Multi-objective coordinated control system and method of flexible on-load tap-changing transformer based on hourly load prediction

Abstract

The application discloses a flexible on-load voltage regulating transformer multi-target cooperative control system and method based on hour-level load prediction, which comprises the following steps: a node parameter acquisition module acquires electrical parameters of each node of a power distribution network; a load prediction module outputs a load prediction sequence in the future time; a multi-target optimization module generates an optimal control optimization strategy according to a cooperative optimization model; an upper prediction optimization layer makes an hour-level optimization decision on tap position and capacitor group switching; a lower real-time adjustment layer outputs a voltage dynamic compensation amount and generates a minute-level equipment adjustment instruction; and an execution module coordinates and controls flexible on-load voltage regulating transformer (FOLTC) tap switching, capacitor group switching and converter action, so that multi-target optimization of voltage deviation, reactive power compensation and equipment loss is realized. The application is suitable for voltage optimization, reactive power compensation and loss control in a conventional load fluctuation scenario of a power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power system automation and power quality control technology, and in particular to a multi-objective collaborative control system and method for flexible on-load tap-changing transformers based on hourly load forecasting, which is applicable to voltage optimization, reactive power compensation and loss control under the scenario of normal load fluctuation in distribution networks. Background Technology

[0002] As distribution networks become increasingly complex, traditional regulation methods are gradually revealing many shortcomings, especially against the backdrop of increasingly severe load fluctuations. The stability and efficiency of power systems are facing unprecedented challenges. Therefore, addressing issues such as voltage fluctuations, reactive power imbalances, and line losses has become crucial for improving the operational quality of distribution networks.

[0003] In traditional power grids, industrial, commercial, and residential loads exhibit significant periodic fluctuations. These fluctuations are not solely caused by seasonal and weather factors but are also influenced by economic activities, social habits, and policy regulations. In particular, the cyclical loads of industrial production tend to concentrate during morning and evening peak hours. This concentrated load characteristic makes it difficult for distribution network voltage regulation systems to cope with sudden load fluctuations. Furthermore, this increased load fluctuation leads to a lag in the response of traditional voltage regulation equipment, making it difficult for the power grid to efficiently and accurately regulate voltage and reactive power, thus further exacerbating grid instability.

[0004] Flexible on-load tap-changing transformers (FOLCTs) offer high regulation flexibility and can adapt to complex load changes through the coordinated action of converters and tap changes. However, current technologies still struggle to achieve predictive regulation of load fluctuations. While current flexible transformers can respond to voltage changes, they often rely on back-end feedback mechanisms, and the regulation process remains limited by insufficient predictability of future load fluctuations. This prevents transformers from making forward-looking adjustments in the face of load fluctuations, leading to regulation lag and low efficiency.

[0005] Furthermore, although flexible transformers offer higher regulation accuracy and faster response capabilities, the lack of an intelligent optimization system means that existing regulation strategies are often simplistic and fail to comprehensively consider multiple objectives such as voltage, reactive power, and line losses. During grid operation, the lack of a multi-objective collaborative optimization mechanism often leads to insufficient attention being paid to other parameters (such as reactive power balance and losses) during voltage regulation. Without a unified optimization framework to coordinate these objectives, simplistic regulation can result in reduced grid efficiency, increased losses, and even system instability.

[0006] While current flexible transformers offer high regulation flexibility, most still lack sufficient intelligent control systems to adaptively adjust based on load forecast data and grid operating conditions. Transformer regulation typically relies on manual settings or experience based on past data, lacking the ability to respond quickly to real-time changes in the grid. To achieve truly intelligent regulation, more intelligent algorithms need to be combined with real-time monitoring data for precise control.

[0007] Distribution networks involve multi-dimensional data sources, including voltage, load, reactive power, and environmental factors. These data require efficient integration and rapid processing for real-time regulation. However, existing optimization algorithms often face challenges such as massive data volumes and high algorithm complexity, making efficient real-time regulation under complex power grid conditions difficult. Especially under disturbances caused by load fluctuations and changes in the external environment, traditional algorithms struggle to cope with unexpected situations.

[0008] Although flexible on-load tap-changing transformers offer high regulation accuracy, the frequent adjustments during operation can place significant stress on the equipment, impacting its long-term stability and reliability. Furthermore, the introduction of intelligent control systems may increase hardware and maintenance costs, posing a challenge to the long-term economic viability of the equipment.

[0009] With the continuous development of power distribution networks, the instability of load demand has brought many challenges to the operation of power systems. Especially in daily operation, frequent load fluctuations not only lead to unstable voltage quality but may also cause reactive power compensation imbalances, thus affecting the energy efficiency and stability of the entire power distribution network. At the same time, excessive line losses and frequent equipment adjustments also increase the system's operating costs and maintenance burden.

[0010] Therefore, how to reduce maintenance costs and extend equipment lifespan while maintaining efficient equipment adjustment remains a challenge for technological development. Summary of the Invention

[0011] The purpose of this invention is to address the problems of unstable voltage quality, reactive power compensation imbalance, and increased line losses caused by load fluctuations in distribution networks. It provides a multi-objective collaborative control system and method for flexible on-load tap-changing transformers based on hourly load forecasting. This system aims to solve a series of problems caused by the significant hourly fluctuations in conventional loads during operation in distribution networks, such as voltage fluctuations, reactive power imbalance, and line losses. By accurately predicting future load fluctuation trends and combining multi-objective collaborative control of grid voltage optimization, reactive power compensation, and line losses, the system can achieve more precise regulation at different time periods of load change.

