A multi-stage reactive power compensation configurable transformer test capacitor compensation method and device

Abstract

This application provides a multi-level reactive power compensation configurable capacitor compensation method and device for transformer testing, belonging to the field of power system transformer testing technology. It addresses the problems of rigid capacitor switching control strategies, difficulty in co-optimizing compensation accuracy, equipment safety, and operational economy in related technologies. The method constructs a unified state representation by fusing real-time data and target parameters, and uses a multi-objective optimization function coupled with trend inference and switching strategies for collaborative solution. Simultaneously, it generates optimal compensation commands, which, after parsing, drive the capacitor bank to execute. The system can also adaptively adjust its internal model based on execution feedback. The corresponding device includes modules for data fusion, collaborative optimization, command execution, and adaptive learning. This application achieves intelligent compensation, simultaneously optimizing tracking accuracy, equipment lifespan, and operational costs.

Description

Technical Field

[0001] This application relates to the field of power system transformer testing technology, and in particular to a capacitor compensation method and device for transformer testing with multi-level reactive power compensation and configurable. Background Technology

[0002] In the factory testing of power transformers, to meet the testing requirements of large-capacity inductive loads, it is often necessary to configure multi-stage parallel capacitor banks for dynamic reactive power compensation to improve the load-carrying capacity and power factor of the test power supply. In existing technologies, the switching control of capacitor banks mainly relies on simple logic or classic PID control strategies based on preset thresholds.

[0003] These traditional methods typically treat reactive power demand detection, strategy decision-making, and execution control as separate processes. Strategy generation often targets only a single power factor objective or ensures safety by combining simple protection thresholds. For the configuration of multi-stage discrete capacitors, a fixed pattern of sequential or grouped switching is often used.

[0004] However, the aforementioned existing technologies have obvious drawbacks: their response and control strategies are relatively rigid, making it difficult to optimize multiple objectives such as compensation accuracy, number of equipment actions, and capacitor bank life balance in real time and in a coordinated manner under rapidly changing test conditions. They also lack the ability to self-optimize using historical data and real-time feedback, resulting in the need to improve the overall test efficiency and economy. Summary of the Invention

[0005] This application provides a multi-level reactive power compensation configurable transformer test capacitor compensation method and device, which can achieve accurate, economical and adaptive control of reactive power compensation during the test in a collaborative optimization manner.

[0006] Firstly, this application provides a multi-level reactive power compensation configurable capacitor compensation method for transformer testing. Real-time operating data and preset target parameters of the transformer testing system are acquired, and the operating data and target parameters are fused to form a unified state representation. Based on this unified state representation, an optimization process simultaneously generates an inference result of the system state change trend and a switching strategy for the multi-level capacitors, forming a joint output. The optimization process is achieved by minimizing a multi-objective optimization function, which is configured to comprehensively evaluate the rationality of the inference result, the tracking effect of the switching strategy on the preset power factor target, the degree of satisfaction of equipment safety operation constraints, and its own operational economy. Based on the switching strategy in the optimized joint output, control commands for specific capacitor units in the multi-level capacitor bank are generated and executed.

[0007] By adopting the above technical solution, this method integrates the real-time state of the system with the target intent through the construction of a unified state representation, providing a comprehensive information foundation for intelligent decision-making. Furthermore, through an integrated multi-objective optimization process, it simultaneously completes the "perception" of the system's future trends and the "decision" of the optimal compensation strategy, achieving deep synergy between perception and decision-making. This method no longer relies on fixed thresholds or simple feedback, but dynamically generates optimal switching commands by comprehensively evaluating compensation accuracy, safety constraints, and economy, thereby achieving more accurate, faster, and more economical reactive power compensation in complex experimental environments.

[0008] Furthermore, the unified state representation is composed of the following dimensions: a physical quantity dimension representing the real-time electrical state of the system, a memory dimension representing the historical actions and operating states of the capacitor bank, and an intention dimension representing the compensation target and safe operating boundary.

[0009] By adopting the above technical solution, the state representation is concretized into a fusion of three dimensions: physical, memory, and intention. This ensures that the optimization decision is based on both the current instantaneous operating conditions and the historical state of the equipment and long-term operating goals, making the decision more forward-looking and holistic.

