Automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization

Abstract

This invention discloses an automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization, comprising the following steps: collecting voltage, current, load, and environmental parameters to form a monitoring dataset and normalizing it to generate optimized input; constructing a power grid topology weighting matrix, setting a constraint set, and establishing a control variable structure and switching cost model; performing multi-scale decomposition on the optimized input and combining it with load forecast results to form residual input; inputting the residual input and related models into an improved Kaczmarz iterative framework, setting an initial control vector and convergence threshold, and initiating iteration; iteratively generating candidate control vectors and performing feasibility checks; projecting the detected candidate control vectors onto the constraint set to form implementable control vectors; and outputting the final control vector when the convergence threshold is met, generating compensation control commands. This invention improves the real-time performance and stability of transformer power compensation.

Description

Technical Field

[0001] This invention relates to the field of power system automation control technology, and in particular to an automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization. Background Technology

[0002] In the operation of power systems, transformers play a vital role in the transmission and distribution of electrical energy. The level of reactive power compensation directly affects voltage stability, power loss, and safe operation of equipment. Currently, most common compensation control methods rely on fixed parameter settings or switching strategies based on a single monitored quantity. For example, they may trigger the switching of capacitor banks and reactors by monitoring voltage deviations, or perform setpoint compensation based on reactive power thresholds. These methods are simple in structure and easy to implement, but they are difficult to guarantee compensation accuracy in complex power grid environments.

[0003] Some existing technologies attempt to improve compensation control by using optimization and prediction methods. For example, linear programming, genetic algorithms or heuristic search methods are used to solve compensation schemes and to predict future operating conditions based on load forecasts. These methods improve the rationality of compensation decisions to some extent, but their modeling objects are relatively simple. They often only consider load changes and ignore multi-source influencing factors such as voltage, current and environmental conditions, resulting in insufficient prediction accuracy. Moreover, the optimization results are difficult to keep stable in dynamic power grids.

[0004] Traditional Kaczmarz iterative methods and commonly used gradient descent algorithms are also used to solve residual minimization problems to obtain control vectors. However, in power grid operation, residual distribution has nonlinear and multi-scale characteristics. Single iterative methods are prone to problems such as slow convergence speed, sensitivity to initial values, and insufficient stability under constraints, thus affecting the real-time performance and reliability of compensation control.

[0005] In terms of candidate control vector generation and detection, existing technologies typically perform constraint verification after iteration, i.e., first solve the control result and then determine whether it meets the voltage range, harmonic limits and switching constraints. This approach results in a large number of intermediate solutions that do not meet the conditions during the calculation process, which increases the amount of computation and delays the output of control commands. Furthermore, it fails to dynamically filter out unqualified solutions during the iteration stage, thus reducing control efficiency.

[0006] Existing methods often employ separate modeling and independent optimization for handling continuous set variables and discrete switching variables. Continuous set variables are typically used to describe the set values ​​of reactive power compensation devices, while discrete switching variables are used to represent the switching states of capacitor banks and reactors. The independent handling of these two types of variables makes it difficult to ensure that the compensation strategy remains coordinated between voltage stability and equipment action constraints. This is especially true in scenarios with frequent switching, which can easily lead to switching action conflicts, shorten device lifespan, and reduce control stability.

[0007] Therefore, how to provide an automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] One objective of this invention is to propose an automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization. This invention fully utilizes the improved Kaczmarz iterative algorithm, power grid topology weighted modeling, load prediction modeling, and constraint projection mechanism. It details the process of generating compensation control commands through multi-source monitoring data acquisition, multi-scale residual decomposition, candidate control vector generation and detection, constraint feasibility processing, and convergence judgment. It has the advantages of strong parameter coordination capability, high convergence stability, complete constraint satisfaction, and real-time control commands.

[0009] The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization according to an embodiment of the present invention includes the following steps:

[0010] Collect transformer operating parameters, including voltage, current, active load, reactive load and environmental parameters, form a monitoring dataset, and complete time synchronization and normalization processing to generate optimized input;

[0011] Based on the optimized input, a power grid topology weighting matrix is ​​constructed, and a set of constraints including voltage range, harmonic limit, minimum switching dwell time and switching dead zone is defined. A control variable structure and switching cost model including continuous set variables and discrete switching variables are established.

[0012] The optimized input is decomposed into short-time and long-time residuals using a multi-scale method, and then combined with the load forecasting results to form the residual input.

[0013] Input the residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model into the improved Kaczmarz iterative framework, set the initial control vector and the convergence threshold, and start the iterative process;

[0014] During the iteration process, the improved Kaczmarz is used to generate candidate control vectors, and the feasibility of the candidate control vectors is checked.

