Data-driven load frequency control method, system, device and medium

Abstract

The application relates to the technical field of automatic control, in particular to a load frequency control method, system, device and medium based on data driving. i The method comprises the following steps: a discrete time model of a first regional power system operation is constructed, a local tracking error is calculated according to a control input parameter; an action network is constructed, an estimated value of an unknown part of the discrete time model is calculated based on the error; an output trigger condition is constructed through a local tracking error change, whether the output trigger condition is met is judged, if the output trigger condition is met, an updated instruction is obtained by processing the estimated value through a sliding mode controller; an input trigger condition is constructed according to a control input parameter change, when the input trigger condition is met, the power system operates according to the updated control input parameter. The application effectively solves the problems of limited communication resources and uncertain system model, compensates for the control performance decline caused by data delay by means of model uncertainty estimation technology, and provides an efficient and reliable solution for load frequency control of a micro-grid system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic control, in particular to a load frequency control method, system, device and medium based on data driving. BACKGROUND

[0002] In the operation of a micro-grid system, load frequency control (LFC) as a key technology aims to maintain the dynamic balance between system load demand and frequency, and plays a crucial role in improving the overall operation performance and stability of the power system. However, due to the high complexity of the topology structure of the power system and the coupling effect of multiple factors in the operating environment, accurate system modeling faces great challenges. Especially in the large-scale micro-grid scenario, the input delay problem is widespread, which further aggravates the difficulty of system control. Therefore, researching a micro-grid load frequency control strategy that can operate under input delay conditions without relying on accurate system models has become an important research topic in the field of power system automation.

[0003] At the same time, considering the scarcity of communication resources in the communication architecture of the micro-grid, the load frequency control strategy needs to take into account the communication efficiency and energy consumption optimization to reduce the number of communications and reduce the communication energy consumption, and to realize the efficient and stable operation of the system. SUMMARY

[0004] To solve the above technical problems, the present application provides a load frequency control method, system, device and medium based on data driving, which effectively solves the problem of limited communication resources and system model uncertainty, and uses model uncertainty estimation technology to compensate for the decline in control performance caused by data delay, providing an efficient and reliable solution for load frequency control of micro-grid systems. The method comprises the following steps:

[0005] Step S1: Constructing a discrete-time model of the power system operation of the i-th region, calculating the local tracking error through the discrete-time model according to the control input parameters received by the current power system, the local tracking error being the deviation between the actual regional control error and the expected regional control error of the system;

[0006] Step S2: Constructing an evaluation network and an action network, iteratively updating the parameters of the action network according to the evaluation network, and calculating the unknown part of the discrete-time model based on the local tracking error through the action network updated by each iteration to obtain an unknown part estimate;

[0007] Step S3: constructing an output trigger condition according to the variation between the local tracking error and the local tracking error at the last output trigger time, and judging whether the current time meets the output trigger condition:

[0008] If not, keeping the unknown part estimation value of the current discrete time model at the last output trigger time and the local tracking error unchanged, and returning to continue executing step S1;

[0009] If yes, processing the unknown part estimation value through the sliding mode controller constructed in advance to obtain an updated control input parameter, and executing step S4;

[0010] Step S4: constructing an input trigger condition according to the variation of the control input parameter at different input times, and judging whether the current time meets the input trigger condition:

[0011] If not, keeping the control input parameter received by the power system in the current area unchanged, and returning to continue executing step S1;

[0012] If yes, sending the updated control input parameter to the power system in the current area, and making the power system in the current area operate according to the updated control input parameter, and returning to continue executing step S1.

[0013] In an embodiment of the present application, in step S1, the method for constructing the discrete time model of the power system in the ith area is as follows:

[0014] The discrete time model of the power system in the ith area is expressed as:

[0015] (1)

[0016] Wherein, T represents a sampling period, h represents an input time delay, 、 、 、 is a system matrix, represents a state vector, k represents a discrete time, represents an actual regional control error at the discrete time k, represents an input time constant, represents a control input parameter, represents a disturbance vector at the discrete time k, In addition, 、 、 、 is defined as

[0017] ,

[0018] , ,

[0019] ,

[0020] Since the internal stability of the system equilibrium point is equivalent to the origin stability, i.e. the perturbation vector , the discrete-time model expressed in equation (1) is converted into an input-delayed system as follows:

[0021] (2)

[0022] For the input-delayed system, if there exists a control input at time k, there exists a bounded input-delayed function such that Substituting it into equation (2), we obtain:

[0023] (3)

[0024] where , are known, are known parts in system matrices , are unknown but bounded parts in system matrices , , Equation (3) is further improved into a discrete-time model containing unknown parts as follows:

[0025] (4)

[0026] where is the unknown part of the model, represents the generator inertia, represents the generator set damping coefficient, represents the synchronous torque coefficient, represents the steam turbine time constant, represents the control time constant, represents the frequency deviation coefficient, j represents the jth regional power system, represents the droop characteristic.

