Power distribution network voltage adaptive model predictive control method based on fuzzy weighting factor

Abstract

The application discloses a power distribution network voltage adaptive model predictive control method based on fuzzy weighting factors and belongs to the technical field of power system operation and control. In view of the reactive power and voltage optimization problem of the power distribution network caused by high-proportion distributed power output fluctuation, firstly, an online evaluation mechanism of model prediction reliability is established, a model fidelity factor is introduced, and the confidence degree of the prediction result is quantitatively evaluated in real time through a sliding time window; a global voltage deviation risk index considering the fidelity and a node regulation potential index integrating the node sensitivity and the equipment fatigue degree are constructed; the fuzzy weighting factor is generated based on the above index, and then a node differentiated weighted MPC rolling optimization model considering the physical constraint is constructed, and the rolling solution is used to guide the reactive power output optimization of each inverter. The method realizes the dynamic switching of the prospective prediction control and the robustness feedback control, improves the node voltage level, considers the compensation equipment operation life, and improves the voltage robustness and regulation economy.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and control technology, specifically relating to a distribution network voltage adaptive model predictive control method based on fuzzy weighted factors, which is particularly suitable for real-time voltage regulation and reactive power optimization of distribution network nodes under high proportion of distributed power generation access. Background Technology

[0002] The penetration rate of distributed generation (DG) power sources such as photovoltaics and wind power in distribution networks is increasing. However, the intermittent and random nature of DG output leads to complex and variable power flow in distribution networks, which can easily cause power quality problems such as node voltage exceeding limits and severe voltage fluctuations. Traditional voltage regulation methods, such as on-load tap changers (OLTCs) and switched capacitor banks (SCBs), are limited in the number of operations and have a slow response speed, making it difficult to cope with rapid voltage fluctuations on the order of seconds or minutes.

[0003] Therefore, utilizing the rapid reactive power regulation capability of grid-connected inverters from distributed generation sources for voltage control has become a current research hotspot. Among these methods, Model Predictive Control (MPC) is widely used in distribution network voltage control due to its ability to effectively handle multivariate constraints and rolling optimization characteristics. Although existing MPC methods have improved voltage levels to some extent, they still face the following key technical bottlenecks in practical applications: First, distorted predictive models lead to decreased control performance. Traditional MPC (Multi-Level Control) relies heavily on the accuracy of predictive models. In actual operation, environmental factors such as cloud cover and sudden wind speed changes often result in significant errors in source load prediction data. When the predictive model mismatches with the actual physical process, open-loop optimization commands based on the predictive data may lead to control overshoot or even system oscillations. Most existing technologies assume model accuracy or employ overly conservative robust control at the expense of economic efficiency, lacking a mechanism for dynamically adjusting the control strategy based on predictive reliability.

[0004] Second, the weighting coefficients are difficult to tune and lack adaptability. The objective function of MPC typically includes voltage deviation penalty terms and control action cost penalty terms, and the weighting coefficients between the two directly determine the control effect. Existing methods mostly use fixed empirical weights or offline trial-and-error methods, which cannot adapt to the time-varying operating conditions of distribution networks. When the system is in a high-risk state of voltage exceeding limits, voltage safety should be prioritized; while when the system is stable, the focus should be on reducing control costs. Fixed weights cannot balance these two opposing needs, resulting in the distribution network's reactive power resources and voltage distribution not reaching an optimal state.

[0005] Third, it fails to consider individual equipment differences and operational lifespan management. Existing reactive power optimization strategies typically aim to minimize the overall network's mathematical objective, often resulting in inverters with high electrical position sensitivity remaining in a state of high-frequency, large-amplitude regulation for extended periods, while other inverters output insufficient power. This control method ignores the differences in regulation potential and equipment fatigue at each node, leading to premature aging or failure of some critical equipment and reducing the operational reliability and economy of the entire power distribution system.

