Power grid impedance regulation system based on adaptive event triggering and hybrid deep learning

Abstract

The application discloses a power grid impedance adjusting system based on adaptive event triggering and hybrid deep learning, and specifically comprises an adaptive event triggering impedance estimation module, an impedance optimization module based on HDRL, a real-time feedback adjusting module, and an integrated mechanism of adaptive event triggering and HDRL, wherein linkage control is realized through data interaction and signal feedback among the adaptive event triggering impedance estimation module, the impedance optimization module based on HDRL, and the real-time feedback adjusting module; when the fluctuation of PCC voltage or current exceeds a dynamic threshold, the adaptive event triggering module activates impedance estimation, and takes the estimation result as one of the input states of HDRL; HDRL determines the optimal impedance control action according to the current PCC state, and transmits the same to the feedback adjusting module; through the joint design of the adaptive event triggering mechanism, the HDRL optimization strategy and the feedback control mechanism, the application realizes real-time optimization and intelligent control of the power grid impedance, and improves the stability and power quality of the power grid.

Description

Technical Field

[0001] This invention relates to the field of power system and power quality management technology, specifically a power grid impedance regulation system based on adaptive event triggering and hybrid deep learning. Background Technology

[0002] With the rapid development of new energy technologies, the integration of distributed energy sources, such as solar and wind power, into the power grid has become a global trend. This integration not only changes the traditional operation of the power grid but also brings a series of power quality problems, such as voltage fluctuations, power fluctuations, and harmonic pollution. To address these issues, power grid control and optimization technologies are constantly evolving. In the field of power grid impedance regulation, researchers have proposed various methods to improve grid stability and power quality. Traditional power grid impedance control methods are generally divided into two categories: passive control methods and active control methods. Passive control methods achieve grid stability and impedance control through the adjustment of physical components, exhibiting good stability but lower accuracy and inability to adapt to dynamically changing grid environments. Active control methods use power electronic devices to estimate and adjust grid impedance in real time, accurately responding to grid changes. However, they often require complex calculations and frequent signal disturbances, which not only increases the computational burden but may also lead to power quality problems. Chinese patent CN117630487B proposes a virtual impedance control method based on distributed power sources. This method is used to improve the dynamic characteristics and voltage quality of distributed power sources by adjusting virtual impedance parameters, thereby optimizing the power flow of the power grid.

[0003] While current adaptive event-triggered mechanisms have certain advantages, such as reducing unnecessary computational burden, their convergence speed is slow and their response to rapidly changing loads is not timely enough. With the introduction of artificial intelligence technology, many studies have begun to try to combine AI technology to solve complex problems in power grid control, especially by using deep learning and reinforcement learning to optimize the estimation and regulation strategies of power grid impedance. However, existing technologies still have shortcomings in terms of real-time performance, adaptability, and control accuracy. Summary of the Invention

[0004] This invention provides a power grid impedance regulation system based on adaptive event triggering and hybrid deep learning. It can effectively solve the problems of poor real-time performance and adaptability of traditional methods mentioned in the background, which are difficult to effectively cope with rapidly changing loads. Most impedance estimation methods rely on frequent calculations or the injection of interference signals, which leads to excessive computational burden and may cause power quality problems. In addition, traditional control methods lack intelligence and are difficult to adapt to complex dynamic load environments, resulting in system response lag and limited contribution to power grid stability. Therefore, the existing technology has obvious shortcomings in impedance control efficiency and accuracy under dynamic environments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a power grid impedance regulation system based on adaptive event triggering and hybrid deep learning, which realizes real-time adjustment of power grid impedance based on adaptive event-driven impedance estimation, intelligent control optimization and dynamic feedback regulation, thereby reducing harmonic distortion, mitigating voltage fluctuations and improving power quality;

[0006] Specifically, it includes an adaptive event-triggered impedance estimation module, an HDRL-based impedance optimization module, a real-time feedback adjustment module, and an integration mechanism between adaptive event triggering and HDRL.

[0007] The adaptive event-triggered impedance estimation module, the HDRL-based impedance optimization module, and the real-time feedback adjustment module achieve linkage control through data interaction and signal feedback. When the fluctuation of PCC voltage or current exceeds the dynamic threshold, the adaptive event-triggered module activates impedance estimation and uses the estimation result as one of the input states of HDRL. HDRL determines the optimal impedance control action based on the current PCC state and transmits it to the feedback adjustment module.

[0008] The real-time feedback adjustment module further adjusts and optimizes the grid impedance through a weighted combination of PID control signals and HDRL control signals. This full-link linkage control structure can ensure the flexibility and stability of the control strategy in scenarios with severe load fluctuations, reduce the transmission effect of load disturbances, and significantly improve power quality and grid stability.

[0009] According to the above technical solution, the adaptive event-triggered impedance estimation module dynamically monitors the voltage and current fluctuations at the common coupling point (PCC), and uses a dynamic threshold adjustment mechanism and a Kalman filter-based estimation method combined with recursive least squares (RLS) to achieve real-time estimation of the grid impedance.

[0010] According to the above technical solution, the impedance optimization module based on HDRL is based on the hybrid deep reinforcement learning HDRL algorithm, which divides the impedance control into a two-layer control structure of discrete actions and continuous actions to cope with the complex load fluctuations and impedance changes in the power grid.

[0011] According to the above technical solution, the real-time feedback adjustment module, under the synergistic effect of the HDRL control signal and the PID control signal, is responsible for quickly correcting and adjusting the voltage and current of the power grid. The PID controller is responsible for handling small load fluctuations, while the HDRL compensation module is responsible for quickly correcting the system's control output when the power grid experiences large load changes, preventing the voltage or current from exceeding the threshold range.