[0012] One aspect of the present invention provides a multi-objective cooperative control system for flexible on-load tap-changing transformers based on hourly load forecasting, comprising:

[0013] The node parameter acquisition module is used to collect the electrical parameters of each node in the distribution network in real time.

[0014] The load forecasting module is used to perform variational mode decomposition and optimization forecasting on the power characteristic curve of the power grid based on the electrical parameters of each node in the distribution network, and output the load forecast sequence for the future time.

[0015] The multi-objective optimization module is used to generate the optimal control optimization strategy for tap position and capacitor bank switching combination based on the constructed collaborative optimization model that includes voltage deviation, reactive power compensation and equipment loss, with the objectives of minimizing voltage deviation, optimizing reactive power compensation and minimizing equipment loss, through dynamic weight adjustment.

[0016] The hierarchical control module includes an upper predictive optimization layer and a lower real-time adjustment layer. The upper predictive optimization layer uses a mixed-integer linear programming (MILP) algorithm, based on the optimal control optimization strategy, to make hourly optimization decisions and adjustment plans for tap positions and capacitor bank switching according to the load forecast sequence and the current state of the equipment. The lower real-time adjustment layer generates minute-level equipment adjustment commands by using the voltage dynamic compensation output by the proportional-integral (PI) controller when adjustment is triggered.

[0017] The execution module is used to coordinate and control the tap switching of the flexible on-load tap changer (FOLTC), the switching of capacitor banks, and the operation of the converter according to the equipment adjustment instructions output by the lower real-time adjustment layer, so as to realize the reactive power output of the converter and achieve multi-objective optimization of voltage deviation, reactive power compensation, and equipment loss.

[0018] Specifically, the load forecasting module includes:

[0019] VMD unit is used to decompose the power characteristic curve of the grid into multiple IMF components;

[0020] The BES optimization unit is used to optimize the penalty parameters and kernel function parameters of the LSSVM prediction unit using the BES optimization algorithm.

[0021] The LSSVM forecasting unit is used to independently forecast and reconstruct each IMF component, outputting a 24-hour load forecast sequence that includes forecast results of different types of load capacity and power factor.

[0022] The objective function of the multi-objective optimization module is:

[0023] minF=ω1f U +ω2fQ +ω3f L ;

[0024] The objective is to minimize voltage deviation;

[0025] The goal is to optimize reactive power compensation.

[0026] f L =C tap ΔT(t)+C cb ΔH(t) is the objective of minimizing equipment losses;

[0027] ω1, ω2, and ω3 are adjustable weighting coefficients, U i (t) represents the voltage at node i, U ref U represents the reference voltage. max and U min These are the maximum and minimum allowable values ​​of the voltage, Q. OLTC (t) is the reactive power compensation of the flexible OLTC converter, Q CB (t) is the reactive power compensation of the capacitor bank. This represents the reactive power demand of the load, indicating the optimal reactive power compensation capacity. ΔT(t) is the change in the number of tap changes, ΔH(t) is the change in the number of capacitor bank changes, and C... tap and C cb These are the costs of each tap changer and capacitor bank operation, respectively.

[0028] The upper-level prediction and optimization layer employs the MILP algorithm for hourly-level optimization decisions and adjustment planning for tap positions and capacitor bank switching, including:

[0029] Pre-adjust tap positions one hour before peak load; prioritize switching capacitor banks; generate tap and capacitor bank operation sequences for the next 24 hours.

[0030] The lower real-time adjustment layer outputs minute-level dynamic voltage compensation through a proportional-integral (PI) controller. The PI controller outputs the dynamic voltage compensation according to the following formula:

[0031] ΔU=K p ·e(t)+K i ·∫e(t)dt;

[0032] Where ΔU is the voltage dynamic compensation amount, e(t) is the real-time voltage deviation, and K p and K i These are the proportional coefficient and the integral coefficient.

[0033] The electrical parameters include one or more of voltage, current, power, and power factor.

[0034] The load forecasting module, in conjunction with the Prophet model to assist in load forecasting, automatically identifies seasonal fluctuation patterns and outputs a load forecast sequence for a preset future time period.

[0035] The lower-level real-time adjustment layer receives real-time data from the power distribution network and calculates real-time voltage deviation and reactive power compensation error through a PI controller. When the real-time voltage deviation or reactive power compensation error exceeds its respective threshold, adjustment is triggered, and equipment adjustment commands are generated.

[0036] When the lower-level real-time adjustment layer generates equipment adjustment commands, it prioritizes generating converter adjustment commands; if the voltage deviation is greater than the threshold and continues for more than a preset time, it generates an emergency tap adjustment command, or if the reactive power compensation error is greater than its threshold, it generates a capacitor bank replenishment command.

[0037] Another aspect of the present invention provides a control method for a multi-objective cooperative control system of a flexible on-load tap-changing transformer based on hourly load forecasting, comprising the following steps:

[0038] Real-time acquisition of electrical parameters at each node of the power distribution network;

[0039] Based on the electrical parameters of each node in the distribution network, the VMD-BES-LSSVM hybrid algorithm is used to perform variational mode decomposition and optimization prediction on the power characteristic curve of the power grid, and output the load prediction sequence for the future time period.