[0006] In summary, a distribution network voltage control method that can sense the fidelity of the prediction model in real time, adaptively adjust the control weights, and take into account both voltage quality and equipment lifespan can effectively make up for the shortcomings of existing methods. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing a distribution network voltage adaptive model predictive control method based on fuzzy weighted factors. This method solves the problems of large voltage fluctuations, poor robustness of prediction mismatch, inability to adapt fixed weights to operating conditions, and rapid lifespan decay due to uneven equipment adjustment under high-proportion distributed power source access.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows: A voltage adaptive model predictive control method for distribution networks based on fuzzy weighted factors includes the following steps: S1: Based on the day-ahead source-load forecast data, solve the optimal power flow problem with the goal of minimizing the overall operating cost, obtain the day-ahead scheduling plan of the on-load tap changer (OLTC) and the switching capacitor bank (CB), and generate the voltage reference benchmark sequence of the entire network nodes; based on this, construct the intraday voltage rolling forecast equation using the linearized sensitivity model, and establish a two-layer collaborative control framework of "day-ahead benchmark setting and intraday tracking optimization"; S2: In the intraday real-time control cycle k, the reliability of the prediction model is first evaluated online by introducing the model fidelity factor. Based on the model fidelity factor, the global voltage deviation risk index and the node regulation potential index are calculated respectively. The two indices are used as inputs to the preset 5×5 fuzzy logic controller to generate the exclusive fuzzy weighting factor for each node. Then, an MPC quadratic programming QP model that takes into account the node differential weights is constructed to solve for the optimal reactive power increment sequence of the distributed power source that satisfies the physical constraints. S3: Based on the rolling time-domain principle of model predictive control, select the first component of the optimal reactive power increment sequence and send it to each distributed power inverter for execution; at the next control time k+1, based on the latest collected measured node voltage value and the initial conditions of the distributed power power state correction prediction model, repeat steps S2 and S3 to realize closed-loop feedback adaptive control of the distribution network voltage.

[0009] Furthermore, the specific method for generating the network-wide node voltage reference sequence in step S1 is as follows: 1) Construct an objective function for day-ahead layer optimization with the goal of minimizing the overall system operating cost throughout the day; 2) Clarify the specific calculation method for each cost item in the objective function, including the cost of mechanical equipment switching actions and the cost of reactive power compensation for distributed power sources; 3) Set day-ahead optimization constraints to ensure that the dispatch plan meets the requirements for safe operation of the power grid; 4) Solve the mixed-integer linear programming (MILP) problem corresponding to the above optimization to obtain the optimal day-ahead scheduling plan for OLTC and CB. The reference voltage sequence for each node over the entire 24 hours was calculated by combining the power flow equations. This is used as a reference value for the intraday MPC optimization objective function.

[0010] Furthermore, the objective function and cost term for the day-ahead optimization are as follows: The objective function expression is:

[0011] Where T=24 is the day-ahead optimization period, and ΔT=1h is the optimization step size. Let t be the marginal price of grid-loss electricity. For the total active power loss of the system, Cost of switching operations on mechanical equipment, The cost of reactive power compensation provided for distributed generation (DG); The Including on-load tap changer (OLTC) tap changer Number of parallel capacitor banks (SCB) in operation Adjustment costs:

[0012] The Characterize the economic cost of distributed generation providing reactive power compensation:

[0013] Where a is the secondary cost coefficient, which represents the rate of increase in internal losses of the inverter due to the increase in reactive current; b is the primary cost coefficient, which represents the inherent marginal cost that is proportional to reactive power output.

[0014] Furthermore, the recently optimized constraints specifically include: 1) Power flow constraints: Based on the linearized sensitivity matrix, the linear influence of source load power injection changes on node voltage is characterized; 2) Voltage safety constraints: Limiting the voltage at each node to within the permissible safe operating range throughout the entire time period; 3) Equipment operation frequency constraints: The number of daily adjustments to the OLTC tap changer and the number of daily switching operations of the parallel capacitor bank are limited to a preset upper limit; 4) DG reactive power capacity constraint: The reactive power output of the distributed generation inverter is limited to its rated capacity range.

[0015] Furthermore, the formula for calculating the model fidelity factor in step S2 is as follows:

[0016] in, The measured voltage value at time k. This represents the predicted value from the previous time step to the current time step. To prevent tiny constants with a denominator of zero; The moving standard deviation of the prediction error is calculated from the historical prediction error sequence within the moving time window. It is used to characterize the severity of recent power grid environment fluctuations and to achieve adaptive perception of environmental uncertainties.

[0017] Where L is the length of the sliding window. This is a historical prediction error sequence.

[0018] Furthermore, the calculation methods for the global voltage deviation risk index and the node regulation potential index in step S2 are as follows: 1) Global voltage deviation risk index, i.e., voltage deviation severity index:

[0019] in, The prediction domain risk term includes both the magnitude and rate of change (acceleration) of the voltage deviation. This is the real-time deviation term at the current moment; through Achieve dynamic soft switching between forward-looking control and robust feedback control; 2) Node adjustment potential index:

[0020] in, Let be the absolute value of the voltage-reactive power sensitivity at node i; The rated reactive power capacity of the inverter; This represents the number of times the device can be significantly adjusted within the current sliding window. This is the preset upper limit threshold for the number of actions.

