures, the present invention integrates the paradigm of power system mechanism equations and control experience knowledge in the process of intelligent model training. On the one hand, it constructs an initial strategy space based on traditional power system analysis methods such as sensitivity formulas to ensure that the auxiliary decision meets the basic control rules of the power grid; on the other hand, it constructs a safe and stable domain space to ensure that the generated auxiliary strategy meets the power grid constraints such as stability regulations. By constructing the initial strategy space and the safe and stable domain space, the situation where the auxiliary decision does not meet the control experience in actual power grid control is avoided, which greatly reduces the workload of dispatching and operating personnel to make adjustments based on their own experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0118] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below; obviously, the drawings described below are some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0119] Figure 1 It is a flowchart of an auxiliary decision-making method for static security risks of a power grid in an embodiment of the present invention;

[0120] Figure 2 is a schematic diagram of the training principle in an embodiment of the present invention;

[0121] Figure 3 is a flowchart of real-time heavy overload adjustment auxiliary decision-making in an embodiment of the present invention;

[0122] Figure 4 is a schematic diagram of a data preprocessing process in an embodiment of the present invention;

[0123] Figure 5 It is a schematic diagram of an auxiliary decision-making system for static safety risks of a power grid in an embodiment of the present invention. DETAILED DESCRIPTION

[0124] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention; it is obvious that the described embodiments and technical solutions are only part of the embodiments of the present invention, not all of the embodiments.

[0125] All other embodiments obtained by those of ordinary skill in the art without creative work based on the technical solutions disclosed in the embodiments of the present invention belong to the scope of protection of the present invention. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0126] See also Figure 1 and Figure 2 In an embodiment of the present invention, a method for assisting decision-making for static security risks of a power grid is provided, comprising the following steps:

[0127] Step 1, obtaining grid model parameters and operation data of the grid;

[0128] Step 2: Based on the power grid model parameters and operation data obtained in step 1, the trained intelligent agent is used to perform simulation operation to obtain auxiliary decision results of the power grid;

[0129] The training steps of the agent include:

[0130] Step 1) obtaining a training sample data set and a strategy space of the power grid; wherein the training sample data set includes power grid model data and a safety and stability boundary;

[0131] Step 2) Input the power grid model data, safety and stability boundaries and strategy space into the intelligent agent built based on the deep reinforcement learning model for training to obtain a trained intelligent agent; the training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy based on the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; uses the power grid regulation simulation function to verify the action strategy, calculates the adjusted equipment load rate, and ends the training when all over-limits are eliminated.

[0132] In the technical solution disclosed in the embodiment of the present invention, by integrating the mechanism, data and knowledge of power grid dispatching and control, a set of auxiliary decision-making solutions for real-time heavy overload adjustment of the power grid is designed, which solves the problems existing in the prior art, such as the difficulty in solving optimization problems, insufficient consideration of actual control experience, and decision results exceeding the stability boundary. Further explanatory, the technical solution of the embodiment of the present invention is based on constructing a data sample set based on data such as offline power grid operation mode and power grid model, using data-driven methods to train the intelligent agent model, generating an initial strategy space based on the traditional mechanism model, and combining control knowledge such as stability regulations and expert experience to calculate the safe and stable boundary, ensuring that the auxiliary decision-making results are within the safe and stable domain; the method of the present invention can adapt to the problems of the continuous growth of the calculation scale of new power systems, the explosion of the number of mode combinations, and the difficulty of solving traditional optimization algorithms, and has improved the calculation speed and scope of application of auxiliary decision-making.

[0133] In a preferred embodiment of the present invention, the power grid model data, the safety and stability boundary and the strategy space are input into the step of training the intelligent agent constructed based on the deep reinforcement learning model.

[0134] State Space It is expressed as:

[0135] ;

[0136] In the formula, for t The active output of hydro-thermal nuclear power units at all times; for t The new energy units are always producing power; for t Always balance the active output of the units; for t Load power at all times; for t Branch load rate at any moment; for t The node voltage load rate at the moment;

[0137] Action Space It is expressed as:

[0138] ;

[0139] In the formula, for t Active output adjustment value of hydro-thermal nuclear power units at any given moment; for t Active output adjustment value of pumped storage unit at any moment; for t Active output adjustment value of new energy units at all times; for t Always balance the active output adjustment value of the unit; for t The DC power adjustment value at the moment; for t The load power adjustment value can be adjusted at any time;

[0140] Reward Function It is expressed as:

[0141] ;

[0142] In the formula, , , , They are t Normalized values ​​of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item at the moment; , , , They are the weights of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item.