[0027] In one embodiment of the present application, the calculation method of the local tracking error is as follows:

[0028] (5)

[0029] where represents the expected regional control error.

[0030] In one embodiment of the present invention, the method for constructing an evaluation network is as follows:

[0031] Based on the control input parameters received by the power system in the current area and the local tracking error, the instantaneous cost value is calculated by constructing the instantaneous cost function:

[0032] (6)

[0033] The performance index function of the power system operation in the i-th region is constructed according to the instantaneous cost value:

[0034] (7)

[0035] Will Simplified to , convert the performance index function of formula (7) into the Bellman equation:

[0036] (8)

[0037] The Bellman equation of formula (8) is converted into a performance indicator function that is iteratively calculated using a neural network:

[0038] (9)

[0039] in, 、 is a positive constant, and represents the performance indicator function, Indicated by 、 The data composed of represents the activation function, represents the weight of the evaluation network, Represents dimension, represents the number of neurons; represents the performance index value obtained by iterative calculation of the evaluation network at discrete time k, is the discount factor.

[0040] In one embodiment of the present invention, the method for iteratively updating the parameters of the action network according to the evaluation network is as follows:

[0041] The temporal difference error of the evaluation network is defined as:

[0042] (10)

[0043] The loss function of the evaluation network is constructed based on the temporal difference error as follows:

[0044] (11)

[0045] The weight of the evaluation network is adjusted by the following formula to minimize the loss function value of the evaluation network:

[0046] (12)

[0047] wherein, represents the weight of the evaluation network obtained by iterative training at discrete time (k+1), represents the weight of the evaluation network obtained by iterative training at discrete time k, represents the weight of the evaluation network obtained by iterative training at discrete time (k-1), represents the loss function value calculated by the evaluation network at discrete time k; represents the performance index value calculated by the evaluation network at discrete time (k-1), represents the learning rate of the evaluation network.

[0048] In an embodiment of the present application, the training method of the action network is:

[0049] The performance index value function of the power system operation of the i-th region constructed by the evaluation network as the error function of the action network , according to the error function the loss function of the action network is constructed:

[0050] (13)

[0051] The weight of the action network is adjusted by the following formula to minimize the loss function value of the action network:

[0052] (14)

[0053] wherein, represents the weight of the action network obtained by iterative calculation at discrete time (k+1), represents the weight of the action network obtained by iterative calculation at discrete time k, represents the unknown part estimate value of the action network obtained by iterative calculation at discrete time k; is the learning rate of the action network, represents the activation function, represents the dimension, represents the number of neurons; represents the data composed of the control input parameters and the local tracking error at discrete time k, represents the activation function, represents the weights of the evaluation network, represents the number of neurons.

[0054] In an embodiment of the present application, the method of obtaining the updated control input parameter is as follows:

[0055] calculating the unknown part estimation value according to the action network :

[0056] (15)

[0057] constructing a sliding mode controller which processes the unknown part estimation value by the following formula to obtain the updated control input parameter:

[0058] (16)

[0059] wherein, , , , is a sliding mode controller parameter and satisfies , is a sliding mode controller parameter greater than zero, , and satisfies , is a set sliding surface, represents the local tracking error.

[0060] In an embodiment of the present application, the output trigger condition is constructed according to the change between the local tracking error and the local tracking error at the last output trigger time, as follows:

[0061] (17)

[0062] wherein, represents the output trigger time of the power system of the i-th region, represents the input trigger time of the power system of the i-th region, , represents the local tracking error at the discrete time k, represents the local tracking error at the last output trigger time of ; is a constant greater than zero, is a measurement standard trigger threshold.

[0063] In an embodiment of the present application, the input trigger condition is constructed according to the change of the control input parameter at different input times, as follows:

[0064] (18)

[0065] wherein, denotes the input trigger time of the power system of the i-th region, denotes the last input trigger time of the power system of the i-th region, denotes the control input parameter received by the system at time , and is a measurement standard trigger threshold, is a trigger coefficient.

