Power distribution network load balancing scheduling method and system based on model predictive control

Abstract

This application relates to the field of predictive control technology, specifically a distribution network load balancing dispatching method and system based on model predictive control, including the following steps: collecting voltage, current, power, photovoltaic irradiance, and wind speed, normalizing them to form predictive inputs and constructing a collaborative relationship; solving for power distribution and voltage levels based on power balance, energy storage, and line constraints to generate control quantities; correcting the rolling step size based on continuous periodic errors and time-series correlations; adjusting energy storage charging and discharging and adjustable load start-up and shutdown according to the difference between load forecast and baseline power to generate instructions; comparing dispatching and forecasting and adjusting the step size weights with mean square error to generate load balancing results. In this application, load and power supply are collaboratively characterized by unified processing of multi-dimensional electrical and environmental quantities; optimization quantities that can be updated with data changes are generated by combining constraint relationships; the rolling rhythm is calibrated in error evaluation; and energy storage and adjustable loads are adjusted under the drive of power difference to keep power flow and voltage stable, thereby improving load balance.

Description

Technical Field

[0001] This application relates to the field of predictive control technology, and in particular to a method and system for load balancing and dispatching of distribution networks based on model predictive control. Background Technology

[0002] Predictive control technology, a branch of automatic control engineering, primarily studies the technical system for predicting future system behavior and calculating optimal control sequences through rolling optimization under known system dynamic models. Core aspects include system modeling, state prediction, constraint handling, and optimization solutions. Predictive control technology is commonly applied in industrial process control, power system dispatching, traffic flow control, and energy management. By establishing a mathematical model of the controlled object and combining it with real-time measurement data, it predicts the system output over a future period, thereby calculating the optimal control quantity at the current moment to achieve precise control and coordinated operation of complex dynamic systems. This has promoted the widespread application of multivariable constraint systems in engineering control and has become an important direction for intelligent control strategies.

[0003] Traditional load balancing dispatching methods for power distribution networks refer to the process of adjusting the switching status of distribution lines and the power allocation ratio based on real-time load monitoring data, either by manually setting dispatching plans or using rule-based dispatching algorithms, to maintain the stability and economy of the power supply system and achieve load balancing across multiple nodes in the power distribution network. These methods typically use load forecasting curves or historical operating data as a basis, employing deterministic power flow calculations to analyze operating parameters such as node voltage, line current, and transformer capacity, and then formulate time-segmented dispatching instructions accordingly to achieve load balancing regulation of the distribution network.

[0004] Traditional load balancing dispatching in distribution networks relies on manual settings and fixed rules, lacking the ability to respond sensitively to dynamic load fluctuations. The analysis method based on deterministic power flow is difficult to handle the timing disturbances and parameter coupling after multiple sources are connected. When faced with real-time data changes, the dispatching instructions are delayed, which can easily cause node voltage deviations and line stress accumulation. In complex operating scenarios, load prediction errors are difficult to correct in a timely manner, resulting in a decrease in dispatching stability and coordination. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the first aspect of this application provides a distribution network load balancing dispatching method based on model predictive control, comprising the following steps: S1: Collect voltage, current and power data of the distribution network and photovoltaic irradiance and wind speed at the distributed power source end and normalize them, establish a load-power source coordination model of the distribution network and calculate the load forecast value for future periods, and generate an optimized input dataset. S2: Based on the optimized input dataset, analyze and solve the power balance equation of the distribution network, the charging and discharging constraint equation of the energy storage device and the line capacity constraint equation, calculate the line power distribution and voltage level, and generate an optimized control variable set; S3: Call the set of optimized control variables, calculate the error of the continuous prediction cycle and determine the temporal correlation, analyze the rolling step size adjustment coefficient and correct the rolling time length of the model predictive control, and generate the step size adjustment instruction; S4: Based on the step size adjustment command, determine the power allocation of the energy storage device and the adjustable load, compare the difference between the load prediction value and the reference power, adjust the charging or discharging command according to the positive or negative value of the difference, determine the start and stop status of the adjustable load, and generate a scheduling command set. S5: Based on the comparison between the scheduling results of the scheduling instruction set and the load forecast value, calculate the mean square error of the scheduling deviation and the forecast error, adjust the step size weight, and generate the load balancing scheduling result.

[0006] As a further embodiment of this application, the optimized input dataset includes load forecast feature items, photovoltaic irradiance feature items, and wind speed feature items; the optimized control variable set includes line power allocation item, voltage level item, and energy storage charging and discharging constraint item; the step size adjustment instruction includes prediction error correlation coefficient, rolling step size correction coefficient, and time series correlation parameter; the scheduling instruction set includes energy storage charging and discharging instruction item, adjustable load start and stop item, and power difference correction item; and the load balancing scheduling result includes scheduling deviation, prediction error root mean square error, and step size weighting coefficient.

[0007] As a further aspect of this application, the specific steps of S1 are as follows: S101: Collect voltage, current and power data of the distribution network, and simultaneously acquire photovoltaic irradiance and wind speed data of the distributed power source. Synchronize and normalize all data in time, analyze the load-power coordination model of the distribution network, and generate a standardized set of operating parameters. S102: Based on the standardized operating parameter set, voltage, current and power are called, and feature vector splicing is performed on the parameter distribution of power distribution and distributed power sources within the same time period. The power distribution features are correlated and compared with irradiance and wind speed features, and the correlation matrix is ​​calculated to generate a load power source coordinated feature set. S103: Based on the load power supply coordination feature set, a sliding window is applied to the time series distribution to extract the load change gradient over multiple time periods. The power trend series over consecutive time periods is then weighted and averaged to calculate the load forecast for future time periods. This forecast is then integrated with the load power supply coordination features to generate an optimized input dataset.

[0008] As a further aspect of this application, the specific steps of S2 are as follows: S201: Based on the optimized input dataset, load forecast values ​​and power distribution data are called to match and calculate the power injection and power flow of multiple branches of the distribution network, calculate the difference between active and reactive power and perform constraint adjustment, analyze the power balance relationship, and generate a power balance parameter set. S202: Based on the power balance parameter set, extract the charging and discharging state variables at the energy storage device end, and compare them with the charging power, discharging power and state of charge change range. If they exceed the range, redefine the corresponding power range, correct the power output for multiple time periods, and generate an energy storage constraint dataset. S203: Based on the energy storage constraint dataset, obtain power parameters and line power data, perform allocation of active and reactive power components of multiple lines, compare the deviation between the calculation results and the line capacity threshold according to the line capacity constraint equation, analyze the line power allocation and voltage level, and generate an optimized control variable set.

[0009] As a further aspect of this application, the line capacity threshold is determined by extracting the measured current values ​​of multiple lines under multiple load conditions and the corresponding allowable current values ​​of the conductors, performing interval statistical calculations on the ratio of the two, and analyzing the average load rate.

[0010] As a further aspect of this application, the specific steps of S3 are as follows: S301: Based on the optimized control variable set, call the continuous prediction cycle power data and the current running data and calculate the difference, extract the multi-time period error sequence, and perform statistical aggregation on the difference sequence to extract the error mean and fluctuation amplitude parameter set of the continuous cycle, and generate the prediction error parameter set; S302: Based on the predicted error parameter set, call the error mean sequence and fluctuation amplitude parameter set, perform correlation analysis on the error change rate of adjacent periods, calculate the autocorrelation coefficient of the error change rate with time, screen time series segments, extract time-related feature factors, and generate a time-related feature factor set. S303: Call the time-series associated feature factor set, obtain time-series weight distribution data and predictive control time parameters, perform proportional analysis on the rolling step length adjustment coefficients of multiple time periods, and correct the rolling time length according to the step length adjustment benchmark value and perform instruction encoding to generate step length adjustment instructions.

[0011] As a further aspect of this application, the step size adjustment benchmark value is determined by extracting the prediction error change rate and the corresponding correlation strength coefficient within multiple rolling cycles, performing normalization on both, and then calculating a weighted average.