[0143] In a preferred embodiment of the present invention, the reward function In the calculation process,

[0144] The limit-crossing judgment conditions are:

[0145] ;

[0146] ;

[0147] ;

[0148] In the formula, F is a weighted crossing function, including the ground state crossing subfunction 、N-1 scan limit-crossing subfunction , Fault set scan limit-crossing sub-function , , , They are weighted coefficients for base state over-limit, N-1 scan over-limit, and fault set scan over-limit respectively; each type of over-limit sub-function is composed of AC line over-limit and the voltage of the computing node exceeds the limit Two components, , are the number of AC lines and the number of computing nodes respectively. i 1. i 2 are respectively i 1 AC line, i 2 compute nodes, , Respectively i The current value of one AC line, i The voltage values ​​of the two computing nodes, , Respectively i 1 AC line limit, i Limit of 2 computing nodes; A very small correction.

[0149] In one embodiment of the present invention, t The calculation steps of the line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item at the moment include:

[0150] t The calculation expression of the line crossing penalty item at time is:

[0151] , only when hour;

[0152] t The calculation expression of the node voltage over-limit penalty term at time is:

[0153] ;

[0154] t The calculation expression of the over-limit penalty item processed by the balancing machine at the moment is:

[0155] ;

[0156] In the formula, To balance the number of units; for t Moment i Active output of a balancing unit; for t Moment i The upper limit of active output of a balancing unit; for t Moment i The lower limit of active output of a balancing unit.

[0157] t The calculation expression of the control experience reward item at the moment is:

[0158] ;

[0159] In the formula, for t Action at all times j Belongs to the original strategy space The number of for t Total number of actions at a time.

[0160] In one embodiment of the present invention,

[0161] In the reward function, the penalty term The range of is (-∞,0), The range of is (-∞,1), The value range is (-1,0), reward items The value range of is [0,1], and the following formulas are used to , , , Perform normalization:

[0162] ;

[0163] After normalization, the value range of the penalty term and the reward term are both within [0,1], and the reward function It is expressed as:

[0164] .

[0165] In the technical solution of the embodiment of the present invention, the intelligent agent model is trained based on the data-driven method, the objective function and reward mechanism are designed in combination with the traditional mechanism model, the model parameters are continuously adjusted and the training process is controlled, and after the model training is completed, different auxiliary decision-making scenarios can be applied to verify the effect. In the present invention, the power system mechanism model is integrated, including power grid analysis application functions such as power flow calculation, sensitivity analysis, and power grid risk assessment, and a deterministic power system mechanism model is used to assist in enhancing the training effect of the artificial intelligence model, which improves the adaptability of auxiliary decision-making on the one hand, and improves the interpretability of the intelligent agent on the other hand.

[0166] See also Figure 3In one embodiment of the present invention, a real-time heavy overload auxiliary decision method integrating mechanism, data and knowledge is specifically provided, which mainly includes: training sample data set management, electronic stability limit, power grid operation mode generation, safety stability domain verification, strategy space generation, auxiliary decision calculation, result verification and optimization, result distribution and other steps; wherein,

[0167] The training sample data set management includes: exporting the state estimation history section from the new generation dispatching technology support system, exporting the typical operation mode section, safety control system action strategy, fault set, etc. from the offline system, and integrating them to form a training sample data set set;

[0168] The electronicization of stability regulations includes: calling the new generation of intelligent section limit calculation service, obtaining the electronic stability regulation logic rules, inputting the power grid model data into the intelligent section limit function in the standard format of the new generation of regulations, generating section limits that match the operation mode of the power grid, and providing calculation services for safety and stability boundary calculation and auxiliary decision-making result verification;

[0169] The generation of power grid operation modes includes: using state estimation, dispatcher flow and other functions to adjust the basic operation mode and flow data, simulating and generating various power grid operation modes, expanding the sample library, and then calling the intelligent section limit service to form a safe and stable boundary under different power grid operation modes;