[0012] According to the above technical solution, the adaptive event-triggered impedance estimation module is responsible for real-time monitoring of voltage and current fluctuations at the power grid's point of common coupling (PCC), and automatically triggers the impedance estimation process based on these fluctuations. The core objective of this module is to reduce the computational burden, respond quickly when the power grid experiences significant light load fluctuations, and improve the stability and power quality of the power grid. It adopts a dynamic threshold adjustment mechanism instead of the traditional fixed threshold mechanism, which can dynamically adjust the event triggering threshold according to real-time voltage and current fluctuations, thereby avoiding over-computation and ensuring that the impedance estimation process is accurately triggered when needed.

[0013] According to the above technical solution, the dynamic threshold adjustment mechanism is based on the voltage and current fluctuation amplitude. It effectively separates the power grid signal and noise through an adaptive filter, automatically identifies effective power grid disturbance signals, and when the power grid fluctuation is small or the background noise is large, the impedance estimation will not be frequently triggered, reducing the computational load of the system and improving the system's response efficiency. The impedance estimation process will only be triggered when the power grid fluctuation reaches a certain level, ensuring the stability of the system.

[0014] To further improve the effect of dynamic threshold adjustment, a sliding window algorithm is proposed. This algorithm dynamically adjusts the threshold by calculating the maximum and minimum voltage values ​​within a set time window and updates the threshold base at each time node. It can effectively cope with common random fluctuations in the power grid and has a high anti-interference capability for light load fluctuations.

[0015] Specifically, assuming within time window T win Within the range, the maximum and minimum values ​​of the voltage signal V(t) are Vmax and Vmin, respectively. max (t) and V min (t), then the sliding window algorithm can calculate the current fluctuation base value V using the following formula. base (t): Based on this fluctuation base value, the system can dynamically adjust the event triggering threshold to ensure that impedance estimation can be triggered in a timely manner when there are large changes in the power grid.

[0016] When the event triggering condition is met, the system will begin the impedance estimation process. To achieve accurate real-time estimation, two algorithms are combined: Recursive Least Squares (RLS) and Kalman filter. The RLS algorithm can quickly update the impedance estimate, while the Kalman filter can effectively suppress noise in the power grid environment, improving the accuracy of the estimation results. Specifically, the mathematical formula for the impedance estimation process is as follows:

[0017] Z est (t)=Z est (t-1)+K(t)[V(t)-I(t)Z est (t-1)];

[0018] Among them, Z est Let V(t) be the impedance estimate at the current time t, V(t) be the current voltage value, I(t) be the current current value, and K(t) be the adaptive gain, dynamically adjusted based on the inverse prediction values ​​of voltage, current, and impedance in real time. est (t-1) represents the impedance estimate from the previous moment, and the gain K(t) is calculated based on the voltage V(t), current I(t), and impedance Z at the current moment. est The prediction error of (t-1) is dynamically calculated, and the adaptive gain can optimize the estimation results based on real-time fluctuations and background noise.

[0019] The introduction of the Kalman filter enables the impedance estimation process to maintain high accuracy even in noisy environments. The specific Kalman gain update formula is as follows:

[0020] K(t) = P(t-1)[P(t-1) + R(t)] -1

[0021] P(t)=[P(t-1) -1 +Q(t)] -1 ;

[0022] Where P(t) is the estimation error covariance matrix, Q(t) is the process noise covariance, and R(t) is the measurement noise covariance. By combining RLS with a Kalman filter, the system can accurately estimate the grid impedance in a noisy environment and has a high real-time response capability.

[0023] According to the above technical solution, the impedance optimization module based on HDRL is the core intelligent control unit, which aims to adjust the grid impedance in real time, thereby optimizing power quality and improving grid stability. The module adopts a hybrid deep reinforcement learning algorithm, which combines discrete and continuous action spaces to ensure that the grid can quickly and accurately perform impedance regulation optimization when the load fluctuates and the grid state changes.

[0024] The HDRL algorithm selects discrete actions through a policy network and refines and optimizes these discrete actions through a value network to achieve the best impedance regulation effect. It can adapt to frequent load changes in the power grid, while avoiding frequent calculation operations, thus improving real-time response capability and overall stability.

[0025] According to the above technical solution, the impedance optimization module based on HDRL needs to first design the state space of HDRL. The design of the state space involves the system's real-time perception and feedback mechanism of the power grid. The system's input vector mainly includes the voltage and current of the common coupling point (PCC) in the power grid, as well as the estimated power grid impedance value. Through the voltage and current data collected by real-time sensors, the system can monitor the dynamic changes of the power grid in real time and adjust the power grid impedance according to the changes. The design of the state space can be expressed by the following formula: s t =[V PCC ,I PCC Z est This mainly includes the voltage and current of the common coupling point (PCC) in the power grid, as well as the estimated power grid impedance value. Through this state-space design, the system can quickly respond to changes in the power grid load and adjust the impedance in a timely manner to ensure the stability of the power grid.

[0026] The action space design of HDRL is one of the key aspects of this module. The action space is divided into two parts: discrete actions and continuous actions. The combination of the two can flexibly respond to various fluctuations in the power grid. Discrete actions are used for large-scale impedance adjustment and are suitable for situations where the power grid load fluctuates drastically or changes suddenly. Each discrete action represents a large adjustment step size for impedance and is usually used to quickly respond to significant fluctuations. Continuous actions are used for fine adjustment to ensure that the power grid can still operate smoothly when the load changes slightly. Continuous actions provide fine-grained control of impedance adjustment, avoid frequent large adjustments, maintain the stability of the power grid, and avoid the negative impact of excessive adjustments.