[0012] As a further aspect of this application, the specific steps of S4 are as follows: S401: Based on the step size adjustment command, obtain the rated power of the energy storage device and the operating power of the adjustable load, perform proportional allocation calculation, analyze the allocation ratio of energy storage power and load power in each time period, and perform weighted processing on the proportional sequence to generate a power allocation parameter set. S402: Call the power allocation parameter set, compare the difference sequence between the load forecast value and the reference power, mark the positive and negative states of the difference as positive deviation segments and negative deviation segments respectively, perform interval aggregation on the difference segments, calculate the segment deviation signal, and generate a power deviation signal set; S403: Based on the power deviation signal set, extract the deviation symbol identifiers for multiple time periods, set the discharge command for the section with positive deviation, set the charging command for the section with negative deviation, determine the start-stop status code of the adjustable load, aggregate the energy storage control command and the load status code, and generate a scheduling command set. The reference power is set by performing difference analysis on the current power data and predicted power data of the load during the operating cycle, and analyzing the sum of the mean of the difference sequence and the standard deviation of a preset multiple.

[0013] As a further aspect of this application, the specific steps of S5 are as follows: S501: Based on the scheduling results and load forecast values ​​of the scheduling instruction set, obtain the instruction execution power value and forecast power value for multiple time periods, calculate the difference vector between the two power sequences, accumulate and calculate the squared deviation values ​​within multiple intervals and perform a weighted average to generate the mean square result of the scheduling deviation. S502: Call the mean square result of the scheduling deviation, retrieve the prediction error data of multiple time periods during the scheduling process and normalize it, compare the difference with the mean square value of the scheduling deviation, calculate the weight adjustment coefficient, and perform a product operation between the adjustment coefficient and the original step size weight parameter to update the step size control parameter and obtain the step size weight parameter set; S503: Based on the set of step size weight parameters, re-invoke the power value and load prediction value, perform weighted fusion on the two power sequences according to the updated step size weight, perform moving average smoothing, extract the power distribution sequence after step size adjustment, and generate load balancing scheduling results.

[0014] A second aspect of this application provides a distribution network load balancing dispatching system based on model predictive control, comprising: The data analysis module collects and normalizes voltage, current and power data of the distribution network and photovoltaic irradiance and wind speed at the distributed power source end, establishes a load-power source coordination model of the distribution network and calculates the load forecast value for future periods, generates an optimized input dataset and passes it to the constraint solving module. The constraint solving module, based on the optimized input dataset, analyzes and solves the power balance equation of the distribution network, the charging and discharging constraint equation of the energy storage device, and the line capacity constraint equation, calculates the line power distribution and voltage level, generates an optimized control variable set, and passes it to the rolling adjustment module. The rolling adjustment module calls the optimized control variable set, calculates the error of the continuous prediction cycle and judges the time correlation, analyzes the rolling step size adjustment coefficient and corrects the rolling time length of the model predictive control, generates the step size adjustment instruction and transmits it to the load scheduling module. The load scheduling module, based on the step size adjustment command, determines the power allocation of the energy storage device and the adjustable load, compares the difference between the load prediction value and the reference power, adjusts the charging or discharging command according to the positive or negative value of the difference, determines the start-stop status of the adjustable load, generates a scheduling command set, and transmits it to the deviation correction module. The deviation correction module compares the scheduling results of the scheduling instruction set with the load forecast values, calculates the root mean square error of the scheduling deviation and the forecast error, adjusts the step size weights, and generates a load balancing scheduling result.

[0015] Compared with the prior art, the beneficial effects of this application are as follows: In this application, load and power supply status are collaboratively characterized by collecting and processing multi-dimensional electrical and environmental quantities. Power balance constraints, energy storage constraints, and capacity constraints are combined to form optimized variables that can be updated with data changes. In the evaluation of prediction error and time series correlation, the rolling time and step size weights are corrected to keep the scheduling rhythm consistent with load changes. Under the power difference drive, the energy storage and adjustable load status are adjusted to keep the voltage distribution and power flow stable, thereby promoting the improvement of the load balance and operation coordination of the distribution network. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the steps of this application; Figure 2 This is a detailed schematic diagram of S1 in this application; Figure 3 This is a detailed schematic diagram of S2 in this application; Figure 4 This is a detailed schematic diagram of S3 in this application; Figure 5 This is a detailed schematic diagram of S4 in this application; Figure 6 This is a detailed schematic diagram of S5 in this application; Figure 7 This is a system module diagram of this application. Detailed Implementation

[0018] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0020] In this text, the term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " in this text generally indicates that the preceding and following related objects have an "or" relationship.

[0021] Please see Figure 1 This application provides a distribution network load balancing scheduling method based on model predictive control, including the following steps: S1: Collect voltage, current and power data of the distribution network and photovoltaic irradiance and wind speed at the distributed power source end and normalize them, establish a load-power source coordination model of the distribution network and calculate the load forecast value for future periods, and generate an optimized input dataset. S2: Based on the optimized input dataset, analyze and solve the power balance equation of the distribution network, the charging and discharging constraint equation of energy storage equipment and the line capacity constraint equation, calculate the line power distribution and voltage level, and generate an optimized control variable set; S3: Call the optimized control variable set, calculate the error of the continuous prediction cycle and determine the temporal correlation, analyze the rolling step size adjustment coefficient and correct the rolling time length of the model predictive control, and generate the step size adjustment instruction. S4: Based on the step size adjustment command, determine the power allocation of energy storage devices and adjustable loads, compare the difference between the load forecast value and the reference power, adjust the charging or discharging command according to the positive or negative value of the difference, determine the start and stop status of the adjustable load, and generate a scheduling command set. S5: Based on the comparison between the scheduling results and load forecast values ​​of the scheduling instruction set, calculate the mean square error of the scheduling deviation and forecast error, adjust the step size weight, and generate the load balancing scheduling results.

[0022] The optimized input dataset includes load forecast feature items, photovoltaic irradiance feature items, and wind speed feature items. The optimized control variable set includes line power allocation item, voltage level item, and energy storage charging and discharging constraint item. The step size adjustment instructions include prediction error correlation coefficient, rolling step size correction coefficient, and time series correlation parameter. The scheduling instruction set includes energy storage charging and discharging instruction item, adjustable load start and stop item, and power difference correction item. The load balancing scheduling results include scheduling deviation, prediction error root mean square error, and step size weighting coefficient.

[0023] Please see Figure 2 The specific steps of S1 are as follows: S101: Collect voltage, current and power data of the distribution network, and simultaneously acquire photovoltaic irradiance and wind speed data of the distributed power source. Synchronize and normalize all data in time, analyze the load-power coordination model of the distribution network, and generate a standardized set of operating parameters. In one specific embodiment, a key node A (rated voltage 10kV) in the distribution network was selected as the monitoring point. This node connects residential loads to a 500kW photovoltaic power station. Data was collected from 10:00 AM to 10:45 AM on November 18, 2024, at 5-minute intervals. Three-phase voltage, current, and active power data were acquired using a power quality analyzer installed at node A. Simultaneously, total irradiance and ambient wind speed on the photovoltaic array surface were collected using irradiance sensors and ultrasonic anemometers deployed within the photovoltaic power station. All acquisition devices were connected to a GPS satellite signal-based time synchronization unit to ensure that the timestamp accuracy of each set of acquired data reached the millisecond level, achieving precise data alignment in the time dimension. The collected raw data is shown in Table 1.

[0024] Table 1. Raw Data Collection Table for Distribution Network Node A and Distributed Power Generation

[0025] To eliminate the influence of different physical dimensions on subsequent analysis, all data needs to be normalized. This process uses the max-min normalization method to linearly map the data to intervals. The reference values ​​(maximum and minimum values) for normalization are determined based on statistical analysis of the node's historical one-year operating data to ensure the stability and representativeness of the reference. The reference ranges for voltage are set as [9.80kV, 10.20kV]; for current, [10A, 50A]; for active power, [100kW, 600kW]; for irradiance, [0W / m², 1000W / m²]; and for wind speed, [0m / s, 15m / s]. Taking the voltage data at 10:00:00 as an example, its value is 10.12kV. The normalization calculation process is: (10.12-9.80) / (10.20-9.80)=0.32 / 0.40=0.800. Similarly, the irradiance data of 855 W / m² at 10:00:00 is normalized: (855-0) / (1000-0) = 0.855. After performing the same normalization calculation on data from all time points and all dimensions, a set of dimensionless parameter sequences is obtained. These sequences together constitute the standardized set of operating parameters.