[0040] Based on the constructed collaborative optimization model that includes voltage deviation, reactive power compensation and equipment loss, with the goals of minimizing voltage deviation, optimizing reactive power compensation and minimizing equipment loss, the optimal control optimization strategy for generating tap positions and capacitor bank switching combinations is achieved through dynamic weight adjustment.

[0041] The upper-level prediction and optimization layer adopts the mixed integer linear programming (MILP) algorithm. Based on the optimal control optimization strategy, it makes hourly optimization decisions and adjustment plans for tap positions and capacitor bank switching according to the load prediction sequence and the current status of the equipment. When the lower-level real-time adjustment layer of the hierarchical control module is triggered, it generates minute-level equipment adjustment commands through the voltage dynamic compensation amount output by the proportional-integral (PI) controller.

[0042] In response to equipment adjustment commands, the flexible on-load tap changer (FOLTC) switches taps, switches capacitor banks, or operates the converter to achieve reactive power output from the converter, thus realizing multi-objective optimization of voltage deviation, reactive power compensation, and equipment losses.

[0043] The system of this invention not only optimizes the accuracy of reactive power compensation but also reduces reliance on frequent equipment operation, thereby reducing equipment wear and maintenance requirements. It enables the distribution network to significantly improve energy utilization efficiency, reduce energy loss, and enhance overall economy and operational reliability while ensuring the safety and stability of power supply. This greatly promotes the improvement of the distribution network's intelligence and automation level and provides strong support for coping with the high load and volatility challenges of future power systems.

[0044] This invention introduces hourly load forecasting to predict the load change trend of the distribution network within a time range of 1 to 24 hours, thereby enabling the formulation of corresponding adjustment strategies before load fluctuations occur. The core advantage of this forward-looking adjustment capability lies in its ability to identify potential voltage fluctuations or reactive power imbalances in advance based on predicted load data, and to adjust them through flexible on-load tap-changing transformers (FOLTC). This avoids the suddenness and lag of voltage deviations, thus overcoming the problems or defects of traditional voltage regulation methods, which typically rely on real-time voltage deviations for response. Adjustments are only made when voltage deviations occur, resulting in delays and lags that lead to persistent voltage deviations and an inability to anticipate and adjust before load changes. This is especially problematic when load fluctuations are large, as voltage fluctuations and over-limit durations increase significantly.

[0045] The technology of this invention is particularly applicable to distribution network scenarios with significant hourly fluctuation characteristics, providing effective technical support for the construction of smart grids. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the single-phase topology of a flexible OLTC.

[0047] Figure 2 This is a schematic diagram of a multi-objective collaborative control system for flexible on-load tap-changing transformers.

[0048] Figure 3 This is a schematic diagram illustrating the training of the prediction model for the load prediction module.

[0049] Figure 4 This is a schematic diagram of the adjustment process of an on-load tap-changing transformer.

[0050] Figure 5 This is a flowchart of a multi-objective collaborative control method for on-load tap-changing transformers.

[0051] Figure 6 This is a schematic diagram of the training process for the prediction model of the load prediction module. Detailed Implementation

[0052] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0053] See Figure 2 As shown in the exemplary embodiment of this application, the flexible on-load tap-changing transformer multi-objective cooperative control system based on hourly load forecasting includes:

[0054] The node parameter acquisition module is used to collect the electrical parameters of each node in the distribution network in real time.

[0055] The load forecasting module is used to perform variational mode decomposition and optimization forecasting on the power characteristic curve of the power grid based on the electrical parameters of each node in the distribution network, and output the load forecast sequence for the future time.

[0056] The multi-objective optimization module is used to generate the optimal control optimization strategy for tap position and capacitor bank switching combination based on the constructed collaborative optimization model that includes voltage deviation, reactive power compensation and equipment loss, with the objectives of minimizing voltage deviation, optimizing reactive power compensation and minimizing equipment loss, through dynamic weight adjustment.

[0057] The hierarchical control module includes an upper predictive optimization layer and a lower real-time adjustment layer. The upper predictive optimization layer uses a mixed-integer linear programming (MILP) algorithm, based on the optimal control optimization strategy, to make hourly optimization decisions and adjustment plans for tap positions and capacitor bank switching according to the load forecast sequence and the current state of the equipment. The lower real-time adjustment layer generates minute-level equipment adjustment commands by using the voltage dynamic compensation output by the proportional-integral (PI) controller when adjustment is triggered.

[0058] The execution module is used to coordinate and control the tap switching of the flexible on-load tap changer (FOLTC), the switching of capacitor banks, and the operation of the converter according to the equipment adjustment instructions output by the lower real-time adjustment layer, so as to realize the reactive power output of the converter and achieve multi-objective optimization of voltage deviation, reactive power compensation, and equipment loss.

[0059] This invention predicts future load change trends and adjusts transformer taps and converter operating states in advance to achieve dynamic optimization control of voltage, ensuring that the distribution network can maintain good voltage stability and reactive power balance under any load conditions.

