CN-121983987-A - Electric power system damping controller setting method and system based on artificial intelligence

CN121983987ACN 121983987 ACN121983987 ACN 121983987ACN-121983987-A

Abstract

The embodiment of the invention provides an artificial intelligence-based electric power system damping controller setting method and system, which belong to the technical field of damping controller setting, and the method comprises the steps of collecting power grid operation data and energy storage unit operation data, inputting a state vector into a deep reinforcement learning model, outputting parameter adjustment quantity of a damping controller by the deep reinforcement learning model, updating proportional gain and integral gain of the damping controller according to the parameter adjustment quantity, generating a control instruction for adjusting active power output of the energy storage unit through an outer damping adjusting ring and an inner energy compensating ring, calculating a reward value based on electric power system response, and updating the deep reinforcement learning model according to the reward value to realize self-adaptive setting of the damping controller. According to the method, the self-adaptive optimization setting of the damping controller parameters is realized through the deep reinforcement learning model, the self-balancing of the power oscillation inhibition and the energy storage energy is realized by combining the cooperation of the double-layer control loop, and the transient stability and the toughness of the power system are remarkably improved.

Inventors

LI TAO
FENG XIUZHU
LING SHURONG
XIE XUAN
XIONG WEI
SHI SHIYI
GONG XUE
HONG YUNFEI
BAI GANG
ZHOU YE
SHI YI
JIANG YIQIANG
HU QING
XIE JIANG

Assignees

国网四川省电力公司宜宾供电公司

Dates

Publication Date: 20260505
Application Date: 20251210

Claims (10)

1. An artificial intelligence-based power system damping controller setting method is characterized by comprising the following steps: Collecting power grid operation data and energy storage unit operation data, and preprocessing the collected data; Based on the preprocessed data, extracting oscillation characteristics of the power system and constructing a state vector representing the stability of the power grid; Inputting the state vector into a deep reinforcement learning model, outputting parameter adjustment quantity of a damping controller by the deep reinforcement learning model, and updating proportional gain and integral gain of the damping controller according to the parameter adjustment quantity; Based on the updated proportional gain and integral gain, generating a control instruction for adjusting the active power output of the energy storage unit through the outer damping adjusting ring and the inner energy compensating ring; And sending the control instruction to an energy storage converter to regulate the active power output of the energy storage unit, calculating a reward value based on the response of the power system, and updating the deep reinforcement learning model according to the reward value to realize the self-adaptive setting of the damping controller.
2. The artificial intelligence based power system damping controller tuning method of claim 1, wherein the grid operation data includes system frequency, active power and short circuit current information, and the energy storage unit operation data includes energy storage state of charge, charge and discharge power, current signals and voltage signals.
3. The method for tuning an artificial intelligence based damping controller of an electrical power system according to claim 1, wherein the extracting oscillation characteristics of the electrical power system and constructing a state vector characterizing grid stability based on the preprocessed data comprises: Screening system frequency deviation data, active power change rate data and energy storage charge state data from the preprocessed data; performing time sequence consistency check and alignment on the screened data to obtain a continuous time sequence data set; determining a low-frequency oscillation frequency range affecting the stability of the power system; Using frequency domain analysis technology to decompose oscillation components of the continuous time sequence data set, and separating a main oscillation mode and an interference mode in the data; And extracting oscillation characteristic parameters corresponding to the main oscillation mode, and carrying out validity check on the oscillation characteristic parameters to remove invalid modes caused by numerical errors or overfitting.
4. The method for tuning an artificial intelligence based damping controller of an electrical power system according to claim 1, wherein the extracting oscillation characteristics of the electrical power system and constructing a state vector characterizing grid stability based on the preprocessed data, further comprises: Constructing a state vector, wherein the dimension of the state vector comprises a direct characterization parameter of power grid oscillation, a dynamic change parameter of a system, an energy storage synergistic effect parameter and a stability trend parameter; The dynamic adaptability verification is carried out on the state vector under various power grid operation conditions, wherein the various power grid operation conditions comprise normal operation, load fluctuation and new energy output change; Redundant parameters irrelevant to oscillation characteristics and stability states in the state vector are eliminated through correlation analysis.
5. The method for setting the damping controller of the electric power system based on the artificial intelligence according to claim 1, wherein the state vector is input into a deep reinforcement learning model, the deep reinforcement learning model outputs parameter adjustment amounts of the damping controller and comprises a double-network architecture formed by a strategy network and a value evaluation network, the state vector is input into the strategy network, and the strategy network outputs the parameter adjustment amounts of the damping controller and the parameter adjustment amounts comprise proportional gain adjustment amounts and integral gain adjustment amounts.
6. The method for tuning an artificial intelligence based damping controller of an electrical power system according to claim 1, wherein updating the proportional gain and the integral gain of the damping controller according to the parameter adjustment amount comprises: Acquiring a proportional gain parameter and an integral gain parameter of the damping controller in a current control period; adding the proportional gain adjustment quantity and the integral gain adjustment quantity output by the deep reinforcement learning model with corresponding gain parameters of the current control period respectively to obtain a proportional gain parameter value and an integral gain parameter value of the next control period; performing amplitude limiting processing on the updated proportional gain parameter value and integral gain parameter value; and loading the proportional gain parameter value and the integral gain parameter value which are subjected to amplitude limiting processing into the damping controller when the next control period starts.
7. The method for tuning an artificial intelligence based damping controller of an electrical power system of claim 1, wherein generating control commands through an outer damping adjustment loop and an inner energy compensation loop comprises: the outer damping adjusting ring calculates damping control signals for restraining power oscillation according to the real-time system frequency deviation and the updated proportional gain and integral gain; The inner-layer energy compensation loop calculates and obtains a compensation control signal for maintaining the energy balance of the energy storage according to the deviation between the actual state of charge and the expected state of charge of the energy storage unit; And carrying out weighted fusion on the damping control signal and the compensation control signal to obtain a control instruction.
8. The artificial intelligence based power system damping controller tuning method of claim 1, wherein the sending the control command to the energy storage converter to adjust the active power output of the energy storage unit and calculating the prize value based on the power system response comprises: the control method comprises the steps that a control instruction is sent to a control unit of an energy storage converter through an industrial communication protocol, and the control unit adjusts active power output of the energy storage converter in real time according to the control instruction; Collecting response data of the power system after the control instruction acts, wherein the response data at least comprises system frequency deviation, active power change rate and energy storage charge state; and calculating the rewarding value of the current control period according to a preset multi-objective rewarding function based on the response data.
9. The artificial intelligence based power system damping controller tuning method of claim 1, wherein the updating the deep reinforcement learning model according to the bonus value comprises: Forming an experience data by the state vector at the current moment, the action output by the deep reinforcement learning model, the rewarding value obtained after the action is executed and the new state vector at the next moment, and storing the experience data into an experience playback buffer area; Randomly sampling a plurality of pieces of experience data from the experience playback buffer; for each piece of sampled experience data, calculating a time sequence difference error by a value evaluation network in the deep reinforcement learning model, and updating a weight parameter of the value evaluation network through a gradient descent algorithm according to the time sequence difference error; for each piece of sampled experience data, updating weight parameters of a strategy network by a strategy network in the deep reinforcement learning model according to a value evaluation signal provided by the value evaluation network through a strategy gradient algorithm; after the parameter updating of the value evaluation network and the strategy network is completed, the corresponding target network parameters are synchronously updated by adopting a soft updating mode.
10. An artificial intelligence based power system damping controller tuning system, characterized in that the system comprises a control module comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the computer program to implement the artificial intelligence based power system damping controller tuning method according to any one of claims 1-9.