[0170] The strategy space generation includes: establishing a power grid control simulation function test environment, calculating and generating the adjustment objects and the amount to be adjusted for heavy overload equipment adjustment based on the sensitivity analysis function, and forming the original strategy space ; Then, according to the security control system action strategy derived from the offline system, the actions in the original strategy space are prioritized, and the adjustable quantities of different devices are calculated according to the fixed values, which are used to calculate the reward scores in the agent training process;

[0171] The result verification and optimization include: adjusting the grid operation status according to the auxiliary decision output by the intelligent agent, calling the intelligent section limit function again, calculating the grid safety and stability boundary under the new mode, and if the new safety and stability domain constraints are met, the verification is passed. If not, the new grid operation mode needs to be updated to the grid operation mode sample set, and the safety and stability domain is updated at the same time, and fed back to the intelligent agent for re-training.

[0172] The result distribution includes: auxiliary decision-making after result verification, setting the affiliated agency and issuing agency of the auxiliary decision-making according to the associated equipment and execution authority, and issuing the auxiliary decision to the responsible unit to achieve closed-loop control.

[0173] In the technical solution of the embodiment of the present invention, a power grid control knowledge base is constructed, knowledge is extracted from various unstructured or structured file materials, and modeling is performed according to the knowledge representation method. Various types of knowledge such as stability regulations, operating procedures, fault plan strategies, security control system strategies, section definitions and limits, fault sets, etc. are included, which are provided to the intelligent agent for feature selection and training, and the effect is verified using the traditional mechanism model.

[0174] See also Figure 4 In one embodiment of the present invention, the step of data preprocessing specifically includes:

[0175] Before training the agent, the sample data can be preprocessed, including five steps: raw data extraction, data cleaning, data integration, data screening, and feature selection.

[0176] Raw data extraction: Obtain power grid model files, state estimation result files, and online static safety analysis calculation result files from the control cloud big data platform;

[0177] Data cleaning: Parse the power grid model file and state estimation result file according to the Protocol Buffers protocol, parse the online static safety analysis calculation result file according to the JSON (JavaScript Object Notation) format, perform timestamp matching verification on the parsed results, exclude the moments with missing data, and obtain the cleaned power grid model file, state estimation result file, and online static safety analysis calculation result file;

[0178] Data integration: Arrange the samples in chronological order to form time series data;

[0179] Data screening: For the static safety risks of the power grid, first screen out the base state heavy overload equipment, N-1 over-limit equipment, and related equipment with a close electrical distance to the heavy overload equipment, and then screen out the model parameters and power flow data of all the above equipment;

[0180] Feature selection: Select key features such as section time, base state over-limit equipment flow value, N-1 over-limit equipment flow value, associated equipment flow value, over-limit equipment stable limit value, network topology parameters, etc. to construct a sample data feature set.

[0181] As one of the auxiliary decision-making scenarios for power grid regulation, real-time overload adjustment of power grid plays an important role in ensuring the safe and stable operation of power grid. At present, the auxiliary decision-making of active power, reactive power, voltage and other overload or over-limit adjustments of power grid equipment mainly relies on traditional mechanism models, such as sensitivity models, safety constraint unit combination models, safety constraint economic dispatch models, etc. However, there are problems such as inaccurate modeling of power grid equipment, difficulty in solving optimization problems, and decision results that do not meet the regulation experience. Grid control decision support is an important support means to ensure the safe and stable operation of the power system. The present invention integrates the mechanism, data and knowledge of grid dispatching and control to design a set of decision support solutions for real-time heavy overload adjustment of the power grid. It mainly includes five technical links, namely sample library construction, knowledge base construction, mechanism model correction, intelligent agent generation and evaluation verification; specifically, the above scheme constructs a data sample set based on data such as offline grid operation mode and grid model, trains the intelligent agent model using a data-driven method, generates an initial strategy space based on the traditional mechanism model, and calculates the safe and stable boundary by combining control knowledge such as stability regulations and expert experience, ensuring that the decision support results are within the safe and stable domain; the decision support method integrating mechanism, data and knowledge can adapt to the characteristics of large calculation scale and complex optimization problems of new power systems, and can be used in provincial and above control centers to build related functions in the new generation of dispatching technology support system; in addition, the method can also be used in other decision support scenarios such as fault handling, balance support and new energy consumption, and has good versatility. In addition, in view of the problem that the decision results of traditional decision-making support methods exceed the safety and stability boundaries, the present invention integrates application tools such as intelligent section limits and safety constraint scheduling to achieve accurate matching of operating modes under different power grid conditions, ensuring that the generated auxiliary strategies meet power grid constraints such as stability regulations, thereby improving the accuracy and applicability of auxiliary decisions.