[0066] Based on the same inventive concept, the present application also provides a data-driven load frequency control system for implementing the steps of the data-driven load frequency control method, the data-driven load frequency control system comprising the following modules:

[0067] a model construction module for constructing a discrete-time model of the power system operation of the i-th region, and calculating a local tracking error through the discrete-time model according to the control input parameter received by the current power system, the local tracking error being the deviation between the actual regional control error and the expected regional control error of the system;

[0068] an unknown part calculation module for constructing an evaluation network and an action network, iteratively updating the parameters of the action network according to the evaluation network, and calculating the unknown part of the discrete-time model through the action network after each iterative update based on the local tracking error to obtain an unknown part estimate;

[0069] a first trigger module for constructing an output trigger condition according to the change between the local tracking error and the local tracking error at the last output trigger time, and determining whether the current time meets the output trigger condition: if not, keeping the unknown part estimate and the local tracking error of the last output trigger time of the current discrete-time model unchanged; if yes, processing the unknown part estimate through a pre-constructed sliding mode controller to obtain an updated control input parameter;

[0070] a second trigger module for constructing an input trigger condition according to the change of the control input parameter at different input times, and determining whether the current time meets the input trigger condition: if not, keeping the control input parameter received by the current regional power system unchanged; if yes, sending the updated control input parameter to the current regional power system, and the current regional power system operating according to the updated control input parameter.

[0071] The application further provides an electronic device, comprising a processor, a memory and a bus system, the processor and the memory being connected through the bus system, the memory being used for storing instructions, and the processor being used for executing the instructions stored in the memory to realize the data-driven load frequency control method.

[0072] The application further provides a computer storage medium, which stores a computer software product, the computer software product comprising a plurality of instructions for enabling a computer device to execute the data-driven load frequency control method.

[0073] The above technical solution of the application has the following advantages compared with the prior art.

[0074] The application deeply couples the intelligent decision-making efficiency of reinforcement learning, the rapid convergence characteristics and strong robustness of the fixed-time non-singular terminal sliding mode control algorithm, and the efficient communication management capability realized by the double-channel dynamic event triggering strategy, and builds a collaborative and efficient integrated control system. This innovative integration not only solves the problems of limited communication resources and uncertain system model, but also significantly improves the system's ability to cope with data delay, providing an efficient, reliable and comprehensive solution for load frequency control of microgrid systems, and having high practical value and broad application prospects. BRIEF DESCRIPTION OF DRAWINGS

[0075] In order to make the content of the application more easily understood, the application will be further described in detail below according to specific embodiments of the application and in conjunction with the drawings, in which,

[0076] Figure 1 is a flow chart of a data-driven load frequency control method provided in an embodiment of the application;

[0077] Figure 2 is a dynamic model structure diagram of a power system of the i-th region;

[0078] Figure 3 is a flow chart of a specific implementation of a power system control method of the i-th region;

[0079] Figure 4 is a three-region interconnected microgrid system;

[0080] Figure 5 is the triggering interval of the input and output channels under the condition of a load with a step change, wherein (a) represents the triggering interval of the input channel, and (b) represents the triggering interval of the output channel;

[0081] Figure 6are the dynamic response conditions of the load frequency deviation under different control strategies of the application, control method A and control method B under the condition of applying a step change load;

[0082] Figure 7 are the trigger intervals of the input and output channels under the condition of applying a dynamic change load, wherein (a) represents the trigger interval of the input channel, and (b) represents the trigger interval of the output channel;

[0083] Figure 8 are the dynamic response conditions of the load frequency deviation under different control strategies of the application, control method A and control method B under the condition of applying a dynamic change load;

[0084] Figure 9 is a structure diagram of a data-driven load frequency control system provided in an embodiment of the application;

[0085] The description of the reference signs in the drawings is as follows: 100, model construction module; 200, unknown part calculation module; 300, first trigger module; 400, second trigger module. DETAILED DESCRIPTION

[0086] The application will be further described in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the application and implement it, but the embodiments are not limiting to the application.

[0087] Embodiment one:

[0088] Referring to Figure 1 The application provides a data-driven load frequency control method, which comprises the following steps:

[0089] Step S1: constructing a discrete-time model of power system operation of the i-th region, calculating a local tracking error through the discrete-time model according to a control input parameter received by the current power system, wherein the local tracking error is a deviation between an actual regional control error and an expected regional control error of the system;

[0090] Step S2: constructing an evaluation network and an action network by using a reinforcement learning algorithm, iteratively updating parameters of the action network according to the evaluation network, calculating an unknown part of the discrete-time model through the action network after each iterative update based on the local tracking error, and obtaining an unknown part estimation value;

[0091] Step S3: constructing an output trigger condition according to a change amount between the local tracking error and a local tracking error at a last output trigger time, and determining whether the current time meets the output trigger condition;