[0010] Furthermore, in the joint output, the inference result of the system state change trend and the switching strategy of the multi-stage capacitor are coupled through an adaptively adjustable correlation model; the correlation model enables the specific state change trend identified in the inference result to be automatically associated with the corresponding compensation adjustment action.

[0011] By adopting the above technical solution, an adjustable correlation model is introduced, which intelligently couples "perception" (inference results) with "action" (switching strategy), enabling the system to learn and establish the most effective compensation response rules under different operating conditions, thereby improving the level of intelligence in control.

[0012] Furthermore, the adaptive adjustment of the correlation model is based on the difference between the actual response of the system and the response predicted by the inference result after the switching strategy is executed; the direction of the adjustment is to reduce the expected difference between the predicted response and the actual response in the joint output under similar conditions in the future.

[0013] By adopting the above technical solution, a model self-learning mechanism based on actual feedback was established. The system can continuously correct its internal association rules according to the effect of strategy execution, thereby continuously improving decision-making accuracy and adapting to changes in equipment characteristics and fluctuations in the test environment.

[0014] Furthermore, the evaluation of motion economy in the multi-objective optimization function includes a comprehensive consideration of the frequency and amplitude of the throwing and cutting motions.

[0015] By adopting the above technical solutions, the frequency and amplitude of actions are incorporated into the economic evaluation, enabling the optimization process to proactively seek strategies to minimize the mechanical and electrical wear of switching devices while meeting compensation requirements, thereby extending the overall lifespan of the equipment.

[0016] Furthermore, the preset target parameters include a steady-state power factor target value, test process stage information, and a dynamic reactive power demand reference determined by the parameters of the transformer under test.

[0017] By adopting the above technical solutions, the stages of the experiment and the characteristics of the test equipment are incorporated into the target setting, so that the compensation control can be matched with the test process and the specific test object, and more refined target tracking can be achieved.

[0018] Furthermore, the step of generating control instructions based on the switching strategy includes: quantifying the strategy into a priority sequence of each candidate switching action; and determining the specific capacitor unit and action to be executed based on the priority sequence and in conjunction with the current state of the multi-level capacitors and the principle of balanced operation.

[0019] By employing the above technical solution, a method is provided to reliably map continuous or abstract optimization strategies into discrete, concrete hardware execution instructions. By introducing priority sorting and device balancing principles, reliable instruction execution and balanced capacitor usage are ensured.

[0020] Furthermore, the evaluation of the degree to which the constraints on safe operation of the equipment are met includes at least consideration of the constraints on the temperature rise of the capacitor bank and the harmonic content of the circuit current.

[0021] By adopting the above technical solutions, key safety operation indicators (temperature rise and harmonics) are clearly incorporated into the optimization objectives as hard constraints, and the risks of equipment overheating and electrical pollution are actively avoided from the algorithm level, ensuring the safety and reliability of the test process.

[0022] Furthermore, the real-time operating data includes at least the three-phase voltage and current of the test circuit, the temperature of the capacitor body, and the switching history of each capacitor unit.

[0023] By adopting the above technical solution, the minimum dataset required to support the unified state representation and optimization decision-making is clarified, ensuring the feasibility and completeness of the technical solution.

[0024] Secondly, this application provides a multi-level reactive power compensation configurable capacitor compensation device for transformer testing. The method described in any one of the first aspects includes: a data fusion module for acquiring real-time operating data and preset target parameters of the transformer testing system, and performing fusion processing to form a unified state representation; a collaborative optimization module for, based on the unified state representation, minimizing a multi-objective optimization function, simultaneously generating an inference result of the system state change trend and a switching strategy for the multi-level capacitors, and outputting a joint result; an instruction generation and execution module for generating and executing specific control instructions for the multi-level capacitor bank according to the switching strategy in the joint result; and an adaptive learning module for adjusting the internal correlation model in the collaborative optimization module according to the actual system response after instruction execution.

[0025] By adopting the above technical solution, the device realizes all the functions of the above method through modular design. The modules work together to form a complete intelligent compensation system with perception, decision-making, execution and learning capabilities.

[0026] In summary, this application has at least the following beneficial effects:

[0027] A method and device for intelligent and collaborative reactive power compensation are provided, which realizes multi-objective integrated optimization of compensation accuracy, equipment safety and operation economy in transformer testing;

[0028] By introducing an adaptive and adjustable association model, the system is able to learn from historical data and optimize the "perception-decision" mapping relationship.