[0021] Furthermore, the specific method for generating the fuzzy weighting factor in step S2 is as follows: 1) Fuzzification: Using a triangular membership function, the global voltage deviation risk index and the node regulation potential index are mapped to five fuzzy subsets {NB, NS, ZO, PS, PB} respectively; 2) Fuzzy Inference: Inference is performed based on a preset 5×5 fuzzy rule matrix. The inference logic is as follows: when the voltage deviation risk is high and the node regulation potential is high, a larger weighting factor is output; when the node regulation potential is low, a smaller weighting factor is output. 3) Defuzzification: The centroid method is used to calculate the precise fuzzy weighting factor corresponding to each node.

[0022] Furthermore, the objective function of the MPC quadratic programming model that takes into account the node differentiation weights in step S2 is constructed as follows:

[0023] Where P is the prediction time domain and M is the control time domain; This is the predicted voltage value for the p-th step in the future. This refers to the day-ahead reference voltage issued in Step 1; This is the predicted reactive power adjustment increment.

[0024] Furthermore, the physical constraints of the MPC quadratic programming model in step S2 specifically include: 1) Inverter capacity limit: The reactive power output of the inverter at any given moment, combined with the current active power output, shall not exceed its rated apparent power range. 2) Regulation rate limit: Limits the upper and lower limits of the change in single-step reactive power regulation to prevent the instantaneous jump in inverter output from impacting the power grid; 3) Voltage safety range: Ensure that the predicted voltage of all nodes in the prediction time domain is within the safe operating range of 0.95-1.05 pu.

[0025] Furthermore, the specific process of closed-loop feedback correction and rolling execution in step S3 is as follows: obtain the optimal reactive power increment sequence obtained in step S2, select only the first component in the sequence as the actual control command and issue it to each distributed power source inverter for execution; at the next sampling time k+1, collect the latest measured voltage values ​​of each node in the distribution network and the active power output status of the distributed power source, use the measured values ​​to correct the initial state of the prediction model, and shift the prediction time window backward by one control step, re-trigger the rolling optimization calculation in step S2, and realize continuous closed-loop adaptive control.

[0026] Beneficial effects: This invention introduces a model fidelity factor, which enables online quantitative evaluation of the reliability of the prediction model. It can automatically achieve dynamic soft switching between forward-looking feedforward control and robust feedback control based on the prediction accuracy, solving the problems of decreased control performance and insufficient robustness when traditional MPC prediction mismatch occurs, and significantly improving the stability of voltage control in scenarios with strong uncertainty.

[0027] This invention is based on two indicators: voltage deviation risk and node regulation potential. It generates node-differentiated weighting factors online through fuzzy logic, eliminating the need for manual offline weight setting. It can adaptively match the time-varying operating conditions of the distribution network, solving the problem that fixed weights cannot simultaneously take into account voltage safety and regulation economy, and achieving optimal allocation of reactive power resources.

[0028] This invention comprehensively considers voltage sensitivity, inverter remaining capacity, and equipment fatigue in the node regulation potential index, which can evenly distribute the regulation tasks of each node, avoid the premature aging problem caused by long-term frequent operation of highly sensitive inverters, effectively extend the operating life of compensation equipment, and improve the operational reliability of the power distribution system.

[0029] This invention adopts a two-layer collaborative control framework of "setting a benchmark day-ahead and tracking and optimizing intraday", which takes into account the global economy of day-ahead scheduling and the rapid response capability of intraday rolling optimization. While ensuring voltage safety, it effectively reduces system network loss and overall operating costs, and its overall performance is significantly better than traditional voltage control methods. Attached Figure Description

[0030] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the distribution network voltage adaptive model predictive control method based on fuzzy weighting factors as described in an embodiment of the present invention.

[0031] Figure 2 This is a comparison curve of the voltage envelope of the entire network under complex operating conditions, comparing the present invention with that of traditional methods. Figure 2 As shown, conventional control exhibits slow response and severe voltage divergence exceeding limits, reaching as high as 1095 times; the fixed-weight MPC strategy is rigid and prone to local oscillations, occurring 539 times. In contrast, this invention, relying on model fidelity and a fuzzy adaptive weighting mechanism, clamps the average voltage of the entire network near the benchmark, with the envelope converging well within the safe range, reducing the number of exceedances to 429 times. This intuitively and powerfully demonstrates the superior voltage control accuracy and strong disturbance rejection robustness of this invention.

[0032] Figure 3 This is a bar chart comparing the average daily adjustment frequency of the inverter using the present invention and the traditional method.

