Power plant automatic voltage control AVC system distribution method and device based on equal-margin same-direction distribution

Abstract

The invention provides a power plant automatic voltage control (AVC) system distribution method and device based on equal-margin same-direction distribution, and the method comprises the steps: predicting the reactive power output and error of each device through extended Kalman filtering, and calculating a dynamic safety margin; according to bus voltage deviation, reactive power is preliminarily distributed according to an equal-margin same-direction principle; a multi-target quadratic programming model with the equipment life, the system loss and the voltage stability as targets is constructed, and the optimal reactive increment is solved; real-time optimization is carried out in a rolling time domain through model prediction control, and an instruction is issued; the weight coefficient is adaptively adjusted according to the equipment health score and the real-time electricity price, and safe, economical and efficient voltage control is realized. The reactive power regulation precision and the response speed of the AVC system of the power plant can be improved, the safety and the economical efficiency of equipment operation are enhanced, and the voltage stability and the capacity of adapting to complex working conditions of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of power system automation and optimization control technology, and more specifically, to a distribution method and apparatus for an automatic voltage control (AVC) system for power plants based on equal margin and unidirectional distribution. Background Technology

[0002] With the increasing proportion of new energy sources such as wind power and photovoltaics connected to the grid, grid voltage fluctuations are becoming more severe, putting greater pressure on reactive power regulation in power plant-side automatic voltage control (AVC) systems. Currently, existing reactive power allocation methods mainly include proportional allocation, sensitivity allocation, and static margin allocation. Proportional allocation simply distributes reactive power demand according to the rated capacity of the equipment. This method can easily lead to some equipment reaching its limits prematurely, failing to fully utilize the potential of other equipment. Sensitivity allocation allocates reactive power based on the equipment's sensitivity to voltage, but this method ignores the equipment's remaining capacity, posing an overload risk. Static margin allocation allocates reactive power according to the equipment's static remaining margin, but this method lacks consideration for future fluctuations and uncertainties, and cannot effectively cope with dynamically changing grid environments.

[0003] In implementing the embodiments of the present invention, the prior art has at least the following problems or defects: the existing methods cannot balance multiple objectives (equipment life, safety margin, system loss, voltage recovery speed) in a single strategy, and do not combine predictive control. When faced with sudden voltage fluctuations, the response is lagging, making it difficult to meet the requirements of modern power systems for high precision, fast response and safe and economical operation of voltage control. Summary of the Invention

[0004] This invention provides a distribution method and apparatus for an automatic voltage control (AVC) system in a power plant based on equal margin and unidirectional distribution.

[0005] In a first aspect of the present invention, a power plant automatic voltage control (AVC) system allocation method based on equal margin unidirectional allocation is provided, comprising: Step 1: Based on the extended Kalman filter, predict the reactive power output and error covariance of each reactive power regulation device, and generate the dynamic reactive power upper limit and dynamic reactive power lower limit of each reactive power regulation device according to the risk coefficient. Step 2: Based on the deviation between the measured bus voltage value and the target bus voltage value, calculate the available safety margin of each reactive power regulation device in the current regulation direction; Step 3: Calculate the preliminary allocation ratio of each reactive power regulation device based on the available safety margin, and perform preliminary reactive power allocation according to the total reactive power regulation demand; Step 4: Construct a multi-objective quadratic programming model with equipment life cost, system reactive power loss and voltage stability as optimization objectives. Under the constraints of meeting the total reactive power regulation demand, the reactive power increment of each reactive power regulation device not exceeding its available safety margin and the regulation direction being consistent, solve for the optimal reactive power increment. Step 5: Based on the linear prediction model, optimize and distribute the optimal reactive power increment in real time through model prediction control within the rolling time domain; Step 6: Based on the health scores of each reactive power regulation device and the real-time electricity price, adaptively adjust the weight coefficients in the multi-objective quadratic programming model.

[0006] Further, step one includes: The reactive power output state of the i-th reactive power regulation device is predicted using an extended Kalman filter to obtain the predicted reactive power output. and prediction error covariance ,in For the predicted reactive power output of the i-th reactive power regulation device, Let be the prediction error covariance of the i-th reactive power regulation device; Based on the prediction error covariance and preset risk coefficient Calculate the dynamic reactive power upper limit. and dynamic reactive power lower limit The calculation formula is:

[0007] in, Let i be the rated minimum reactive power output of the i-th reactive power regulating device. Let i be the rated maximum reactive power output of the i-th reactive power regulating device. Let i be the preset risk coefficient for the i-th reactive power regulation device. Let i be the dynamic reactive power limit of the i-th reactive power regulation device. Let be the lower limit of dynamic reactive power for the i-th reactive power regulation device.

[0008] Furthermore, the extended Kalman filtering process includes: Updated in time:

[0009] in, Let be the process noise covariance of the i-th reactive power regulation device. for Let t be the a priori predicted reactive power output of the i-th reactive power regulating device. For the prediction cycle, for The prior prediction error covariance of the i-th reactive power regulation device at time i. Let be the prediction error covariance of the i-th reactive power regulation device at time t; Observation Update:

[0010] in, Let be the measurement noise covariance of the i-th reactive power regulation device. Let Kalman gain be the value of the i-th reactive power regulation device. Let i be the measured reactive power output of the i-th reactive power regulating device. for The posterior prediction of reactive power output of the i-th reactive power regulation device at time i. for The posterior prediction error covariance of the i-th reactive power regulation device at time i.

[0011] Furthermore, in step two: The bus voltage deviation ,in This is the measured value of the bus voltage. The target value for bus voltage. This refers to the bus voltage deviation. The available safety margin The calculation method is as follows: when hour, ; when hour, ,in Let represent the available safety margin of the i-th reactive power regulation device.

[0012] Furthermore, in step three: The preliminary allocation ratio ,in >0 indicates that the j-th reactive power regulating device has a usable safety margin in the current regulating direction. Let i be the initial allocation ratio for the i-th reactive power regulating device. The available safety margin for the j-th reactive power regulation device; The total reactive power regulation demand ,in This is the scheduling ratio coefficient. To meet the total reactive power regulation demand.