[0060] In this embodiment of the invention, the load forecasting module performs variational mode decomposition and optimization forecasting on the power grid power characteristic curve using the VMD-BES-LSSVM method, predicts the power grid power characteristics and the capacity of each type of load, and outputs a load forecast sequence for a preset future time period, which includes:

[0061] VMD unit is used to decompose the power characteristic curve of the grid into multiple IMF components;

[0062] The BES optimization unit is used to optimize the penalty parameters and kernel function parameters of the LSSVM prediction unit using the BES optimization algorithm.

[0063] The LSSVM forecasting unit is used to independently forecast and reconstruct each IMF component, outputting a 24-hour load forecast sequence that includes forecast results of different types of load capacity and power factor.

[0064] The model building process of the load forecasting module is as follows: Figure 3 As shown, electrical parameter data is first collected and divided into training and test sets. Then, the Variational Mode Decomposition (VMD) method is used to decompose the power characteristic curve data of a single hour in the training set, extracting various load characteristics and capacities to decompose the original data. The VMD method is applied to the power characteristic curve of a single hour to decompose the complex power curve into multiple sub-signals with different frequency components. These sub-signal components represent different types of load characteristics and capacities in the power grid. Through this decomposition, the load patterns of the power grid at different time scales can be identified, and the accuracy and stability of load forecasting can be improved by reducing nonlinear components.

[0065] In the VMD method, the modal parameter K is set according to the data processing needs. The magnitude of the modal parameter indicates that the signal will be decomposed into multiple independent modal components. A penalty factor α controls the smoothness of the modal components; a larger α value makes the modes smoother, thereby reducing the occurrence of high-frequency components. Furthermore, an adaptive stopping criterion is set to ensure the accuracy of the decomposition process by setting a threshold for the correlation coefficient between modes, avoiding over-decomposition. The core objective of VMD is to optimize the frequency center of each mode by minimizing a variational problem, ensuring that each modal component can independently reflect a periodic component in the signal. VMD efficiently optimizes this variational problem using the Alternating Direction Multiplier Method (ADMM). ADMM decomposes the problem into multiple sub-problems, progressively optimizing the frequency and time characteristics of each modal component. Ultimately, VMD can successfully decompose several modal components with a clear frequency center in the frequency domain, effectively capturing different frequency characteristics of the signal.

[0066] Secondly, the penalty parameters and kernel function parameters in LSSVM are optimized using BES (Binary Eagle Strategy) to construct a BES-LSSVM prediction model. The BES optimization algorithm is employed to optimize the penalty parameters and kernel function parameters of the LSSVM (Least Squares Support Vector Machine) model. Through fine-tuning of these parameters, LSSVM can more effectively fit the power grid power characteristic curve, thereby improving the prediction accuracy of load and capacity. LSSVM is then used to predict the capacity and power factor of different types of loads through regression analysis.

[0067] Specifically, when optimizing the penalty parameters and kernel function parameters in LSSVM using BES, the BES parameters (penalty parameters and kernel function parameters) are initialized, the operating parameters (penalty parameters and kernel function parameters) for a single type of load are calculated, the operating parameters (penalty parameters and kernel function parameters) for a single type of load are updated, and the operating parameters (penalty parameters and kernel function parameters) for calculating the total load are updated. When the maximum number of iterations is reached, the optimal values ​​of the parameters (penalty parameters and kernel function parameters) are output; otherwise, the parameters (penalty parameters and kernel function parameters) for a single type of load are updated.

[0068] Finally, after optimizing the penalty and kernel function parameters in LSSVM using BES, each IMF component from the VMD-decomposed training set is input into LSSVM for independent prediction, and the final prediction result is reconstructed. Each IMF component after VMD decomposition is used as input, forming a sub-sequence, which is then input into the LSSVM model for prediction, and the prediction result is output. Each IMF component represents the characteristics of different load categories, such as daily load fluctuations and peak loads. By predicting the capacitance and power factor of each IMF component, the prediction results of each sub-signal are reconstructed to obtain the overall power grid load prediction result, including the capacitance and power factor of each load category.

[0069] After training with the training set, the test set is used to test and obtain the predicted results of the capacity and power factor of the required different types of loads, that is, the load prediction series for a future predetermined time such as 24 hours.

[0070] By combining VMD-BES-LSSVM, not only can the grid load be accurately decomposed, but the capacity and power factor of each load category can also be effectively predicted, thus providing a more accurate load forecast basis for the operation and management of the grid.

[0071] In this embodiment of the invention, the multi-objective optimization module, through its multi-objective collaborative optimization model, performs multi-objective optimization with the objectives of "minimizing voltage deviation, optimizing reactive power compensation, and minimizing equipment loss," and constructs the multi-objective function as minF=ω1f. U +ω2f Q +ω3f L ,like Figure 2 As shown, each objective is assigned a weight coefficient ω1, ω2, ω3, representing the importance of each objective. In practical applications, these weight coefficients can be adjusted according to actual needs to achieve different optimization objectives.

[0072] Among them, f U This represents the goal of minimizing voltage deviation, used to minimize the voltage U at each node. i (t) is within a preset range, typically 0.95-1.05 pu. To achieve f U The objective is to first calculate the voltage deviation of each node relative to the reference voltage U. ref The difference is expressed by the following formula:

[0073]

[0074] Among them, U i (t) is the voltage at node i, U ref It is the reference voltage, U max and U min These are the maximum and minimum allowable values ​​for the voltage, respectively. The goal is to minimize the voltage deviation and make the voltage as close as possible to the reference voltage.