[0033] like Figure 3 As shown, traditional strategies overuse DG1, DG2, and DG4, while DG3 is almost idle. This invention, relying on an adaptive weighted mechanism, proactively and smoothly transfers the adjustment burden to the highly capable DG3. Although the number of DG3 actions slightly increases to three, it successfully results in a significant reduction in the losses of other critical nodes. The distribution of actions across the entire network becomes more balanced, the total number of losses is significantly reduced, and a better balance is achieved between efficient voltage management and full lifecycle protection of equipment.

[0034] Figure 4 This is a comparison chart of the total active power loss index of the present invention with other strategies throughout the day.

[0035] like Figure 4 As shown, conventional control suffers from the highest network loss due to limited reactive power support capacity; after introducing MPC, the network loss is significantly reduced to 1201.1 kWh. Building upon this, this invention further optimizes the reactive power output distribution of the inverter cluster through fuzzy adaptive weighting, promoting local reactive power balance and reducing total losses to 1106.9 kWh. The data objectively demonstrates that this method, while mitigating voltage fluctuations, reduces reactive power flow transmission through reasonable reactive power allocation, effectively improving the overall operational economy of the system. Detailed Implementation

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0037] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] Example 1 See Figure 1-4 A voltage adaptive model predictive control method for distribution networks based on fuzzy weighted factors includes the following steps: S1: Based on the day-ahead source-load forecast data, solve the optimal power flow problem with the goal of minimizing the overall operating cost, obtain the day-ahead scheduling plan for on-load tap changers (OLTC) and switched capacitor banks (CB), and generate a network-wide node voltage reference benchmark sequence; based on this, construct the intraday voltage rolling forecast equation using a linearized sensitivity model, and establish a two-layer collaborative control framework of "day-ahead benchmark setting and intraday tracking optimization"; S2: In the intraday real-time control cycle k, the reliability of the prediction model is first evaluated online by introducing the model fidelity factor. Based on the model fidelity factor, the global voltage deviation risk index and the node regulation potential index are calculated respectively. The two indices are fed into the preset 5×5 fuzzy logic controller to generate the exclusive fuzzy weighting factor for each node. Then, an MPC quadratic programming (QP) model that takes into account the node differential weights is constructed to solve for the optimal reactive power increment sequence of distributed power sources that meets the physical constraints. S3: Based on the rolling time-domain principle of model predictive control, select the first component of the optimal reactive power increment sequence and send it to each distributed power inverter for execution; at the next control time k+1, based on the latest collected measured node voltage value and the initial conditions of the distributed power power state correction prediction model, repeat steps S2 and S3 to realize closed-loop feedback adaptive control of the distribution network voltage.

[0039] This embodiment constructs a two-layer collaborative control architecture of "setting a benchmark before the day and tracking and optimizing within the day"; within the day, the prediction accuracy is evaluated by the model fidelity factor, the voltage deviation risk and node adjustment potential are calculated, a weighting factor is generated by fuzzy logic, and a differentiated weighted model predictive control (MPC) model is constructed to solve for the optimal reactive power increment; instructions are issued and closed-loop correction is performed according to the rolling time domain principle.

[0040] This embodiment forms a complete adaptive voltage control closed loop, realizing dynamic switching between predictive feedforward and feedback control. It comprehensively solves problems such as voltage fluctuation, exceeding limits, and uneven equipment adjustment under high proportion of distributed power supply access, and significantly improves control robustness and adaptability.

[0041] In a specific example, the method for generating the network-wide node voltage reference sequence in step S1 is as follows: 1) Construct an objective function for day-ahead layer optimization with the goal of minimizing the overall system operating cost throughout the day; 2) Clarify the specific calculation method for each cost item in the objective function, including the cost of mechanical equipment switching actions and the cost of reactive power compensation for distributed power sources; 3) Set day-ahead optimization constraints to ensure that the dispatch plan meets the requirements for safe operation of the power grid; 4) Solve the mixed-integer linear programming (MILP) problem corresponding to the above optimization to obtain the optimal day-ahead scheduling plan for OLTC and CB. The reference voltage sequence for each node over the entire 24 hours was calculated by combining the power flow equations. This is used as a reference value for the intraday MPC optimization objective function.

[0042] This embodiment generates a network-wide node voltage reference sequence by establishing a day-ahead optimization objective, clarifying cost composition, setting operational constraints, and solving a mixed-integer linear programming problem.

[0043] This embodiment provides a stable and economical voltage reference target for rapid intraday rolling optimization, avoiding blind adjustment without a reference and ensuring the economy and rationality of overall control.