[0015] The candidate control vectors that pass the feasibility test are projected onto the constraint set to obtain implementable control vectors that satisfy the voltage range, harmonic limit, minimum switching dwell time and switching dead zone.

[0016] When the control vector meets the convergence threshold, the final control vector is output, generating the automatic switching control command for transformer power compensation.

[0017] Optionally, the improved Kaczmarz includes:

[0018] During the iteration process, adaptive row selection is performed based on the ratio of the magnitude of the residual input to the switching cost model;

[0019] The topological residuals are generated by spreading the residuals of the selected rows using the power grid topology weighting matrix.

[0020] Nonlinear residuals are obtained by performing nonlinear mapping on topological residuals;

[0021] The control variables are updated and candidate control vectors are generated based on nonlinear residuals. The continuous set variables are updated through continuous projection operators, and the discrete cut variables are updated through level alignment and minimum dwell time correction.

[0022] During the iteration process, the residual weights are dynamically adjusted according to different stages to form a convergence path.

[0023] Optionally, the construction of the constraint set and control variable structure specifically includes:

[0024] A power grid topology weighting matrix is ​​established based on the optimization input. The power grid topology weighting matrix is ​​used to represent the connection relationship between nodes and branches in the power grid. Each matrix element represents the branch weight between the corresponding nodes.

[0025] Define a set of constraints, which includes voltage operating range constraints, harmonic limit constraints, minimum switch dwell time constraints, and switch dead zone constraints. Each constraint limits the range of values ​​for the control vector under the corresponding conditions.

[0026] A control variable structure is established, which consists of continuous set variables and discrete switching variables. The continuous set variables are used to represent the set values ​​of the reactive power compensation device, and the discrete switching variables are used to represent the switching stages of the capacitor bank and the reactor.

[0027] A switching cost model is constructed, which consists of a set of cost values. Each cost value corresponds to a switching action. The cost value is used to quantify the switching consumption and lifetime impact generated by the switching action.

[0028] Optionally, the generation of the residual input specifically includes:

[0029] A load prediction model is constructed using historical active load, reactive load, voltage, current and environmental parameters from the monitoring dataset. The input of the load prediction model is historical operating data, and the output is the predicted load sequence for future time. The prediction model is run on the sampling interval and time index consistent with the optimization input to obtain the load prediction results.

[0030] The optimized input is decomposed into short-time and long-time components by multi-scale decomposition, and the load forecasting results are decomposed into short-time and long-time forecasting components by multi-scale decomposition in the same way. The two types of components correspond point by point on a unified time index.

[0031] Calculate the short-time residual and the long-time residual. The short-time residual is the short-time component minus the short-time prediction component and recorded according to the time index. The long-time residual is the long-time component minus the long-time prediction component and recorded according to the time index.

[0032] The short-time residuals and long-time residuals are weighted and fused on a unified time axis to form the residual input. The weighting coefficients take values ​​in the range of zero to one and remain unchanged within one iteration cycle.

[0033] Optionally, the execution of the iterative framework specifically includes:

[0034] Set an initial control vector, which consists of continuous set variables and discrete switching variables. The initial values ​​of the continuous set variables are derived from the operating parameters of the reactive power compensation device, and the initial values ​​of the discrete switching variables are derived from the switching states of the capacitor bank and the reactor.

[0035] Set a convergence threshold, which limits the amplitude range of the residual input. The termination condition is met when the amplitude of the residual input does not exceed the convergence threshold.

[0036] The residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model are input into the improved Kaczmarz iterative framework. In each iteration, the row index is determined according to the adaptive row selection strategy, and the topology-weighted row vector and nonlinear residual are calculated. The intermediate control vector is calculated according to the following formula:

[0037] ;

[0038] in, For the first The control vector for the next iteration For the first The intermediate control vector updated in the next iteration. For the first The step size of the next iteration. For the first The next iteration is in the row index. The residual input at the location, For residual input at row index The corresponding row vector at that position, This is a linear operator representation of the power grid topology weighting matrix. This is the result of topologically weighting the row vectors. The dot product of the topologically weighted row vector and itself. This is a function that performs a nonlinear mapping on the residual input;

[0039] The intermediate control vector is constrained and made feasible. The continuous set variables are continuously projected according to the voltage operating range and harmonic limits. The discrete switching variables are aligned according to the switching level difference and timed according to the minimum dwell time and dead zone of the switch. After processing, the control vector for entering the next iteration is obtained.

[0040] During the iteration process, a convergence judgment is performed. When the magnitude of the residual input does not exceed the convergence threshold, the iteration stops and the control vector is output. When the magnitude of the residual input exceeds the convergence threshold, the iteration continues.