[0075] Secondly, the optimal objective of reactive power compensation is f. Q By calculating the optimal reactive power compensation capacity ( This optimizes the efficiency of reactive power compensation. The formula is:

[0076]

[0077] Among them, Q OLTC (t) is the reactive power compensation of the flexible OLTC converter, Q CB (t) is the reactive power compensation of the capacitor bank. This refers to the reactive power demand of the load. Q The goal is to make the total reactive power compensation amount close to the expected reactive power demand of the load. This improves compensation efficiency.

[0078] The last one is to calculate the minimum loss target of the computing equipment (f) L The equipment loss objective is to minimize the operating costs of tap changers and capacitor banks, as shown in the formula:

[0079] f L=C tap ΔT(t)+C cb ΔH(t)

[0080] Where ΔT(t) is the change in the number of tap changes, ΔH(t) is the change in the number of capacitor bank changes, and C tap and c cb These are the costs of each tap changer operation and each capacitor bank operation. The goal of equipment depreciation is to reduce the number of equipment operations, thereby reducing costs and equipment depreciation.

[0081] Ultimately, the three optimization objectives mentioned above—minimizing voltage deviation, optimizing reactive power compensation, and minimizing equipment losses—are combined through the aforementioned multi-objective optimization function.

[0082] In this embodiment of the invention, after load forecasting and target optimization for future prediction times, a hierarchical control strategy is employed for further optimization, including upper-level load forecasting optimization and lower-level real-time voltage regulation, to improve the stability of the power grid under load fluctuations. Specifically, by dividing the power system regulation task into an upper-level forecasting optimization layer and an upper-level real-time regulation layer, mixed-integer linear programming (MILP) algorithm and proportional-integral (PI) controller are used respectively to optimize the load distribution and voltage regulation of the power grid, thereby achieving efficient and stable operation of the power grid.

[0083] The upper-level prediction and optimization layer (hourly level) first generates a load forecast sequence for the next 24 hours based on historical data and the prediction model of the load forecast module. Based on the optimal control optimization strategy output by the multi-objective optimization module, it performs hourly optimization decisions and adjustment planning for tap positions and capacitor bank switching, including: pre-adjusting tap positions one hour before the load peak; prioritizing capacitor bank switching; and generating a tap and capacitor bank action sequence for the next 24 hours. When the lower-level real-time adjustment layer triggers adjustment, it generates minute-level equipment adjustment commands through the voltage dynamic compensation output by the proportional-integral PI controller.

[0084] The upper-level prediction and optimization layer, based on the prediction of load change trends, can identify peak load periods in advance. It employs a mixed-integer linear programming (MILP) method to perform multi-objective optimization for load fluctuations and voltage regulation needs. The optimization process generates tap positions (T(t)) and capacitor bank switching combinations (H(t)) to reduce reactive power compensation requirements and improve grid stability. The optimization objectives of this upper-level prediction and optimization layer include adjusting tap positions in advance to cope with peak loads. For example, one hour before the predicted load increase, by raising the tap position, the impact of the load surge on the grid is mitigated; and by prioritizing the use of capacitor banks for reactive power compensation, frequent tap operations are reduced, extending the lifespan of power equipment and reducing system energy losses.

[0085] In this embodiment of the invention, the lower real-time adjustment layer (minute-level) generates a minute-level device adjustment command by using the voltage dynamic compensation amount output by the proportional-integral (PI) controller when adjustment is triggered; wherein, the formula for adjusting the output voltage dynamic compensation amount based on the real-time voltage deviation is as follows:

[0086] ΔU=K p ·e(t)+K i ·∫e(t)dt

[0087] Where e(t) is the real-time voltage deviation, K p and K i These are the proportional and integral coefficients. The PI controller dynamically adjusts the voltage by outputting a voltage compensation amount based on real-time voltage deviation, ensuring the grid voltage remains within the set range.

[0088] In this embodiment of the invention, the electrical parameters include one or more of voltage, current, power, and power factor.

[0089] In this embodiment of the invention, the load forecasting module further combines the Prophet model to assist load forecasting, automatically identifies seasonal fluctuation patterns, and outputs a load forecast sequence for a future preset time period.

[0090] In this embodiment of the invention, the lower real-time adjustment layer receives real-time data from the power distribution network and calculates the real-time voltage deviation and reactive power compensation error through a PI controller. When the real-time voltage deviation or reactive power compensation error exceeds its respective threshold (e.g., voltage deviation controlled within ±0.02pu; reactive power compensation error ≤4%), adjustment is triggered, and equipment adjustment commands are generated.

[0091] In this embodiment of the invention, when the lower real-time adjustment layer generates equipment adjustment commands, it prioritizes generating converter adjustment commands; if the voltage deviation is greater than the threshold and continues for more than a preset time, it generates an emergency tap changer adjustment command, or if the reactive power compensation error is greater than its threshold, it generates a capacitor bank replenishment command.