[0026] S102: Based on the standardized operating parameter set, voltage, current and power are called to perform feature vector concatenation on the parameter distribution of power distribution and distributed power sources within the same time period. The power distribution characteristics are correlated and compared with irradiance and wind speed characteristics and the correlation matrix is ​​calculated to generate a load power source coordinated feature set. The standardized operating parameter set is invoked based on the generated normalized data. To perform multi-dimensional feature aggregation, each normalized parameter at the same timestamp is first constructed into a feature vector. For example, at 10:05:00, the corresponding feature vector is [normalized voltage value, normalized current value, normalized power value, normalized irradiance value, normalized wind speed value]. The specific values ​​are: normalized voltage (10.15kV) is 0.875, normalized current (29.1A) is 0.478, normalized power (512.3kW) is 0.825, normalized irradiance (870W / m²) is 0.870, and normalized wind speed (2.3m / s) is 0.153. The feature vector at this time is [0.875, 0.478, 0.825, 0.870, 0.153]. Repeat this process for all 10 data points within the selected time period (10:00:00 to 10:45:00) to form a feature vector matrix. The specific action of multidimensional feature aggregation is to calculate the mean and variance of this matrix column by column (i.e., by feature) within the time period to reflect the central tendency and degree of fluctuation of the parameters.

[0027] Next, the power distribution characteristics are compared and correlated with irradiance and wind speed characteristics. This comparison is a quantitative coupling analysis, not a qualitative judgment. First, parameter segments are defined: normalized irradiance values ​​in the range [0.8, 1.0] are defined as high irradiance periods; normalized active power values ​​in the range [0.8, 1.0] are defined as high power output periods. At 10:05:00, the normalized irradiance value is 0.870, belonging to a high irradiance period, and the corresponding normalized power value is 0.825, belonging to a high power output period. A preliminary assessment indicates a positive correlation between the two. To accurately analyze the data coupling degree, the coupling coefficient between the normalized power sequence and the normalized irradiance sequence is calculated.

[0028] It should be noted that this process does not directly call a specific algorithm name, but describes its operation steps: extract the values ​​of two sequences from 10:00 to 10:10: power sequence P=[0.797, 0.825, 0.811], irradiance sequence I=[0.855, 0.870, 0.862].

[0029] Calculate the average values ​​of the two sequences: the average power is 0.811 and the average irradiance is 0.862.

[0030] Calculate the deviation of each data point from its respective mean: the power deviation sequence is [-0.014, 0.014, 0.000], and the irradiance deviation sequence is [-0.007, 0.008, 0.000]. Calculate the sum of the products of the corresponding deviations: (-0.014 * -0.007) + (0.014 * 0.008) + (0.000 * 0.000) = 0.000098 + 0.000112 = 0.00021.

[0031] The sum of squares of the respective deviations is calculated: the sum of squares of power deviation is 0.000392, and the sum of squares of irradiance deviation is 0.000113. The coupling coefficient is calculated to be 0.998. This coupling coefficient of 0.998 is one of the load-source coordination characteristics, which, together with other aggregated characteristics (such as mean and variance), generates the load-source coordination characteristic set.

[0032] S103: Based on the load power supply coordination feature set, a sliding window is applied to the time series distribution to extract the load change gradient over multiple time periods. The power trend series over consecutive time periods is then weighted and averaged to calculate the load forecast for future time periods. This forecast is then integrated with the load power supply coordination features to generate an optimized input dataset. Based on the generated load-power coordination feature set, the original active power time series (unit: kW) is selected as the basic data for a sliding window operation. The width of the sliding window is set to 4 time points (i.e., 15 minutes), and the sliding step size is 1 time point (5 minutes). Taking the data from 10:00:00 to 10:15:00 as an example, the power sequence within the window is [498.5, 512.3, 505.4, 515.0]. Within this window, the load change gradients over multiple time periods are extracted, i.e., the power difference between adjacent time points is calculated: Gradient 1 (10:05-10:00) = 512.3 - 498.5 = 13.8kW. Gradient 2 (10:10-10:05) = 505.4 - 512.3 = -6.9kW. Gradient 3 (10:15-10:10) = 515.0 - 505.4 = 9.6kW. The power trend series over the continuous period is [13.8, -6.9, 9.6]. A weighted average of this trend series is used to calculate the load forecast for future periods. The weighting coefficients are not arbitrarily assigned but determined through repeated testing and verification on historical datasets.

[0033] The experimental procedure is as follows: Select the power data of the distribution network node for three months in the past, and test multiple sets of weights with the same window width (4 time points) and prediction method (the sum of the weights is 1, and the gradient weight is higher the closer to the current time).

[0034] By comparing the mean absolute error of prediction under different weight settings, it was found that the prediction error was minimized when the weight set was {0.6, 0.25, 0.15} (corresponding to gradient 3, gradient 2, and gradient 1, respectively). Therefore, this weight set was chosen. The predicted power change trend was calculated as follows: Predicted trend = (9.6 * 0.6) + (-6.9 * 0.25) + (13.8 * 0.15) = 5.76 - 1.725 + 2.07 = 6.105 kW. The predicted load value for the future time period (10:20:00) was obtained by adding the load value at the last moment (10:15:00) to the predicted trend: Predicted load value = 515.0 + 6.105 = 521.105 kW. The predicted value (521.105kW) is integrated with the load power supply coordination characteristics generated in S102 (such as coupling coefficient 0.998, mean and variance of each parameter) to form a structured dataset, which is the optimized input dataset.

[0035] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the optimized input dataset, load forecast values ​​and power distribution data are called to match and calculate the power injection and power flow of multiple branches of the distribution network, calculate the difference between active and reactive power and perform constraint adjustment, analyze the power balance relationship, and generate a power balance parameter set. Based on the generated optimized input dataset, load forecast and power distribution data are retrieved at 10:20:00. The retrieved load forecast is 521.105kW (from S103). The power distribution data is sourced from the distribution network topology database and distributed generation monitoring, determining that the distribution network at this time contains two main branches: branch L1 (residential load) and branch L2 (industrial load and photovoltaic power). The power distribution data (predicted load) for branch L1 is 200.0kW active power and 50.0kVar reactive power. The power distribution data (predicted load) for branch L2 is 321.105kW active power and 121.28kVar reactive power. The predicted output of the photovoltaic power source (power injection) on branch L2 is 530.0kW, its inverter is set to operate at unity power factor, and its reactive power output is 0kVar. Matching calculations are performed on the power injection and power flow of multiple branches in the distribution network. Here, power flow refers to the total load demand in the network. Total power flow (active) = Active power of branch L1 + Active power of branch L2 = 200.0 + 321.105 = 521.105kW. Total power flow (reactive) = Reactive power of branch L1 + Reactive power of branch L2 = 50.0 + 121.28 = 171.28kVar. Total power injection (active) = Photovoltaic output of branch L2 = 530.0kW. Total power injection (reactive) = Photovoltaic output of branch L2 = 0kVar. Calculate the difference between active and reactive power: The active power difference is calculated by subtracting the total power flow (active) from the total power injection: 530.0 - 521.105 = +8.895kW. The reactive power difference is calculated by subtracting the total power flow (reactive power) from the total power injection: 0 - 171.28 = -171.28 kVar. This result indicates an internal active power surplus of 8.895 kW and a reactive power deficit of 171.28 kVar. Next, constraint adjustments are performed. Key operational constraints include the voltage constraint at node A and the main grid exchange power constraint. The node voltage reference range is set according to S101 as [9.80 kV, 10.20 kV]. The main grid exchange power constraint is set according to the grid dispatch protocol, with limits of [-1000 kW, 1000 kW] (negative values ​​indicate absorption from the main grid, positive values ​​indicate transmission to the main grid). Analysis of the power balance relationship: Before adjustment, the +8.895 kW active power surplus and the -171.28 kVar reactive power deficit will cause node voltage fluctuations and trigger main grid exchange. The goal of the adjustment is to internally balance these differences. The target values ​​for balancing the power balance equation are -8.895kW (absorption) of active power and +171.28kVar (injection) of reactive power. These target values ​​constitute the power balance parameter set.