[0027] The action space can be defined by the following formula: α(t)={α dis (t),α con (t)};

[0028] Where, α dis (t) represents discrete actions, α con (t) represents continuous actions;

[0029] To enable the HDRL model to learn the optimal impedance regulation strategy, a reward function is designed. The goal of the reward function is to minimize voltage and current fluctuations in the power grid, thereby improving power quality and grid stability. The specific formula is as follows:

[0030] R(s,a)=-α|V target -V PCC |-β|I target -I PCC |;

[0031] Here, α and β are the stability weight coefficients of voltage and current, respectively, which can be adjusted according to actual needs to ensure power quality and grid stability. The learning process of HDRL adopts a hierarchical neural network structure, including a policy network and a value network. The policy network is responsible for selecting appropriate discrete actions from the current state space, which is suitable for dealing with large-scale load changes. The value network further refines the discrete actions selected by the policy network and optimizes the accuracy of impedance adjustment. This refinement process enables the system to make fine adjustments through continuous actions when there are small load changes, ensuring the stable operation of the grid.

[0032] After each adjustment, the system calculates the reward and updates the network parameters based on the reward signal to optimize the impedance regulation strategy. Through the reward feedback mechanism, the HDRL model continuously learns and adjusts its strategy, enabling the power grid to maintain efficient and stable operation in dynamic load environments.

[0033] During training, the Q-learning algorithm of reinforcement learning is used for parameter updates, and the update formula is as follows:

[0034] Q(s t ,α t )=Q(s t ,α t )+α[R(t)+γmaxQ(s t+1 ,α')-Q(s t ,α t )];

[0035] Where Q(s) t ,α t ) represents state s t Take action α t The Q-value, where α is the learning rate, γ is the discount factor, and maxQ(s) t+1 α') represents the maximum Q value in the next state. This learning algorithm enables the HDRL model to learn the optimal impedance regulation strategy step by step, thereby adaptively adjusting the grid impedance in practical applications.

[0036] According to the above technical solution, the real-time feedback adjustment module is responsible for making fine dynamic corrections to impedance adjustment during power grid operation to cope with rapidly changing load environment and complex power grid fluctuations. It adopts a control strategy that combines a classic PID controller with a compensation term based on hybrid deep reinforcement learning (HDRL), which combines the fast response capability of traditional control methods with the optimization capability of intelligent control methods, significantly improving the real-time performance, robustness and adaptability of the system.

[0037] The real-time feedback regulation module consists of two key parts: a PID controller and an HDRL compensation term. The PID controller is used to initially regulate the grid impedance, respond to rapid load fluctuations, and perform basic dynamic compensation. Through simple and efficient proportional, integral, and derivative control, the PID controller ensures that the system can quickly stabilize near the target value.

[0038] The HDRL compensation term, based on the PID controller, provides more precise regulation capabilities to the system by learning complex nonlinear dynamic characteristics. This compensation term is generated in real-time based on the HDRL algorithm and can correct for parts that the PID controller cannot fully compensate for. The combination of the two is achieved through the following control formula:

[0039] u(t)=u PID (t)+u HDRL (t);

[0040] Where u(t) is the total output control signal, u PID (t) represents the output of the PID controller, u HDRL (t) is the output of the HDRL compensation term. This combined control strategy utilizes the high response speed of PID control and fully leverages the adaptive learning capability of HDRL in complex scenarios, achieving fast and accurate impedance regulation.

[0041] in, e(t) = Z ref (t)-Z est (t) represents the error between the target impedance and the estimated impedance, where K p K i and K d These are proportional, integral, and differential gains, respectively.

[0042] The HDRL compensation term is designed based on a hybrid deep reinforcement learning algorithm, which generates the compensation signal u in real time through a policy network and a value network. HDRL The design goal of the compensation term (t) is to eliminate the complex nonlinear characteristics that the PID controller cannot handle and the insufficient regulation caused by the dynamic changes of the system.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] 1. Through a dynamic threshold adjustment mechanism, voltage and current fluctuations at the common coupling point are monitored in real time, reducing unnecessary computational load. The sliding window algorithm combines the statistical characteristics of voltage and current to dynamically set triggering conditions, improving the accuracy and real-time performance of event triggering. A combination of discrete and continuous action spaces is adopted, with a policy network for significant adjustments and a value network for fine optimization, ensuring the accuracy of grid impedance regulation. The reward function combines grid voltage fluctuations, current fluctuations, and triggering frequency to optimize system stability and efficiency. A PID controller is introduced for initial response, while the HDRL module provides dynamically optimized compensation signals to achieve intelligent control of complex load fluctuations. The dual feedback mechanism enhances the system's robustness and real-time response capability.

[0045] 2. The event-triggered mechanism is used to activate the HDRL module, achieving a combination of efficient triggering and intelligent impedance optimization. The linkage mechanism significantly reduces computational resource consumption while improving the system's response speed and adjustment accuracy. The collaborative design of the event-triggered mechanism and the HDRL optimization module, the combination of adaptive dynamic threshold adjustment and sliding window algorithm, the feedback regulation system composed of PID controller and HDRL module, the power grid impedance estimation method combining recursive least squares method and Kalman filtering, the application of hybrid deep reinforcement learning algorithm in impedance optimization, especially the combination of discrete and continuous action space and hierarchical network structure, the dynamic threshold adjustment algorithm and its implementation in the event-triggered mechanism.