[0092] Please see Figure 4 As shown, when applying the technology of this invention for adjustment, after the start of operation, the voltage of each node is monitored at any time. If the voltage exceeds the limit, the coordinated adjustment mechanism is immediately activated. The transformer responds to the adjustment command at the minute level. After the voltage is detected to recover, the current state of the transformer is maintained. Otherwise, the OLTC tap changer adjustment (hourly level) is activated. If the tap changer action conditions are met, the transformer adjustment is returned. If the tap changer action conditions are not met, the tap changer is switched and the transformer adjustment reference is updated. After the coordinated adjustment is completed, the voltage of the node is monitored again to see if it exceeds the limit.

[0093] Another aspect of the present invention provides a control method for a multi-objective cooperative control system of a flexible on-load tap-changing transformer based on hourly load forecasting, comprising the following steps, see below. Figure 5 As shown:

[0094] Data acquisition: Real-time acquisition of electrical parameters at each node of the power distribution network;

[0095] Model prediction: Based on the electrical parameters of each node in the distribution network, the VMD-BES-LSSVM hybrid algorithm is used to perform variational mode decomposition and optimization prediction on the power characteristic curve of the power grid, and output the load prediction sequence for the future time period.

[0096] Multi-objective optimization: Based on the constructed collaborative optimization model that includes voltage deviation, reactive power compensation and equipment loss, the optimal control optimization strategy for generating tap positions and capacitor bank switching combinations is achieved through dynamic weight adjustment, with the goals of minimizing voltage deviation, optimizing reactive power compensation and minimizing equipment loss.

[0097] MILP optimization: The upper predictive optimization layer adopts the mixed integer linear programming (MILP) algorithm. Based on the optimal control optimization strategy, it makes hourly optimization decisions and adjustment plans for tap position and capacitor bank switching according to the load prediction sequence and the current status of the equipment. When the lower real-time adjustment layer of the hierarchical control module is triggered, it generates minute-level equipment adjustment commands through the voltage dynamic compensation amount output by the proportional-integral (PI) controller.

[0098] Layered execution: Responding to equipment adjustment commands, the FOLTC (Flexible On-Load Tap Changer) switches taps, capacitor banks are switched on or off, or the converter operates, enabling reactive power output from the converter and achieving multi-objective optimization of voltage deviation, reactive power compensation, and equipment losses.

[0099] The present invention will now be described using a 17-slot flexible on-load tap changer as an example.

[0100] Specifically, the system comprises a data acquisition layer, edge computing nodes, and execution devices connected in sequence. The data acquisition layer, through the deployment of a PMU (Phasor Measurement Unit) and a SCADA system, collects electrical parameters such as voltage, current, and power factor of each node in the distribution network in real time, with a sampling period of 1 minute, serving as the node parameter acquisition module in this embodiment. The edge computing nodes employ industrial-grade control computers (equipped with an Intel Xeon quad-core processor and 32GB of memory) to deploy the load forecasting module, multi-objective optimization module, and hierarchical control module of this embodiment. The execution devices, serving as the execution modules in this embodiment, include a flexible on-load tap changer (FOLTC) (with an IGBT converter, a capacity of 1MVar, a 17-position vacuum on-load tap changer, and an adjustment range of ±10%) and intelligent capacitor banks (modular design, each with a capacity of 50kVar, supporting remote switching control). The communication network adopts IEC 61850 standard fiber optic Ethernet to ensure that the transmission delay of control commands is less than 50ms. This hardware architecture design guarantees the efficient operation and real-time response capability of the power system.

[0101] Based on the aforementioned hardware architecture of sequentially communicating data acquisition layer, edge computing nodes, and execution devices, when performing multi-objective collaborative control processing of flexible on-load tap-changing transformers, the data acquisition layer first collects target data. Then, further data preprocessing is performed. This involves inputting historical data from the power system, including electrical parameters such as voltage, current, power, and power factor of each node in the distribution network. Data cleaning is then performed, such as using the 3σ criterion to remove outliers, ensuring data quality. For missing data, time series interpolation methods are used to fill in the gaps, ensuring data continuity and accuracy. For the feature engineering part, multiple features are constructed to improve the model's predictive ability, including time-series features (such as 24-hour moving averages) and periodic features (fundamental frequency information extracted through Fourier transform).

[0102] After the collected data is preprocessed, it enters the load forecasting stage. Variational mode decomposition (VMD) is performed through the VMD unit to decompose the input signal x(t) into several mode components xk(t) in order to extract different frequency components of the signal and decompose the power characteristic curve of the power grid into multiple IMF components.

[0103] Parameter settings: number of modes K = 5, penalty factor α = 2000, and an adaptive stopping criterion (intermodal correlation coefficient < 0.05) is used to construct a variational problem for the input power curve x(t).

[0104]

[0105] st∑_k u_k=x(t)

[0106] The IMF components u_1 to u_5 are obtained by solving the problem using the Alternating Direction Multiplier Method (ADMM), and are then used for...

[0107] For training the BES-LSSVM model, the first step is to define the parameter optimization space, where the penalty parameter C has a value range of [0.1, 1000] and the RBF kernel parameter σ has a value range of [0.01, 10].