[0044] In a specific example, the objective function and cost term for day-ahead layer optimization are as follows: The objective function expression is:

[0045] Where T=24 is the day-ahead optimization period, and ΔT=1h is the optimization step size. Let t be the marginal price of grid-loss electricity. For the total active power loss of the system, Cost of switching operations on mechanical equipment, The cost of reactive power compensation provided for distributed generation (DG); The Including on-load tap changer (OLTC) taps Number of parallel capacitor banks (SCB) in operation Adjustment costs:

[0046] The Characterize the economic cost of distributed generation providing reactive power compensation:

[0047] Where a is the secondary cost coefficient, which represents the rate of increase in internal losses of the inverter due to the increase in reactive current; b is the primary cost coefficient, which represents the inherent marginal cost that is proportional to reactive power output.

[0048] This embodiment clarifies the day-ahead optimization objective function, comprehensively considering network loss costs, equipment operation costs, and distributed power source reactive power compensation costs to minimize the total daily operating cost. This allows for precise quantification of system operating costs, enabling the day-ahead scheduling plan to balance low network loss, fewer equipment operations, and low reactive power compensation costs, thereby improving overall operational economy.

[0049] In a specific instance, the constraints for day-ahead optimization include: 1) Power flow constraints: Based on the linearized sensitivity matrix, the linear influence of source load power injection changes on node voltage is characterized;

[0050] in, For each node voltage vector; The voltage setting value for the root node of the distribution network; and These are the voltage sensitivity matrices to active and reactive power, respectively, used to characterize the linearization effect of changes in source load power injection on node voltage. and These are the active power output prediction vector and the node active power load prediction vector of the distributed generation at time t, respectively. and These are the reactive power output decision vector of the distributed power inverter and the reactive power switching vector of the parallel capacitor bank at time t, respectively. Let be the node reactive load prediction vector at time t; 2) Voltage safety constraints: Limiting the voltage at each node to within the permissible safe operating range throughout the entire time period;

[0051] 3) Equipment operation frequency constraints: The number of daily adjustments to the OLTC tap changer and the number of daily switching operations of the parallel capacitor bank are limited to a preset upper limit;

[0052]

[0053] 4) DG reactive power capacity constraint: The reactive power output of the distributed generation inverter is limited to its rated capacity range. .

[0054] This embodiment sets power flow constraints, voltage safety constraints, equipment operation frequency constraints, and distributed power source reactive power capacity constraints in the day-ahead optimization, thereby ensuring that the day-ahead dispatch plan meets the requirements for safe grid operation, does not exceed limits, does not exceed equipment capacity, and does not operate frequently, thus possessing engineering feasibility.

[0055] In a specific example, the formula for calculating the model fidelity factor in step S2 is:

[0056] in, The measured voltage value at time k. This represents the predicted value from the previous time step to the current time step. To prevent tiny constants with a denominator of zero; The moving standard deviation of the prediction error is calculated from the historical prediction error sequence within the moving time window. It is used to characterize the severity of recent power grid environment fluctuations and to achieve adaptive perception of environmental uncertainties.

[0057] Where L is the length of the sliding window. This is a historical prediction error sequence.

[0058] This embodiment calculates the model fidelity factor online based on the error between the measured and predicted voltage values ​​and the moving standard deviation of the prediction error, and quantifies the reliability of the prediction model in real time, thereby achieving adaptive perception of prediction accuracy. This provides a basis for dynamic soft switching between predictive control and feedback control, and significantly improves control stability in prediction mismatch scenarios.

[0059] In a specific example, the calculation methods for the global voltage deviation risk index and the node regulation potential index in step S2 are as follows: 1) Global voltage deviation risk index, i.e., voltage deviation severity index:

[0060] in, The prediction domain risk term includes both the magnitude and rate of change (acceleration) of the voltage deviation. This is the real-time deviation term at the current moment; through Achieve dynamic soft switching between forward-looking control and robust feedback control; When the model has high recent prediction accuracy, the coefficients It will approach 1. The formula becomes The system will primarily rely on a predictive feedforward mechanism to drive future actions. When environmental changes occur abruptly (such as a sudden shift in photovoltaic output due to cloud cover), the model predictions will show significant bias, and the coefficients will... It will approach 0, and the formula becomes The system will automatically switch to the real-time feedback mode, and ensure stable operation of the system through dynamic compensation for deviations.

[0061]

[0062] Normalization formula:

[0063] The first term represents the risk of amplitude deviation, and the second term represents the risk of rate of change. , To adjust the weights.

[0064] 2) Node adjustment potential index:

[0065] in, Let be the absolute value of the voltage-reactive power sensitivity at node i; The rated reactive power capacity of the inverter; This represents the number of times the device can be significantly adjusted within the current sliding window. This is the preset upper limit threshold for the number of actions.