[0013] Furthermore, in step four: The objective function of the multi-objective quadratic programming model is:

[0014] in, For each reactive power regulation device, the reactive power increment vector is... Let n be the reactive power increment of the i-th reactive power regulating device, and n be the total number of reactive power regulating devices. Let H be the objective function value of the multi-objective quadratic programming model, H be the coefficient matrix of the quadratic terms, and f be the coefficient vector of the linear terms. This is the weighting coefficient for equipment life cost. For equipment life cost matrix, This is the system reactive power loss weighting coefficient. This is the system's reactive power loss sensitivity matrix. This is the voltage stability weighting coefficient. This is the voltage stability weight matrix. Here, g(V) is the voltage deviation weighting coefficient, and g(V) is the mapping function between voltage deviation and reactive power demand. Let g(V) be the partial derivative of the mapping function g(V) with respect to the voltage V. This refers to the bus voltage deviation. The constraints include:

[0015] in, For equipment life cost matrix, This is the system's reactive power loss sensitivity matrix. This is the voltage stability weighting matrix; Here, g(V) represents the weighting coefficient; g(V) is the mapping function between voltage deviation and reactive power demand. To meet the total reactive power regulation demand, The available safety margin for the i-th reactive power regulation device. Bus voltage deviation The sign function (when) hour, =1; when hour, ;when hour, ).

[0016] Furthermore, in step five: The linear prediction model is as follows:

[0017] Where 'a' is the autoregressive coefficient of the bus voltage deviation. Let be the influence coefficient of the reactive power increment of the i-th reactive power regulation device on the bus voltage deviation, and w(t) be the residual noise at time t. Let be the bus voltage deviation at time t+1. Let be the bus voltage deviation at time t. Let be the reactive power increment of the i-th reactive power regulating device at time t, and n be the total number of reactive power regulating devices; The model prediction control minimizes the objective function within the prediction time domain length N:

[0018] in, Here, N is the control penalty coefficient, N is the prediction time domain length, and k is the prediction step size. Let be the bus voltage deviation at time t+k+1. Let be the reactive power increment vector of each reactive power regulation device at time t+k. This refers to the 2-norm operation.

[0019] Furthermore, in step six: The health score The results were obtained based on multi-source sensor data and machine learning model evaluation, among which... Give the health score to the i-th reactive power regulation device; The weighting coefficients are adjusted as follows:

[0020] in, For health-adaptive gain, p(t) is the real-time electricity price at time t. This serves as the benchmark value for weighting equipment life-cycle costs. Let be the life cost weighting coefficient for the i-th reactive power regulation device. This is the system reactive power loss weighting coefficient. This is the voltage deviation weighting coefficient. This indicates a direct proportional relationship.

[0021] Furthermore, the process noise covariance in the extended Kalman filter and measurement noise covariance Adaptive updates are performed based on online error statistics, whereby... Let be the process noise covariance of the i-th reactive power regulation device. Let be the measurement noise covariance of the i-th reactive power regulation device.

[0022] In a second aspect of the invention, a power plant automatic voltage control (AVC) system distribution device based on equal margin unidirectional distribution is provided, comprising: The data acquisition module is used to acquire bus voltage measurement values, reactive power output measurement values ​​of each reactive power regulation device, real-time electricity price, and equipment health status data; The EKF prediction module is used to perform extended Kalman filter prediction and outputs the predicted reactive power output and the prediction error covariance. The dynamic limit module is used to generate the dynamic upper limit and dynamic lower limit of reactive power for each reactive power regulation device based on the prediction error covariance and the preset risk coefficient. The preliminary allocation module is used to calculate the bus voltage deviation, available safety margin, preliminary allocation ratio, and total reactive power regulation demand, and to complete the preliminary reactive power allocation. The QP optimization module is used to construct and solve a multi-objective quadratic programming model and output the optimal reactive power increment. The MPC control module is used to optimize control through model prediction and issue the optimal reactive power increment in the rolling time domain. The weight adaptive module is used to adjust the weight coefficients in the multi-objective quadratic programming model based on the equipment health score and real-time electricity price.

[0023] The embodiments of the present invention have at least the following beneficial effects: 1. By using dynamic safety margin sensing technology, extended Kalman filtering is used to predict reactive power output and error covariance, and dynamic upper and lower limits of reactive power are generated based on risk coefficients. This can prevent equipment from exceeding the safe range during operation due to prediction uncertainty, improve the safety and reliability of equipment operation, and enhance the system's adaptability to uncertainty.

[0024] 2. The equal margin and unidirectional initial allocation method is adopted. The available safety margin of each reactive power regulation device is calculated based on the bus voltage deviation, and the initial allocation is carried out proportionally. This ensures the consistency of reactive power regulation direction, avoids reactive power regulation conflicts between devices, improves the overall coordination of the system and the accuracy of voltage control, and solves the problem of overload or underutilization of potential of some devices due to unreasonable allocation in the existing technology.

[0025] 3. A multi-objective quadratic programming model was constructed, comprehensively considering multiple optimization objectives such as equipment lifespan cost, system reactive power loss, and voltage stability. Through model predictive control, reactive power increments were optimized in real-time within the rolling time domain, achieving globally optimal reactive power allocation. Simultaneously, weighting coefficients were adaptively adjusted based on equipment health scores and real-time electricity prices, enabling the system to automatically optimize towards safer and more economical directions under different operating conditions. This improved system operating efficiency and economic benefits, and enhanced the system's adaptability and flexibility. Attached Figure Description

[0026] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example, not limitation, in which: Figure 1 A schematic flowchart of an automatic voltage control (AVC) system allocation method for power plants based on equal margin unidirectional allocation according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the distribution device of the automatic voltage control (AVC) system for power plants based on equal margin and unidirectional distribution, provided in an embodiment of the present invention. Detailed Implementation

[0027] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make the invention more thorough and complete, and to fully convey the scope of the invention to those skilled in the art.

[0028] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0029] It should be noted that the number of any elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinction only and has no limiting meaning.

