Intelligent reactive power automatic compensation control device and method

Abstract

The invention discloses an intelligent reactive power automatic compensation control device and method, and relates to the technical field of reactive power compensation of a power system. The device comprises a power supply input module, a sampling and signal conditioning module, a high-speed ADC module-block, an FPGA high-speed processing unit, a data storage module, an AI decision unit, a control driving module, a multi-channel capacitor switching device and a capacitor bank module. Through combination of the multi-channel capacitor switching device and an intelligent dynamic control technology, matrix-type accurate switching control of the capacitor bank is realized, the corresponding capacitor bank can be flexibly switched based on reactive power data dynamically acquired in real time, instant accurate compensation of reactive power capacity is achieved, and the reactive power capacity of the capacitor bank is accurately compensated. And the adaptability and the flexibility of compensation under different load working conditions are ensured.

Description

Technical Field

[0001] This invention relates to the field of reactive power compensation technology in power systems. More specifically, this invention relates to an intelligent automatic reactive power compensation control device and method. Background Technology

[0002] In industrial and residential power supply systems, the extensive use of inductive loads leads to increased reactive power and a decreased power factor in the power grid. This not only wastes electrical energy but can also cause voltage fluctuations, affecting power quality. Power supply departments typically conduct rigorous assessments of users' power factors and implement reward and penalty mechanisms. For example, there are no rewards or penalties when the power factor is equal to 0.9; penalties increase as the power factor decreases below 0.9; and rewards increase as the power factor increases above 0.9. Whether the power factor meets the standards directly affects the user's economic rewards and penalties, thus impacting electricity costs.

[0003] To address this issue, existing automatic reactive power compensation devices generally employ fixed-capacity capacitor banks, controlling their switching based on power factor or reactive power measurements. However, these devices have significant limitations: when active power in the circuit is low (e.g., at night, during factory shutdowns, or when equipment is on standby), the reactive current generated by inductive loads is small, and the minimum switching capacity of conventional compensation capacitors may be far greater than the actual requirement, easily leading to overcompensation. Alternatively, the device may lock out the compensation function to avoid frequent switching oscillations caused by mismatched capacitors, ultimately failing to achieve effective compensation and making it difficult to meet the power factor requirements. Even under normal load conditions, the switching strategy of traditional devices is relatively fixed, unable to precisely match rapidly changing load characteristics, resulting in a lag in compensation response and a single compensation capacity step, making it difficult to achieve the optimal power factor improvement effect.

[0004] To address some of the aforementioned issues, the applicant previously filed an application on August 27, 2014, for an automatic reactive power compensation control device (application number / patent number: 201410427011.8). This device, by adding a small-capacity reactive power control circuit, improved the power factor to some extent when there was a small inductive current in the circuit. However, with the development of power grid technology and electricity consumption conditions, this device still cannot meet modern needs. For example, factories and other electricity consumption scenarios often have equipment such as computerized knitting machines. To prevent data loss, these devices need to remain in standby mode during non-working hours, resulting in low active power and high reactive power in the power grid. While the patented device can handle power conditions, it still cannot achieve instantaneous automatic compensation under such conditions. Furthermore, its switching strategy lacks intelligence and falls short in balancing compensation continuity and adaptability, failing to meet the demands of modern power grids for energy conservation, power quality improvement, and cost reduction. In addition, existing compensation devices have a fixed number of control output ports (e.g., 20 or 40), making it difficult to flexibly adapt to the reactive power compensation control cabinet designs corresponding to transformers of different capacity levels (e.g., from 250KVA to 3000KVA). This necessitates users customizing different core controllers for different application scenarios, increasing equipment costs and maintenance complexity.

[0005] Furthermore, existing compensation devices are only designed for three-phase balanced loads and cannot address the common three-phase power factor imbalance problems in industrial and residential scenarios. For example, equipment such as two-phase welding machines in factories and single-phase inductive loads in residential areas can cause uneven three-phase currents and phase differences. Traditional devices lack targeted compensation methods, and compensation by three-phase capacitor banks alone is insufficient to ensure that the power factor of each phase meets the standard, affecting power supply quality and assessment results.