[0092] If no, keep the unknown part estimation value and local tracking error at the last output trigger time on the current discrete-time model unchanged, and return to continue to execute step S1;

[0093] If yes, process the unknown part estimation value by the fixed-time nonsingular terminal sliding mode controller constructed in advance to obtain an updated control input parameter, and execute step S4;

[0094] Step S4: Construct an input trigger condition according to the variation of the control input parameter at different input times, and determine whether the input trigger condition is met at the current time:

[0095] If no, keep the control input parameter received by the power system in the current area unchanged, and return to continue to execute step S1;

[0096] If yes, send the updated control input parameter to the power system in the current area, and the power system in the current area operates according to the updated control input parameter, and return to continue to execute step S1.

[0097] From the above technical solutions, it can be known that the present application is based on the fact that there are many factors in the power system that are difficult to accurately model, such as random fluctuations of loads, uncertainties of equipment parameters and the like, and an action network is constructed by using a reinforcement learning algorithm to calculate and obtain an estimation value of the unknown part, which can effectively estimate the unknown part in the continuous interaction process, thereby making up for the deficiency of the model and improving the reliability of control.

[0098] The input trigger condition and the output trigger condition are constructed by using a double-channel dynamic event trigger strategy in steps S3 and S4, the output trigger condition is constructed according to the variation of the local tracking error, unnecessary updating of the control input parameter is avoided, and only when the error variation reaches a certain degree, the updating of the unknown part estimation value and the adjustment of the control input parameter are triggered, the waste of computing resources is reduced, and the control efficiency is improved. At the same time, when the trigger condition is not met, the last estimation value and the error are kept unchanged, system fluctuations caused by frequent adjustment are avoided, and the stability of the system is enhanced. The input trigger condition is constructed according to the variation of the control input parameter at different input times, and the number of sending of the control input parameter is further reduced. Only when the variation of the control input parameter reaches a certain threshold, the updated control input parameter is sent to the power system, the communication burden and the control cost are reduced. This dynamic trigger mechanism can fully utilize the limited communication resources on the premise of ensuring the control effect, and improve the overall performance of the system.

[0099] As shown in Figure 2 , the dynamic model of the power system in the i-th area is constructed as:

[0100] (1)

[0101] wherein, denotes the load frequency deviation, denotes the tie-line power deviation, denotes the generator output power deviation, denotes the load variation, denotes the governor valve position deviation, denotes the load disturbance and control input; denotes the generator inertia, denotes the generator set damping coefficient, denotes the control time constant, denotes the synchronous torque coefficient, denotes the steam turbine time constant, denotes the input time constant, denotes the frequency deviation coefficient, denotes the actual area control error, j denotes the power system of the jth area, denotes the droop characteristic.

[0102] Further, in the actual power system operation scene, due to the difficulty in obtaining the accurate model thereof, and the frequent network delay and limited communication resources, the load frequency control faces great challenges. In view of these problems, the application innovatively designs a fixed-time non-singular terminal sliding mode control algorithm based on reinforcement learning. At the same time, a double-channel dynamic event triggering strategy is developed. Through accurate estimation of model uncertainty, the negative effects caused by data delay are effectively compensated, thereby solving the problems of limited communication resources and system model uncertainty. The specific implementation process is shown in Figure 3 .

[0103] Based on the Euler approximation law, the method for converting the dynamic model of the above power system into a discrete-time model is as follows:

[0104] The discrete-time model of the power system operation of the ith area is represented as:

[0105] (2)

[0106] wherein, T represents a sampling period, h represents an input time delay, 、 、 、 is a system matrix, denotes a state vector, k denotes a discrete time, denotes the actual area control error at the discrete time k, denotes the control input parameter at the discrete time k, denotes the disturbance vector at the discrete time k, , 、 、 、 is defined as:

[0107] ,

[0108] , ,

[0109] ,

[0110] Since the internal stability of the system equilibrium point is equivalent to the origin stability, i.e. the perturbation vector The discrete-time model representation conversion of formula (2) into an input-delay system is as follows:

[0111] (3)

[0112] For the input-delay system, if there is a control input at time k, there is a bounded input-delay function such that Substituting it into formula (3), we get:

[0113] (4)

[0114] where , are the known parts of the system matrix , is the unknown but bounded part of the system matrix , , , formula (4) is further improved into a discrete-time model containing unknown parts as follows:

[0115] (5)

[0116] where is the unknown part of the model.

[0117] In this embodiment, the calculation method of the local tracking error is defined as follows:

[0118] (6)

[0119] where represents the expected area control error, which is usually set to 0.