[0030] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating a power plant automatic voltage control (AVC) system allocation method based on equal margin unidirectional allocation, according to an embodiment of the present invention. Figure 1 As shown, a power plant automatic voltage control (AVC) system allocation method based on equal margin and unidirectional allocation includes: Step 1: Based on the extended Kalman filter, predict the reactive power output and error covariance of each reactive power regulation device, and generate the dynamic reactive power upper limit and dynamic reactive power lower limit of each reactive power regulation device according to the risk coefficient. Step 2: Based on the deviation between the measured bus voltage value and the target bus voltage value, calculate the available safety margin of each reactive power regulation device in the current regulation direction; Step 3: Calculate the preliminary allocation ratio of each reactive power regulation device based on the available safety margin, and perform preliminary reactive power allocation according to the total reactive power regulation demand; Step 4: Construct a multi-objective quadratic programming model with equipment life cost, system reactive power loss and voltage stability as optimization objectives. Under the constraints of meeting the total reactive power regulation demand, the reactive power increment of each reactive power regulation device not exceeding its available safety margin and the regulation direction being consistent, solve for the optimal reactive power increment. Step 5: Based on the linear prediction model, optimize and distribute the optimal reactive power increment in real time through model prediction control within the rolling time domain; Step 6: Based on the health scores of each reactive power regulation device and the real-time electricity price, adaptively adjust the weight coefficients in the multi-objective quadratic programming model.

[0031] The Extended Kalman Filter (EKF) is a nonlinear filtering algorithm used to predict the reactive power output and its error covariance of various reactive power regulation devices. Reactive power regulation devices refer to equipment capable of regulating reactive power, such as generators and capacitor banks. Dynamic reactive power upper and lower limits are dynamic limits on the reactive power output of the equipment, calculated based on the prediction error covariance and risk coefficient, to ensure that the equipment operates within a safe range. Bus voltage deviation refers to the difference between the measured bus voltage value and the target value, reflecting the deviation between the current voltage state and the desired state. The preliminary allocation ratio is calculated based on the available safety margin of each device and is used to initially determine the reactive power allocation for each device. The multi-objective quadratic programming model is an optimization model used to weigh and optimize multiple objectives. Model predictive control (MPC) is an advanced control strategy that optimizes control quantities in the rolling time domain through a predictive model. These technical features collectively constitute the core content of this invention, aiming to improve the reactive power regulation accuracy and safety of power plant AVC systems.

[0032] Specifically, the reactive power regulation equipment in this invention refers to equipment used in power plants to regulate reactive power, including but not limited to generators, capacitor banks, and reactors. The reactive power output of these devices can be predicted using an Extended Kalman Filter (EKF). The EKF predicts the reactive power output and its error covariance through two steps: time update and observation update. In the time update step, the prior predicted reactive power output is updated based on the predicted reactive power output at the current moment, plus the process noise covariance. In the observation update step, the Kalman gain is used to combine the measured and predicted values ​​to update the posterior predicted reactive power output and error covariance. The dynamic upper and lower limits of reactive power are calculated based on the predicted error covariance and a preset risk coefficient, used to limit the reactive power output of the equipment and ensure that the equipment operates within a safe range. The bus voltage deviation is the difference between the measured bus voltage value and the target value, used to determine whether the current voltage state requires adjustment. The preliminary allocation ratio is calculated based on the available safety margin of each device, reflecting the remaining capacity of each device in the current adjustment direction. The total reactive power regulation demand is calculated based on the bus voltage deviation and the dispatch ratio coefficient, and is used to determine the total reactive power that needs to be allocated. The objective function of the multi-objective quadratic programming model includes three optimization objectives: equipment lifetime cost, system reactive power loss, and voltage stability, which are weighed using weighting coefficients. Constraints ensure that the total reactive power regulation demand is met, the reactive power increment of each device does not exceed its available safety margin, and the regulation direction is consistent. Model predictive control (MPC) optimizes the control quantity in the rolling time domain using a linear predictive model, minimizing the future voltage deviation energy and control quantity energy. These steps and parameter settings ensure that the system's reactive power regulation is both safe and efficient.

[0033] Furthermore, the extended Kalman filter (EKF) model construction steps in this invention include initializing the reactive power output and error covariance of the equipment, and then performing time updates and observation updates at each time step. In the time update step, the prior predicted reactive power output equals the predicted reactive power output at the current time, plus the error covariance and the process noise covariance. In the observation update step, the Kalman gain is used to combine the measured and predicted values ​​to update the posterior predicted reactive power output and error covariance. The dynamic reactive power upper and lower limits are calculated as follows: the dynamic reactive power upper limit equals the rated maximum reactive power output of the equipment minus the risk factor multiplied by the square root of the error covariance; the dynamic reactive power lower limit equals the rated minimum reactive power output of the equipment plus the risk factor multiplied by the square root of the error covariance. The bus voltage deviation is calculated as follows: the bus voltage deviation equals the measured bus voltage value minus the target value. The initial allocation ratio is calculated as follows: the initial allocation ratio equals the available safety margin of the equipment divided by the sum of the available safety margins of all equipment. The formula for calculating total reactive power regulation demand is: Total reactive power regulation demand equals bus voltage deviation multiplied by the dispatch ratio coefficient. In the objective function of the multi-objective quadratic programming model, the quadratic term coefficient matrix is ​​obtained by weighted summation of the equipment life cost matrix, the system reactive power loss sensitivity matrix, and the voltage stability weight matrix, while the linear term coefficient vector is linearly related to the voltage deviation gradient. Constraints include satisfying the total reactive power regulation demand, ensuring that the reactive power increment of each device does not exceed its available safety margin, and maintaining consistent regulation direction. Model predictive control (MPC) optimizes the control quantity in the rolling time domain through a linear predictive model, minimizing future voltage deviation energy and control quantity energy. Only the first step of control quantity is executed in each step, followed by a rolling window update. These refined operational steps and parameter settings further improve the reactive power regulation accuracy and response speed of the system.