[0066] Normalization process:

[0067] Node voltage sensitivity can be obtained from the inverse of the Jacobian matrix, which is derived from offline power flow calculations. The specific calculation process is as follows: power flow calculations using Newton's method yield the following formula.

[0068]

[0069] The voltage sensitivity matrix to reactive power can be obtained:

[0070] Then the sensitivity matrix It can be obtained by the following formula .

[0071] This embodiment constructs a voltage deviation severity index that integrates predicted risk and real-time deviation, as well as a node regulation potential index that integrates sensitivity, remaining capacity, and equipment fatigue. This allows for the accurate identification of high-risk voltage nodes and nodes with regulation capabilities, providing scientific indicators to support differentiated weighted control.

[0072] In a specific example, the method for generating the fuzzy weighting factor in step S2 is as follows: 1) Fuzzification: Using a triangular membership function, the global voltage deviation risk index and the node regulation potential index are mapped to five fuzzy subsets {NB, NS, ZO, PS, PB} respectively; The membership function expression for a triangle is:

[0073] Where a, b, and c are the left boundary, center point, and right boundary of the fuzzy set, respectively, and The domain of discourse is [-1,1]. The domain of discourse is [0,1].

[0074] 2) Fuzzy Inference: Inference is performed based on a preset 5×5 fuzzy rule matrix. The inference logic is as follows: when the voltage deviation risk is high and the node regulation potential is high, a larger weighting factor is output; when the node regulation potential is low, a smaller weighting factor is output. 3) Defuzzification: The centroid method is used to calculate the precise fuzzy weighting factor corresponding to each node.

[0075] It should be noted that the fuzzy weighting factor The formula for calculation is:

[0076] Where M is the total number of activation rules, Let m be the trigger strength of the m-th rule. To output the center value of the membership function, To mitigate voltage deviation risk, To adjust the potential of nodes.

[0077] The preset 5×5 fuzzy rule matrix is ​​as follows: .

[0078] This embodiment uses fuzzification, fuzzy inference, and centroid method to defuzzify and generate node-specific fuzzy weighting factors online. This embodiment does not require manual offline weight adjustment and can adapt to the time-varying operating conditions of the distribution network to achieve balanced regulation of "prioritizing control of high-risk nodes and minimizing the operation of fatigued equipment".

[0079] In a specific example, the objective function of the MPC quadratic programming model that takes into account the node differentiation weights in step S2 is constructed as follows:

[0080] Where P is the prediction time domain and M is the control time domain; This is the predicted voltage value for the p-th step in the future. This refers to the day-ahead reference voltage issued in Step 1; This is the predicted reactive power adjustment increment.

[0081] This embodiment introduces a fuzzy weighting factor into the model predictive control objective function, while minimizing voltage deviation and reactive power regulation increment. This embodiment dynamically balances voltage tracking accuracy and control stability, avoids control overshoot and frequent actions, and improves voltage control quality.

[0082] In a specific example, the physical constraints of the MPC quadratic programming model in step S2 include: 1) Inverter capacity limit: The reactive power output of the inverter at any given moment, combined with the current active power output, shall not exceed its rated apparent power range. Specifically, it must satisfy: ,in, The rated apparent power of the inverter, This is the reactive power baseline value at the current moment.

[0083] 2) Regulation rate limit: Limits the upper and lower limits of the change in single-step reactive power regulation to prevent the instantaneous jump in inverter output from impacting the power grid; ,in, This represents the change amount adjusted in a single step.

[0084] 3) Voltage safety range: Ensure that the predicted voltage of all nodes in the prediction time domain is within the safe operating range of 0.95-1.05 pu.

[0085] ,in, For predicting voltage.

[0086] In this embodiment, inverter capacity constraints, regulation rate constraints, and voltage safety constraints are set in model predictive control to ensure that control commands meet the physical limitations of the equipment and the requirements for safe operation of the power grid, without impacting the power grid, exceeding limits, or damaging the equipment.

[0087] In a specific example, the closed-loop feedback correction and rolling execution process in step S3 is as follows: obtain the optimal reactive power increment sequence obtained in step S2, select only the first component in the sequence as the actual control command and issue it to each distributed power source inverter for execution; at the next sampling time k+1, collect the latest measured voltage values ​​of each node in the distribution network and the active power output status of the distributed power source, use the measured values ​​to correct the initial state of the prediction model, and shift the prediction time window backward by one control step, re-trigger the rolling optimization calculation in step S2, and realize continuous closed-loop adaptive control.