[0046] 3. Through the adaptive event triggering mechanism, the triggering conditions can be dynamically adjusted according to the real-time fluctuations of the power grid, and impedance estimation and adjustment can be performed only when necessary, thereby reducing unnecessary computational interference and ensuring the stability of power grid operation. The hybrid deep reinforcement learning module can quickly respond to drastic load fluctuations and avoid power grid instability caused by impedance mismatch.

[0047] Furthermore, by combining the sliding window algorithm and the dynamic threshold adjustment mechanism, the fluctuation characteristics of voltage and current are monitored in real time to ensure the accuracy and timeliness of event triggering. The hierarchical action space design of the HDRL algorithm realizes the combination of fast and fine adjustment, which significantly improves the system's response efficiency. Through the adaptive event triggering mechanism, the impedance estimation and optimization process is only started when there are significant fluctuations, avoiding frequent resource calls and greatly reducing the computational burden. The combination of recursive least squares method and Kalman filtering further optimizes the impedance estimation algorithm, enabling the system to achieve a balance between computational accuracy and efficiency.

[0048] 4. By adopting the HDRL algorithm and combining the division of labor between the policy network and the value network, the system can autonomously learn the optimal impedance adjustment strategy according to the actual load environment, reducing the need for human intervention. The system continuously optimizes the adjustment strategy through a reward mechanism, enabling intelligent impedance adjustment in complex dynamic load environments. Furthermore, the impedance optimization module can significantly reduce voltage and current fluctuations, reduce harmonic distortion and power loss, and improve power transmission efficiency. The joint compensation strategy of the PID controller and HDRL module maintains output stability while responding quickly to load changes, ensuring power quality.

[0049] In summary, intelligent control methods are used to achieve real-time impedance regulation of the power grid under complex load conditions, thereby effectively improving power quality. By introducing an event triggering mechanism, the system can adaptively estimate impedance when significant load fluctuations are detected, reducing unnecessary calculations. The HDRL algorithm further optimizes the regulation strategy to cope with frequent load changes, improve response speed and adaptability, ensure voltage and current stability, and overcome the shortcomings of existing technologies in real-time response, load adaptability, and power quality assurance.

[0050] Furthermore, in response to the problems of power quality fluctuations and grid impedance changes under dynamic conditions such as distributed energy access and complex load fluctuations, a full-link impedance optimization scheme from "adaptive sensing" to "intelligent control" is proposed. Through the joint design of adaptive event triggering mechanism, HDRL optimization strategy and feedback control mechanism, real-time optimization and intelligent control of grid impedance are realized, thereby improving grid stability and power quality. Attached Figure Description

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0052] In the attached diagram:

[0053] Figure 1 This is a schematic diagram of the working process of the power grid impedance regulation system of the present invention. Detailed Implementation

[0054] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0055] Example: Figure 1As shown, the present invention provides a technical solution, a power grid impedance regulation system based on adaptive event triggering and hybrid deep learning, which uses adaptive event-driven impedance estimation, intelligent control optimization and dynamic feedback regulation to achieve real-time adjustment of power grid impedance, thereby reducing harmonic distortion, alleviating voltage fluctuations and improving power quality.

[0056] Specifically, it includes an adaptive event-triggered impedance estimation module, an HDRL-based impedance optimization module, a real-time feedback adjustment module, and an integration mechanism between adaptive event triggering and HDRL.

[0057] The adaptive event-triggered impedance estimation module, the HDRL-based impedance optimization module, and the real-time feedback adjustment module achieve linkage control through data interaction and signal feedback. When the fluctuation of PCC voltage or current exceeds the dynamic threshold, the adaptive event-triggered module activates impedance estimation and uses the estimation result as one of the input states of HDRL. HDRL determines the optimal impedance control action based on the current PCC state and transmits it to the feedback adjustment module. The PCC state includes voltage, current, and estimated impedance.

[0058] The real-time feedback adjustment module further adjusts and optimizes the grid impedance through a weighted combination of PID control signals and HDRL control signals. This full-link linkage control structure can ensure the flexibility and stability of the control strategy in scenarios with severe load fluctuations, reduce the transmission effect of load disturbances, and significantly improve power quality and grid stability.

[0059] Based on the above technical solution, the adaptive event-triggered impedance estimation module dynamically monitors the voltage and current fluctuations at the point of common coupling (PCC). It employs a dynamic threshold adjustment mechanism and a Kalman filter-based estimation method combining recursive least squares (RLS) to achieve real-time estimation of the grid impedance. Unlike traditional fixed threshold triggering methods, the event-triggered mechanism dynamically adjusts the triggering conditions according to the load fluctuations at the PCC, enabling immediate impedance estimation during significant load changes while reducing unnecessary triggering during stable load conditions. This effectively reduces computational burden and energy consumption. Furthermore, by combining the adaptive filter with dynamic threshold control, robustness to noise and high-frequency disturbances is further improved. The combination of RLS and Kalman filtering ensures high-frequency dynamic response, allowing rapid convergence to a new impedance value under sudden load changes, thereby improving the stability and accuracy of the estimation.