[0034] In some embodiments, step one includes: The reactive power output state of the i-th reactive power regulation device is predicted using an extended Kalman filter to obtain the predicted reactive power output. and prediction error covariance ,in For the predicted reactive power output of the i-th reactive power regulation device, Let be the prediction error covariance of the i-th reactive power regulation device; Based on the prediction error covariance and preset risk coefficient Calculate the dynamic reactive power upper limit. and dynamic reactive power lower limit The calculation formula is:

[0035] in, Let i be the rated minimum reactive power output of the i-th reactive power regulating device. Let i be the rated maximum reactive power output of the i-th reactive power regulating device. Let i be the preset risk coefficient for the i-th reactive power regulation device. Let i be the dynamic reactive power limit of the i-th reactive power regulation device. Let be the lower limit of dynamic reactive power for the i-th reactive power regulation device.

[0036] In practical applications, process noise covariance and measurement noise covariance are crucial parameters affecting filtering accuracy. Process noise covariance reflects the uncertainty of the system model, while measurement noise covariance reflects errors in the measurement process. By adaptively updating these noise covariances through online error statistics, the filter's adaptability to dynamic system changes can be improved, thereby more accurately predicting the reactive power output and its error covariance of reactive power regulation equipment. This adaptive update mechanism ensures that the filter maintains high prediction accuracy under different operating conditions, thus improving the overall performance of the AVC system.

[0037] Specifically, process noise covariance and measurement noise covariance are two important parameters in the extended Kalman filter. (Process noise covariance) This represents the uncertainty caused by factors such as inaccurate system models or external interference during the operation of reactive power regulation equipment. It also includes the measurement of noise covariance. This indicates the error of the measuring equipment when measuring reactive power output. Accurate settings of these two parameters are crucial for filter performance. In this invention, these parameters are adaptively updated using an online error statistics method. Online error statistics refer to collecting the error between predicted and actual measured values ​​in real time during filter operation and adjusting the noise covariance based on these errors. This method can dynamically reflect the actual operating state of the system, thereby improving the adaptability and accuracy of the filter.

[0038] Furthermore, the specific steps for adaptively updating the process noise covariance and measurement noise covariance are as follows: First, at each time step, calculate the residual between the predicted reactive power output and the actual measured value. Then, based on the statistical characteristics of the residuals, such as the mean and variance, adjust the process noise covariance and measurement noise covariance. For example, if the variance of the residuals is large, it indicates high uncertainty in the system, and the process noise covariance can be appropriately increased; if the measurement error is large, the measurement noise covariance can be increased. Through this adaptive update mechanism, the filter can better adapt to the dynamic changes of the system and improve the accuracy of reactive power output prediction. This dynamic adjustment not only improves the system's response speed but also enhances its robustness, enabling it to operate stably in complex power grid environments.

[0039] In some embodiments, the extended Kalman filtering process includes: Updated in time:

[0040] in, Let be the process noise covariance of the i-th reactive power regulation device. for Let t be the a priori predicted reactive power output of the i-th reactive power regulating device. For the prediction cycle, for The prior prediction error covariance of the i-th reactive power regulation device at time i. Let be the prediction error covariance of the i-th reactive power regulation device at time t; Observation Update:

[0041] in, Let be the measurement noise covariance of the i-th reactive power regulation device. Let Kalman gain be the value of the i-th reactive power regulation device. Let i be the measured reactive power output of the i-th reactive power regulating device. for The posterior prediction of reactive power output of the i-th reactive power regulation device at time i. for The posterior prediction error covariance of the i-th reactive power regulation device at time i.

[0042] In a multi-objective quadratic programming model, the linear relationship between the first-order term vector and the voltage deviation gradient ensures that the model can dynamically adjust its optimization objective based on changes in voltage deviation. This linear relationship allows the model to respond more flexibly to voltage changes, thereby achieving a better reactive power allocation strategy under different operating conditions. The voltage deviation gradient reflects the sensitivity of voltage deviation to reactive power demand; by linearly relating it to the first-order term vector, the optimization process can be ensured to better align with actual voltage regulation requirements.

[0043] Specifically, in the quadratic programming model, the linear term vector is the linear part of the objective function, directly affecting the direction and magnitude of the optimization result. The voltage deviation gradient is the partial derivative of the voltage deviation-reactive demand mapping function with respect to voltage, representing the rate of change of reactive demand when the voltage deviation changes. In this invention, the linear term vector is linearly related to the voltage deviation gradient, meaning that the value of the linear term vector changes according to the voltage deviation gradient. This linear relationship can be achieved through a scaling factor, i.e., the linear term vector equals the scaling factor multiplied by the voltage deviation gradient and then multiplied by the voltage deviation. Here, the scaling factor is the voltage deviation weighting factor, used to adjust the importance of the voltage deviation in the optimization objective. The voltage deviation is the difference between the measured bus voltage value and the target value, reflecting the deviation between the current voltage state and the desired state.

[0044] Furthermore, when constructing the quadratic programming model, the first step is to determine the mapping function between voltage deviation and reactive power demand. This function can be designed based on the actual power grid model and equipment characteristics; it is typically a nonlinear function reflecting the relationship between voltage deviation and reactive power demand. Then, the partial derivative of this function with respect to voltage is calculated to obtain the voltage deviation gradient. Next, based on the voltage deviation gradient and voltage deviation weighting coefficients, a first-order term vector is calculated. For example, if the voltage deviation gradient is large, it indicates that the voltage deviation has a significant impact on reactive power demand. In this case, the weighting coefficients of the voltage deviation can be increased to enhance its weight in the optimization objective, thereby more effectively regulating the voltage. In this way, the quadratic programming model can dynamically adjust the reactive power allocation strategy according to changes in voltage deviation, ensuring the system's voltage stability and the economy of reactive power regulation.

[0045] In some embodiments, in step two: The bus voltage deviation ,in This is the measured value of the bus voltage. The target value for bus voltage. This refers to the bus voltage deviation. The available safety margin The calculation method is as follows: when hour, ; when hour, ,in Let represent the available safety margin of the i-th reactive power regulation device.

[0046] Model predictive control (MPC) effectively addresses system dynamics and uncertainties by optimizing control inputs over a rolling time domain. In minimizing future voltage deviation energy and control input energy, MPC ensures that the system adjusts the output of reactive power regulation equipment in the optimal manner while meeting constraints, thereby achieving stable voltage control and optimized reactive power distribution.