[0006] In summary, neither traditional fixed-capacity compensation devices nor the aforementioned 2014 patented device can achieve fast, accurate, and adaptive reactive power compensation under a wide load range (especially under low active power conditions). How to overcome this technical bottleneck has become an urgent problem to be solved in this field. Summary of the Invention

[0007] This invention provides an intelligent automatic reactive power compensation control device and method. By combining a multi-channel capacitor switch with intelligent dynamic control technology, it achieves matrix-style precise switching control of capacitor banks. Based on real-time dynamically acquired reactive power data, it can flexibly switch corresponding capacitor banks to achieve instantaneous and precise compensation of reactive capacity. At the same time, it calculates and matches the required capacitor capacity specifications by combining the type and capacity characteristics of the inductive load carried by the transformer, and can predefine multiple switching combination strategies to ensure the adaptability and flexibility of compensation under different load conditions.

[0008] To achieve these objectives and other advantages according to the present invention, an intelligent automatic reactive power compensation control device is provided, comprising: The power input module is used to connect to an AC power source and provide operating power. The sampling and signal conditioning module is connected to the power input module and is used to collect voltage analog signals, current analog signals, current waveform signals and voltage-current phase difference signals at the transformer load end in real time from the power grid, and to isolate and filter the signals to output a conditioned low-voltage safety analog signal. A high-speed ADC module, which is connected to the sampling and signal conditioning module and the power input module, is used to convert low-voltage safety analog signals into digital signals; The FPGA high-speed processing unit is connected to the high-speed ADC module and the power input module. It is used to receive digital signals, perform instantaneous reactive power calculation and harmonic analysis, and calculate the type and capacity of the load carried by the transformer based on the current waveform signal and the voltage-current phase difference signal, and generate real-time reactive power status data, load type information and load capacity information. A data storage module, which is connected to the power input module, is used to store a variety of predefined capacitor switching combination strategies; The AI ​​decision unit is connected to the FPGA high-speed processing unit, the data storage module, and the power input module, respectively. It receives real-time reactive power status data, load type information, and load capacity information. It then calls the switching combination strategy in the data storage module, calculates the required capacitor capacity based on the load type and capacity information, performs the calculation using a pre-trained AI model, and outputs a matrix-style capacitor switching command. The AI ​​decision unit is also configured to: identify three-phase power factor imbalance based on the three-phase power factor data provided by the FPGA high-speed processing unit; and generate a single-phase compensation switching command for a specific phase when an imbalance is detected. The control drive module is connected to the AI ​​decision unit, the FPGA high-speed processing unit and the power input module. It is used to prioritize receiving the matrix capacitor switching command issued by the AI ​​decision unit, and to receive the backup switching command issued by the FPGA high-speed processing unit when the AI ​​decision unit is abnormal, thereby generating the corresponding multi-channel drive signal. A multi-channel capacitor switch, which is connected to the control drive module and the power input module; and The capacitor bank module is connected to the multi-channel capacitor switch and the power input module. The capacitor bank module includes multiple independently controlled capacitor branches, and the capacity specifications of each branch include at least three different levels. The intelligent automatic reactive power compensation control device has no less than 80 configurable switching ports. The multi-channel capacitor switch controls the multi-channel capacitor branches to dynamically switch and combine them according to the multi-channel drive signals through the configurable switching ports, forming a matrix switching structure to achieve instantaneous and accurate compensation of reactive power capacity.

[0009] Preferably, the FPGA high-speed processing unit is further configured to automatically switch to the built-in backup control logic when the confidence level of the output instruction of the AI ​​decision unit is lower than a preset threshold or the AI ​​decision unit malfunctions, and directly issue a backup switching instruction to the control drive module; the backup control logic is a PID control algorithm or fuzzy control logic.

[0010] Preferably, it also includes: The fault detection and alarm module is connected to the FPGA high-speed processing unit and the power input module. The fault detection and alarm module is configured to: monitor the system harmonic distortion rate and power factor in real time, and output an alarm signal and send fault information to the external monitoring system when the harmonic distortion rate is greater than 5% or the power factor is lower than 0.99.

[0011] Preferably, it also includes: The fault diagnosis and self-healing module is connected to the FPGA high-speed processing unit, the control drive module, and the power input module. The fault diagnosis and self-healing module is configured to monitor the switching status and health status of each branch in the capacitor bank module. When a fault is detected in a specific branch, the information is fed back to the AI ​​decision unit. The AI ​​decision unit dynamically adjusts the subsequent switching combination strategy to bypass the faulty branch and recombines the remaining healthy branches to achieve the compensation target.