[0036] S202: Based on the power balance parameter set, extract the charging and discharging state variables at the energy storage device end, and compare them with the charging power, discharging power and state of charge change range. If they exceed the range, redefine the corresponding power range, correct the power output for multiple time periods, and generate an energy storage constraint dataset. Based on the generated power balance parameter set, the active power balance target is -8.895kW, and the reactive power balance target is +171.28kVar. The operation is then executed. First, the real-time state variables of the energy storage device at access node A are extracted. These variables are acquired through the local controller of the BESS (Battery Energy Storage System): Current State of Charge (SOC) = 70.0%, Current Operating State = 0 (idle), Rated Capacity = 500kWh, Rated Charge / Discharge Power = 100kW, and Inverter Rated Apparent Power = 150kVA. The balance target is then compared with the energy storage constraints. The operating constraints of the energy storage device are determined by the manufacturer's specifications and verified experimentally. The SOC range is set to [20.0%, 90.0%]. Cycle life tests are conducted on batteries from the same batch to compare the capacity decay rate across different SOC ranges. The attenuation rate in the range [20.0%, 90.0%] (4.5%) is significantly lower than that in [10.0%, 90.0%] (8.2%) and [20.0%, 100.0%] (7.9%), while also considering a larger available capacity. Therefore, [20.0%, 90.0%] was selected as the operating range. Active power comparison: Target -8.895kW (indicating charging). The absolute value of this value, 8.895kW, is less than the rated charging power of 100kW. SOC comparison: Charging at -8.895kW for 5 minutes (0.0833 hours), charging energy = 8.895 * 0.0833 = 0.741kWh. SOC change = (0.741 / 500) * 100% = 0.148%. Predicted SOC for the next moment = 70.0% + 0.148% = 70.148%. The value is within the range of [20.0%, 90.0%]. The active power constraint is met. Comparison of reactive and apparent power: target reactive power + 171.28 kVar (indicating reactive power output). The energy storage inverter must simultaneously meet both active and reactive power targets. The square of the required apparent power is calculated by adding the square of the active power (-8.895 squared) to the square of the reactive power (171.28 squared): (79.12 + 29336.7) = 29415.8. The square of the inverter capacity (150 squared) is 22500. Comparison result: 29415.8 > 22500. The required apparent power exceeds the inverter capacity limit. At this point, the operation of "re-limiting the corresponding power range if exceeded" is executed. According to the scheduling strategy, reactive power (voltage support) has a higher priority than active power (power balance). Reactive power is allocated first. Reactive power limit: The active power command is temporarily set to 0kW. At this time, the maximum reactive power that the inverter can provide is 150kVar (equal to its rated apparent power). The corrected reactive power command = +150kVar (less than the target 171.28kVar).Active power constraint: After issuing 150kVar reactive power, the inverter's remaining active power capacity is 0kW (obtained by subtracting the square root of 150 from its square). Corrected active power command = 0kW. Corrections are applied to power output over multiple time periods to generate an energy storage constraint dataset. This dataset contains: {BESS active power command = 0kW, BESS reactive power command = +150kVar}.

[0037] S203: Based on the energy storage constraint dataset, obtain power parameters and line power data, perform allocation of active and reactive power components of multiple lines, compare the deviation between the calculation results and the line capacity threshold according to the line capacity constraint equation, analyze the line power allocation and voltage level, and generate an optimized control variable set. Based on the generated energy storage constraint dataset, power parameters and line power data are obtained. The obtained power parameters are the execution commands of the energy storage system {BESS active power command = 0kW, BESS reactive power command = +150kVar}. Simultaneously, the unmet balance targets in S202 are: residual active power imbalance = -8.895 - 0 = -8.895kW; residual reactive power imbalance = +171.28 - 150 = +21.28kVar. The line power data are the load and power supply data of branches L1 and L2 defined in S201. The active and reactive power components of multiple lines are allocated. This allocation process calculates the power flow of each line and the main grid connection point after the application of the energy storage commands.

[0038] Total active power injection at node A = 530.0 kW (PV) + 0 kW (BESS) = 530.0 kW; Total active power consumption at node A = 200.0 kW (L1) + 321.105 kW (L2) = 521.105 kW; Active power imbalance at node A (requires grid balancing) = 521.105 - 530.0 = -8.895 kW (i.e., grid absorption of 8.895 kW); Total active power consumption at node A Reactive power injection = 0 kVar (PV) + 150 kVar (BESS) = +150 kVar; Total reactive power consumption of node A = 50.0 kVar (L1) + 121.28 kVar (L2) = +171.28 kVar; Reactive power imbalance of node A (requires grid balancing) = 171.28 - 150.0 = +21.28 kVar (i.e., 21.28 kVar injected by the grid).

[0039] Next, the deviation between the calculated results and the line capacity threshold is compared based on the line capacity constraint equation. First, the line capacity threshold is set.

[0040] Table 2. Test Results of Thermal Capacity Limitation for YJV-10kV-150mm² Cables

[0041] As shown in Table 2, through load tests on the same type of cable used in L1 and L2, the long-term safe current carrying capacity threshold corresponding to the conductor's maximum operating temperature of 90°C at an ambient temperature of 25°C was determined to be 350A. At a nominal voltage of 10kV, the corresponding apparent power capacity threshold (calculated using 1.732 * 10.0kV * 350A) is 6062kVA. The apparent power of each branch is calculated as follows: Apparent power of branch L1, calculated by taking the square root of the sum of the squares of active power (200.0kW) and reactive power (50.0kVar): 206.16kVA. Net apparent power of branch L2 (after photovoltaic load offset): Net active power = 321.105 - 530.0 = -208.895kW; Net reactive power = 121.28kVar. The apparent power of branch L2 is calculated as 241.55 kVA by taking the square root of the sum of the squares of the active power of -208.895 kW and the reactive power of 121.28 kVar.

[0042] Comparison results: 206.16kVA < 6062kVA; 241.55kVA < 6062kVA. The deviation (margin) is greater than 5800kVA in both cases. Analysis of line power distribution and voltage level: At 10:15:00, the voltage at node A is 10.15kV. At 10:20:00, without energy storage intervention, a reactive power deficit of -171.28kVar would cause the voltage to drop to 9.90kV. After the BESS (Body Energy Storage System) injects +150kVar of reactive power, the remaining reactive power deficit is only +21.28kVar (supplied by the main grid), while active power is fed back to -8.895kW.

[0043] Based on power flow calculations, the voltage at node A stabilizes at 10.10 kV. This voltage value falls within the constraint range of [9.80 kV, 10.20 kV]. A power optimization control variable set is generated: {BESS active power command = 0 kW, BESS reactive power command = +150 kVar, main grid switching active power = -8.895 kW, main grid switching reactive power = +21.28 kVar, predicted voltage at node A = 10.10 kV}.

[0044] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the optimized control variable set, call the continuous prediction cycle power data and the current running data and calculate the difference, extract the multi-time period error sequence, perform statistical aggregation on the difference sequence, extract the error mean and fluctuation amplitude parameter set of the continuous cycle, and generate the prediction error parameter set; Based on the generated set of optimized control variables corresponding to the time of 10:20:00, the prediction error assessment begins. First, the continuous prediction cycle power data is retrieved. This data refers to the load prediction values ​​generated by the prediction process of S103 in previous time periods (e.g., 10:00:00, 10:05:00, 10:10:00, and 10:15:00). Simultaneously, the current operating data is retrieved, i.e., the actual active power value of distribution network node A actually monitored and recorded by the data acquisition system of S101 at the same time points mentioned above. Next, the difference operation is performed on the two sets of data. This calculation subtracts the corresponding continuous prediction cycle power data value from the actual operating data value for the same timestamp. For example, at 10:15:00, the prediction value generated by the S103 process at 10:10:00 is 513.0 kW, while the actual operating data collected by S101 at 10:15:00 is 515.0 kW. The difference between the two (prediction error) is calculated as: 515.0 - 513.0 = +2.0kW. This operation is performed on the selected four consecutive periods (10:00:00 to 10:15:00) to extract the multi-period error sequence.

[0045] Table 3 Calculation of Prediction Error Sequence

[0046] As shown in Table 3, the extracted multi-period error sequence is [+3.5, +2.3, -2.6, +2.0] (unit: kW). Subsequently, statistical aggregation is performed on this difference sequence. This aggregation operation is used to extract the mean error and fluctuation amplitude parameter set for continuous periods.