[0029] By constructing a multidimensional state representation that integrates physics, memory, and intention, the decision-making process can be made to be real-time, based on historical experience, and goal-oriented.

[0030] It should be understood that the description in the Summary Section is not intended to limit the key or essential features of the embodiments of this application, nor is it intended to restrict the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0031] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0032] Figure 1 A schematic diagram of an exemplary operating environment in which embodiments of this application can be implemented is shown.

[0033] Figure 2 A flowchart of a multi-level reactive power compensation configurable transformer test capacitor compensation method is shown in an embodiment of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0036] This application provides a multi-level reactive power compensation configurable capacitor compensation method and device for transformer testing. It integrates multi-source data to form a unified state representation and utilizes a self-learning collaborative optimization mechanism to simultaneously achieve compensation accuracy tracking, equipment safety protection, and operational economy optimization, thereby significantly improving the intelligence level, operating efficiency, and equipment lifespan of transformer testing.

[0037] Figure 1 A schematic diagram of an exemplary operating environment in which embodiments of this application can be implemented is shown.

[0038] Reference Figure 1 The operating environment includes a complete transformer test station system, which provides the necessary physical hardware and data interaction foundation for the implementation of the multi-level reactive power compensation configurable capacitor compensation method. This operating environment is mainly composed of the following hardware support systems working together.

[0039] First, there is the test power supply and load system. This system includes a high-capacity adjustable test power supply, the transformer under test, and related load equipment. Its function is to provide the required high-voltage and high-current test conditions for various transformer tests and to generate dynamically changing reactive power demands, which constitute the object of compensation control.

[0040] Secondly, there is the multi-level capacitor compensation execution system. This system consists of multiple independent, switchable reactive power compensation capacitor units, usually arranged in group cabinets or tower structures. Its function is to act as a controllable reactive power source, which can be quickly switched on or off according to control commands to dynamically compensate for the inductive reactive power in the test circuit. It is the final execution mechanism of the method.

[0041] Third is the comprehensive state perception system. This system includes high-precision voltage and current transformers deployed in the test circuit, temperature sensors installed on the capacitor body, and state sensors used to monitor switch positions. Its function is to collect the electrical parameters of the system, equipment temperature, and switch status in real time, providing real-time, multi-dimensional data input for intelligent decision-making.

[0042] Fourth is the centralized control and computing system. The core of this system is an industrial control computer or a high-performance programmable logic controller. Its function is to carry and run the data fusion, collaborative optimization, and adaptive learning algorithms, process the sensed data, generate optimization strategies, and finally issue switching control commands. It is the computational and control core of the entire intelligent compensation method.

[0043] Fifth, there is the high-speed data communication system. This system consists of industrial Ethernet switches, fieldbus networks, and corresponding communication interfaces. Its function is to reliably and in real-time connect the aforementioned sensing, control, and execution systems, ensuring the uploading of monitoring data and the issuance of control commands. It is a key link in realizing closed-loop real-time control.

[0044] In summary, the aforementioned hardware systems are organically integrated through electrical connections and communication networks, forming a complete closed loop from state perception and intelligent decision-making to precise execution. The perception system transmits the real-time state of the test system to the control and computing system, which processes the data using algorithms to generate optimization strategies. The communication system then sends control commands to the compensation execution system, thereby driving the capacitor bank to operate and ultimately providing hardware support for the capacitor compensation method used in the transformer test.

[0045] Secondly, embodiments of this application disclose a capacitor compensation method for transformer testing with configurable multi-level reactive power compensation.

[0046] Figure 2 A flowchart of a multi-level reactive power compensation configurable transformer test capacitor compensation method is shown in an embodiment of this application.

[0047] Reference Figure 2 The implementation of this method relies on a complete transformer testing hardware environment and is executed within the computational control unit of that environment. The core inventive concept lies in abandoning the traditional separate detection, calculation, and execution model, and instead constructing an intelligent collaborative optimization framework integrating state perception, trend inference, strategy decision-making, and online learning. This framework, through a unified embedded model, deeply integrates and optimizes the system's physical dynamics, historical experience, safety boundaries, and compensation objectives in real time. This enables automatic, accurate, economical, and safe control of multi-stage capacitor switching under complex and changing testing conditions, significantly improving the automation level and overall efficiency of the entire testing system.