[0012] Preferably, the pre-trained AI model is a long short-term memory network model, which performs time series analysis on historical load data to capture the dynamic characteristics of load changes.

[0013] Preferably, the specifications of the three-phase capacitor branches of the capacitor bank module include 4kvar, 16kvar, and 25kvar, and the specifications of the single-phase capacitor branches include 1kvar and 2kvar. The three-phase and single-phase capacitor branches of different specifications can be combined through the configurable switching ports of no less than 80 channels, and the capacity can be expanded by connecting multiple channels of the same specifications in parallel, providing a continuously adjustable compensation capacity from 4kvar upwards to adapt to the reactive power compensation requirements of transformer loads from 250KVA to 3000KVA.

[0014] Preferably, the high-speed ADC module has a resolution of 16 bits. After converting the low-voltage safety analog signal into a digital signal, the high-speed ADC module transmits it to the FPGA high-speed processing unit for the FPGA high-speed processing unit to calculate real-time reactive power status data, load type information, and load capacity information.

[0015] The intelligent automatic reactive power compensation control method, applying the aforementioned intelligent automatic reactive power compensation control device, includes the following steps: The sampling and signal conditioning module acquires voltage analog signals, current analog signals, current waveform signals and voltage-current phase difference signals from the power grid in real time, as well as transformer load terminals, and isolates and filters the signals to output a conditioned low-voltage safety analog signal. Low-voltage safety analog signals are converted into digital signals using a high-speed ADC module; The system receives digital signals through the FPGA high-speed processing unit, performs instantaneous reactive power calculation and harmonic analysis, and generates real-time reactive power status data, load type information and load capacity information of the system based on current waveform signals and voltage-current phase difference signals. The AI ​​decision unit receives real-time reactive power status data, load type information, and load capacity information, calls pre-stored capacitor switching combination strategies, and outputs matrix capacitor switching instructions through a pre-trained AI model. The control drive module generates multiple drive signals based on the matrix capacitor switching command. The multi-channel capacitor switcher controls the dynamic switching combination of multiple capacitor branches, including at least three different capacity specifications, through at least 80 configurable switching ports based on multiple drive signals to form a matrix switching structure, thereby achieving instantaneous and accurate compensation of reactive power capacity.

[0016] Preferably, the FPGA high-speed processing unit also performs trend extrapolation on the load characteristic information to generate load change prediction data; The AI ​​decision-making unit combines load change prediction data to output matrix capacitor switching commands to achieve precise compensation.

[0017] Preferably, the pre-stored capacitor switching combination strategy includes a harmonic-sensitive strategy; When the FPGA high-speed processing unit analyzes and finds that the harmonic distortion rate exceeds a preset threshold, it triggers the AI ​​decision unit to call the harmonic sensitive strategy to generate a switching command that can suppress the resonance of the dominant harmonic frequency.

[0018] The present invention has at least the following beneficial effects: First, this invention, through FPGA high-speed processing, AI intelligent decision-making, and matrix-type multi-capacity capacitor banks, breaks through the compensation blind zone of traditional devices under low active power conditions, realizes fine-grained stepless adjustment of compensation capacity over a wide load range, can quickly respond to load changes, and dynamically stabilizes the power factor of the power grid at a high level of 0.99 but less than 1, effectively improving power supply quality, reducing power waste, and fully meeting the energy-saving and power quality requirements of modern power grids.

[0019] Secondly, by setting up a backup control path based on classical control algorithms and an automatic switching strategy, when the confidence level of the AI ​​decision unit's output instruction is insufficient or a fault occurs, the FPGA can automatically switch to PID or fuzzy control logic, seamlessly degrading to provide stable compensation. This not only leverages the intelligent advantages of AI, but also ensures uninterrupted system operation through the backup path, greatly enhancing the applicability and long-term operational reliability in harsh industrial environments.

[0020] Third, this invention integrates external performance monitoring and internal hardware diagnostics. On the one hand, it monitors harmonic distortion rate and power factor in real time, and automatically alarms and uploads fault information when they fail to meet the standards. On the other hand, it can accurately identify capacitor branch faults, dynamically adjust the switching strategy through AI to bypass faulty branches, and use healthy branches to reorganize to achieve the compensation target, reduce manual operation and maintenance intervention, and ensure the long-term stability of the system.