[0047] The calculation process for the mean error is as follows: Add all the values ​​in the error sequence (3.5 + 2.3 + -2.6 + 2.0) to obtain a total of 5.2. Then divide this total of 5.2 by the number of data points in the sequence, 4. The result is 1.3kW. The extraction process for the fluctuation amplitude parameter group is as follows: The fluctuation amplitude is defined here as the average absolute deviation of the error sequence relative to its mean. Calculate the deviation of each error value from the mean (1.3kW): Deviation 1: 3.5 - 1.3 = 2.2kW, Deviation 2: 2.3 - 1.3 = 1.0kW, Deviation 3: -2.6 - 1.3 = -3.9kW, Deviation 4: 2.0 - 1.3 = 0.7kW. Obtain the absolute values ​​of the deviations: [2.2, 1.0, 3.9, 0.7]. Calculate the average of the absolute deviations: (2.2 + 1.0 + 3.9 + 0.7) to obtain a total of 7.8. Divide the total of 7.8 by the number of data points, 4. The calculated result is 1.95kW. This {mean error = 1.3kW, fluctuation range = 1.95kW} constitutes the parameter set for the current cycle (10:15:00), which, together with the parameter set stored in the historical cycles, generates the prediction error parameter set.

[0048] S302: Based on the prediction error parameter set, call the error mean sequence and fluctuation amplitude parameter set, perform correlation analysis on the error change rate of adjacent periods, calculate the autocorrelation coefficient of the error change rate with time, screen time series segments, extract time-related feature factors, and generate a time-related feature factor set. Based on the generated set of prediction error parameters, the mean error sequence and fluctuation amplitude parameter set are retrieved. The mean error sequence is a sequence consisting of the mean error (1.3kW) calculated in S301 and its historical values. To perform correlation analysis, the mean error sequence (unit: kW) for the most recent four periods is retrieved: [1.375, 1.700, 0.675, 1.300]. Simultaneously, the corresponding normalized irradiance sequence (10:00 to 10:15) from S101 is retrieved: [0.855, 0.870, 0.862, 0.880] (data for 10:15 is added to match the sequence length). First, the error change rate for adjacent periods is calculated. This process calculates the difference between the time-T and time-T-1 values ​​of the error mean sequence: rate of change 1 (10:05) = 1.700 - 1.375 = +0.325, rate of change 2 (10:10) = 0.675 - 1.700 = -1.025, rate of change 3 (10:15) = 1.300 - 0.675 = +0.625, resulting in the error rate of change sequence X = [+0.325, -1.025, +0.625]. Next, the mutual information between the error rate of change sequence (sequence X) and the time series (here, the normalized irradiance sequence Y = [0.870, 0.862, 0.880] is calculated. Sequence X is discretized. The interval division criteria are set as: [less than 0] (labeled A), [greater than or equal to 0] (labeled B). The discretization result of sequence X is: [B, A, B]. Discretize the sequence Y. Set the interval division criteria: [less than 0.875] (labeled C), [greater than or equal to 0.875] (labeled D). The discretized result of sequence Y is: [C, C, D]. Calculate the marginal probabilities of each label: P(A) = 1 / 3, P(B) = 2 / 3, P(C) = 2 / 3, P(D) = 1 / 3. Calculate the joint probability distribution: P(B, C)(10:05) = 1 / 3, P(A, C)(10:10) = 1 / 3, P(B, D)(10:15) = 1 / 3. Calculate the mutual information. This process involves calculating, for each joint probability event P(x, y), P(x, y) multiplied by [the logarithm of P(x, y) divided by (P(x) multiplied by P(y)) (base 2)], and then summing the results: Item 1 (B, C): (1 / 3) multiplied by [the logarithm of (1 / 3) divided by (2 / 3 multiplied by 2 / 3)] = (1 / 3) multiplied by log2(0.75) = -0.138. Item 2 (A, C): (1 / 3) multiplied by [the logarithm of (1 / 3) divided by (1 / 3 multiplied by 2 / 3)] = (1 / 3) multiplied by log2(1.5) = +0.195. Item 3 (B, D): (1 / 3) multiplied by [the logarithm of (1 / 3) divided by (2 / 3 multiplied by 1 / 3)] = (1 / 3) multiplied by log2(1.5) = +0.195. Mutual information = -0.138 + 0.195 + 0.195 = 0.252.The same mutual information calculation was performed on the error change rate and other time series such as wind speed and load, and time series segments were filtered. Through backtesting with historical data, a mutual information threshold of 0.5 was set. The associated sequences [irradiance, mutual information 0.252, corresponding prediction model MAE 2.51; wind speed, mutual information 0.050, corresponding prediction model MAE 2.58; load gradient, mutual information 0.810, corresponding prediction model MAE 1.92] were used. This threshold was determined by comparing the mutual information of different associated sequences and their impact on the mean absolute error (MAE) of the prediction model. When the MI value is greater than 0.5, introducing this factor can reduce the MAE by more than 20%, so 0.5 was selected. The filtering result: the mutual information of the error change rate and load gradient is 0.810, which is greater than 0.5. Therefore, "load gradient" is extracted as a time-related feature factor with a value of 0.810, and together they generate a time-related feature factor set.

[0049] S303: Call the time-series correlation feature factor set, obtain time-series weight distribution data and predictive control time parameters, perform proportional analysis on the rolling step length adjustment coefficient of multiple time periods, and correct the rolling time length according to the step length adjustment benchmark value and perform instruction encoding to generate step length adjustment instructions. The generated time-series correlation feature factor set is called, and the obtained factor is "load gradient," with a mutual information value of 0.810. Simultaneously, the prediction control time parameter is obtained, i.e., the basic rolling time length (step size) used in the current S103 prediction process, set to 5 minutes (300 seconds). Next, the time-series weight distribution data is obtained. This data defines the mapping relationship between the mutual information value and the rolling step size adjustment coefficient. This relationship is determined through experimental verification, and the step size adjustment benchmark value is set based on the following: Through historical data simulation, the mutual information (MI) is divided into three intervals. Tests show that when the MI value is in the (0.60, 1.00) interval, it indicates that the prediction error is highly correlated with this factor (load gradient). At this time, if the prediction step size coefficient is adjusted to 0.5 (i.e., shorten the step size), the mean absolute error (MAE) of the prediction can be reduced from 2.58kW to 1.85kW. Proportional analysis is performed on the rolling step size adjustment coefficient for multiple time periods. The specific steps of this analysis are as follows: The mutual information value of 0.810 obtained by S302 is compared with the "mutual information (MI) interval". The comparison result is that 0.810 is within the interval (0.60, 1.00). Based on this comparison result, the corresponding "step size adjustment coefficient" is extracted, with a value of 0.5. Subsequently, the rolling time length is corrected based on the step size adjustment baseline value (coefficient 0.5). The calculation process is as follows: the original rolling time length (5 minutes) is multiplied by the step size adjustment coefficient (0.5). The calculation result is: 5 minutes multiplied by 0.5 = 2.5 minutes. This 2.5 minutes (150 seconds) is the corrected rolling time length. Finally, the corrected time length is encoded into an instruction. The encoding process generates a data structure containing new parameters. The generated step size adjustment instruction is: target module: [S101, S103], parameters: [acquisition interval, prediction step size], set value: [150 seconds, 150 seconds].

[0050] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the step size adjustment command, obtain the rated power of the energy storage device and the operating power of the adjustable load, perform proportional allocation calculation, analyze the allocation ratio of energy storage power and load power in each time period, and perform weighted processing on the proportional sequence to generate a power allocation parameter set; Based on the generated step size adjustment instruction (setting the acquisition interval and prediction step size to 150 seconds), power allocation parameters are prepared for the next scheduling cycle (starting at 10:20:00). First, the rated power of the energy storage device and the operating power of the adjustable loads in the grid are obtained. The device database is retrieved via S101, and the rated charging and discharging power of the BESS (energy storage device) at node A is obtained as 100kW (derived from S202). Simultaneously, the list of adjustable loads connected to node A and their operating power are obtained: adjustable load AL1 (industrial ventilation system) has an operating power of 50kW, and adjustable load AL2 (intelligent temperature-controlled air conditioner) has an operating power of 80kW. The total adjustable power base is calculated as: 100 + 50 + 80 = 230kW. Next, a proportional allocation calculation is performed on the allocation ratio between energy storage power and load power. This calculation is based on the proportion of each device's rated operating power to the total power base: BESS proportion = 100 / 230 = 0.435, AL1 proportion = 50 / 230 = 0.217, AL2 proportion = 80 / 230 = 0.348. The allocation proportion for each time period is analyzed to form a power capacity-based proportion sequence: [0.435, 0.217, 0.348]. Subsequently, a weighted processing is performed on this proportion sequence. This weighting process aims to adjust the allocation proportion based on the adjustment response rate (i.e., adjustment priority) of different resources. Historical response time data for each resource are retrieved, as shown in Table 4.