[0120] Further, an evaluation network is constructed, including:

[0121] S21: Based on the control input parameters received by the power system of the current area and the local tracking error, the instantaneous cost value is calculated by constructing the instantaneous cost function:

[0122] (7)

[0123] S22: According to the instantaneous cost value, the performance index function of the power system operation of the i-th area is constructed:

[0124] (8)

[0125] S23: Simplify to , and convert the performance index function of formula (8) into Bellman equation:

[0126] (9)

[0127] S24: According to the universal approximation property of neural network, the Bellman equation of formula (9) is converted into the performance index function calculated by iterative neural network:

[0128] (10)

[0129] wherein, , is a constant, and denotes the performance index function, denotes the local tracking error, denotes the data composed of , denotes the activation function, denotes the weight of the evaluation network, denotes the dimension, denotes the number of neurons; denotes the performance index value calculated by iterative evaluation network at discrete time k, is the discount factor. According to the method for iterative updating the parameters of the action network by the evaluation network:

[0130] S25: The performance index value function of the power system operation of the i-th area constructed by the evaluation network

[0131] is taken as the error function of the action network, and the loss function of the action network is constructed according to the error function :

[0132] (11)

[0133] S26: Adjust the weight of the action network by using the following formula, aiming to minimize the loss function value of the action network:

[0134] (12)

[0135] wherein, represents the weight of the action network obtained by iterative calculation at discrete time (k+1), represents the weight of the action network obtained by iterative calculation at discrete time k, represents the unknown part estimate value of the action network obtained by iterative calculation at discrete time k; is the learning rate of the action network, represents the activation function, represents the number of neurons; represents data composed of control input parameters and local tracking errors at discrete time k, represents the activation function, represents the weight of the evaluation network, represents the number of neurons.

[0136] Further, the training method of the evaluation network is:

[0137] The time difference error of the evaluation network is defined as:

[0138] (13)

[0139] The loss function of the evaluation network is constructed based on the time difference error, as follows:

[0140] (14)

[0141] Adjust the weight of the evaluation network by using the following formula, aiming to minimize the loss function value of the evaluation network:

[0142] (15)

[0143] wherein, represents the weight of the evaluation network obtained by iterative training at discrete time (k+1), represents the weight of the evaluation network obtained by iterative training at discrete time k, represents the weight of the evaluation network obtained by iterative training at discrete time (k-1), represents the loss function value of the evaluation network calculated at discrete time k; represents the performance index value of the evaluation network obtained by iterative calculation at discrete time (k-1), represents the learning rate of the evaluation network.

[0144] Further, in step S3, an output trigger condition is constructed according to the variation between the local tracking error and the local tracking error at the last output trigger time, as follows:

[0145] (16)

[0146] wherein, represents the output trigger time of the power system of the i-th region, represents the input trigger time of the power system of the i-th region, , represents the local tracking error at discrete time k, represents the local tracking error at the last output trigger time of is a constant greater than zero, is a measurement standard trigger threshold. When the condition of is met, the system can achieve output and input synchronous triggering.

[0147] Further, based on the constructed output trigger condition, it is determined whether the current time meets the output trigger condition:

[0148] If not, the unknown part estimate value and the local tracking error at the last output trigger time of the current discrete time model remain unchanged, and the step S1 is returned to continue to be executed.

[0149] If yes, the unknown part estimate value is processed by the sliding mode controller constructed in advance to obtain the updated control input parameter, and the step S4 is executed. The method of obtaining the updated control input parameter is as follows:

[0150] The unknown part estimate value is calculated according to the action network :

[0151] (17)

[0152] A fixed-time non-singular terminal sliding mode controller is constructed, which processes the unknown part estimate value to obtain the updated control input parameter:

[0153] (18)

[0154] wherein, , , , is a sliding mode controller parameter and satisfies , is a sliding mode controller parameter greater than zero, , , and satisfy , is a set sliding surface.

[0155] In step S4, an input trigger condition is constructed according to the variation of the control input parameter at different input time instants, as follows:

[0156] (19)

[0157] wherein, denotes the input trigger time instant of the power system of the i-th region, denotes the last input trigger time instant of , denotes the control input parameter received by the system at time instant , is a measurement standard trigger threshold, is a trigger coefficient.

[0158] Further, according to the constructed input trigger condition, it is determined whether the current time instant satisfies the input trigger condition:

[0159] If not, the control input parameter received by the power system of the current region is kept unchanged, i.e. , , and the step S1 is returned to continue to be executed;

[0160] If yes, the updated control input parameter is sent to the power system of the current region, and the power system of the current region operates according to the updated control input parameter, and the step S1 is returned to continue to be executed.