[0047] Specifically, Model Predictive Control (MPC) is a model-based control method that uses a predictive model of the system to predict future system behavior and optimizes the control input within the prediction time domain. In minimizing the future voltage deviation energy and control input energy, the goal of MPC is to minimize the sum of squares of the voltage deviation and the sum of squares of the control input changes over a future period. The voltage deviation energy reflects the difference between the voltage and the target value, while the control input energy reflects the magnitude of the control action. By minimizing these two energies, MPC can reduce drastic changes in control actions while ensuring voltage stability, thereby improving system stability and economy. In implementation, it is necessary to set the prediction time domain length, i.e., the future time range considered by MPC; the control input penalty coefficient, used to balance the importance of voltage deviation and control input changes; and the rolling time domain control step size, i.e., the time step forward after each optimization.

[0048] Furthermore, when constructing the Model Predictive Control (MPC) model, the first step is to determine the system's predictive model, which is typically a linear or nonlinear model built upon the system's physical characteristics and historical data. Input parameters include the current voltage deviation, the current state of the reactive power regulation equipment, and the system's dynamic characteristic parameters. During optimization, the MPC calculates the optimal control quantity for each time step within the predictive time domain and then applies the control quantity of the first time step to the actual system. Subsequently, as the system state is updated, the MPC rolls to the next time step, repeating the optimization process. This rolling optimization approach enables the MPC to respond to dynamic changes in the system in real time, ensuring the real-time performance and effectiveness of the control strategy. Simultaneously, by appropriately setting the predictive time domain length and the control quantity penalty coefficient, control performance can be further optimized, improving the system's voltage stability and the economy of reactive power distribution.

[0049] In some embodiments, in step three: The preliminary allocation ratio ,in >0 indicates that the j-th reactive power regulating device has a usable safety margin in the current regulating direction. Let i be the initial allocation ratio for the i-th reactive power regulating device. The available safety margin for the j-th reactive power regulation device; The total reactive power regulation demand ,in This is the scheduling ratio coefficient. To meet the total reactive power regulation demand.

[0050] Equipment health score is a comprehensive indicator used to assess the current health status of reactive power regulation equipment. By combining multi-source sensor data and machine learning models, the health status of equipment can be assessed in real time and accurately, providing important reference for reactive power allocation strategies. This assessment method can effectively prevent equipment failures and improve system reliability and economy.

[0051] Specifically, equipment health scoring assesses the health status of reactive power regulation equipment by collecting and analyzing data from multiple sensors. Multi-source sensor data includes operating parameters such as temperature, vibration, current, and voltage, which reflect the equipment's operating condition and potential faults. Machine learning models are used to process this data, training the model to identify patterns and trends in the equipment's health status. These models can be Support Vector Machines (SVMs), neural networks, or other suitable algorithms. During implementation, the model needs to be trained using historical data and known equipment states as the training set. After training, the model can be fed real-time multi-source sensor data, and the model will output a health score, typically a value between 0 and 1, with higher values ​​indicating healthier equipment.

[0052] Furthermore, when constructing an equipment health scoring model, the first step is to select appropriate sensors to collect data, ensuring its comprehensiveness and accuracy. Then, the collected data undergoes preprocessing, including data cleaning and normalization, to improve the model's training effectiveness. Next, a suitable machine learning algorithm is selected to build the model, and historical data is used for training and validation. During model training, parameters such as the learning rate and regularization parameters need to be adjusted to optimize performance. After training, the model can be deployed to a real-world system, receiving multi-source sensor data in real time and outputting equipment health scores. This approach enables real-time monitoring and evaluation of equipment health status, providing a scientific basis for reactive power allocation strategies and ultimately improving the overall system performance and reliability.

[0053] In some embodiments, in step four: The objective function of the multi-objective quadratic programming model is:

[0054] in, For each reactive power regulation device, the reactive power increment vector is... Let n be the reactive power increment of the i-th reactive power regulating device, and n be the total number of reactive power regulating devices. Let H be the objective function value of the multi-objective quadratic programming model, H be the coefficient matrix of the quadratic terms, and f be the coefficient vector of the linear terms. This is the weighting coefficient for equipment life cost. For equipment life cost matrix, This is the system reactive power loss weighting coefficient. This is the system's reactive power loss sensitivity matrix. This is the voltage stability weighting coefficient. This is the voltage stability weight matrix. Here, g(V) is the voltage deviation weighting coefficient, and g(V) is the mapping function between voltage deviation and reactive power demand. Let g(V) be the partial derivative of the mapping function g(V) with respect to the voltage V. This refers to the bus voltage deviation. The constraints include:

[0055] in, For equipment life cost matrix, This is the system's reactive power loss sensitivity matrix. This is the voltage stability weighting matrix; Here, g(V) represents the weighting coefficient; g(V) is the mapping function between voltage deviation and reactive power demand. To meet the total reactive power regulation demand, The available safety margin for the i-th reactive power regulation device. Bus voltage deviation The sign function (when) hour, =1; when hour, ;when hour, ).

[0056] Specifically, the data acquisition module is responsible for acquiring bus voltage measurements, reactive power output measurements of each reactive power regulation device, real-time electricity prices, and equipment health status data, which form the basis for subsequent control strategies. The EKF prediction module uses the extended Kalman filter algorithm to predict the reactive power output status of the reactive power regulation devices, outputting predicted reactive power output and prediction error covariance to provide data support for the dynamic limit module. The dynamic limit module generates dynamic upper and lower limits for each reactive power regulation device based on the prediction error covariance and preset risk coefficients, ensuring that the equipment operates within a safe range. The preliminary allocation module calculates the bus voltage deviation, available safety margin, preliminary allocation ratio, and total reactive power regulation demand, and completes the preliminary reactive power allocation. The QP optimization module constructs and solves a multi-objective quadratic programming model, outputting the optimal reactive power increment. The MPC control module optimizes and distributes the optimal reactive power increment through model predictive control in the rolling time domain. The weight adaptive module adjusts the weight coefficients in the multi-objective quadratic programming model based on equipment health scores and real-time electricity prices, enabling the system to automatically optimize towards a safer or more economical direction under different operating conditions. The coordinated operation of these modules ensures efficient system operation and optimized control.