[0051] Table 4 Experimental Table of Adjusting Resource Response Time

[0052] As shown in Table 4, the time it takes for each resource to reach 90% response target is measured by applying a 10kW power step command. BESS responds the fastest (0.5 seconds), followed by AL1 (2.0 seconds), and AL2 is the slowest (5.0 seconds). Weighting coefficients are set based on the reciprocals of the response times. First, the sum of the reciprocals of the response times is calculated: (1 / 0.5) + (1 / 2.0) + (1 / 5.0) = 2.0 + 0.5 + 0.2, with a total sum of 2.7. Calculate the normalized weight coefficients: BESS weight = 2.0 / 2.7 = 0.741, AL1 weight = 0.5 / 2.7 = 0.185, AL2 weight = 0.2 / 2.7 = 0.074. The specific weighting process is as follows: Multiply the power ratio sequence [0.435, 0.217, 0.348] calculated in S401 with the weight coefficient sequence [0.741, 0.185, 0.074] item by item: Weight 1 = 0.435 * 0.741 = 0.322, Weight 2 = 0.217 * 0.185 = 0.040, Weight 3 = 0.348 * 0.074 = 0.026. Calculate the total weighted value: 0.322 + 0.040 + 0.026 = 0.388. Finally, divide each weighted value by the sum of the weighted values ​​(normalization process) to obtain the power allocation parameters: BESS allocation parameter = 0.322 / 0.388 = 0.830, AL1 allocation parameter = 0.040 / 0.388 = 0.103, AL2 allocation parameter = 0.026 / 0.388 = 0.067, generating the power allocation parameter set: {BESS-Alloc = 0.830, AL1-Alloc = 0.103, AL2-Alloc = 0.067.

[0053] S402: Call the power allocation parameter set, compare the difference sequence between the load forecast value and the reference power, mark the positive and negative states of the difference as positive deviation segments and negative deviation segments respectively, perform interval aggregation on the difference segments, calculate the segment deviation signal, and generate a power deviation signal set; The generated power allocation parameter set is invoked, and the deviation calculation process begins. First, the difference sequence between the load forecast and the baseline power is compared. The load forecast is provided by S103 (adjusted to a 150-second step size according to instruction S303). The predicted load sequence (unit: kW) for four consecutive periods starting at 10:20:00 is retrieved: [521.1, 525.0, 523.0, 518.0]. The baseline power is set according to the load smoothing target for the period from 10:00 to 11:00 in S102.

[0054] The baseline power was calculated as 522.0 kW by using a 30-minute moving average of the historical load curve for this period. The difference sequence (load forecast minus baseline power) was calculated as follows: Difference 1 (10:20:00) = 521.1 - 522.0 = -0.9 kW, Difference 2 (10:22:30) = 525.0 - 522.0 = +3.0 kW, Difference 3 (10:25:00) = 523.0 - 522.0 = +1.0 kW, Difference 4 (10:27:30) = 518.0 - 522.0 = -4.0 kW, resulting in the difference sequence: [-0.9, +3.0, +1.0, -4.0]. The positive and negative values ​​of the differences were labeled as positive and negative deviation segments, respectively. The determination process was as follows: Difference 1, -0.9, is less than 0 and is labeled as "negative deviation". Difference 2, +3.0 > 0, is marked as "positive deviation". Difference 3, +1.0 > 0, is marked as "positive deviation". Difference 4, -4.0 < 0, is marked as "negative deviation". The resulting label sequence is: [negative, positive, positive, negative]. Interval aggregation is performed on the difference segments. This process aggregates consecutive identical labels in the label sequence into a single segment: Segment 1 (10:20:00): negative deviation; Segment 2 (10:22:30 to 10:25:00): positive deviation; Segment 3 (10:27:30): negative deviation. Finally, the segment deviation signal is calculated. This calculation sums the differences within each aggregated segment: Segment 1 deviation signal = -0.9kW; Segment 2 deviation signal = +3.0 + 1.0 = +4.0kW; Segment 3 deviation signal = -4.0kW. Generate a power deviation signal set: {Time period 1 (10:20:00): -0.9kW; Time period 2 (10:22:30-10:25:00): +4.0kW; Time period 3 (10:27:30): -4.0kW}.

[0055] S403: Based on the power deviation signal set, extract the deviation symbol identifiers for multiple time periods, set the discharge command for the section with positive deviation, set the charging command for the section with negative deviation, determine the start-stop status code of the adjustable load, aggregate the energy storage control command and the load status code, and generate a dispatch command set. Based on the generated power deviation signal set, specific scheduling instructions are then generated.

[0056] First, extract the deviation sign identifiers for multiple time periods; Time period 1 (10:20:00): deviation signal -0.9kW, negative sign; Time period 2 (10:22:30-10:25:00): deviation signal +4.0kW, positive sign; Time period 3 (10:27:30): deviation signal -4.0kW, negative sign. Set a discharge command for the segments with positive deviations and a charging command for the segments with negative deviations.

[0057] The specific actions in this process are as follows: When the deviation is positive (e.g., +4.0kW in period 2), it indicates that the predicted load is higher than the baseline, and the net load needs to be reduced. The action is set as follows: the energy storage device (BESS) discharges, and the adjustable loads (AL1, AL2) perform power reduction (Curtail). When the deviation is negative (e.g., -0.9kW in period 1), it indicates that the predicted load is lower than the baseline, and the net load needs to be increased. The action is set as follows: the energy storage device (BESS) charges, and the adjustable loads (AL1, AL2) perform power increase (Start / Increase).

[0058] The power allocation parameter set {BESS-Alloc=0.830, AL1-Alloc=0.103, AL2-Alloc=0.067} of S401 is called to allocate the absolute value of the deviation signal: Period 1 (absolute value 0.9kW, action: charging / increasing): BESS charging power = 0.9 * 0.830 = 0.747kW. AL1 increasing power = 0.9 * 0.103 = 0.093kW. AL2 increasing power = 0.9 * 0.067 = 0.060kW. Period 2 (absolute value 4.0kW, action: discharging / reducing): BESS discharging power = 4.0 * 0.830 = 3.320kW. AL1 reducing power = 4.0 * 0.103 = 0.412kW. AL2 reducing power = 4.0 * 0.067 = 0.268kW. Time period 3 (absolute value 4.0kW, action: charging / increasing): BESS charging power = 4.0 * 0.830 = 3.320kW. AL1 increasing power = 4.0 * 0.103 = 0.412kW. AL2 increasing power = 4.0 * 0.067 = 0.268kW. Determine the start / stop status code of the adjustable load.

[0059] The calculated load power is converted into binary start / stop commands (0 for stop / reduce, 1 for start / increase); Time Period 1 Command: AL1 increases by 0.093kW, coded as 1. AL2 increases by 0.060kW, coded as 1. Time Period 2 Command: AL1 decreases by 0.412kW, coded as 0. AL2 decreases by 0.268kW, coded as 0. Time Period 3 Command: AL1 increases by 0.412kW, coded as 1. AL2 increases by 0.268kW, coded as 1. Aggregated energy storage control commands and load status codes: Time period 1 (10:20:00) command: {BESS command = -0.747kW, AL1 command = 1, AL2 command = 1} Time period 2 (10:22:30) command: {BESS command = +3.320kW, AL1 command = 0, AL2 command = 0} Time period 3 (10:27:30) command: {BESS command = -3.320kW, AL1 command = 1, AL2 command = 1}, generating a dispatch command set.

[0060] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the scheduling results and load forecast values ​​of the scheduling instruction set, obtain the instruction execution power value and forecast power value for multiple time periods, calculate the difference vector between the two power sequences, accumulate and calculate the squared deviation values ​​within multiple intervals and perform weighted averaging to generate the mean square result of the scheduling deviation. Based on the scheduling results of the generated scheduling instruction set (10:20:00 to 10:27:30) and the corresponding load forecast values ​​generated by S402, the scheduling deviation assessment begins. First, the instruction execution power values ​​for multiple time periods are acquired. This data is collected by the data acquisition module of S101 at the end of a 150-second (2.5-minute) step after the execution of the S403 instruction, monitoring the actual total active power load of node A. The acquired sequence (instruction execution power values) is as follows: Time Period 1 (end of 10:22:30): 521.2kW; Time Period 2 (end of 10:25:00): 523.0kW; Time Period 3 (end of 10:27:30): 522.0kW; Time Period 4 (end of 10:30:00): 518.1kW. Simultaneously, the corresponding predicted power value sequence (unit: kW) from S402 is acquired: [521.1, 525.0, 523.0, 518.0].