[0161] The effectiveness of the method of the present application is verified by using specific experiments, and the power system structure shown in FIG. 1 is taken as an example, and the main parameters of the model are designed as follows: Figure 4

[0162] , , , , , , ,

[0163] , ,

[0164] , ,

[0165] , ,

[0166] .

[0167] The main parameters of the controller are designed as:

[0168] , , , , , , , , , , , , , , , , , , , , .

[0169] In addition, the initial weights of the action network are randomly selected from (0, 1), and the initial weights of the evaluation network are 0. The activation functions of the evaluation network and the action network are selected as and .

[0170] Experiment 1:

[0171] The expected regional control error of the power system of region 1, region 2 and region 3 is set as , and a step change load is applied. In (a) and (b) of Figure 5 , the event triggered moment and the corresponding trigger interval are shown, and the height of each point represents the time difference between the current trigger moment and the last trigger moment. Among a total of 7200 sampling moments, 2819 times are triggered in total, saving 39.15% of energy. In Figure 6 , the load frequency deviation is shown. Under the action of the control method of the present application, it quickly converges to 0, and is compared with the existing data-driven control method A and the model-based control method B, which embodies the advantages of the method in convergence speed and elimination of steady-state error.

[0172] Experiment 2:

[0173] The expected regional control error of the power system of region 1, region 2 and region 3 is set as , and a constantly changing load is applied. The method proposed in the present application is still effective for the changing load. In Figure 7In (a) and (b), the event-triggered instants and the corresponding trigger intervals are shown, where the height of each dot represents the time difference between the current trigger instant and the last trigger instant. Among the total 7200 sampling instants, 3197 triggers are generated, saving 44.40% of energy. In Figure 8 In (a) and (b), the event-triggered instants and the corresponding trigger intervals are shown, where the height of each dot represents the time difference between the current trigger instant and the last trigger instant. Among the total 7200 sampling instants, 3197 triggers are generated, saving 44.40% of energy. In Under the action of the control method of the application, it quickly converges to 0 and is compared with the existing data-driven control method A and model-based control method B, which embodies the advantages of the method in convergence speed and elimination of steady-state error.

[0174] Embodiment Two:

[0175] Based on the same inventive concept as Embodiment One, the application also provides a data-driven load frequency control system for implementing the steps of the data-driven load frequency control method described in Embodiment One. As shown in Figure 9 The data-driven load frequency control system comprises the following modules:

[0176] A model construction module 100 is configured to construct a discrete-time model of the power system operation of the i-th region, and calculate a local tracking error between the actual output and the expected output of the system by the discrete-time model according to the control input parameters received by the current power system;

[0177] An unknown part calculation module 200 is configured to construct an evaluation network and an action network, iteratively update the parameters of the action network according to the evaluation network, and calculate the unknown part of the discrete-time model based on the local tracking error through the action network updated each time to obtain an unknown part estimate value;

[0178] A first trigger module 300 is configured to construct an output trigger condition according to the change between the local tracking error and the local tracking error at the last output trigger time, and determine whether the current time meets the output trigger condition: if not, keep the unknown part estimate value and the local tracking error of the last output trigger time of the current discrete-time model unchanged; if yes, process the unknown part estimate value through a pre-constructed sliding mode controller to obtain an updated control input parameter;

[0179] A second trigger module 400 is configured to construct an input trigger condition according to the change of the control input parameter at different input instants, and determine whether the current time meets the input trigger condition: if not, keep the control input parameter received by the power system of the current region unchanged; if yes, send the updated control input parameter to the power system of the current region, and the power system of the current region operates according to the updated control input parameter.

[0180] The embodiment proposes a data-driven load frequency control system for implementing the aforementioned data-driven load frequency control method, and therefore the specific embodiments of the data-driven load frequency control system can be found in the embodiment part of the aforementioned data-driven load frequency control method, for example, the model construction module 100, the unknown part calculation module 200, the first triggering module 300 and the second triggering module 400, which are respectively used to correspondingly implement steps S1, S2 and S3 in the data-driven load frequency control method in the embodiment one, and therefore the specific embodiments can be referred to the description of the respective part embodiments, and in order to avoid redundancy, will not be repeated here.

[0181] Embodiment three:

[0182] The embodiment also provides an electronic device, which comprises a processor, a memory and a bus system, the processor and the memory are connected through the bus system, the memory is used for storing instructions, and the processor is used for executing the instructions stored in the memory to implement the data-driven load frequency control method in the embodiment one.