[0182] The following are device embodiments of the present invention, which can be used to implement the method embodiments of the present invention. For details not disclosed in the device embodiments, please refer to the method embodiments of the present invention.

[0183] See also Figure 5 In an embodiment of the present invention, an auxiliary decision-making system for static safety risks of a power grid is provided, comprising:

[0184] A data acquisition module, used to acquire grid model parameters and operation data of the grid;

[0185] The simulation operation module is used to perform simulation operation based on the acquired power grid model parameters and operation data using the trained intelligent agent to obtain auxiliary decision-making results of the power grid;

[0186] The training steps of the agent include:

[0187] Acquire a training sample data set and a strategy space of a power grid; wherein the training sample data set includes power grid model data and a safety and stability boundary;

[0188] The training sample data set and strategy space of the power grid are input into the intelligent agent built based on the deep reinforcement learning model and trained to obtain a trained intelligent agent; the training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy according to the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; the action strategy is verified by using the power grid regulation simulation function, and the adjusted equipment load rate is calculated. The training ends when all over-limits are eliminated.

[0189] In one embodiment of the present invention, a computer device is provided, the computer device comprising a processor and a memory, the memory being used to store a computer program, the computer program comprising program instructions, and the processor being used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions in a computer storage medium to implement a corresponding method flow or a corresponding function; the processor described in the embodiment of the present invention can be used to perform operations of an auxiliary decision-making method for static security risks of a power grid.

[0190] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM (Random Access Memory) memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the auxiliary decision-making method for static security risks of the power grid in the above embodiment.

[0191] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program codes.

[0192] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, 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 generate 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 flowchart and / or block diagram. 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.

[0193] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0194] 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 instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0195] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An auxiliary decision-making method for static security risks of power grid, characterized in that: include: Obtaining grid model parameters and operation data of the power grid; Based on the acquired power grid model parameters and operation data, the trained intelligent agent is used to perform simulation operation to obtain auxiliary decision-making results of the power grid; The training steps of the agent include: Acquire a training sample data set and a strategy space of a power grid; wherein the training sample data set includes power grid model data and a safety and stability boundary; The training sample data set and strategy space of the power grid are input into the intelligent agent built based on the deep reinforcement learning model and trained to obtain a trained intelligent agent; the training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy based on the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; the action strategy is verified using the power grid control simulation function, and the adjusted equipment load rate is calculated. The training ends when all over-limits are eliminated; in, In the step of inputting the training sample data set and strategy space of the power grid into the intelligent agent constructed based on the deep reinforcement learning model and training, The state space of the agent It is expressed as: ； In the formula, for t The active output of hydro-thermal nuclear power units at all times; for t The new energy units are always producing power; for t Always balance the active output of the units; for t Load power at all times; for t Branch load rate at any moment; for t The node voltage load rate at the moment; The action space of the agent It is expressed as: ； In the formula, for t Active output adjustment value of hydro-thermal nuclear power units at any given moment; for t Active output adjustment value of pumped storage unit at any moment; for t Active output adjustment value of new energy units at all times; for t Always balance the active output adjustment value of the unit; for t The DC power adjustment value at the moment; for t The load power adjustment value can be adjusted at any time; The agent's reward function It is expressed as: ； In the formula, , , , They are t Normalized values ​​of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item at the moment; , , , They are the weights of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item; The agent's reward function In the calculation process, The limit-crossing judgment condition is expressed as: ； In the formula, F is the weighted crossing function; is the ground state crossing subfunction; It is the N-1 scanning limit-crossing subfunction; Scan the out-of-limit subfunction for the fault set; , , They are the weighting coefficients of base state over-limit, N-1 scan over-limit, and fault set scan over-limit respectively; Among them, the weighted limit function F Each type of over-limit sub-function in the above is composed of AC line over-limit and calculation node voltage exceeds the limit It consists of two items; ； ； In the formula, , are the number of AC lines and the number of computing nodes respectively; , Respectively i The current value of one AC line, i The voltage values ​​of 2 computing nodes; , Respectively i 1 AC line limit, i Limit of 2 computing nodes; A very small correction amount.