[0061] Next, the difference vector between the two power sequences is calculated. This calculation involves subtracting the predicted power value from the command execution power value for the same time period. Difference 1 = 521.2 - 521.1 = +0.1kW The difference 2 = 523.0 - 525.0 = -2.0 kW. The difference 3 = 522.0 - 523.0 = -1.0 kW. The difference 4 = 518.1 - 518.0 = +0.1kW, and the difference vector is [+0.1, -2.0, -1.0, +0.1]. Iterative analysis is performed on the squared deviation values ​​within multiple intervals.

[0062] Perform a squaring operation on each element in the difference vector: square value 1 = 0.1 * 0.1 = 0.01, square value 2 = -2.0 * -2.0 = 4.0, square value 3 = -1.0 * -1.0 = 1.0, square value 4 = 0.1 * 0.1 = 0.01. The sequence of squared deviation values ​​is [0.01, 4.0, 1.0, 0.01].

[0063] Finally, a weighted average is applied to the squared deviation values. The weighting coefficients are set with reference to the absolute values ​​of the original deviation signals for each time period in S402 [0.9, 3.0, 1.0, 4.0]. This setting is based on the magnitude of the scheduling action; intervals with larger action magnitudes (such as time period 4) receive higher weight in the evaluation.

[0064] Calculate the total weights: 0.9 + 3.0 + 1.0 + 4.0 = 8.9; Calculate the normalized weights: W1 = 0.9 / 8.9 = 0.101, W2 = 3.0 / 8.9 = 0.337, W3 = 1.0 / 8.9 = 0.112, W4 = 4.0 / 8.9 = 0.449; Calculate the weighted average: (0.01 * 0.101) + (4.0 * 0.337) + (1.0 * 0.112) + (0.01 * 0.449) = 0.00101 + 1.348 + 0.112 + 0.00449 = 1.46552. This 1.46552 is the mean square result of the generated scheduling deviation.

[0065] S502: Call the mean square result of scheduling deviation, retrieve the prediction error data of multiple time periods during the scheduling process and normalize it, compare the difference with the mean square value of scheduling deviation, calculate the weight adjustment coefficient, and perform a product operation between the adjustment coefficient and the original step size weight parameter to update the step size control parameter and obtain the step size weight parameter set. The generated mean squared result of the scheduling deviation is retrieved, with a value of 1.46552. Next, prediction error data for multiple time periods during the scheduling process is retrieved. This data refers to the metric used in S301 to evaluate the original prediction performance, specifically the mean squared error of the error sequence [+3.5, +2.3, -2.6, +2.0] shown in Table 3 of S301.

[0066] The calculation process is as follows: mean error = 1.3kW; sum of squared deviations = 21.54; mean squared error of prediction (MSE) = 21.54 / 4 = 5.385. This 5.385 is the retrieved prediction error data. Next, the numerical difference between the mean squared result of scheduling deviations and the prediction error data is compared.

[0067] It should be noted that this comparison is performed by calculating the ratio between the two: Error Ratio = Mean Square of Scheduling Deviation divided by Prediction Error Data = 1.46552 / 5.385 = 0.272. The weight adjustment factor is calculated based on this error ratio. The determination of the weight adjustment factor is shown in Table 5.

[0068] Table 5 Experimental Table of Weight Adjustment Coefficients

[0069] As shown in Table 5, the adjustment coefficient was determined through 1000 historical backtesting scheduling cycles. When the error ratio is in the range of (0.20, 0.40), it indicates that the scheduling execution has significantly improved the prediction uncertainty, but there is still room for optimization. In this case, the adjustment coefficient is set to 0.9 to slightly shorten the step size weight. The currently calculated error ratio is 0.272, which is within the range of (0.20, 0.40).

[0070] Therefore, a weight adjustment coefficient of 0.9 is selected. Finally, this adjustment coefficient (0.9) is multiplied by the original step size weight parameter (i.e., the "step size adjustment coefficient" of 0.5 calculated in S303) to update the step size control parameter. The updated step size weight = 0.5 * 0.9 = 0.45. This 0.45 is used as the updated step size weight and stored in the generated step size weight parameter set.

[0071] S503: Based on the step size weight parameter set, re-invoke the instruction to execute the power value and load forecast value, perform weighted fusion on the two types of power sequences according to the updated step size weight, perform moving average smoothing, extract the power distribution sequence after step size adjustment, and generate the load balancing scheduling result. Based on the generated step-size weight parameter set, obtain the updated step-size weight (fusion weight), which has a value of 0.45. Recall the instructions in S501 to execute the power values ​​(actual load sequence) [521.2, 523.0, 522.0, 518.1] (unit kW), and the load forecast values ​​(predicted load sequence) [521.1, 525.0, 523.0, 518.0] (unit kW) referenced in S501.

[0072] The two power sequences are weighted and fused according to the updated step size weight (0.45). In this fusion calculation, the weight of the predicted load sequence is 0.45, and the weight of the command execution power value sequence is (1-0.45)=0.55.

[0073] The fusion calculation process is as follows: Fusion value 1 = (521.1 * 0.45) + (521.2 * 0.55) = 234.495 + 286.66 = 521.155 kW; Fusion value 2 = (525.0 * 0.45) + (523.0 * 0.55) = 236.25 + 287.65 = 523.9 kW; Fusion value 3 = (523.0 * 0.45) + (522.0 * 0.55) = 235.35 + 287.1 = 522.45 kW; Fusion value 4 = (518.0 * 0.45) + (518.1 * 0.55) = 233.1 + 284.955 = 518.055 kW; The resulting weighted fusion sequence is [521.155, 523.9, 522.45, 518.055]. Next, interval smoothing is performed on this fusion sequence.

[0074] This smoothing is calculated using a two-point simple moving average (SMA): smoothing value 1 = 521.155 (the first data point has no prior value and remains unchanged), smoothing value 2 = (521.155 + 523.9) / 2 = 522.5275 kW, smoothing value 3 = (523.9 + 522.45) / 2 = 523.175 kW, smoothing value 4 = (522.45 + 518.055) / 2 = 520.2525 kW. The power distribution sequence after step size adjustment is extracted, i.e., the smoothed sequence [521.155, 522.5275, 523.175, 520.2525]. The above sequence is the generated distribution network load balancing dispatch result.

[0075] Please see Figure 7 The model predictive control-based distribution network load balancing dispatching system provided in this application includes: The data analysis module collects and normalizes voltage, current and power data of the distribution network and photovoltaic irradiance and wind speed at the distributed power source end, establishes a load-power source coordination model of the distribution network and calculates the load forecast value for future periods, generates an optimized input dataset and passes it to the constraint solving module. The constraint solving module analyzes and solves the power balance equation of the distribution network, the charging and discharging constraint equation of energy storage equipment, and the line capacity constraint equation based on the optimized input dataset. It calculates the line power distribution and voltage level, generates an optimized control variable set, and passes it to the rolling adjustment module. The rolling adjustment module calls the optimized control variable set, calculates the error of the continuous prediction cycle and judges the time correlation, analyzes the rolling step size adjustment coefficient and corrects the rolling time length of the model predictive control, generates the step size adjustment instruction and transmits it to the load scheduling module. The load dispatching module, based on step adjustment instructions, determines the power allocation of energy storage devices and adjustable loads, compares the difference between the load forecast value and the baseline power, adjusts the charging or discharging instructions according to the positive or negative value of the difference, determines the start-stop status of the adjustable load, generates a dispatching instruction set, and transmits it to the deviation correction module. The deviation correction module compares the scheduling results based on the scheduling instruction set with the load forecast values, calculates the root mean square error of the scheduling deviation and the forecast error, adjusts the step size weights, and generates the load balancing scheduling results.