[0183] Embodiment four:

[0184] The embodiment also provides a computer storage medium, which stores a computer software product, and the computer software product comprises a plurality of instructions for enabling a computer device to execute the data-driven load frequency control method in the embodiment one.

[0185] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media containing computer usable program codes (including but not limited to disk storage, CD-ROM, optical storage, etc.).

[0186] The present application is described with reference to flowcharts and / or block diagrams according to the method, device (system) and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows 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 the 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 produce a machine that implements the functions described in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocksFigure 1 means for performing the function specified in the block or blocks.

[0187] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 flow or flows and / or blocks Figure 1 means for performing the function specified in the block or blocks.

[0188] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 flow or flows and / or blocks Figure 1 steps of means for performing the function specified in the block or blocks.

[0189] Obviously, the above-described embodiments are only examples and are not intended to limit the present application. Other variations and modifications can be possible based on the above description, which can be made by those skilled in the art without departing from the spirit of the present application. It is not necessary to recite all of the embodiments, and obvious changes or modifications made from the above description are still within the scope of the present application.

Claims

1. A data-driven load frequency control method, characterized in that: The following steps are involved: Step S1: Construct a discrete-time model of the power system operation in the i-th region, and calculate the local tracking error through the discrete-time model based on the control input parameters received by the current power system. The local tracking error is the deviation between the actual regional control error and the expected regional control error of the system; Step S2: constructing an evaluation network and an action network, iteratively updating the parameters of the action network based on the evaluation network, and calculating the unknown part of the discrete-time model through the action network after each iterative update based on the local tracking error to obtain an estimated value of the unknown part; wherein the method for constructing the evaluation network is as follows: Based on the control input parameters received by the power system in the current area and the local tracking error, the instantaneous cost value is calculated by constructing the instantaneous cost function: (6) The performance index function of the power system operation in the i-th region is constructed according to the instantaneous cost value: (7) Will Simplified to , convert the performance index function of formula (7) into the Bellman equation: (8) The Bellman equation of formula (8) is converted into a performance indicator function that is iteratively calculated using a neural network: (9) in, 、 is a positive constant, and represents the performance indicator function, represents the local tracking error, Indicates the control input parameters, Indicated by 、 The data composed of represents the activation function, represents the weight of the evaluation network, Represents the dimension, represents the number of neurons; represents the performance index value obtained by iterative calculation of the evaluation network at discrete time k, is the discount factor; Step S3: constructing an output trigger condition based on the change between the local tracking error and the local tracking error at the last output trigger moment, and judging whether the output trigger condition is satisfied at the current moment: If not, keep the unknown part estimate value and local tracking error of the last output trigger moment of the current discrete-time model unchanged, and return to step S1; If yes, the unknown part estimation value is processed by the sliding mode controller constructed in advance to obtain the updated control input parameter, and step S4 is executed; Step S4: construct an input trigger condition based on the change of the control input parameter at different input moments, and determine whether the input trigger condition is met at the current moment: If not, keep the control input parameters received by the power system in the current area unchanged, and return to step S1; If so, the updated control input parameters are sent to the power system in the current area, and the power system in the current area operates according to the updated control input parameters, and the process returns to step S1 .

2. The data-driven load frequency control method according to claim 1, characterized in that: In step S1, the method for constructing the discrete-time model of the power system operation in the i-th region is as follows: The discrete-time model of the power system operation in the i-th region is expressed as: (1) Where T represents the sampling period, h represents the input time delay, 、 、 、 is the system matrix, represents the state vector at discrete time k, represents the actual area control error at discrete time k, represents the input time constant, Indicates the control input parameters, represents the disturbance vector at discrete time k, ,also, 、 、 、 is defined as: ， ， ， ， Since the internal stability of the system equilibrium point is equivalent to the origin stability, that is, the disturbance vector , the discrete-time model representation described in formula (1) is converted into an input delay system: (2) For the input delay system, if there is a control input at time k , then there exists a bounded input delay function Make , substituting it into formula (2), we get: (3) in, , The system matrix 、 The known part of is the system matrix 、 The unknown but bounded part in , formula (3) is further improved to a discrete time model including the unknown part, as follows: (4) in, is the unknown part of the model, represents the generator inertia, represents the damping coefficient of the generator set, represents the synchronous torque coefficient, represents the turbine time constant, represents the control time constant, represents the frequency deviation coefficient, j represents the power system of the jth region, Indicates droop characteristics.

3. The data-driven load frequency control method according to claim 2, wherein: The local tracking error The calculation method is as follows: (5) in, represents the expected regional control error.