[0057] Furthermore, when constructing the AVC system distribution unit, it is first necessary to ensure the accuracy of the data acquisition module by installing high-precision sensors and measuring equipment to obtain reliable bus voltage and reactive power output measurements. The implementation of the EKF prediction module requires setting appropriate initial parameters and noise covariance based on the characteristics of the reactive power regulation equipment to improve prediction accuracy. In the dynamic limit module, the risk coefficient can be adjusted according to the importance of the equipment and the operating environment to balance equipment safety and utilization. In the preliminary allocation module, the dispatch ratio coefficient can be set according to the actual needs of the power grid to ensure a reasonable allocation of total reactive power regulation demand. In the QP optimization module, the weight coefficients of the objective function can be dynamically adjusted according to the health status of the equipment and real-time electricity prices to achieve multi-objective optimization. In the MPC control module, the prediction time domain length and control penalty coefficient can be optimized according to the dynamic characteristics of the system to improve control response speed and stability. Through the meticulous design and optimization of these modules, efficient, safe, and economical control of the power plant's AVC system can be achieved.

[0058] In some embodiments, in step five: The linear prediction model is as follows:

[0059] Where 'a' is the autoregressive coefficient of the bus voltage deviation. Let be the influence coefficient of the reactive power increment of the i-th reactive power regulation device on the bus voltage deviation, and w(t) be the residual noise at time t. Let be the bus voltage deviation at time t+1. Let be the bus voltage deviation at time t. Let be the reactive power increment of the i-th reactive power regulating device at time t, and n be the total number of reactive power regulating devices; The model prediction control minimizes the objective function within the prediction time domain length N:

[0060] in, Here, N is the control penalty coefficient, N is the prediction time domain length, and k is the prediction step size. Let be the bus voltage deviation at time t+k+1. Let be the reactive power increment vector of each reactive power regulation device at time t+k. This refers to the 2-norm operation.

[0061] Process noise covariance and measurement noise covariance are two important parameters in the extended Kalman filter (AVC), reflecting the uncertainty of the system model and the error of the measurement process, respectively. Adaptive updates through online error statistics improve the filter's adaptability to dynamic system changes, thereby more accurately predicting the reactive power output and its error covariance of reactive power regulation equipment. This adaptive update mechanism ensures that the filter maintains high prediction accuracy under different operating conditions, thus improving the overall performance of the AVC system.

[0062] Specifically, process noise covariance is a parameter describing the uncertainty of the system model. It reflects the uncertainty caused by factors such as inaccurate system models or external interference during the operation of reactive power regulation equipment. Measurement noise covariance, on the other hand, is a parameter describing the measurement process error. It reflects the error of the measuring equipment when measuring reactive power output. Online error statistics refer to the real-time collection of errors between predicted and actual measured values ​​during filter operation, and the adjustment of the noise covariance based on these errors. This method can dynamically reflect the actual operating state of the system, thereby improving the adaptability and accuracy of the filter. In implementation, an error statistics window needs to be set to collect prediction errors over a certain period of time, and the noise covariance is updated based on the statistical characteristics of these errors, such as mean and variance.

[0063] Furthermore, when implementing adaptive updates to the process noise covariance and measurement noise covariance, the size of the error statistics window must first be determined, which typically depends on the system's dynamic characteristics and measurement frequency. An excessively large error statistics window may cause the filter to react sluggishly to rapidly changing system states, while an excessively small window may introduce excessive noise. Next, the process noise covariance and measurement noise covariance are updated by calculating the mean and variance of the prediction errors within the error statistics window. For example, if the error variance is large, it indicates high system uncertainty, and the process noise covariance can be appropriately increased; if the measurement error is large, the measurement noise covariance can be increased. Through this adaptive update mechanism, the filter can better adapt to the dynamic changes of the system, improving the accuracy of reactive power output prediction. This dynamic adjustment not only improves the system's response speed but also enhances its robustness, enabling it to operate stably in complex power grid environments.

[0064] In some embodiments, in step six: The health score The results were obtained based on multi-source sensor data and machine learning model evaluation, among which... Give the health score to the i-th reactive power regulation device; The weighting coefficients are adjusted as follows:

[0065] in, For health-adaptive gain, p(t) is the real-time electricity price at time t. This serves as the benchmark value for weighting equipment life-cycle costs. Let be the life cost weighting coefficient for the i-th reactive power regulation device. This is the system reactive power loss weighting coefficient. This is the voltage deviation weighting coefficient. This indicates a direct proportional relationship.

[0066] The AVC allocation device comprises the following modules: A data acquisition module collects bus voltage measurements, reactive power output measurements from reactive power regulation equipment, real-time electricity prices, and equipment health status data, which form the basis for subsequent control strategies. An EKF prediction module uses an extended Kalman filter algorithm to predict the reactive power output of reactive power regulation equipment, outputting the predicted reactive power output and prediction error covariance, providing data support for the dynamic limit module. The dynamic limit module generates dynamic upper and lower limits for each reactive power regulation device based on the prediction error covariance and preset risk coefficients, ensuring that the equipment operates within a safe range. A preliminary allocation module calculates the bus voltage deviation, available safety margin, preliminary allocation ratio, and total reactive power regulation demand, and completes the preliminary reactive power allocation. A QP optimization module constructs and solves a multi-objective quadratic programming model, outputting the optimal reactive power increment. An MPC control module optimizes and distributes the optimal reactive power increment through model predictive control in the rolling time domain. A weighted adaptive module adjusts the weight coefficients in the multi-objective quadratic programming model based on equipment health scores and real-time electricity prices, enabling the system to automatically optimize towards a safer or more economical direction under different operating conditions. The coordinated operation of these modules ensures efficient system operation and optimized control.