[0041] Optionally, the generation and detection of the candidate control vector specifically includes:

[0042] During the iteration process, the improved Kaczmarz update control vector is used to generate candidate control vectors, which consist of continuously set variables and discrete switching variables.

[0043] Feasibility testing is performed on candidate control vectors. During the testing process, the voltage operating range constraint and harmonic limit constraint in the constraint set are used to determine whether the continuously set variables are within the allowable range. If the continuously set variables exceed the allowable range, the constraint conditions are not met.

[0044] During the detection process, the switching sequence of discrete switching variables is determined based on the minimum dwell time constraint and the dead zone constraint in the constraint set. When the switching interval is less than the minimum dwell time or the dead zone is triggered, it is determined that the constraint conditions are not met.

[0045] During the detection process, the judgment results of continuous set variables and discrete cut variables are combined. When all constraints are satisfied, the candidate control vector is retained. When there are constraints that are not satisfied, the candidate control vector is discarded and the iterative calculation is restarted.

[0046] Optionally, the generation of the implementable control vector specifically includes:

[0047] The candidate control vectors obtained through feasibility testing are input into the constraint set, and projection processing is performed on the continuously set variables. When the continuously set variables exceed the voltage operating range, they are truncated according to the upper and lower limit boundaries. When the continuously set variables exceed the harmonic limit, they are corrected according to the limit value corresponding to the harmonic constraint.

[0048] Projection processing is performed on discrete switching variables. When the switching interval of the discrete switching variable is less than the minimum dwell time of the switch, the holding time is extended to the minimum dwell time. When the discrete switching variable triggers the switch dead zone, the current switching state is maintained until the switch dead zone ends.

[0049] After completing the continuous setting variable projection processing and the discrete switching variable projection processing, an implementable control vector that satisfies the voltage operating range constraint, harmonic limit constraint, minimum switching dwell time constraint and switching dead zone constraint is generated and recorded under a unified time index.

[0050] Optionally, the output of the final control vector specifically includes:

[0051] During the iteration process, the amplitude of the residual input is calculated and compared with the convergence threshold. When the amplitude of the residual input is less than the convergence threshold, the convergence condition is considered to be met; when the amplitude of the residual input is greater than the convergence threshold, the convergence condition is considered not met.

[0052] When the convergence condition is met, the control vector obtained in the current iteration is determined as the final control vector. The final control vector consists of continuous set variables and discrete switching variables. The continuous set variables represent the set values ​​of the reactive power compensation device, and the discrete switching variables represent the switching states of the capacitor bank and the reactor.

[0053] The final control vector is converted into automatic switching control instructions for transformer power compensation. The control instructions include output instructions for continuously set quantities and execution instructions for discrete switching actions, and are sent to the control module to perform power compensation operations under a unified time index.

[0054] The beneficial effects of this invention are:

[0055] This invention introduces a multi-parameter collaborative optimization framework into the compensation control process, unifying the modeling of operating parameters such as voltage, current, active load, reactive load, and environmental quantities, and constructing residual inputs by combining load forecasting results. Compared with traditional compensation methods that rely on a single variable, this invention can more comprehensively reflect the power grid operating status, thereby ensuring the accuracy and stability of the compensation strategy.

[0056] An improved Kaczmarz algorithm is adopted in the iterative solution stage. The control vector is dynamically updated through adaptive row selection, topology weighted diffusion, nonlinear residual mapping and constraint projection processing. It converges quickly under the constraints of voltage operating range, harmonic limit, minimum switching dwell time and switching dead zone, effectively solving the problems of slow convergence speed, sensitivity to initial value and insufficient stability of existing methods.

[0057] This invention integrates continuously set variables and discrete switching variables into an iterative framework for collaborative optimization through a candidate control vector generation and feasibility detection mechanism. In the feasibility detection stage, solutions that do not meet the constraints are promptly eliminated. Then, combined with constraint projection, an implementable control vector is formed, and finally, an automatic conversion control command is output, thus achieving the real-time performance and reliability of compensation control. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 The flowchart shows the automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization proposed in this invention.

[0060] Figure 2 This is a schematic diagram of the power grid topology modeling and constraint construction for the automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization proposed in this invention.

[0061] Figure 3 This is a schematic diagram of the improved Kaczmarz iterative calculation of the automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization proposed in this invention. Detailed Implementation

[0062] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0063] refer to Figure 1-3 The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization includes the following steps:

[0064] Collect transformer operating parameters, including voltage, current, active load, reactive load and environmental parameters, form a monitoring dataset, and complete time synchronization and normalization processing to generate optimized input;

[0065] Based on the optimized input, a power grid topology weighting matrix is ​​constructed, and a set of constraints including voltage range, harmonic limit, minimum switching dwell time and switching dead zone is defined. A control variable structure and switching cost model including continuous set variables and discrete switching variables are established.