[0108] To optimize the penalty parameter C and the RBF kernel parameter σ (i.e., the kernel function), the Bald Eagle Search (BES) algorithm is used. The BES process begins by initializing the population, which contains 20 individuals, each representing a (C,σ) pair. During the selection phase, the top 30% of individuals with the best fitness are retained; these are calculated using a fitness function with the root mean square error (RMSE) of 5-fold cross-validation as the evaluation criterion. In the search phase, a Levy flight update strategy is used to adjust the position of each individual. Levy flight has strong global search capabilities, which helps improve the parameter optimization effect. Finally, iteration stops when convergence conditions are met, including reaching 50 iterations or a change in RMSE less than 1 × 10⁻⁴, at which point parameter optimization is considered complete. Through these steps, the BES-LSSVM model can effectively optimize parameters and improve the model's prediction accuracy. The optimized model is then used to predict future loads, forming a load prediction sequence.

[0109] The above describes the training process for the prediction model of the load prediction module. After training, the model is used for prediction and evaluation. Once the evaluation meets the preset requirements, the model training is complete. The pre-trained model can then be used in the multi-objective cooperative control system of this embodiment for multi-objective cooperative control. The prediction model training process is described in [link to documentation]. Figure 6 As shown.

[0110] In the multi-objective optimization module, a mixed-integer linear programming (MILP) model is established. Decision variables include integer variables (such as tap position T(t)∈[1,17] and capacitor bank switching number H(t)∈[0,20]) and continuous variables (such as converter reactive power output Q). OLTC (t)∈[-1,1]MVar).

[0111] The constraints include a voltage range limit: 0.95 ≤ Ui(t) ≤ 1.05 pu, and a power limit: Q min ≤

[0112] Q OLTC (t)+Q CB (t)≤Q maxAnd the gear change limit: |T(t)-T(t-1)|≤2.

[0113] The mixed-integer linear programming (MILP) model is solved using the branch and bound method, with a gap of 0.1% and a computation time limit of 5 minutes per hour. To handle different operating conditions, a weight adjustment strategy is employed.

[0114] Under normal operating conditions, the weights are set to ω1 = 0.6, ω2 = 0.3, and ω3 = 0.1.

[0115] Under the emergency voltage condition (U<0.93pu), the weights are adjusted to ω1=0.9, v2=0.1, ω3=0;

[0116] Under the economic model, the weights are ω1 = 0.3, ω2 = 0.2, and ω3 = 0.5.

[0117] Through the above steps, the relevant parameters for multi-objective optimization were set.

[0118] In hierarchical control, the upper-level predictive optimization layer is set to start every hour on the hour. Its input data includes the latest 24-hour load forecast and the current equipment status. Using this input data, the upper-level predictive optimization layer calculates the tap changer operation plan and capacitor bank switching plan for the next 24 hours to facilitate subsequent control operations. The lower-level real-time adjustment layer operates on a 1-minute control cycle, making adjustments in real time. The lower-level real-time adjustment layer employs a PI control algorithm with a proportional coefficient of Kp = 0.8 and an integral coefficient of Ki = 0.05. To avoid control system saturation, an integral separation strategy is adopted: when the error value |e(t)| is greater than 0.03pu, integration stops to ensure stable system operation.

[0119] In the implementation of hierarchical control, the coordinated control of tap changers and converters is first carried out at the hourly level. For example, one hour before the load increases, the tap position is adjusted in advance based on load forecast data (e.g., if the predicted load increase is 5%, the tap position is increased by one level in advance). At the same time, the decision to switch capacitor banks is made based on the predicted value of cosφ to maintain the power factor between 0.95 and 1.0, ensuring the stability of system operation.

[0120] In the implementation of hierarchical control, in addition to hourly adjustments for the coordinated control of tap changers and converters, a minute-level compensation strategy is also adopted. When the voltage deviation is greater than 0.03 pu, the converter is adjusted first to compensate; if the voltage deviation is greater than 0.03 pu for more than 5 minutes, the emergency adjustment of the tap changer is triggered.

[0121] To ensure system safety, several protection mechanisms are in place: such as mandatory locking of tap changer operation intervals for ≥2 minutes, triggering an alarm if the number of tap changer operations exceeds 15 times per day, and requiring capacitor bank switching intervals to be ≥30 seconds to avoid instability caused by frequent adjustments.

[0122] This invention, through a collaborative hardware and software design, successfully achieves closed-loop control of prediction, optimization, and execution, making it particularly suitable for distribution network scenarios with load fluctuation rates greater than 15%. The system can also be expanded to include coordinated control functions for distributed power sources such as photovoltaics and energy storage, further enhancing the intelligence and adaptability of the distribution network.

[0123] The above test examples demonstrate that the system and method of this application can reduce line losses by 8.5% and reduce the number of capacitor bank operations by 30%.

[0124] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0125] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the claims be included within the invention.