[0067] Furthermore, when constructing the AVC distribution unit, the data acquisition module needs to be equipped with high-precision sensors and measuring devices to ensure data accuracy and reliability. The implementation of the EKF prediction module requires setting appropriate initial parameters and noise covariance based on the characteristics of the reactive power regulation equipment to improve prediction accuracy. In the dynamic limit module, the risk coefficient can be adjusted according to the importance of the equipment and the operating environment to balance equipment safety and utilization. In the preliminary allocation module, the dispatch ratio coefficient can be set according to the actual needs of the power grid to ensure a reasonable allocation of total reactive power regulation demand. In the QP optimization module, the weight coefficients of the objective function can be dynamically adjusted according to the health status of the equipment and real-time electricity prices to achieve multi-objective optimization. In the MPC control module, the prediction time domain length and control penalty coefficient can be optimized according to the dynamic characteristics of the system to improve control response speed and stability. Through the meticulous design and optimization of these modules, efficient, safe, and economical control of the power plant's AVC system can be achieved.

[0068] In some embodiments, the process noise covariance in the extended Kalman filter and measurement noise covariance Adaptive updates are performed based on online error statistics, whereby... Let be the process noise covariance of the i-th reactive power regulation device. Let be the measurement noise covariance of the i-th reactive power regulation device.

[0069] It should be noted that the process noise covariance and measurement noise covariance in the Extended Kalman Filter (EKF) mentioned in this invention are adaptively updated based on online error statistics. This technical feature ensures that the filter can dynamically adapt to system changes. Process noise covariance and measurement noise covariance are two important parameters in the Extended Kalman Filter, reflecting the uncertainty of the system model and the error of the measurement process, respectively. Adaptive updating through online error statistics improves the filter's adaptability to dynamic system changes, thereby more accurately predicting the reactive power output and its error covariance of reactive power regulation equipment. This adaptive update mechanism ensures that the filter maintains high prediction accuracy under different operating conditions, thus improving the performance of the entire AVC system.

[0070] Specifically, process noise covariance is a parameter describing the uncertainty of the system model. It reflects the uncertainty caused by factors such as inaccurate system models or external interference during the operation of reactive power regulation equipment. Measurement noise covariance, on the other hand, is a parameter describing the measurement process error. It reflects the error of the measuring equipment when measuring reactive power output. Online error statistics refer to the real-time collection of errors between predicted and actual measured values ​​during filter operation, and the adjustment of the noise covariance based on these errors. This method can dynamically reflect the actual operating state of the system, thereby improving the adaptability and accuracy of the filter. In implementation, an error statistics window needs to be set to collect prediction errors over a certain period of time, and the noise covariance is updated based on the statistical characteristics of these errors, such as mean and variance.

[0071] Furthermore, when implementing adaptive updates to the process noise covariance and measurement noise covariance, the size of the error statistics window must first be determined, which typically depends on the system's dynamic characteristics and measurement frequency. An excessively large error statistics window may cause the filter to react sluggishly to rapidly changing system states, while an excessively small window may introduce excessive noise. Next, the process noise covariance and measurement noise covariance are updated by calculating the mean and variance of the prediction errors within the error statistics window. For example, if the error variance is large, it indicates high system uncertainty, and the process noise covariance can be appropriately increased; if the measurement error is large, the measurement noise covariance can be increased. Through this adaptive update mechanism, the filter can better adapt to the dynamic changes of the system, improving the accuracy of reactive power output prediction. This dynamic adjustment not only improves the system's response speed but also enhances its robustness, enabling it to operate stably in complex power grid environments.

[0072] like Figure 2 As shown in some embodiments, a power plant automatic voltage control (AVC) system distribution device based on equal margin unidirectional distribution includes: Data acquisition module 201 is used to acquire bus voltage measurement value, reactive power output measurement value of each reactive power regulation device, real-time electricity price and equipment health status data; EKF prediction module 202 is used to perform extended Kalman filter prediction and output predicted reactive power output and prediction error covariance. The dynamic limit module 203 is used to generate the dynamic upper limit and dynamic lower limit of reactive power for each reactive power regulation device based on the prediction error covariance and the preset risk coefficient. The preliminary allocation module 204 is used to calculate the bus voltage deviation, available safety margin, preliminary allocation ratio and total reactive power regulation demand, and to complete the preliminary reactive power allocation. QP optimization module 205 is used to construct and solve a multi-objective quadratic programming model and output the optimal reactive power increment. MPC control module 206 is used to optimize control through model prediction in the rolling time domain and issue the optimal reactive power increment. The weight adaptive module 207 is used to adjust the weight coefficients in the multi-objective quadratic programming model based on the equipment health score and the real-time electricity price.

[0073] It is understandable that the modules described in the power plant automatic voltage control (AVC) system distribution device based on equal margin and unidirectional distribution are similar to those in the reference system. Figure 1The steps described correspond to those in the power plant automatic voltage control (AVC) system allocation method based on equal margin and unidirectional allocation. Therefore, the operations, features, and beneficial effects described above for the power plant automatic voltage control (AVC) system allocation method based on equal margin and unidirectional allocation are also applicable to the power plant automatic voltage control (AVC) system allocation device and its components based on equal margin and unidirectional allocation, and will not be repeated here.

[0074] The above description is merely an explanation of some preferred embodiments of the present invention and the technical principles employed. Those skilled in the art should understand that the scope of the invention as described in the embodiments of the present invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A power plant automatic voltage control (AVC) system distribution method based on equal-marginal co-directional distribution, characterized in that, The method comprises the following steps: Step one, based on extended Kalman filter, the reactive power output of each reactive power regulation device and error covariance are predicted, and the dynamic reactive power upper limit and dynamic reactive power lower limit of each reactive power regulation device are generated according to the risk coefficient; Step two, according to the deviation of bus voltage measurement value and bus voltage target value, the available safety margin of each reactive power regulation device in the current regulation direction is calculated; Step three, the preliminary allocation proportion of each reactive power regulation device is calculated based on the available safety margin, and preliminary reactive power allocation is carried out according to the total reactive power regulation demand; Step four, a multi-objective quadratic programming model is constructed with device life cost, system reactive power loss and voltage stability as optimization objectives, and the optimal reactive power increment is solved under the constraint conditions that the total reactive power regulation demand is met, the reactive power increment of each reactive power regulation device does not exceed the available safety margin, and the regulation direction is consistent; Step five, based on the linear prediction model, the optimal reactive power increment is optimized and issued in real time through model predictive control in the rolling time domain; Step six, the weight coefficient in the multi-objective quadratic programming model is adjusted adaptively according to the health score of each reactive power regulation device and the real-time electricity price.