[0060] Based on the above technical solution, the impedance optimization module based on HDRL uses the hybrid deep reinforcement learning HDRL algorithm to divide impedance control into a two-layer control structure of discrete and continuous actions to cope with complex load fluctuations and impedance changes in the power grid. Specifically, the discrete action space is responsible for large-scale impedance adjustment to quickly respond to sudden load fluctuations, while the continuous action space is responsible for fine-tuning the impedance to ensure the stability of the power grid operation. The HDRL algorithm is based on the architecture of policy network Actor and value network Critic. It continuously adjusts and optimizes the strategy through a reward and punishment mechanism to ensure efficient response in scenarios of large load fluctuations and small load adjustments. Compared with traditional PID control and support vector machine (SVM) methods, HDRL has self-learning and adaptive capabilities. Even in complex power grid fluctuation scenarios, it can generate the optimal impedance adjustment strategy in real time through the policy network, thereby significantly improving the system's adaptability and response speed.

[0061] Based on the above technical solution, the real-time feedback regulation module, under the synergistic effect of the HDRL control signal and the PID control signal, is responsible for the rapid correction and regulation of the voltage and current of the power grid. The PID controller is responsible for handling small load fluctuations, while the HDRL compensation module is responsible for quickly correcting the system's control output when the power grid experiences large load changes, preventing the voltage or current from exceeding the threshold range. The PID controller calculates the control increment using the voltage and current deviation at the PCC point. The compensation term generated by the HDRL uses the optimal control action generated by the HDRL strategy network as a supplementary signal to the PID controller, enabling the system to respond quickly when the load changes abruptly. This dual control signal fusion design significantly improves the system's response speed and stability, especially in scenarios with frequent load fluctuations.

[0062] Based on the above technical solution, the adaptive event-triggered impedance estimation module is responsible for real-time monitoring of voltage and current fluctuations at the power grid's point of common coupling (PCC), and automatically triggers the impedance estimation process based on these fluctuations. The core objective of this module is to reduce the computational burden, respond quickly when the power grid experiences significant light load fluctuations, and improve the stability and power quality of the power grid. It adopts a dynamic threshold adjustment mechanism instead of the traditional fixed threshold mechanism, which can dynamically adjust the event triggering threshold according to real-time voltage and current fluctuations, thereby avoiding over-computation and ensuring that the impedance estimation process is accurately triggered when needed.

[0063] Based on the above technical solution, the dynamic threshold adjustment mechanism is based on the voltage and current fluctuation amplitude. It effectively separates the grid signal and noise through an adaptive filter, automatically identifies the effective grid disturbance signal, and when the grid fluctuation is small or the background noise is large, the impedance estimation will not be frequently triggered, reducing the computational load of the system and improving the system response efficiency. The impedance estimation process will only be triggered when the grid fluctuation reaches a certain level, ensuring the stability of the system.

[0064] To further improve the effect of dynamic threshold adjustment, a sliding window algorithm is proposed. This algorithm dynamically adjusts the threshold by calculating the maximum and minimum voltage values ​​within a set time window and updates the threshold base at each time node. It can effectively cope with common random fluctuations in the power grid and has a high anti-interference capability for light load fluctuations.

[0065] Specifically, assuming within time window T win Within the range, the maximum and minimum values ​​of the voltage signal V(t) are Vmax and Vmin, respectively. max (t) and V min (t), then the sliding window algorithm can calculate the current fluctuation base value V using the following formula. base (t): Based on this fluctuation base value, the system can dynamically adjust the event triggering threshold to ensure that impedance estimation can be triggered in a timely manner when there are large changes in the power grid.

[0066] When the event triggering condition is met, the system will begin the impedance estimation process. To achieve accurate real-time estimation, two algorithms are combined: Recursive Least Squares (RLS) and Kalman Filter. The RLS algorithm can quickly update the impedance estimate, while the Kalman Filter can effectively suppress noise in the power grid environment, improving the accuracy of the estimation results. Specifically, the mathematical formula for the impedance estimation process is as follows:

[0067] Z est (t)=Z est (t-1)+K(t)[V(t)-I(t)Z est (t-1)];

[0068] Among them, Z est Let V(t) be the impedance estimate at the current time t, V(t) be the current voltage value, I(t) be the current current value, and K(t) be the adaptive gain, dynamically adjusted based on the inverse prediction values ​​of voltage, current, and impedance in real time. est (t-1) represents the impedance estimate from the previous moment, and the gain K(t) is calculated based on the voltage V(t), current I(t), and impedance Z at the current moment. est The prediction error of (t-1) is dynamically calculated, and the adaptive gain can optimize the estimation results based on real-time fluctuations and background noise.

[0069] The introduction of the Kalman filter enables the impedance estimation process to maintain high accuracy even in noisy environments. The specific Kalman gain update formula is as follows:

[0070] K(t) = P(t-1)[P(t-1) + R(t)] -1

[0071] P(t)=[P(t-1) -1 +Q(t)] -1 ;

[0072] Where P(t) is the estimation error covariance matrix, Q(t) is the process noise covariance, and R(t) is the measurement noise covariance. By combining RLS with a Kalman filter, the system can accurately estimate the grid impedance in a noisy environment and has a high real-time response capability.

[0073] Based on the above technical solution, the impedance optimization module based on HDRL is the core intelligent control unit, which aims to adjust the grid impedance in real time, thereby optimizing power quality and improving grid stability. This module adopts a hybrid deep reinforcement learning algorithm, combining discrete and continuous action spaces to ensure that the grid can quickly and accurately perform impedance regulation optimization when the load fluctuates and the grid state changes.