[0066] The optimized input is decomposed into short-time and long-time residuals using a multi-scale method, and then combined with the load forecasting results to form the residual input.

[0067] Input the residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model into the improved Kaczmarz iterative framework, set the initial control vector and the convergence threshold, and start the iterative process;

[0068] During the iteration process, the improved Kaczmarz is used to generate candidate control vectors, and the feasibility of the candidate control vectors is checked.

[0069] The candidate control vectors that pass the feasibility test are projected onto the constraint set to obtain implementable control vectors that satisfy the voltage range, harmonic limit, minimum switching dwell time and switching dead zone.

[0070] When the control vector meets the convergence threshold, the final control vector is output, generating the automatic switching control command for transformer power compensation.

[0071] This invention collects operating parameters such as voltage, current, active load, reactive load, and environmental parameters, and performs time synchronization and normalization processing. It converts data from different sources into inputs that can be directly used in optimization calculations. In the monitoring and compensation control of power systems, it avoids the problems of inconsistent data dimensions and timestamp misalignment, ensuring that the input basis for the subsequent optimization process is accurate and reliable. This enables the optimization model to take into account the influence of multiple operating state variables at the same time, thereby maintaining a stable compensation strategy generation capability under different load fluctuations and environmental conditions.

[0072] In this embodiment, the improved Kaczmarz includes:

[0073] During the iteration process, adaptive row selection is performed based on the ratio of the magnitude of the residual input to the switching cost model;

[0074] The topological residuals are generated by spreading the residuals of the selected rows using the power grid topology weighting matrix.

[0075] Nonlinear residuals are obtained by performing nonlinear mapping on topological residuals;

[0076] The control variables are updated and candidate control vectors are generated based on nonlinear residuals. The continuous set variables are updated through continuous projection operators, and the discrete cut variables are updated through level alignment and minimum dwell time correction.

[0077] During the iteration process, the residual weights are dynamically adjusted according to different stages to form a convergence path.

[0078] This invention introduces the connection relationship between nodes and branches by constructing a power grid topology weighted matrix, and on this basis defines a set of constraints including voltage range, harmonic limits, minimum switching dwell time and switching dead zone. At the same time, the set value of the reactive power compensation device is used as a continuous set variable, and the switching stages of capacitor banks and reactors are used as discrete switching variables. Combined with the switching cost model, the action cost is quantified, realizing the unified modeling of multi-dimensional constraints and control variables. This makes the compensation optimization not only consider the operation effect, but also the equipment action cost.

[0079] In this embodiment, the construction of the constraint set and control variable structure specifically includes:

[0080] A power grid topology weighting matrix is ​​established based on the optimization input. The power grid topology weighting matrix is ​​used to represent the connection relationship between nodes and branches in the power grid. Each matrix element represents the branch weight between the corresponding nodes.

[0081] Define a set of constraints, which includes voltage operating range constraints, harmonic limit constraints, minimum switch dwell time constraints, and switch dead zone constraints. Each constraint limits the range of values ​​for the control vector under the corresponding conditions.

[0082] A control variable structure is established, which consists of continuous set variables and discrete switching variables. The continuous set variables are used to represent the set values ​​of the reactive power compensation device, and the discrete switching variables are used to represent the switching stages of the capacitor bank and the reactor.

[0083] A switching cost model is constructed, which consists of a set of cost values. Each cost value corresponds to a switching action. The cost value is used to quantify the switching consumption and lifetime impact generated by the switching action.

[0084] This invention obtains short-term and long-term components by multi-scale decomposition of the optimization input, and generates a prediction sequence by combining it with a load forecasting model built based on historical operating data. Short-term and long-term residuals are formed under a unified time index, and then further weighted and fused to form the residual input. This processing method enables the optimization process to capture both short-term disturbances and long-term trends, ensuring that the compensation control can respond quickly to sudden fluctuations while taking into account the overall operating trend, thereby avoiding the imbalance problem caused by relying on only a single scale data in traditional methods.

[0085] In this embodiment, the generation of the residual input specifically includes:

[0086] A load prediction model is constructed using historical active load, reactive load, voltage, current and environmental parameters from the monitoring dataset. The input of the load prediction model is historical operating data, and the output is the predicted load sequence for future time. The prediction model is run on the sampling interval and time index consistent with the optimization input to obtain the load prediction results.

[0087] The optimized input is decomposed into short-time and long-time components by multi-scale decomposition, and the load forecasting results are decomposed into short-time and long-time forecasting components by multi-scale decomposition in the same way. The two types of components correspond point by point on a unified time index.