[0021] Fourth, this invention uses a long short-term memory network model to analyze historical load data, accurately capture the periodic and trend-based changes in load, and combine it with the load trend extrapolation capability of FPGA. AI can generate forward-looking switching instructions to achieve accurate compensation. When faced with rapid load changes or periodic fluctuations, the compensation action is smoother, the power factor fluctuation is smaller, and the dynamic response speed and stability are significantly improved.

[0022] Fifth, the present invention has a preset harmonic-sensitive switching strategy. When the FPGA detects that the harmonic distortion rate exceeds the standard, the AI ​​will prioritize calling the corresponding strategy to avoid capacitor combinations that may cause resonance of the dominant harmonic frequency. While improving the power factor, it actively suppresses harmonic risks, achieves comprehensive optimization of power quality, and enhances the safety of power grid operation.

[0023] Sixth, this invention enables real-time data interaction and remote control with external monitoring systems through Ethernet or CAN bus communication interfaces. This facilitates remote acquisition of operational data and issuance of control commands by maintenance personnel, while also providing support for centralized monitoring, big data analysis, and global energy optimization, thereby improving the level of intelligent equipment management and operational efficiency.

[0024] Seventh, this invention, as a universal core control device for reactive power compensation control cabinets, achieves excellent scalability and adaptability by setting no fewer than 80 configurable switching ports. For compensation cabinets with small-capacity transformers (such as 500KVA), only some ports need to be activated to control a small number of capacitor banks; for compensation cabinets with large-capacity transformers (such as 3000KVA), all ports can be fully utilized to drive more capacitor banks. A single core device can cover the compensation needs of a wide range of loads, greatly improving product versatility and reducing the R&D, production costs, and inventory management difficulties of multi-specification compensation cabinets.

[0025] Eighth, this invention adds a single-phase capacitor bank to the capacitor bank module, sharing no fewer than 80 configurable switching ports with the three-phase capacitor branch, which can specifically solve the problem of three-phase power factor imbalance. When the AI ​​decision unit identifies an imbalance in the three-phase phase difference, it can accurately switch the single-phase compensation capacitor through matrix coding without the need for additional port resources. This function is suitable for three-phase load imbalance scenarios such as factories, substations, and residential areas, such as compensating for single-phase inductive loads like two-phase welding machines, further broadening the application scope of the device and improving the overall optimization capability of power supply quality.

[0026] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the structure of one technical solution of the present invention. Detailed Implementation

[0028] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0029] It should be understood that terms such as "having," "comprising," and "including" as used herein do not exclude the presence or addition of one or more other elements or combinations thereof. It should be noted that the experimental methods described in the following embodiments, unless otherwise specified, are conventional methods and are commercially available, and therefore should not be construed as limiting the invention.