[0074] The HDRL algorithm selects discrete actions through a policy network, which refers to large-scale impedance adjustments. These discrete actions are then refined and optimized through a value network to achieve the optimal impedance regulation effect. This allows the algorithm to adapt to frequent load changes in the power grid while avoiding frequent calculation operations, thus improving real-time response capability and overall stability.

[0075] Based on the above technical solution, the impedance optimization module based on HDRL needs to first design the state space of HDRL. The design of the state space involves the system's real-time sensing and feedback mechanism of the power grid. The system's input vector mainly includes the voltage and current of the common coupling point (PCC) in the power grid, as well as the estimated power grid impedance value. Through the voltage and current data collected by real-time sensors, the system can monitor the dynamic changes of the power grid in real time and adjust the power grid impedance according to the changes. The design of the state space can be expressed by the following formula: s t =[V PCC ,I PCC Z est This mainly includes the voltage and current of the common coupling point (PCC) in the power grid, as well as the estimated power grid impedance value. Through this state-space design, the system can quickly respond to changes in the power grid load and adjust the impedance in a timely manner to ensure the stability of the power grid.

[0076] The action space design of HDRL is one of the key aspects of this module. The action space is divided into two parts: discrete actions and continuous actions. The combination of the two can flexibly respond to various fluctuations in the power grid. Discrete actions are used for large-range impedance adjustment, which is suitable for situations where the power grid load fluctuates drastically or changes suddenly. Each discrete action represents a large adjustment step of impedance and is usually used to quickly respond to significant fluctuations. If the power grid load increases sharply, the discrete actions will make a large adjustment to the impedance to restore the power grid state to a stable level as soon as possible. Continuous actions are used for fine adjustment to ensure that the power grid can still operate smoothly when the load changes slightly. Continuous actions provide fine-grained control of impedance adjustment, avoid frequent large adjustments, maintain the stability of the power grid, and avoid the negative impact of excessive adjustments.

[0077] The action space can be defined by the following formula: α(t)={α dis (t),α con (t)};

[0078] Where, α dis (t) represents discrete action, which refers to large-scale adjustment, α con (t) represents continuous motion, which refers to fine adjustment;

[0079] To enable the HDRL model to learn the optimal impedance regulation strategy, a reward function is designed. The goal of the reward function is to minimize voltage and current fluctuations in the power grid, thereby improving power quality and grid stability. The specific formula is as follows:

[0080] R(s,a)=-α|V target -V PCC |-β|I target -I PCC |;

[0081] Here, α and β are the stability weight coefficients of voltage and current, respectively, which can be adjusted according to actual needs to ensure power quality and grid stability. The learning process of HDRL adopts a hierarchical neural network structure, including a policy network and a value network. The policy network is responsible for selecting appropriate discrete actions from the current state space, that is, making a large-scale adjustment to the impedance, which is suitable for dealing with large-scale load changes. The value network further refines the discrete actions selected by the policy network and optimizes the accuracy of impedance adjustment. This refinement process enables the system to make fine adjustments through continuous actions when there are small load changes, ensuring the stable operation of the grid.

[0082] After each adjustment, the system calculates the reward and updates the network parameters based on the reward signal to optimize the impedance regulation strategy. Through the reward feedback mechanism, the HDRL model continuously learns and adjusts its strategy, enabling the power grid to maintain efficient and stable operation in dynamic load environments.

[0083] During training, the Q-learning algorithm of reinforcement learning is used for parameter updates, and the update formula is as follows:

[0084] Q(s t ,α t )=Q(s t ,α t )+α[R(t)+γmaxQ(s t+1 ,α')-Q(s t ,α t )];

[0085] Where Q(s) t ,α t ) represents state s t Take action α t The Q-value, where α is the learning rate, γ is the discount factor, and maxQ(s) t+1 ,α') is the maximum Q value in the next state. This learning algorithm enables the HDRL model to learn the optimal impedance regulation strategy step by step, thereby adaptively adjusting the grid impedance in practical applications.

[0086] The impedance optimization module based on HDRL has the following advantages: fast response and efficient regulation: by combining discrete and continuous action space, HDRL can quickly make large-scale adjustments when the load fluctuates drastically and make fine adjustments when the load changes slightly, thereby ensuring the stable operation of the power grid.

[0087] Intelligent learning capability: HDRL continuously optimizes impedance regulation strategy through a reward mechanism, enabling it to adaptively adjust impedance in complex power grid environments, thereby improving power grid stability and power quality;

[0088] Reduced computational burden: Through reasonable action space design and reward function optimization, HDRL avoids excessive computational operations, reduces the computational burden of the system, and improves the real-time response capability of the system.

[0089] Based on the above technical solution, the real-time feedback regulation module is responsible for making fine dynamic corrections to impedance regulation during power grid operation to cope with rapidly changing load environment and complex power grid fluctuations. It adopts a control strategy that combines the classic PID controller with compensation terms based on hybrid deep reinforcement learning (HDRL), which combines the fast response capability of traditional control methods with the optimization capability of intelligent control methods, significantly improving the real-time performance, robustness and adaptability of the system.

[0090] The real-time feedback regulation module consists of two key parts: a PID controller and an HDRL compensation term. The PID controller is used to initially regulate the grid impedance, respond to rapid load fluctuations, and perform basic dynamic compensation. Through simple and efficient proportional, integral, and derivative control, the PID controller ensures that the system can quickly stabilize near the target value.