2. The method of claim 1, wherein, The step one comprises: The reactive power output state of the ith reactive power regulation device is predicted by using an extended Kalman filter to obtain a predicted reactive power output and a prediction error covariance wherein is the predicted reactive power output of the ith reactive power regulation device, is the prediction error covariance of the ith reactive power regulation device; based on the prediction error covariance and a preset risk coefficient , the dynamic reactive power upper limit and the dynamic reactive power lower limit , the calculation formula is: wherein, is the rated minimum reactive power output of the i-th reactive power regulating device, is the rated maximum reactive power output of the i-th reactive power regulating device, is the preset risk coefficient of the i-th reactive power regulating device, is the dynamic upper limit of reactive power of the i-th reactive power regulating device, is the dynamic lower limit of reactive power of the i-th reactive power regulating device.

3. The method of claim 2, wherein, The process of the extended Kalman filter comprises: Time update: wherein, is the process noise covariance of the ith reactive power regulating device, is the process noise covariance of the ith reactive power regulating device, is the prior predictive reactive power output of the ith reactive power regulating device at time t, t is the current time, is the prediction period, is the process noise covariance of the ith reactive power regulating device, is the prior predictive error covariance of the ith reactive power regulating device at time t, is the predictive error covariance of the ith reactive power regulating device at time t. Observation update: wherein, is the measurement noise covariance of the i-th reactive power regulating device, is the Kalman gain of the i-th reactive power regulating device, is the reactive power output measurement of the i-th reactive power regulating device, is the is the posterior predictive reactive power output of the i-th reactive power regulating device at time instant, is the is the posterior predictive error covariance of the i-th reactive power regulating device at time instant.

4. The method of claim 1, wherein, In the step two: bus voltage deviation wherein is a bus voltage measurement value, is a bus voltage target value, is a bus voltage deviation; The available safety margin is calculated as When Time, ; When Time, where is the available safety margin for the ith reactive power regulating device.

5. The method of claim 4, wherein, In the step three: said preliminary allocation ratio wherein > 0 indicates that the jth reactive power regulating device has an available safety margin in the current regulating direction, is a preliminary allocation ratio for the ith reactive power regulating device, is an available safety margin for the jth reactive power regulating device; the total reactive power regulation demand wherein is a dispatching proportionality coefficient, is the total reactive power regulation demand.

6. The method of claim 1, wherein, In the step four: The objective function of the multi-objective quadratic programming model is: wherein, is the reactive power increment vector of each reactive power regulating device, is the reactive power increment of the i-th reactive power regulating device, and n is the total number of reactive power regulating devices, is the objective function value of the multi-objective quadratic programming model, H is the quadratic term coefficient matrix, and f is the linear term coefficient vector, is the device life cost weight coefficient, is the device life cost matrix, is the system reactive power loss weight coefficient, is the system reactive power loss sensitivity matrix, is the voltage stability weight coefficient, is the voltage stability weight matrix, is the voltage deviation weight coefficient, and g(V) is the voltage deviation and reactive power demand mapping function, is the partial derivative of the mapping function g(V) with respect to the voltage V, is the bus voltage deviation; The constraint conditions comprise: wherein, is the device lifetime cost matrix, is the system reactive loss sensitivity matrix, is the voltage stability weight matrix; is the weight coefficient; g(V) is the voltage deviation and reactive demand mapping function, is the total reactive regulation demand, is the available safety margin of the ith reactive regulation device, is the bus voltage deviation is the sign function (when , = 1; when , ; when , ).

7. The method of claim 1, wherein, In the step five: The linear prediction model is: wherein a is an autoregressive coefficient of bus voltage deviation, is the influence coefficient of reactive power increment of the ith reactive power regulating device on bus voltage deviation, and w(t) is residual noise at time t, is bus voltage deviation at time t+1, is bus voltage deviation at time t, is reactive power increment of the ith reactive power regulating device at time t, and n is the total number of reactive power regulating devices. The model predictive control minimizes the objective function in the prediction time domain length N: wherein, is a control quantity penalty coefficient, N is a prediction time domain length, k is a prediction step, is a bus voltage deviation at t+k+1, is a reactive power increment vector of each reactive power regulating device at t+k, is a two-norm operation.

8. The method of claim 1, wherein, In the step six: the health score is evaluated based on multi-source sensing data and a machine learning model, wherein is the health score of the i-th reactive power regulation device; The adjustment mode of the weight coefficient is: wherein, is the health adaptive gain, p(t) is the real-time electricity price at time t, is the baseline value of the equipment life cost weight, is the life cost weight coefficient of the i-th reactive power regulating device, is the system reactive power loss weight coefficient, is the voltage deviation weight coefficient, is expressed in a proportional relationship.

9. The method of claim 2 or 3, wherein, process noise covariance in the extended Kalman filter and measurement noise covariance Adaptive update according to online error statistics, where is the process noise covariance for the i-th reactive power regulating device, is the measurement noise covariance for the i-th reactive power regulating device.

10. An automatic voltage control (AVC) system distribution device for implementing the method of any one of claims 1 to 9, characterized in that, Comprise: The data acquisition module is used for acquiring bus voltage measurement value, reactive power output measurement value of each reactive power regulation device, real-time electricity price and device health state data; The EKF prediction module is used for executing extended Kalman filter prediction, and outputting predicted reactive power output and predicted error covariance; The dynamic limit value module is used for generating the dynamic reactive power upper limit and the dynamic reactive power lower limit of each reactive power regulation device according to the predicted error covariance and the preset risk coefficient; The preliminary allocation module is used for calculating bus voltage deviation, available safety margin, preliminary allocation proportion and total reactive power regulation demand, and completing preliminary reactive power allocation; The QP optimization module is used for constructing and solving the multi-objective quadratic programming model, and outputting the optimal reactive power increment; The MPC control module is used for optimizing and issuing the optimal reactive power increment through model predictive control in the rolling time domain; The weight adaptive module is used for adjusting the weight coefficient in the multi-objective quadratic programming model according to the device health score and the real-time electricity price.

Images