[0030] Traditional automatic reactive power compensation devices employ a fixed-capacity capacitor bank + single-core serial processing architecture. This architecture suffers from problems such as compensation capacity mismatch (overcompensation or blocking) under low active power conditions, delayed compensation command response, and lack of intelligent adaptability in switching strategies. Consequently, compensation accuracy is low, power factor is difficult to achieve consistently, and the device cannot meet compensation requirements across a wide load range. Figure 1 As shown, the present invention provides an intelligent automatic reactive power compensation control device, comprising: The power input module is used to connect to AC 220V / 380V power supply. It is equipped with an internal isolation transformer and provides working power to all power modules and units. The sampling and signal conditioning module, connected to the power input module, performs raw signal acquisition and preprocessing through voltage transformers (PT) and current transformers (CT). Specifically, it is used to acquire in real time the voltage analog signal, current analog signal, current waveform signal, and voltage-current phase difference signal at the transformer load end of the power grid. The module is equipped with an opto-isolator to achieve signal isolation and is configured with a finite impulse response (FIR) low-pass filter circuit to isolate and filter the signal to eliminate signal interference and output a conditioned low-voltage safety analog signal (0~5V). The high-speed ADC module, which is connected to the sampling and signal conditioning module and the power input module, has multi-channel synchronous sampling capability and high conversion rate. It converts low-voltage safety analog signals into digital signals through a sample-and-hold circuit and an analog-to-digital converter. The digital signals represent a discrete digital sequence of the original power grid waveform. The Field Programmable Gate Array (FPGA) high-speed processing unit is connected to the high-speed ADC module and the power input module. It receives digital signals via a high-speed parallel bus and performs instantaneous reactive power calculation (based on acquired voltage and current signals) and harmonic analysis (identifying major harmonics) using built-in algorithms. It calculates instantaneous active power, instantaneous reactive power, apparent power, power factor, and the content of each harmonic in real time. Based on current waveform signals (e.g., whether the waveform is sinusoidal and the degree of distortion to understand the nonlinear characteristics of the load) and voltage-current phase difference signals (a positive phase difference indicates an inductive load, a negative phase difference indicates a capacitive load), it calculates the type and capacity of the load carried by the transformer, generating real-time reactive power status data, load type information, and load capacity information. Specifically, it executes the following in sequence: ① Signal Input ① The FPGA receives the digital signal converted by the high-speed ADC module. This digital signal is obtained by the sampling module through isolation filtering and 16-bit ADC analog-to-digital conversion to ensure that the original signal is interference-free and highly accurate; ② Parallel operation: The reactive power calculation unit calculates reactive power synchronously through instantaneous voltage and current values ​​based on the pq instantaneous reactive power theory, distinguishing between inductive and capacitive reactive power types. The harmonic analysis unit decomposes the current signal through the FFT algorithm to identify the dominant harmonic frequency and distortion rate. The load identification unit calculates the load capacity based on the current waveform characteristics (sine wave, distortion degree) and voltage-current phase difference (positive / negative values ​​determine inductive / capacitive), combined with the effective values ​​of voltage and current; ③ Data output: The calculation results of each unit are integrated into a data packet containing real-time reactive power + load type + load capacity, which is transmitted to the AI ​​decision unit through a high-speed bus. The data storage module, which is connected to the power input module, is used to store a variety of predefined capacitor switching combination strategies. These strategies can be preset by engineers based on typical load scenarios or can be learned and optimized by the system during long-term operation. The AI ​​decision-making unit, connected to the FPGA high-speed processing unit, the data storage module, and the power input module, has learned a large number of mapping relationships between grid states and optimal compensation actions. During decision-making, it receives real-time reactive power status data, load type information, and load capacity information, and calls the switching combination strategy in the data storage module. Combining the load type information (e.g., inductive / resistive / capacitive) and load capacity information, it calls a baseline strategy set matching the current load type and capacity range to calculate the required capacitor capacity specifications. This calculation is performed using a pre-trained AI model, outputting a matrix-style capacitor switching command (specifying the capacitor branch number and quantity to be switched). Specifically, it executes the following sequentially: ① Data interaction: The AI ​​decision-making unit receives real-time reactive power, load type, and capacity information output by the FPGA, and simultaneously... The SPI interface calls the pre-stored multi-specification switching combination strategy (covering different operating conditions of low / medium / high load) in the data storage module; ② Decision logic: The AI ​​model aims to accurately match the compensation capacity with the reactive power demand of the load. It determines the compensation direction based on the load type (inductive / capacitive), calculates the total reactive power capacity to be put into operation based on the load capacity, and matches the optimal branch combination method from the pre-stored strategy. The AI ​​decision unit is also configured to: identify the three-phase power factor imbalance state based on the three-phase power factor data provided by the FPGA high-speed processing unit. When an imbalance is determined, it generates a single-phase compensation switching instruction for a specific phase, specifying the single-phase capacitor branch to be switched and the number; ③ Instruction output: Generate a matrix switching instruction including capacitor branch number + switching quantity, specifying the switching status of each capacitor. The instruction is transmitted to the control drive module. The control drive module is connected to the AI ​​decision unit, the FPGA high-speed processing unit, and the power input module. The control drive module has a built-in instruction priority determination unit. The AI ​​instruction has a higher priority than the FPGA backup instruction (the FPGA has built-in basic compensation logic, which directly calculates the minimum adaptive compensation capacity based on the real-time reactive power value and generates a backup switching instruction). The control drive module prioritizes receiving the matrix capacitor switching instruction issued by the AI ​​decision unit. When the AI ​​decision unit is abnormal (such as communication interruption or illegal output value), it receives the backup switching instruction issued by the FPGA high-speed processing unit, and then generates the corresponding multi-channel drive signal (high level is the switching signal, low level is the disconnect signal) to directly drive the semiconductor switching device, and transmits it to the multi-channel capacitor switcher to ensure that the compensation operation is not interrupted. A multi-channel capacitor switch, connected to the control drive module and the power input module, includes multiple sets of solid-state relays or thyristors. Each channel is controlled by a drive signal output from the control drive module. Upon receiving the drive signal, it controls the on / off state of the corresponding branch of the capacitor bank module. The capacitor bank module, connected to the multi-channel capacitor switch and the power input module, includes multiple independently controlled three-phase capacitor branches. The specifications of the three-phase capacitor branches include, for example, 4kvar, 16kvar, and 25kvar, while the specifications of the single-phase capacitor branches include, for example, 1kvar and 2kvar. This device, as the control core, has no fewer than 80 configurable switching ports. Each branch is connected in series with a fuse and a status sensor, and is connected to the corresponding configurable switching port through the multi-channel capacitor switch to achieve independent on / off control. The switching combination of different capacitor branches is dynamically selected and controlled from these ports based on drive signals (e.g., only small-capacity branches are switched on under low load, and under medium load...). By combining small-capacity branches for medium-load switching and large-capacity branches for heavy-load switching, a flexible matrix switching structure can be formed. Through the configurable switching ports of no less than 80 channels, three-phase and single-phase capacitor branches of different specifications can be combined and expanded in number. It can provide continuously adjustable compensation capacity from the smallest specification (such as 4kvar or 1kvar) upwards. This design allows a single device to adapt to compensation cabinets of different sizes: for a compensation cabinet for a 500KVA transformer, it may only be necessary to configure and enable a few ports to control a capacitor bank with a total capacity of about 300kvar; for a compensation cabinet for a 3000KVA transformer, more capacitors can be configured and all ports can be enabled for control, thereby completely solving the adaptability problem of traditional fixed-port devices under different load scales. The capacitor bank module also includes at least one single-phase capacitor branch for reactive power compensation of a specific phase to address the three-phase power factor imbalance problem. The capacity of the single-phase capacitor branch (e.g., 1kvar, 2kvar, etc.) can be flexibly configured according to the target application scenario (e.g., the capacity of a common two-phase welding machine) and is connected to a pool of at least 80 configurable switching ports along with the three-phase capacitor branch. When designing the compensation cabinet, users can allocate ports as needed to connect three-phase or single-phase capacitors without requiring any modifications to the core control device.