[0048] The implementation of the method begins with the synchronous acquisition and deep fusion processing of multi-source heterogeneous data from the transformer test system. High-precision voltage and current transformers deployed in the test circuit measure the instantaneous values ​​of the three-phase voltages in real time. With the instantaneous value of three-phase current These form the basis for calculating the electrical state. Temperature sensors attached to the heat sinks of each capacitor periodically collect the body temperature values ​​of each capacitor unit. , This is used to monitor the thermal status of the equipment. The control system's registers record the switching sequence of each capacitor cell within a recent time window, forming a historical record. ,in The value can be either 1 (input) or 0 (removal). These data collectively constitute the real-time operational dataset. .

[0049] At the same time, the method receives a preset target parameter set. These parameters are derived from the test procedures and the characteristics of the device under test, specifically including the target steady-state power factor value specified by the test standard. ; Stage variables that indicate the current test progress (such as "no-load loss measurement", "induced withstand voltage", etc.). This variable determines different control priorities and safety thresholds; and a dynamic reactive power demand reference trajectory that varies over time, pre-generated through offline simulation or empirical formulas based on parameters such as the rated capacity and short-circuit impedance of the transformer under test. This trajectory provides a predicted profile of reactive load changes during the test.

[0050] The core task of data fusion processing is to construct a unified high-dimensional state representation vector. This representation is not simply a compilation of raw data, but rather, through feature extraction and organization, it forms three logically distinct dimensions. The physical quantity dimension reflects the instantaneous energy state of the system, obtained through real-time calculation: the total active power of the system. Total reactive power ,in Phase difference between voltage and current of each phase; current power factor ; and the average temperature of the capacitor bank The memory dimension encodes historical information related to equipment lifespan and decision-making inertia, such as the cumulative number of switching operations for each capacitor since the last maintenance. The duration of operation in the current switching state And the exponentially weighted moving average of the power factor tracking error over a period of time. This value reflects recent control accuracy. The intent dimension clarifies the current optimization direction and hard boundaries, transforming the target parameters into driving signals and constraints: combined with and Calculate the reactive power difference required to achieve the target power factor. At the same time, according to and safety regulations, generating data including the maximum permissible temperature of the capacitor. Maximum permissible current harmonic distortion rate and maximum inrush flow limit per operation constraint set .final, This representation structurally integrates transient, historical, and future intentions, providing a complete decision-making context for subsequent intelligent optimization.

[0051] Based on this unified state representation, the method generates a joint output by solving a carefully constructed optimization problem. This output is a pair of tuples. ,in It is a quantitative inference of the trend of key state changes of the system in a short time domain in the future (such as the next control cycle), such as the predicted change in reactive power. Temperature rise of the hottest spot in the capacitor bank ; Then it is a policy vector defined on the action space of a multi-level capacitor, and each component of it is a policy vector. This corresponds to a propensity score for a specific combination of throwing and cutting actions. The optimization objective is to minimize a multi-objective function. This function comprehensively evaluates the rationality of the inference, the effectiveness of the strategy, its security, and its economy.

[0052] Multi-objective function It consists of four sub-items with clear physical and engineering significance. (Inference Reasonableness Item) To ensure that trend inferences conform to physical laws and historical patterns, one feasible form is... .here, Based on state Prior inferences generated by a simplified online physics model (such as estimating temperature rise using thermal time constants) or a lightweight time series prediction model (such as AR models); It is a diagonal weight matrix whose elements are determined based on the reciprocal of the measurement noise of each state variable or the reciprocal of the variance of the historical prediction error. This function is to punish absurd inferences that deviate from common sense and historical statistical laws, thus ensuring the reliability of the "perception" process.

[0053] Power factor tracking term Directly driven strategies aim to meet core compensation objectives, and their core principle is establishing a mapping relationship between strategy actions and reactive power compensation effects. Define the function. policy vector Mapped to the expected amount of reactive power compensation available This mapping relationship can be pre-established based on a lookup table of capacitor nameplate capacitance and system impedance parameters, or approximated by an online estimated linear model. This term can then be defined as... Its function is to make the compensation provided by the strategy as close as possible to the current required reactive power difference, which is the core driving force for achieving precise control.