[0088] It should be noted that the optimal reactive power increment sequence can be represented as: ,in, This is the first component in the sequence.

[0089] In this embodiment, only the first component of the optimal control sequence is executed. At the next moment, the latest measured data is used to correct the prediction model and perform rolling optimization, thereby forming a real-time closed-loop feedback mechanism to continuously correct prediction errors, quickly adapt to the output fluctuations of distributed power sources, and further improve control accuracy and stability.

[0090] To verify the effectiveness of this invention, simulation verification was conducted in the IEEE 33-node distribution network system. A photovoltaic power output fluctuation scenario was set up, and the method of this invention was compared with the traditional fixed-weight MPC method and the conventional OLTC+capacitor bank voltage control method. The core comparison data are shown in the table below:

[0091] Simulation results show that the method of the present invention can effectively suppress voltage fluctuations and limit exceedances caused by high proportion of distributed power supply access. It is significantly better than traditional methods in terms of voltage control accuracy, equipment life protection, system economy and disturbance rejection robustness, and has outstanding practicality and advanced features.

[0092] To verify the independent contribution of each innovative module in this invention, four comparison schemes were set up: Case 1: Conventional voltage control (without MPC, without fuzzy weighting) Case 2: Traditional fixed-weight MPC (no fidelity, no fuzzy weighting) Case 3: MPC + Model Fidelity (without fuzzy weighting) Case 4: The method of this invention (MPC + fidelity + fuzzy weighting)

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 voltage adaptive model predictive control method for distribution networks based on fuzzy weighted factors, characterized in that, Includes the following steps: S1: Based on the day-ahead source-load forecast data, solve the optimal power flow problem with the goal of minimizing the overall operating cost, obtain the day-ahead scheduling plan of the on-load tap changer (OLTC) and the switching capacitor bank (CB), and generate the voltage reference benchmark sequence of the entire network nodes; based on this, construct the intraday voltage rolling forecast equation using the linearized sensitivity model, and establish a two-layer collaborative control framework of "day-ahead benchmark setting and intraday tracking optimization"; S2: In the intraday real-time control cycle k, the reliability of the prediction model is first evaluated online by introducing the model fidelity factor. Based on the model fidelity factor, the global voltage deviation risk index and the node regulation potential index are calculated respectively. The two indices are used as inputs to the preset 5×5 fuzzy logic controller to generate the exclusive fuzzy weighting factor for each node. Then, an MPC quadratic programming QP model that takes into account the node differential weights is constructed to solve for the optimal reactive power increment sequence of the distributed power source that satisfies the physical constraints. S3: Based on the rolling time-domain principle of model predictive control, select the first component of the optimal reactive power increment sequence and send it to each distributed power inverter for execution; at the next control time k+1, based on the latest collected measured node voltage value and the initial conditions of the distributed power power state correction prediction model, repeat steps S2 and S3 to realize closed-loop feedback adaptive control of the distribution network voltage.

2. The adaptive model predictive control method for distribution network voltage based on fuzzy weighted factors according to claim 1, characterized in that, The specific method for generating the network-wide node voltage reference sequence in step S1 is as follows: 1) Construct an objective function for day-ahead layer optimization with the goal of minimizing the overall system operating cost throughout the day; 2) Clarify the specific calculation method for each cost item in the objective function, including the cost of mechanical equipment switching actions and the cost of reactive power compensation for distributed power sources; 3) Set day-ahead optimization constraints to ensure that the dispatch plan meets the requirements for safe operation of the power grid; 4) Solve the mixed integer linear programming (MILP) problem corresponding to the above optimization to obtain the optimal day-ahead scheduling plan for the on-load tap-changing transformer (OLTC) and the switching capacitor bank (CB). The reference voltage sequence for each node over the entire 24 hours is calculated by combining the power flow equations. This is used as a reference value for the intraday MPC optimization objective function.

3. The adaptive model predictive control method for distribution network voltage based on fuzzy weighting factors according to claim 2, characterized in that, The objective function and cost term for the current layer optimization are as follows: The objective function expression is: Where T=24 is the day-ahead optimization period, and ΔT=1h is the optimization step size. Let t be the marginal price of grid-loss electricity. For the total active power loss of the system, Cost of switching operations on mechanical equipment, The cost of reactive power compensation provided for distributed power sources; The Including on-load tap changer (OLTC) tap changer Number of parallel capacitor banks (SCB) in operation Adjustment costs: The Characterize the economic cost of distributed generation providing reactive power compensation: Where a is the secondary cost coefficient, which represents the rate of increase in internal losses of the inverter due to the increase in reactive current; b is the primary cost coefficient, which represents the inherent marginal cost that is proportional to reactive power output.