2. The auxiliary decision-making method for static security risks of power grid according to claim 1, characterized in that: The agent's reward function In the calculation process, t The calculation expression of the line crossing penalty item at time is: ;in, ; t The calculation expression of the node voltage over-limit penalty term at time is: ； t The calculation expression of the over-limit penalty item processed by the balancing machine at the moment is: ； In the formula, To balance the number of units; for t Moment i Active output of a balancing unit; for t Moment i The upper limit of active output of a balancing unit; for t Moment i The lower limit of active output of a balancing unit; t The calculation expression of the control experience reward item at the moment is: ； In the formula, for t Action at all times j Belongs to the original strategy space The number of for t Total number of actions at a time.

3. The auxiliary decision-making method for static security risks of power grid according to claim 2, characterized in that: The agent's reward function In the calculation process, Line over-limit penalty The value range is (-∞,0), and the node voltage over-limit penalty term The value range is (-∞,1), and the balancing machine handles the over-limit penalty item The value range is (-1,0), and the control experience reward item The value range of is [0,1]; right , , , The normalized expressions are: ； , , , The value range of is in [0,1].

4. The auxiliary decision-making method for static security risks of power grid according to claim 1, characterized in that: The steps of obtaining the training sample data set of the power grid include: raw data extraction, data cleaning, data integration, data screening and feature selection; among them, The original data extraction includes: obtaining the power grid model file, state estimation result file, and online static safety analysis calculation result file from the control cloud big data platform; Data cleaning includes: parsing the power grid model file and state estimation result file according to the Protocol Buffers protocol, parsing the online static safety analysis calculation result file according to the JSON format, and performing timestamp matching verification on the parsed results to exclude the moments of data missing, and obtaining three types of data after data cleaning: power grid model, state estimation, and online static safety analysis calculation result; Data integration includes: arranging three types of data based on the cleaned power grid model, state estimation, and online static safety analysis calculation results according to the order of data time to obtain time series data; Data screening includes: based on time series data, targeting the static safety risks of the power grid, first screening out the base state heavy overload equipment, N-1 over-limit equipment, and related equipment with a close electrical distance to the heavy overload equipment, and then screening out the model parameters and power flow data of all equipment; Feature selection includes: based on the model parameters and flow data obtained by screening, selecting section time, base state over-limit equipment flow value, N-1 over-limit equipment flow value, associated equipment flow value, over-limit equipment stable limit value, and network topology parameters to construct a training sample data set.