[0054] Safety constraints Transform equipment protection requirements into soft constraints or penalties in an optimization problem. It comprises at least two key components: first, temperature rise safety; ,in , It is based on inference The predicted maximum temperature rise was obtained from the analysis. The first is a large penalty coefficient; the second is harmonic and inrush safety. ,in and Based on strategy The harmonic distortion rate and inrush current predicted by the network impedance model. This item functions as a "safety guardian," proactively avoiding dangerous operations that could lead to equipment overheating or excessive electrical stress during the optimization process.

[0055] Motion economy item This focuses on the long-term execution cost of the control strategy and the wear and tear on equipment lifespan. It is specifically manifested as a comprehensive penalty for the frequency and magnitude of actions: The first term uses the L1 norm to compute the current policy. The strategy executed in the previous time step The difference, whose value is proportional to the number of capacitors undergoing state changes, directly penalizes the frequency of switching actions; the second term uses the total magnitude of the L2 norm squared penalty strategy vector, tending to choose to use fewer, smaller capacitor cells to achieve the goal, thereby reducing the cumulative electrical stress of the device. Weighting coefficients and It can be adjusted according to the equipment life cost and maintenance cycle. This function is to achieve long-term operation economy and extend the overall life of the equipment.

[0056] Thus, the optimization problem is formalized as follows: Solving this problem will yield the optimal trend prediction for the current moment. With pitching and cutting strategies It needs to be emphasized that inference... With strategy They are not independent optimizations; they are performed through the objective function. Tight coupling. For example, Requires the generation of strategies Matching is determined based on the current state and trend inferences. ; Temperature rise predictions in China rely directly on inference. However, this inference may be influenced by strategy. Constraints on the impact on future states. This coupled design is one of the core innovations of this method, enabling the decision-making process to be based on future projections, thereby achieving forward-looking control rather than simple feedback correction.

[0057] Obtaining abstract strategies Next, it needs to be parsed into specific control instructions that can be executed by hardware. This process first involves... This is interpreted as a utility score for each alternative throwing action. For example, based on a mapping model. It is possible to calculate the reactive power increment that can be provided by every possible single capacitor switching action. , and its relationship to the policy vector Contribution or matching degree of direction Subsequently, based on the scores... Sort all available actions in descending order to form a priority list.

[0058] When finalizing the execution instruction, the principle of balanced equipment operation must be introduced. This principle requires that, among multiple alternative actions with similar priority scores, the one with the highest cumulative number of switching operations should be selected first. Less, or cumulative running time The action corresponding to a shorter capacitor cell. This can be achieved by introducing a slight bias related to historical information into the final decision. Finally, the instruction generation module outputs a specific set of operations, such as "at time..." "Close circuit breaker CB_5" and "Disconnect contactor K_8". This step ensures that the abstract optimization results can be reliably and fairly translated into actions in the physical world, and is a key link in achieving method closure.

[0059] The optimization model upon which this method relies (such as the mapping function) Prediction model parameters, weighting coefficients (etc.) It possesses the ability to self-evolve online. The adaptive learning mechanism is built upon feedback on the effects after policy execution. (Note:) The strategy executed at each moment ultimately resolves to the following actions: The system is The actual response at time t is reflected in the new state representation. And in Inference in joint output at time Implies a Prediction of state at time step Define the prediction bias vector. .

[0060] The goal of learning is to adjust the model parameters. This minimizes the expected prediction bias in similar future scenarios. This can be achieved online using stochastic gradient descent. For example, for mapping models... parameters Its update rules can be ,in It is based on and The actual compensation effect obtained by reverse calculation The learning rate is used. By continuously integrating this feedback, the system can automatically correct model errors and adapt to slow time-varying factors such as capacitor capacitance drift and network impedance changes, thus making the entire intelligent control system more accurate and robust over time. This self-learning closed loop is another core feature that distinguishes this method from static optimization algorithms, giving the system true "intelligence" and long-term adaptability.

[0061] In summary, this embodiment of the method details how to solve the complex control problem of multi-stage capacitor compensation in transformer testing by utilizing advanced optimization theory and learning algorithms through a complete closed loop from data fusion to instruction execution and model learning. Each technical step has a clear mathematical description and engineering explanation, which together constitute a clear, complete solution that can be implemented by those skilled in the art.