Description

Electric power system damping controller setting method and system based on artificial intelligence Technical Field The invention relates to the technical field of damping controller setting, in particular to an artificial intelligence-based power system damping controller setting method and system. Background The existing energy storage application is steady-state scheduling with multiple focusing and long time scales, the core function is limited to peak clipping and valley filling, and the dynamic adjustment potential (such as enhancing the inertia of a power grid and improving the toughness of a system) is not effectively mined. Aiming at the physical mechanism of improving the transient stability margin of a novel power system by energy storage, the existing research still has the defects, and an effective technical support is difficult to form. In addition, when energy storage participates in power grid stable control through active interaction, the self charge state can fluctuate, and in order to avoid rapid power response capacity loss caused by deep charge and discharge, larger installed capacity is required to be configured, so that energy storage construction cost is obviously increased, and further large-scale application of the energy storage is limited. Disclosure of Invention The invention aims to provide an artificial intelligence-based power system damping controller setting method and system, which are used for suppressing power oscillation and maintaining energy self-balance through artificial intelligence setting of energy storage damping controller parameters, enhancing power system stability and reducing application cost. The invention provides an artificial intelligence-based power system damping controller setting method, which comprises the steps of collecting power grid operation data and energy storage unit operation data, preprocessing the collected data, extracting oscillation characteristics of a power system and constructing a state vector representing power grid stability based on the preprocessed data, inputting the state vector into a deep reinforcement learning model, outputting parameter adjustment quantity of a damping controller by the deep reinforcement learning model, updating proportional gain and integral gain of the damping controller according to the parameter adjustment quantity, generating a control instruction for adjusting active power output of an energy storage unit through an outer damping adjusting ring and an inner energy compensating ring based on the updated proportional gain and integral gain, sending the control instruction to an energy storage converter to adjust active power output of the energy storage unit, calculating a rewarding value based on power system response, and updating the deep reinforcement learning model according to the rewarding value to achieve self-adaptive setting of the damping controller. Optionally, the power grid operation data includes system frequency, active power and short-circuit current information, and the energy storage unit operation data includes energy storage charge state, charge and discharge power, current signals and voltage signals. The method comprises the steps of selecting system frequency deviation data, active power change rate data and energy storage charge state data from the preprocessed data, carrying out time sequence consistency check and alignment on the selected data to obtain a continuous time sequence data set, determining a low-frequency oscillation frequency range affecting the stability of the power system, carrying out oscillation component decomposition on the continuous time sequence data set by using a frequency domain analysis technology, separating a main oscillation mode and an interference mode in the data, extracting oscillation characteristic parameters corresponding to the main oscillation mode, and carrying out validity check on the oscillation characteristic parameters to eliminate invalid modes caused by numerical errors or overfitting. Optionally, the method comprises the steps of extracting oscillation characteristics of a power system based on the preprocessed data and constructing a state vector representing the stability of a power grid, constructing the state vector, wherein the dimension of the state vector comprises a direct power grid oscillation characterization parameter, a system dynamic change parameter, an energy storage synergistic effect parameter and a stability trend parameter, dynamically adaptively verifying the state vector under various power grid operation conditions, wherein the various power grid operation conditions comprise normal operation, load fluctuation and new energy output change, and eliminating redundant parameters irrelevant to the oscillation characteristics and the stability state in the state vector through correlation analysis. Optionally, the state vector is input into a deep reinforcement learning model, the deep reinforcement learning