4. The data-driven load frequency control method according to claim 2, characterized in that: The method for iteratively updating the parameters of the action network according to the evaluation network is: The temporal difference error of the evaluation network is defined as: (10) The loss function of the evaluation network is constructed based on the temporal difference error as follows: (11) With the goal of minimizing the loss function value of the evaluation network, the weight of the evaluation network is adjusted using the following formula: (12) in, represents the weight obtained by iterative training of the evaluation network at discrete time (k+1), represents the weight obtained by iterative training of the evaluation network at discrete time k, represents the weight obtained by iterative training of the evaluation network at discrete time (k-1), represents the loss function value calculated by the evaluation network at discrete time k; It represents the performance index value obtained by iterative calculation of the evaluation network at discrete time (k-1), Represents the learning rate of the evaluation network.

5. The data-driven load frequency control method according to claim 4, characterized in that: The training method of the action network is: The performance index value function of the power system operation in the i-th region constructed by the evaluation network As the error function of the action network , according to the error function Construct the loss function of the action network: (13) With the goal of minimizing the loss function value of the action network, the weight of the action network is adjusted using the following formula: (14) in, represents the weight obtained by iterative calculation of the action network at discrete time (k+1), represents the weight obtained by iterative calculation of the action network at discrete time k, Represents the estimated value of the unknown part obtained by iterative calculation of the action network at discrete time k; is the learning rate of the action network, represents the activation function, Represents the dimension, represents the number of neurons; represents the data consisting of the control input parameters and local tracking error at discrete time k, represents the activation function, Represents the weight of the evaluation network.

6. The data-driven load frequency control method according to claim 5, characterized in that: The method to obtain the updated control input parameters is as follows: Calculate the unknown part estimate based on the action network : (15) Construct a sliding mode controller, which processes the unknown part estimate by the following formula , and get the updated control input parameters: (16) in, 、 、 、 are the sliding mode controller parameters and satisfy , is a sliding mode controller parameter greater than zero, , , and meet , is the set sliding surface, represents the local tracking error.

7. The data-driven load frequency control method according to claim 6, characterized in that: The output trigger condition is constructed based on the change between the local tracking error and the local tracking error at the last output trigger time, as follows: (17) in, represents the output triggering time of the power system in the ith region, represents the input triggering time of the power system in the ith region, , represents the local tracking error at discrete time k, express The local tracking error at the last output trigger moment; is a constant greater than zero, is the measurement standard trigger threshold.

8. The data-driven load frequency control method according to claim 7, characterized in that: The input trigger conditions are constructed based on the changes in the control input parameters at different input times, as follows: (18) in, represents the input triggering time of the power system in the ith region, express The last input trigger time of Indicates time The control input parameters received by the system when is the measurement standard trigger threshold, is the trigger coefficient.

9. A data-driven load frequency control system, characterized in that: The steps for implementing the data-driven load frequency control method according to any one of claims 1 to 8, wherein the data-driven load frequency control system comprises the following modules: a model building module, configured to construct a discrete-time model of the power system operation in the i-th region, and calculate a local tracking error using the discrete-time model based on the control input parameters currently received by the power system, wherein the local tracking error is the deviation between the actual regional control error of the system and the expected regional control error; an unknown part calculation module, configured to construct an evaluation network and an action network, iteratively update parameters of the action network according to the evaluation network, and calculate the unknown part of the discrete-time model through the action network after each iterative update based on the local tracking error to obtain an estimated value of the unknown part; a first triggering module, configured to construct an output triggering condition based on a change between the local tracking error and the local tracking error at the last output triggering moment, and determine whether the output triggering condition is satisfied at the current moment; if not, maintain the unknown part estimate and the local tracking error at the last output triggering moment of the current discrete-time model unchanged; If so, the unknown part estimate is processed by a sliding mode controller constructed in advance to obtain an updated control input parameter; The second trigger module is used to construct an input trigger condition based on the change of the control input parameter at different input moments, and determine whether the input trigger condition is met at the current moment: if not, keep the control input parameter received by the power system in the current area unchanged; If so, the updated control input parameters are sent to the power system in the current area, and the power system in the current area operates according to the updated control input parameters.

10. An electronic device, characterized in that: The electronic device includes a processor, a memory and a bus system, the processor and the memory are connected through the bus system, the memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the data-driven load frequency control method described in any one of claims 1 to 8.

11. A computer storage medium, characterized in that The computer storage medium stores a computer software product, and the computer software product includes several instructions for enabling a computer device to execute the data-driven load frequency control method according to any one of claims 1 to 8.

Images