[0088] Calculate the short-time residual and the long-time residual. The short-time residual is the short-time component minus the short-time prediction component and recorded according to the time index. The long-time residual is the long-time component minus the long-time prediction component and recorded according to the time index.

[0089] The short-time residuals and long-time residuals are weighted and fused on a unified time axis to form the residual input. The weighting coefficients take values ​​in the range of zero to one and remain unchanged within one iteration cycle.

[0090] This invention introduces residual input, power grid topology weighting matrix, constraint set, control variable structure, and switching cost model into an improved Kaczmarz iterative framework. It performs iterative calculations under the condition of setting an initial control vector and a convergence threshold. Each iteration combines adaptive row selection and nonlinear residual mapping to update the intermediate control vector and performs feasibility processing under constraints, thereby achieving dynamic convergence of the control vector under complex constraints and improving the speed and stability of optimization solutions.

[0091] In this embodiment, the execution of the iterative framework specifically includes:

[0092] Set an initial control vector, which consists of continuous set variables and discrete switching variables. The initial values ​​of the continuous set variables are derived from the operating parameters of the reactive power compensation device, and the initial values ​​of the discrete switching variables are derived from the switching states of the capacitor bank and the reactor.

[0093] Set a convergence threshold, which limits the amplitude range of the residual input. The termination condition is met when the amplitude of the residual input does not exceed the convergence threshold.

[0094] The residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model are input into the improved Kaczmarz iterative framework. In each iteration, the row index is determined according to the adaptive row selection strategy, and the topology-weighted row vector and nonlinear residual are calculated. The intermediate control vector is calculated according to the following formula:

[0095] ;

[0096] in, For the first The control vector for the next iteration For the first The intermediate control vector updated in the next iteration. For the first The step size of the next iteration. For the first The next iteration is in the row index. The residual input at the location, For residual input at row index The corresponding row vector at that position, This is a linear operator representation of the power grid topology weighting matrix. This is the result of topologically weighting the row vectors. The dot product of the topologically weighted row vector and itself. This is a function that performs a nonlinear mapping on the residual input;

[0097] The intermediate control vector is constrained and made feasible. The continuous set variables are continuously projected according to the voltage operating range and harmonic limits. The discrete switching variables are aligned according to the switching level difference and timed according to the minimum dwell time and dead zone of the switch. After processing, the control vector for entering the next iteration is obtained.

[0098] During the iteration process, a convergence judgment is performed. When the magnitude of the residual input does not exceed the convergence threshold, the iteration stops and the control vector is output. When the magnitude of the residual input exceeds the convergence threshold, the iteration continues.

[0099] This invention generates candidate control vectors and performs feasibility checks during the iteration process. Voltage range constraints and harmonic limit constraints are used to determine continuously set variables, while minimum switch dwell time constraints and switch dead zone constraints are used to determine discrete switching variables. Candidate solutions are directly discarded when the conditions are not met. This detection mechanism ensures that infeasible solutions will not enter the next step of calculation, avoiding invalid iterations and delays. At the same time, it ensures that the generated candidate control vectors are always within the executable range, improving computational efficiency and result reliability.

[0100] In this embodiment, the generation and detection of the candidate control vector specifically includes:

[0101] During the iteration process, the improved Kaczmarz update control vector is used to generate candidate control vectors, which consist of continuously set variables and discrete switching variables.

[0102] Feasibility testing is performed on candidate control vectors. During the testing process, the voltage operating range constraint and harmonic limit constraint in the constraint set are used to determine whether the continuously set variables are within the allowable range. If the continuously set variables exceed the allowable range, the constraint conditions are not met.

[0103] During the detection process, the switching sequence of discrete switching variables is determined based on the minimum dwell time constraint and the dead zone constraint in the constraint set. When the switching interval is less than the minimum dwell time or the dead zone is triggered, it is determined that the constraint conditions are not met.

[0104] During the detection process, the judgment results of continuous set variables and discrete cut variables are combined. When all constraints are satisfied, the candidate control vector is retained. When there are constraints that are not satisfied, the candidate control vector is discarded and the iterative calculation is restarted.

[0105] This invention generates an implementable control vector that satisfies all constraints by projecting the detected candidate control vector onto a constraint set, performing projection processing of voltage operating range and harmonic limits on the continuously set variable part, and performing alignment of switching levels and correction of minimum dwell time and dead zone on the discrete switching variable part, and recording it under a unified time index. This method ensures that the final control vector can be directly applied to the physical device, avoiding the delay and inconsistency problems caused by post-processing correction.

[0106] In this embodiment, the generation of the implementable control vector specifically includes:

[0107] The candidate control vectors obtained through feasibility testing are input into the constraint set, and projection processing is performed on the continuously set variables. When the continuously set variables exceed the voltage operating range, they are truncated according to the upper and lower limit boundaries. When the continuously set variables exceed the harmonic limit, they are corrected according to the limit value corresponding to the harmonic constraint.