[0091] The HDRL compensation term, based on the PID controller, provides more precise regulation capabilities to the system by learning complex nonlinear dynamic characteristics. This compensation term is generated in real-time based on the HDRL algorithm and can correct for parts that the PID controller cannot fully compensate for. The combination of the two is achieved through the following control formula:

[0092] u(t)=u PID (t)+u HDRL (t);

[0093] Where u(t) is the total output control signal, u PID (t) represents the output of the PID controller, u HDRL (t) is the output of the HDRL compensation term. This combined control strategy utilizes the high response speed of PID control and fully leverages the adaptive learning capability of HDRL in complex scenarios, achieving fast and accurate impedance regulation.

[0094] in, e(t) = Z ref (t)-Z est (t) represents the error between the target impedance and the estimated impedance, where K p K i and K d These are proportional, integral, and differential gains, respectively.

[0095] The HDRL compensation term is designed based on a hybrid deep reinforcement learning algorithm, which generates the compensation signal u in real time through a policy network and a value network. HDRL The design goal of the compensation term (t) is to eliminate the complex nonlinear characteristics that the PID controller cannot handle and the insufficient regulation caused by the dynamic changes of the system.

[0096] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A power grid impedance regulation system based on adaptive event triggering and hybrid deep learning, characterized in that: Based on adaptive event-driven impedance estimation, intelligent control optimization and dynamic feedback regulation, the real-time adjustment of grid impedance is realized, the harmonic distortion is reduced, the voltage fluctuation is alleviated, and the power quality is improved. Specifically, it includes an adaptive event-triggered impedance estimation module, an impedance optimization module based on HDRL, a real-time feedback regulation module, and an integrated mechanism of adaptive event triggering and HDRL. Through data interaction and signal feedback, the adaptive event-triggered impedance estimation module, the impedance optimization module based on HDRL, and the real-time feedback regulation module realize linkage control. When the PCC voltage or current fluctuation exceeds the dynamic threshold, the adaptive event-triggering module activates impedance estimation, and the estimation result is used as one of the input states of HDRL. HDRL determines the optimal impedance control action according to the current PCC state and transmits it to the feedback regulation module. The real-time feedback regulation module further adjusts and optimizes the grid impedance through the weighted combination of PID control signals and HDRL control signals. This is a full-linkage control structure.

2. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 1, wherein: The adaptive event-triggered impedance estimation module dynamically monitors the voltage and current fluctuations of the public coupling point PCC, uses a dynamic threshold adjustment mechanism combined with the recursive least squares method RLS and the Kalman filter estimation method to estimate the real-time impedance of the grid.

3. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 1, wherein: The impedance optimization module based on HDRL divides the impedance control into a two-level control structure of discrete actions and continuous actions based on the hybrid deep reinforcement learning HDRL algorithm.

4. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 1, wherein: Under the synergistic action of HDRL control signals and PID control signals, the real-time feedback regulation module is responsible for quickly correcting and regulating the voltage and current of the grid. The PID controller is responsible for handling small-scale load fluctuations, while the HDRL compensation module is responsible for quickly correcting the system control output when the grid experiences large-scale load changes.

5. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 2, wherein: The adaptive event-triggered impedance estimation module is responsible for real-time monitoring of the voltage and current fluctuations of the grid public coupling point PCC and automatically triggering the impedance estimation process according to these fluctuations. The core goal of this module is to reduce the computational burden and quickly respond to significant light load fluctuations. It uses a dynamic threshold adjustment mechanism instead of the traditional fixed threshold mechanism, which can dynamically adjust the event-triggering threshold based on real-time voltage and current fluctuations.

6. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 5, wherein: The dynamic threshold adjustment mechanism is based on the voltage and current fluctuation amplitude, effectively separates the grid signal and noise through an adaptive filter, and automatically identifies valid grid disturbance signals. In the case of small grid fluctuations or high background noise, the system's computational load is reduced. To further improve the effect of dynamic threshold adjustment, a sliding window algorithm is proposed. This algorithm dynamically adjusts the threshold by calculating the maximum and minimum values of the voltage within a set time window and updates the threshold base at each time node. It can effectively deal with common random fluctuations in the grid and has high anti-interference ability for light load fluctuations. Specifically, assume that in a time window T win , the maximum and minimum values of the voltage signal V(t) are V max (t) and V min (t), respectively, then the sliding window algorithm can calculate the current fluctuation base value V base (t) by the following formula: According to this fluctuation base value, the system can dynamically adjust the event-triggered threshold value to ensure that impedance estimation can be triggered in time when the power grid changes greatly. When the event trigger condition is met, the system will start the impedance estimation process. In order to achieve accurate real-time estimation, both recursive least squares (RLS) and Kalman filter algorithms are combined. The RLS algorithm can quickly update the impedance estimation value, and the Kalman filter can effectively suppress the noise in the power grid environment, improving the accuracy of the estimation result. Specifically, the mathematical formula of the impedance estimation process is as follows: Z est (t) = Z est (t - 1) + K(t)[V(t) - I(t)Z est (t - 1)]; wherein Z est is the impedance estimation value at the current time t, V(t) is the current voltage value, I(t) is the current current value, K(t) is an adaptive gain, which is dynamically adjusted based on the real-time voltage, current and impedance back prediction value, Z est (t-1) is the impedance estimation value at the previous time, and the gain K(t) is dynamically calculated according to the prediction error of the current time voltage V(t), current I(t) and impedance Z est (t-1), and the adaptive gain can optimize the estimation result according to real-time fluctuations and background noise. The introduction of the Kalman filter enables the impedance estimation process to maintain high estimation accuracy in the presence of noise. The specific Kalman gain update formula is as follows: K(t) = P(t-1) [P(t-1) + R(t)] -1 P(t) = [P(t-1) -1 + Q(t)] -1 ; Where P(t) is the estimation error covariance matrix, Q(t) is the process noise covariance, and R(t) is the measurement noise covariance. Through the combination of RLS and Kalman filter, the system can accurately estimate the grid impedance in a noisy environment.

7. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 3, wherein: The impedance optimization module based on HDRL is the core intelligent control unit, aiming to adjust the grid impedance in real time to optimize power quality and improve the stability of the grid. This module uses a hybrid deep reinforcement learning (HDRL) algorithm, combining discrete and continuous action spaces. The HDRL algorithm selects discrete actions through a policy network and refines these actions through a value network.

8. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 7, wherein: The impedance optimization module based on HDRL requires the design of the HDRL's state space. This design involves the system's real-time sensing and feedback mechanism of the power grid. The system's input vector mainly includes the voltage and current of the common coupling point (PCC) in the power grid, as well as the estimated grid impedance. Through real-time voltage and current data collected by sensors, the system can monitor the dynamic changes in the power grid in real time and adjust the grid impedance accordingly. The state space design can be expressed by the following formula: s t =[V PCC ,I PCC Z est This mainly includes the voltage and current of the common coupling point (PCC) in the power grid, as well as the estimated power grid impedance value, through this state-space design; The action space design of HDRL is one of the keys to this module. The action space is divided into discrete and continuous actions, which can flexibly respond to various fluctuations in the grid. Discrete actions are used for large-scale impedance adjustment and are suitable for situations where the grid load fluctuates sharply or there is a sudden load change. Each discrete action represents a large adjustment step for the impedance, which is usually used to quickly respond to significant fluctuations. Continuous actions are used for fine adjustment to ensure smooth operation of the grid during small load changes. Continuous actions provide fine-grained control of impedance adjustment, avoiding frequent large adjustments and maintaining the stability of the grid while avoiding negative effects caused by excessive adjustments. The action space can be defined by the following equation: a(t) = {a dis (t), a con (t)}; wherein a dis (t) is a discrete action, a con (t) is a continuous action; To enable the HDRL model to learn the optimal impedance adjustment strategy, a reward function is designed. The goal of the reward function is to minimize voltage and current fluctuations in the grid, improving power quality and grid stability. The specific formula is as follows: R(s, a) = -aV target - V PCC | - βI target - I PCC |; Where alpha and beta are the stability weight coefficients of voltage and current, respectively, which can be adjusted according to actual needs to ensure power quality and grid stability. The learning process of HDRL uses a hierarchical neural network structure, including a policy network and a value network. The policy network is responsible for selecting appropriate discrete actions from the current state space, suitable for responding to large-scale load changes. The value network further refines the discrete actions selected by the policy network to optimize the precision of impedance adjustment. This refinement allows the system to perform fine adjustments through continuous actions during small load changes. After each adjustment, the system calculates the reward and updates the network parameters based on the reward signal to optimize the impedance adjustment strategy. Through the reward feedback mechanism, the HDRL model continuously learns and adjusts its strategy. During the training process, the Q-learning algorithm of reinforcement learning is used for parameter updating, with the update formula as follows: Q(s t ,α t ) = Q(s t ,α t ) + α[R(t) + γmaxQ(s t+1 ,α') - Q(s t ,α t )]; where Q(s t , a t ) is the Q-value of taking action a t in state s t , a is the learning rate, g is the discount factor, and maxQ(s t+1 , a') is the maximum Q-value in the next state. This learning algorithm enables the HDRL model to gradually learn the optimal impedance regulation strategy, thereby adaptively regulating the grid impedance in practical applications.

9. The adaptive event-triggered and hybrid deep learning based power grid impedance conditioning system of claim 4, wherein: The real-time feedback adjustment module is responsible for fine dynamic correction of impedance adjustment during power grid operation to cope with rapidly changing load environment and complex power grid fluctuation. A control strategy combining a classic PID controller and a compensation term based on hybrid deep reinforcement learning (HDRL) is adopted to combine the fast response capability of traditional control methods with the optimization capability of intelligent control methods. The real-time feedback adjustment module consists of two key parts: a PID controller and an HDRL compensation term. The PID controller is used for preliminary adjustment of power grid impedance, responds to rapid load fluctuations, and performs basic dynamic compensation. The PID controller ensures that the system can quickly and stably reach the target value through simple and efficient proportional, integral, and derivative control. The HDRL compensation term provides more accurate adjustment capability for the system by learning complex nonlinear dynamic characteristics based on the PID controller. This compensation term is generated in real time based on the HDRL algorithm and can correct the parts that the PID controller cannot fully compensate. The combination of the two is achieved through the following control formula: u(t) = u PID (t) + u HDRL (t); Wherein, u(t) is the total output control signal, u PID (t) is the PID controller output, u HDRL (t) is the HDRL compensation item output, this combination control strategy not only utilizes the high response speed of the PID control, but also fully gives play to the adaptive learning ability of the HDRL in the complex scene, realizes the fast and accurate impedance adjustment; wherein, e(t) = Z ref (t) - Z est (t), is the error between the target impedance and the estimated impedance, where, K p , K i and K d are the proportional, integral and derivative gains, respectively. The design of the HDRL compensation term is based on a hybrid deep reinforcement learning algorithm, which generates the compensation signal u in real time through a policy network and a value network HDRL (t), the design goal of the compensation term is to eliminate the complex nonlinear characteristics that the PID controller cannot handle and the regulation deficiency caused by the dynamic changes of the system.

Images