5. An auxiliary decision-making system for static safety risks of power grid, characterized in that: include: A data acquisition module, used to acquire grid model parameters and operation data of the grid; The simulation operation module is used to perform simulation operation based on the acquired power grid model parameters and operation data using the trained intelligent agent to obtain auxiliary decision-making results of the power grid; The training steps of the agent include: Acquire a training sample data set and a strategy space of a power grid; wherein the training sample data set includes power grid model data and a safety and stability boundary; The training sample data set and strategy space of the power grid are input into the intelligent agent built based on the deep reinforcement learning model and trained to obtain a trained intelligent agent; the training process includes: initializing the environmental state of the power grid flow simulation environment; the intelligent agent generates an action strategy based on the environmental state and the reward function; the environment executes the action strategy and updates the state according to the action strategy, calculates the reward score based on the reward function and feeds back the state and reward score to the intelligent agent; the action strategy is verified using the power grid control simulation function, and the adjusted equipment load rate is calculated. The training ends when all over-limits are eliminated; in, In the step of inputting the training sample data set and strategy space of the power grid into the intelligent agent constructed based on the deep reinforcement learning model and training, The state space of the agent It is expressed as: ； In the formula, for t The active output of hydro-thermal nuclear power units at all times; for t The new energy units are always producing active power; for t Always balance the active output of the units; for t Load power at all times; for t Branch load rate at any moment; for t The node voltage load rate at the moment; The action space of the agent It is expressed as: ； In the formula, for t Active output adjustment value of hydro-thermal nuclear power units at any given moment; for t Active output adjustment value of pumped storage unit at any moment; for t Active output adjustment value of new energy units at all times; for t Always balance the active output adjustment value of the unit; for t The DC power adjustment value at the moment; for t The load power adjustment value can be adjusted at any time; The agent's reward function It is expressed as: ； In the formula, , , , They are t Normalized values ​​of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item at the moment; , , , They are the weights of line over-limit penalty item, node voltage over-limit penalty item, balancing machine processing over-limit penalty item, and regulation experience reward item; The agent's reward function In the calculation process, The limit-crossing judgment condition is expressed as: ； In the formula, F is the weighted crossing function; is the ground state crossing subfunction; It is the N-1 scanning limit-crossing subfunction; Scan the out-of-limit subfunction for the fault set; , , They are the weighting coefficients of base state over-limit, N-1 scan over-limit, and fault set scan over-limit respectively; Among them, the weighted limit function F Each type of over-limit sub-function in the above is composed of AC line over-limit and calculation node voltage exceeds the limit It consists of two items; ； ； In the formula, , are the number of AC lines and the number of computing nodes respectively; , Respectively i The current value of one AC line, i 2 voltage values ​​of the calculation nodes; , Respectively i 1 AC line limit, i Limit of 2 computing nodes; A very small correction.

6. The auxiliary decision-making system for static safety risks of power grid according to claim 5, characterized in that: The agent's reward function In the calculation process, t The calculation expression of the line crossing penalty item at time is: ;in, ; t The calculation expression of the node voltage over-limit penalty term at time is: ； t The calculation expression of the over-limit penalty item processed by the balancing machine at the moment is: ； In the formula, To balance the number of units; for t Moment i Active output of a balancing unit; for t Moment i The upper limit of active output of a balancing unit; for t Moment i The lower limit of active output of a balancing unit; t The calculation expression of the control experience reward item at the moment is: ； In the formula, for t Action at all times j Belongs to the original strategy space The number of for t Total number of actions at a time.

7. The auxiliary decision-making system for static safety risks of power grid according to claim 6, characterized in that: The agent's reward function In the calculation process, Line over-limit penalty The value range is (-∞,0), and the node voltage over-limit penalty term The value range is (-∞,1), and the balancing machine handles the over-limit penalty item The value range is (-1,0), and the control experience reward item The value range of is [0,1]; right , , , The normalized expressions are: ； , , , The value range of is in [0,1].

8. The auxiliary decision-making system for static safety risks of power grid according to claim 5, characterized in that: The steps of obtaining the training sample data set of the power grid include: raw data extraction, data cleaning, data integration, data screening and feature selection; among them, The original data extraction includes: obtaining the power grid model file, state estimation result file, and online static safety analysis calculation result file from the control cloud big data platform; Data cleaning includes: parsing the power grid model file and state estimation result file according to the Protocol Buffers protocol, parsing the online static safety analysis calculation result file according to the JSON format, and performing timestamp matching verification on the parsed results to exclude the moments of data missing, and obtaining three types of data after data cleaning: power grid model, state estimation, and online static safety analysis calculation result; Data integration includes: arranging three types of data based on the cleaned power grid model, state estimation, and online static safety analysis calculation results according to the order of data time to obtain time series data; Data screening includes: based on time series data, targeting the static safety risks of the power grid, first screening out the base state heavy overload equipment, N-1 over-limit equipment, and related equipment with a close electrical distance to the heavy overload equipment, and then screening out the model parameters and power flow data of all equipment; Feature selection includes: based on the model parameters and flow data obtained by screening, selecting section time, base state over-limit equipment flow value, N-1 over-limit equipment flow value, associated equipment flow value, over-limit equipment stable limit value, and network topology parameters to construct a training sample data set.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, an auxiliary decision-making method for static security risks of a power grid as described in any one of claims 1 to 4 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the auxiliary decision-making method for static security risks of a power grid as claimed in any one of claims 1 to 4 is implemented.

Images