[0108] Projection processing is performed on discrete switching variables. When the switching interval of the discrete switching variable is less than the minimum dwell time of the switch, the holding time is extended to the minimum dwell time. When the discrete switching variable triggers the switch dead zone, the current switching state is maintained until the switch dead zone ends.

[0109] After completing the continuous setting variable projection processing and the discrete switching variable projection processing, an implementable control vector that satisfies the voltage operating range constraint, harmonic limit constraint, minimum switching dwell time constraint and switching dead zone constraint is generated and recorded under a unified time index.

[0110] In this invention, convergence is determined by comparing the residual input amplitude with the convergence threshold during the iteration process. When the condition is met, the current control vector is determined as the final control vector and further converted into a compensation instruction. The instruction contains the output of a continuously set quantity and the execution information of discrete switching actions, which is sent to the control module under a unified time index. This process ensures that the compensation control can be quickly implemented when the convergence conditions are clear, avoiding the problems of result fluctuation and action lag in traditional methods.

[0111] In this embodiment, the output of the final control vector specifically includes:

[0112] During the iteration process, the amplitude of the residual input is calculated and compared with the convergence threshold. When the amplitude of the residual input is less than the convergence threshold, the convergence condition is considered to be met; when the amplitude of the residual input is greater than the convergence threshold, the convergence condition is considered not met.

[0113] When the convergence condition is met, the control vector obtained in the current iteration is determined as the final control vector. The final control vector consists of continuous set variables and discrete switching variables. The continuous set variables represent the set values ​​of the reactive power compensation device, and the discrete switching variables represent the switching states of the capacitor bank and the reactor.

[0114] The final control vector is converted into automatic switching control instructions for transformer power compensation. The control instructions include output instructions for continuously set quantities and execution instructions for discrete switching actions, and are sent to the control module to perform power compensation operations under a unified time index.

[0115] This invention introduces adaptive row selection, topological weighted diffusion, nonlinear residual mapping, and dynamic weight adjustment into the improved Kaczmarz algorithm. By associating residuals with the switching cost, the control variables are continuously modified during the iterative update process. This allows continuously set variables to be updated through continuous projection operators, while discrete switching variables remain stable through level alignment and temporal correction. Furthermore, the residual weights are dynamically adjusted at different stages to form a convergence path. This approach enhances the flexibility and adaptability of the iterative solution, making the compensation strategy more efficient and reliable.

[0116] Example 1:

[0117] To verify the feasibility of this invention in practice, it was applied to a regional power grid scenario. The main transformer of this regional power grid has a capacity of 50MVA and a voltage level of 110kV / 35kV. It is connected to large reactive load equipment and distributed photovoltaic devices. During the alternation of peak and off-peak loads, the power grid often experiences frequent voltage fluctuations, high harmonic distortion rate, decreased power factor, and frequent switching of compensation devices. Traditional control methods typically use the voltage deviation threshold triggering method, which automatically switches capacitor banks or reactors when the voltage deviates from the rated value by more than ±5%. This method is simple in structure but cannot take into account both load forecasting and multi-parameter constraints, which can easily lead to insufficient or excessive compensation, and can cause frequent switching when the load fluctuates drastically.

[0118] When applying the method of this invention in this scenario, voltage and current data are first collected from the primary and secondary sides of the transformer, and combined with historical active and reactive load data, as well as ambient temperature, humidity and photovoltaic output information, to form a monitoring dataset and then a load prediction model is established. The active and reactive power prediction curves for the next 15 minutes are output under a unified time step, and the prediction results are decomposed into short-term and long-term trends, which are then combined with the optimization input to form residual input as the driving force for iterative optimization.

[0119] In the modeling phase, a power grid topology weighted matrix is ​​constructed based on the optimization input, incorporating the connection relationships between nodes and branches into a unified framework. The voltage range is set to 5% above and below the rated value, the upper limit of harmonic voltage distortion rate is 5%, the minimum switching dwell time of capacitor banks and reactors is 300 seconds, and the switching dead zone is 120 seconds. At the same time, the set value of the reactive power compensation device is used as a continuous set variable, and the switching stages of capacitor banks and reactors are used as discrete switching variables. A switching cost model is established to quantify the lifetime consumption and energy loss of each action.

[0120] In the iterative calculation phase, the residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model are input into the improved Kaczmarz framework. In each iteration, the framework selects the update direction based on the ratio of the residual magnitude to the switching cost and uses the topology diffusion mechanism to propagate the residual. Subsequently, the residual is corrected and the control vector is updated through nonlinear mapping. Continuously set variables are kept within the voltage and harmonic range through projection, and discrete switching variables are kept stable through step alignment and minimum dwell time correction. When a candidate control vector fails the constraint detection, it is directly eliminated to avoid invalid solutions from entering the next round. Finally, when the residual magnitude is lower than the convergence threshold, the final control vector is output and converted into a compensation command.