4. The adaptive model predictive control method for distribution network voltage based on fuzzy weighting factors according to claim 2, characterized in that, The recently optimized constraints specifically include: 1) Power flow constraints: Based on the linearized sensitivity matrix, the linear influence of source load power injection changes on node voltage is characterized; 2) Voltage safety constraints: Limiting the voltage at each node to within the permissible safe operating range throughout the entire time period; 3) Equipment operation frequency constraints: The number of daily adjustments to the OLTC tap changer and the number of daily switching operations of the parallel capacitor bank are limited to a preset upper limit; 4) DG reactive power capacity constraint: The reactive power output of the distributed generation inverter is limited to its rated capacity range.

5. The adaptive model predictive control method for distribution network voltage based on fuzzy weighted factors according to claim 1, characterized in that, The formula for calculating the model fidelity factor in step S2 is as follows: in, The measured voltage value at time k. This represents the predicted value from the previous time step to the current time step. To prevent tiny constants with a denominator of zero; The moving standard deviation of the prediction error is calculated from the historical prediction error sequence within the moving time window. It is used to characterize the severity of recent power grid environment fluctuations and to achieve adaptive perception of environmental uncertainties. Where L is the length of the sliding window. This is a historical prediction error sequence.

6. The adaptive model predictive control method for distribution network voltage based on fuzzy weighting factors according to claim 1, characterized in that, The calculation methods for the global voltage deviation risk index and the node regulation potential index in step S2 are as follows: 1) Global voltage deviation risk index, i.e., voltage deviation severity index: in, The prediction domain risk term includes both the magnitude and rate of change of voltage deviation. This is the real-time deviation term at the current moment; through Achieve dynamic soft switching between forward-looking control and robust feedback control; 2) Node adjustment potential index: in, Let be the absolute value of the voltage-reactive power sensitivity at node i; The rated reactive power capacity of the inverter; This represents the number of times the device can be significantly adjusted within the current sliding window. This is the preset upper limit threshold for the number of actions.

7. The adaptive model predictive control method for distribution network voltage based on fuzzy weighted factors according to claim 1, characterized in that, The specific method for generating the fuzzy weighting factor in step S2 is as follows: 1) Fuzzification: Using a triangular membership function, the global voltage deviation risk index and the node regulation potential index are mapped to five fuzzy subsets {NB, NS, ZO, PS, PB} respectively; 2) Fuzzy Inference: Inference is performed based on a preset 5×5 fuzzy rule matrix. The inference logic is as follows: when the voltage deviation risk is high and the node regulation potential is high, a larger weighting factor is output; when the node regulation potential is low, a smaller weighting factor is output. 3) Defuzzification: The centroid method is used to calculate the precise fuzzy weighting factor corresponding to each node.

8. The adaptive model predictive control method for distribution network voltage based on fuzzy weighted factors according to claim 1, characterized in that, The objective function of the MPC quadratic programming model that takes into account the node differentiation weights in step S2 is constructed as follows: Where P is the prediction time domain and M is the control time domain; This is the predicted voltage value for the p-th step in the future. This refers to the day-ahead reference voltage issued in Step 1; This is the predicted reactive power adjustment increment.

9. The adaptive model predictive control method for distribution network voltage based on fuzzy weighting factors according to claim 8, characterized in that, The physical constraints of the MPC quadratic programming model in step S2 specifically include: 1) Inverter capacity limit: The reactive power output of the inverter at any given moment, combined with the current active power output, shall not exceed its rated apparent power range. 2) Regulation rate limit: Limits the upper and lower limits of the change in single-step reactive power regulation to prevent the instantaneous jump in inverter output from impacting the power grid; 3) Voltage safety range: Ensure that the predicted voltage of all nodes in the prediction time domain is within the safe operating range of 0.95-1.05 pu.

10. The adaptive model predictive control method for distribution network voltage based on fuzzy weighted factors according to claim 1, characterized in that, The specific process of closed-loop feedback correction and rolling execution in step S3 is as follows: Obtain the optimal reactive power increment sequence obtained in step S2, and select only the first component in the sequence as the actual control command to be issued to each distributed power source inverter for execution; at the next sampling time k+1, collect the latest measured voltage values ​​of each node in the distribution network and the active power output status of the distributed power source, use the measured values ​​to correct the initial state of the prediction model, and shift the prediction time window backward by one control step, and re-trigger the rolling optimization calculation in step S2 to achieve continuous closed-loop adaptive control.

Images