[0062] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0063] Secondly, embodiments of this application disclose a capacitor compensation device for transformer testing with configurable multi-level reactive power compensation. This device is used to execute the method described in any one of the first aspects, comprising: a data fusion module for acquiring real-time operating data and preset target parameters of the transformer testing system, and performing fusion processing to form a unified state representation; a collaborative optimization module for, based on the unified state representation, minimizing a multi-objective optimization function, simultaneously generating an inference result of the system state change trend and a switching strategy for the multi-level capacitors, and outputting a joint result; an instruction generation and execution module for generating and executing specific control instructions for the multi-level capacitor bank according to the switching strategy in the joint result; and an adaptive learning module for adjusting the internal correlation model in the collaborative optimization module according to the actual system response after instruction execution.

[0064] By adopting the above technical solution, the device realizes all the functions of the above method through modular design. The modules work together to form a complete intelligent compensation system with perception, decision-making, execution and learning capabilities.

[0065] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0066] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for compensating capacitance for transformer test with multi-stage reactive power compensation configurable, characterized in that, Includes the following steps: The real-time operating data and preset target parameters of the transformer test system are acquired, and the operating data and target parameters are fused together to form a unified state representation. Based on the unified state representation, an optimization process is used to simultaneously generate the inference results of the system state change trend and the switching strategy of multi-stage capacitors, forming a joint output. The optimization process is achieved by minimizing a multi-objective optimization function, which is configured to comprehensively evaluate the rationality of the inference result, the tracking effect of the switching strategy on the preset power factor target, the degree of satisfaction of the equipment safety operation constraints, and its own operational economy. Based on the switching strategy in the optimized joint output, control instructions for specific capacitor units in the multi-level capacitor bank are generated and executed.

2. The multi-level reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, The unified state representation is composed of the following dimensions: physical quantity dimension representing the real-time electrical state of the system, memory dimension representing the historical actions and operating states of the capacitor bank, and intention dimension representing the compensation target and safe operating boundary.

3. The multi-stage reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, In the joint output, the inference result of the system state change trend and the switching strategy of the multi-stage capacitor are coupled through an adaptively adjustable correlation model. The correlation model automatically associates the specific state change trends identified in the inference results with the corresponding compensation and adjustment actions.

4. The multi-stage reactive power compensation configurable transformer test capacitor compensation method according to claim 3, characterized in that, The adaptive adjustment of the correlation model is based on the difference between the actual response of the system and the response predicted by the inference results after the switching strategy is executed. The adjustment aims to reduce the expected difference between the predicted response and the actual response in the joint output under similar future conditions.

5. The multi-level reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, The evaluation of motion economy in the multi-objective optimization function includes a comprehensive consideration of the frequency and amplitude of the throwing and cutting motions.

6. The multi-level reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, The preset target parameters include the target value of steady-state power factor, information on the test process stages, and a reference for dynamic reactive power demand determined by the parameters of the transformer under test.

7. The multi-stage reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, The step of generating control instructions based on the throwing and switching strategy includes: quantifying the strategy into a priority sequence of each alternative throwing and switching action; Based on the priority sequence and in combination with the current state of the multi-level capacitors and the principle of balanced operation, the specific capacitor unit and action to be executed are determined.

8. The multi-level reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, The evaluation of the degree to which the constraints on safe operation of the equipment are met includes at least consideration of the constraints on the temperature rise of the capacitor bank and the harmonic content of the circuit current.

9. The multi-stage reactive power compensation configurable transformer test capacitor compensation method according to claim 1, characterized in that, The real-time operating data includes at least the three-phase voltage and current of the test circuit, the temperature of the capacitor body, and the switching history of each capacitor unit.

10. A multi-stage reactive compensation configurable transformer test capacitive compensation device, characterized by, For performing the method according to any one of claims 1 to 9, comprising: The data fusion module is used to acquire real-time operating data and preset target parameters of the transformer test system, and to perform fusion processing to form a unified state representation. The collaborative optimization module is used to generate, based on the unified state representation, a multi-objective optimization function by minimizing it, and simultaneously generate inference results of system state change trends and switching strategies of multi-stage capacitors, and output joint results. The instruction generation and execution module is used to generate and execute specific control instructions for the multi-stage capacitor bank based on the switching strategy in the joint result. The adaptive learning module is used to adjust the internal correlation model in the collaborative optimization module based on the actual system response after the instruction is executed.

Images