[0121] To compare the effectiveness of the traditional voltage deviation threshold triggering method with the method of this invention under different operating conditions, tests were conducted during two typical periods: peak load and off-peak load. The average reactive power demand during peak load was approximately +15 Mvar, while the average reactive power demand during off-peak load was approximately -10 Mvar. The harmonic distortion rate exceeded 6% during peak load. The data obtained from the comparative experiments are shown in Table 1.

[0122] Table 1 Comparison of Control Method Operation Effects

[0123]

[0124] As can be seen from Table 1, the method of the present invention is significantly superior to the traditional voltage deviation threshold triggering method in terms of voltage control, harmonic suppression and power factor improvement. The voltage qualification rate is increased to over 99%, the harmonic distortion rate is reduced to less than 5%, the power factor is stabilized at 0.98, the reactive power loss is reduced from 12.4 Mvar·h to 8.1 Mvar·h, the average number of switch actions is reduced to less than 4 times per hour, and there are no more switch overheating alarms.

[0125] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A transformer power compensation automatic switching control method based on multi-parameter collaborative optimization, characterized in that, Includes the following steps: Collect transformer operating parameters, including voltage, current, active load, reactive load and environmental parameters, form a monitoring dataset, and complete time synchronization and normalization processing to generate optimized input; Based on the optimized input, a power grid topology weighting matrix is ​​constructed, and a set of constraints including voltage range, harmonic limit, minimum switching dwell time and switching dead zone is defined. A control variable structure and switching cost model including continuous set variables and discrete switching variables are established. The optimized input is decomposed into short-time and long-time residuals using a multi-scale method, and then combined with the load forecasting results to form the residual input. Input the residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model into the improved Kaczmarz iterative framework, set the initial control vector and the convergence threshold, and start the iterative process; During the iteration process, the improved Kaczmarz is used to generate candidate control vectors, and the feasibility of the candidate control vectors is checked. The candidate control vectors that pass the feasibility test are projected onto the constraint set to obtain implementable control vectors that satisfy the voltage range, harmonic limit, minimum switching dwell time and switching dead zone. When the control vector meets the convergence threshold, the final control vector is output, and the automatic switching control command for transformer power compensation is generated. The improved Kaczmarz includes: During the iteration process, adaptive row selection is performed based on the ratio of the magnitude of the residual input to the switching cost model; The topological residuals are generated by spreading the residuals of the selected rows using the power grid topology weighting matrix. Nonlinear residuals are obtained by performing nonlinear mapping on topological residuals; The control variables are updated and candidate control vectors are generated based on nonlinear residuals. The continuous set variables are updated through continuous projection operators, and the discrete cut variables are updated through level alignment and minimum dwell time correction. During the iteration process, the residual weights are dynamically adjusted according to different stages to form a convergence path; The generation and detection of the candidate control vectors specifically include: During the iteration process, the improved Kaczmarz update control vector is used to generate candidate control vectors, which consist of continuously set variables and discrete switching variables. Feasibility testing is performed on candidate control vectors. During the testing process, the voltage operating range constraint and harmonic limit constraint in the constraint set are used to determine whether the continuously set variables are within the allowable range. If the continuously set variables exceed the allowable range, the constraint conditions are not met. During the detection process, the switching sequence of discrete switching variables is determined based on the minimum dwell time constraint and the dead zone constraint in the constraint set. When the switching interval is less than the minimum dwell time or the dead zone is triggered, it is determined that the constraint conditions are not met. During the detection process, the judgment results of continuous set variables and discrete cut variables are combined. When all constraints are satisfied, the candidate control vector is retained. When there are constraints that are not satisfied, the candidate control vector is discarded and the iterative calculation is restarted.

2. The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization according to claim 1, characterized in that, The construction of the constraint set and control variable structure specifically includes: A power grid topology weighting matrix is ​​established based on the optimization input. The power grid topology weighting matrix is ​​used to represent the connection relationship between nodes and branches in the power grid. Each matrix element represents the branch weight between the corresponding nodes. Define a set of constraints, which includes voltage operating range constraints, harmonic limit constraints, minimum switch dwell time constraints, and switch dead zone constraints. Each constraint limits the range of values ​​for the control vector under the corresponding conditions. A control variable structure is established, which consists of continuous set variables and discrete switching variables. The continuous set variables are used to represent the set values ​​of the reactive power compensation device, and the discrete switching variables are used to represent the switching stages of the capacitor bank and the reactor. A switching cost model is constructed, which consists of a set of cost values. Each cost value corresponds to a switching action. The cost value is used to quantify the switching consumption and lifetime impact generated by the switching action.