[0126] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting, characterized in that, include: The node parameter acquisition module is used to collect the electrical parameters of each node in the distribution network in real time. The load forecasting module is used to perform variational mode decomposition and optimization forecasting on the power characteristic curve of the power grid based on the electrical parameters of each node in the distribution network, and output the load forecast sequence for the future time. The multi-objective optimization module is used to minimize voltage deviation, optimize reactive power compensation, and minimize equipment loss based on the constructed collaborative optimization model that includes voltage deviation, reactive power compensation, and equipment loss. Each objective is assigned an adjustable weight, and the optimal control optimization strategy for generating tap position and capacitor bank switching combination is realized through dynamic weight adjustment. The hierarchical control module includes an upper predictive optimization layer and a lower real-time adjustment layer. The upper predictive optimization layer uses a mixed-integer linear programming (MILP) algorithm, based on the optimal control optimization strategy, to make hourly optimization decisions and adjustment plans for tap positions and capacitor bank switching according to the load forecast sequence and the current state of the equipment. The lower real-time adjustment layer generates minute-level equipment adjustment commands by using the voltage dynamic compensation output by the proportional-integral (PI) controller when adjustment is triggered. The execution module is used to coordinate and control the tap switching of the flexible on-load tap changer (FOLTC), the switching of capacitor banks, and the operation of the converter according to the equipment adjustment instructions output by the lower real-time adjustment layer, so as to realize the reactive power output of the converter and achieve multi-objective optimization of voltage deviation, reactive power compensation, and equipment loss. The upper-level prediction and optimization layer uses the MILP algorithm to perform hourly-level optimization decisions and adjustment planning for tap positions and capacitor bank switching, including: Pre-adjust tap positions one hour before peak load; prioritize switching capacitor banks; generate tap and capacitor bank operation sequences for the next 24 hours.

2. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting according to claim 1, characterized in that, The load forecasting module specifically includes: VMD unit is used to decompose the power characteristic curve of the grid into multiple IMF components; The BES optimization unit is used to optimize the penalty parameters and kernel function parameters of the LSSVM prediction unit using the BES optimization algorithm. The LSSVM forecasting unit is used to independently forecast and reconstruct each IMF component, outputting a 24-hour load forecast sequence that includes forecast results of different types of load capacity and power factor.

3. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting as described in claim 1, characterized in that, The objective function of the multi-objective optimization module is: ； The objective is to minimize voltage deviation; The goal is to optimize reactive power compensation. The goal is to minimize equipment wear and tear; , , These are adjustable weighting coefficients. This represents the voltage at node i. Indicates the reference voltage. and These are the maximum and minimum allowable values ​​for voltage, respectively. It is the reactive power compensation of the flexible OLTC converter. It is the reactive power compensation of the capacitor bank. It represents the reactive power demand of the load, indicating the optimal reactive power compensation capacity. It is the change in the number of tap changes. It is the change in the number of times the capacitor bank operates. and These are the costs of each tap changer and capacitor bank operation, respectively.

4. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting according to claim 1, characterized in that, The lower real-time adjustment layer outputs minute-level dynamic voltage compensation through a proportional-integral (PI) controller. The PI controller outputs the dynamic voltage compensation according to the following formula: ； in, This is the voltage dynamic compensation amount. For real-time voltage deviation, and These are the proportional coefficient and the integral coefficient.

5. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting according to claim 1, characterized in that, The electrical parameters include one or more of voltage, current, power, and power factor.

6. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting according to claim 1, characterized in that, The load forecasting module, combined with the Prophet model to assist in load forecasting, automatically identifies seasonal fluctuation patterns and outputs a load forecast sequence for the next 24 hours.

7. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting according to claim 1, characterized in that, The lower-level real-time adjustment layer receives real-time data from the power distribution network and calculates real-time voltage deviation and reactive power compensation error through a PI controller. When the real-time voltage deviation or reactive power compensation error exceeds its respective threshold, adjustment is triggered, and equipment adjustment commands are generated.

8. The multi-objective collaborative control system for flexible on-load tap-changing transformers based on hourly load forecasting according to claim 1, characterized in that, When the lower-level real-time adjustment layer generates equipment adjustment commands, it prioritizes generating converter adjustment commands; if the voltage deviation is greater than the threshold and continues for more than a preset time, it generates an emergency tap adjustment command, or if the reactive power compensation error is greater than its threshold, it generates a capacitor bank replenishment command.

9. The control method for a multi-objective cooperative control system of a flexible on-load tap-changing transformer based on hourly load forecasting as described in any one of claims 1-8, characterized in that, Includes the following steps: Real-time acquisition of electrical parameters at each node of the power distribution network; Based on the electrical parameters of each node in the distribution network, the VMD-BES-LSSVM hybrid algorithm is used to perform variational mode decomposition and optimization prediction on the power characteristic curve of the power grid, and output the load prediction sequence for the future time period. Based on the constructed collaborative optimization model that includes voltage deviation, reactive power compensation and equipment loss, the goal is to minimize voltage deviation, optimize reactive power compensation and minimize equipment loss. Each goal is assigned an adjustable weight, and the optimal control optimization strategy for generating tap positions and capacitor bank switching combinations is realized through dynamic weight adjustment. The upper-level prediction and optimization layer adopts the mixed integer linear programming (MILP) algorithm. Based on the optimal control optimization strategy, it makes hourly optimization decisions and adjustment plans for tap positions and capacitor bank switching according to the load prediction sequence and the current status of the equipment. When the lower-level real-time adjustment layer triggers adjustment, it generates minute-level equipment adjustment commands through the voltage dynamic compensation amount output by the proportional-integral (PI) controller. In response to equipment adjustment commands, the flexible on-load tap changer (FOLTC) switches taps, switches capacitor banks, or operates the converter to achieve reactive power output from the converter, thus realizing multi-objective optimization of voltage deviation, reactive power compensation, and equipment losses.

Images