[0076] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A distribution network load balancing dispatching method based on model predictive control, characterized in that, Includes the following steps: S1: Collect voltage, current and power data of the distribution network and photovoltaic irradiance and wind speed at the distributed power source end and normalize them, establish a load-power source coordination model of the distribution network and calculate the load forecast value for future periods, and generate an optimized input dataset. S2: Based on the optimized input dataset, analyze and solve the power balance equation of the distribution network, the charging and discharging constraint equation of the energy storage device and the line capacity constraint equation, calculate the line power distribution and voltage level, and generate an optimized control variable set; S3: Call the set of optimized control variables, calculate the error of the continuous prediction cycle and determine the temporal correlation, analyze the rolling step size adjustment coefficient and correct the rolling time length of the model predictive control, and generate the step size adjustment instruction; S4: Based on the step size adjustment command, determine the power allocation of the energy storage device and the adjustable load, compare the difference between the load prediction value and the reference power, adjust the charging or discharging command according to the positive or negative value of the difference, determine the start and stop status of the adjustable load, and generate a scheduling command set. S5: Based on the comparison between the scheduling results of the scheduling instruction set and the load forecast value, calculate the mean square error of the scheduling deviation and the forecast error, adjust the step size weight, and generate the load balancing scheduling result.

2. The distribution network load balancing dispatching method based on model predictive control according to claim 1, characterized in that, The optimized input dataset includes load forecast feature items, photovoltaic irradiance feature items, and wind speed feature items. The optimized control variable set includes line power allocation item, voltage level item, and energy storage charging and discharging constraint item. The step size adjustment instruction includes prediction error correlation coefficient, rolling step size correction coefficient, and time series correlation parameter. The scheduling instruction set includes energy storage charging and discharging instruction item, adjustable load start and stop item, and power difference correction item. The load balancing scheduling result includes scheduling deviation, prediction error root mean square error, and step size weighting coefficient.

3. The distribution network load balancing dispatching method based on model predictive control according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collect voltage, current and power data of the distribution network, and simultaneously acquire photovoltaic irradiance and wind speed data of the distributed power source. Synchronize and normalize all data in time, analyze the load-power coordination model of the distribution network, and generate a standardized set of operating parameters. S102: Based on the standardized operating parameter set, voltage, current and power are called, and feature vector splicing is performed on the parameter distribution of power distribution and distributed power sources within the same time period. The power distribution features are correlated and compared with irradiance and wind speed features, and the correlation matrix is ​​calculated to generate a load power source coordinated feature set. S103: Based on the load power supply coordination feature set, a sliding window is applied to the time series distribution to extract the load change gradient over multiple time periods. The power trend series over consecutive time periods is then weighted and averaged to calculate the load forecast for future time periods. This forecast is then integrated with the load power supply coordination features to generate an optimized input dataset.

4. The distribution network load balancing dispatching method based on model predictive control according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the optimized input dataset, load forecast values ​​and power distribution data are called to match and calculate the power injection and power flow of multiple branches of the distribution network, calculate the difference between active and reactive power and perform constraint adjustment, analyze the power balance relationship, and generate a power balance parameter set. S202: Based on the power balance parameter set, extract the charging and discharging state variables at the energy storage device end, and compare them with the charging power, discharging power and state of charge change range. If they exceed the range, redefine the corresponding power range, correct the power output for multiple time periods, and generate an energy storage constraint dataset. S203: Based on the energy storage constraint dataset, obtain power parameters and line power data, perform allocation of active and reactive power components of multiple lines, compare the deviation between the calculation results and the line capacity threshold according to the line capacity constraint equation, analyze the line power allocation and voltage level, and generate an optimized control variable set.

5. The distribution network load balancing dispatching method based on model predictive control according to claim 4, characterized in that, The line capacity threshold is determined by extracting the measured current values ​​of multiple lines under multiple load conditions and the corresponding allowable current values ​​of the conductors, performing interval statistical calculations on the ratio of the two, and analyzing the average load rate.

6. The distribution network load balancing dispatching method based on model predictive control according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the optimized control variable set, call the continuous prediction cycle power data and the current running data and calculate the difference, extract the multi-time period error sequence, and perform statistical aggregation on the difference sequence to extract the error mean and fluctuation amplitude parameter set of the continuous cycle, and generate the prediction error parameter set; S302: Based on the predicted error parameter set, call the error mean sequence and fluctuation amplitude parameter set, perform correlation analysis on the error change rate of adjacent periods, calculate the autocorrelation coefficient of the error change rate with time, screen time series segments, extract time-related feature factors, and generate a time-related feature factor set. S303: Call the time-series associated feature factor set, obtain time-series weight distribution data and predictive control time parameters, perform proportional analysis on the rolling step length adjustment coefficients of multiple time periods, and correct the rolling time length according to the step length adjustment benchmark value and perform instruction encoding to generate step length adjustment instructions.

7. The distribution network load balancing dispatching method based on model predictive control according to claim 6, characterized in that, The step size adjustment benchmark value is determined by extracting the prediction error change rate and the corresponding correlation strength coefficient within multiple rolling cycles, performing normalization on the two, and then calculating the weighted average.

8. The distribution network load balancing dispatching method based on model predictive control according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the step size adjustment command, obtain the rated power of the energy storage device and the operating power of the adjustable load, perform proportional allocation calculation, analyze the allocation ratio of energy storage power and load power in each time period, and perform weighted processing on the proportional sequence to generate a power allocation parameter set. S402: Call the power allocation parameter set, compare the difference sequence between the load forecast value and the reference power, mark the positive and negative states of the difference as positive deviation segments and negative deviation segments respectively, perform interval aggregation on the difference segments, calculate the segment deviation signal, and generate a power deviation signal set; S403: Based on the power deviation signal set, extract the deviation symbol identifiers for multiple time periods, set the discharge command for the section with positive deviation, set the charging command for the section with negative deviation, determine the start-stop status code of the adjustable load, aggregate the energy storage control command and the load status code, and generate a scheduling command set. The reference power is set by performing difference analysis on the current power data and predicted power data of the load during the operating cycle, and the sum of the mean of the difference sequence and the standard deviation of the preset multiple is analyzed.

9. The distribution network load balancing dispatching method based on model predictive control according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the scheduling results and load forecast values ​​of the scheduling instruction set, obtain the instruction execution power value and forecast power value for multiple time periods, calculate the difference vector between the two power sequences, accumulate and calculate the squared deviation values ​​within multiple intervals and perform a weighted average to generate the mean square result of the scheduling deviation. S502: Call the mean square result of the scheduling deviation, retrieve the prediction error data of multiple time periods during the scheduling process and normalize it, compare the difference with the mean square value of the scheduling deviation, calculate the weight adjustment coefficient, and perform a product operation between the adjustment coefficient and the original step size weight parameter to update the step size control parameter and obtain the step size weight parameter set; S503: Based on the set of step size weight parameters, re-invoke the power value and load prediction value, perform weighted fusion on the two power sequences according to the updated step size weight, perform moving average smoothing, extract the power distribution sequence after step size adjustment, and generate load balancing scheduling results.

10. A distribution network load balancing dispatching system based on model predictive control, characterized in that, The system is used to implement the distribution network load balancing dispatching method based on model predictive control as described in any one of claims 1-9, wherein the distribution network load balancing dispatching system comprises: The data analysis module collects and normalizes voltage, current and power data of the distribution network and photovoltaic irradiance and wind speed at the distributed power source end, establishes a load-power source coordination model of the distribution network and calculates the load forecast value for future periods, generates an optimized input dataset and passes it to the constraint solving module. The constraint solving module, based on the optimized input dataset, analyzes and solves the power balance equation of the distribution network, the charging and discharging constraint equation of the energy storage device, and the line capacity constraint equation, calculates the line power distribution and voltage level, generates an optimized control variable set, and passes it to the rolling adjustment module. The rolling adjustment module calls the optimized control variable set, calculates the error of the continuous prediction cycle and judges the time correlation, analyzes the rolling step size adjustment coefficient and corrects the rolling time length of the model predictive control, generates the step size adjustment instruction and transmits it to the load scheduling module. The load scheduling module, based on the step size adjustment command, determines the power allocation of the energy storage device and the adjustable load, compares the difference between the load prediction value and the reference power, adjusts the charging or discharging command according to the positive or negative value of the difference, determines the start-stop status of the adjustable load, generates a scheduling command set, and transmits it to the deviation correction module. The deviation correction module compares the scheduling results of the scheduling instruction set with the load forecast values, calculates the root mean square error of the scheduling deviation and the forecast error, adjusts the step size weights, and generates a load balancing scheduling result.

Images