3. The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization according to claim 1, characterized in that, The generation of the residual input specifically includes: A load prediction model is constructed using historical active load, reactive load, voltage, current and environmental parameters from the monitoring dataset. The input of the load prediction model is historical operating data, and the output is the predicted load sequence for future time. The prediction model is run on the sampling interval and time index consistent with the optimization input to obtain the load prediction results. The optimized input is decomposed into short-time and long-time components by multi-scale decomposition, and the load forecasting results are decomposed into short-time and long-time forecasting components by multi-scale decomposition in the same way. The two types of components correspond point by point on a unified time index. Calculate the short-time residual and the long-time residual. The short-time residual is the short-time component minus the short-time prediction component and recorded according to the time index. The long-time residual is the long-time component minus the long-time prediction component and recorded according to the time index. The short-time residuals and long-time residuals are weighted and fused on a unified time axis to form the residual input. The weighting coefficients take values ​​in the range of zero to one and remain unchanged within one iteration cycle.

4. The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization according to claim 1, characterized in that, The execution of the iterative framework specifically includes: Set an initial control vector, which consists of continuous set variables and discrete switching variables. The initial values ​​of the continuous set variables are derived from the operating parameters of the reactive power compensation device, and the initial values ​​of the discrete switching variables are derived from the switching states of the capacitor bank and the reactor. Set a convergence threshold, which limits the amplitude range of the residual input. The termination condition is met when the amplitude of the residual input does not exceed the convergence threshold. The residual input, the power grid topology weighting matrix, the constraint set, the control variable structure, and the switching cost model are input into the improved Kaczmarz iterative framework. In each iteration, the row index is determined according to the adaptive row selection strategy, and the topology-weighted row vector and nonlinear residual are calculated. The intermediate control vector is calculated according to the following formula: ； in, For the first The control vector for the next iteration For the first The intermediate control vector updated in the next iteration. For the first The step size of the next iteration. For the first The next iteration is in the row index. The residual input at the location, For residual input at row index The corresponding row vector at that position, This is a linear operator representation of the power grid topology weighting matrix. This is the result of topologically weighting the row vectors. The dot product of the topologically weighted row vector and itself. This is a function that performs a nonlinear mapping on the residual input; The intermediate control vector is constrained and made feasible. The continuous set variables are continuously projected according to the voltage operating range and harmonic limits. The discrete switching variables are aligned according to the switching level difference and timed according to the minimum dwell time and dead zone of the switch. After processing, the control vector for entering the next iteration is obtained. During the iteration process, a convergence judgment is performed. When the magnitude of the residual input does not exceed the convergence threshold, the iteration stops and the control vector is output. When the magnitude of the residual input exceeds the convergence threshold, the iteration continues.

5. The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization according to claim 1, characterized in that, The generation of the implementable control vector specifically includes: The candidate control vectors that pass the feasibility test are input into the constraint set, and projection processing is performed on the continuous set variables. When the continuous set variables exceed the voltage operating range, they are truncated according to the upper and lower limit boundaries. When the continuous set variables exceed the harmonic limit, they are corrected according to the limit value corresponding to the harmonic constraint. Projection processing is performed on discrete switching variables. When the switching interval of the discrete switching variable is less than the minimum dwell time of the switch, the holding time is extended to the minimum dwell time. When the discrete switching variable triggers the switch dead zone, the current switching state is maintained until the switch dead zone ends. After completing the continuous setting variable projection processing and the discrete switching variable projection processing, an implementable control vector that satisfies the voltage operating range constraint, harmonic limit constraint, minimum switching dwell time constraint and switching dead zone constraint is generated and recorded under a unified time index.

6. The automatic switching control method for transformer power compensation based on multi-parameter collaborative optimization according to claim 1, characterized in that, The output of the final control vector specifically includes: During the iteration process, the amplitude of the residual input is calculated and compared with the convergence threshold. When the amplitude of the residual input is less than the convergence threshold, the convergence condition is considered to be met; when the amplitude of the residual input is greater than the convergence threshold, the convergence condition is considered not met. When the convergence condition is met, the control vector obtained in the current iteration is determined as the final control vector. The final control vector consists of continuous set variables and discrete switching variables. The continuous set variables represent the set values ​​of the reactive power compensation device, and the discrete switching variables represent the switching states of the capacitor bank and the reactor. The final control vector is converted into automatic switching control instructions for transformer power compensation. The control instructions include output instructions for continuously set quantities and execution instructions for discrete switching actions, and are sent to the control module to perform power compensation operations under a unified time index.

Images