[0031] To enable data interaction and remote control capabilities with external monitoring systems, a communication interface module is preferably included to enhance the device's interconnectivity and remote management capabilities. This communication interface module is connected to the FPGA high-speed processing unit and the AI ​​decision-making unit, and supports Ethernet or CAN bus protocols. The Ethernet protocol refers to industrial Ethernet conforming to the IEEE 802.3 standard, supporting the TCP / IP protocol stack, for data interaction and remote control with external monitoring systems. Regarding data interaction, the communication interface module receives system operating data (such as real-time reactive power, harmonic distortion rate, and power factor) output by the FPGA high-speed processing unit and switching command records output by the AI ​​decision-making unit in real time. It encapsulates the data into standard TCP / IP data packets (Ethernet) or CAN messages (CAN bus) and transmits them to external monitoring systems (such as power grid dispatch centers or factory power management systems). Simultaneously, it receives control commands (such as adjusting the power factor target value or modifying the switching strategy) issued by the external monitoring system, parses them, and transmits them to the FPGA high-speed processing unit or the AI ​​decision-making unit.

[0032] In the above technical solution, by combining multi-channel capacitor switchers with intelligent dynamic control technology, and relying on the hardware expansion capabilities provided by no less than 80 configurable switching ports, matrix-style precise switching control of capacitor banks is achieved. This design enables the device to flexibly configure capacitor scale and port usage strategies according to the target application (from 250KVA to 3000KVA transformers). Based on real-time dynamically acquired reactive power data, the corresponding capacitor banks can be flexibly switched to achieve instantaneous and precise compensation of reactive capacity. At the same time, the required capacitor capacity specifications are calculated and matched based on the type and capacity characteristics of the inductive load driven by the transformer, and multiple switching combination strategies can be predefined to ensure the adaptability and flexibility of compensation under different load conditions. In addition, the FPGA high-speed parallel processing architecture significantly improves the signal processing and decision response speed, and the main control-backup dual control link ensures the continuity of compensation. The whole system can operate stably under a wide load range (especially under low active power conditions), improving the power factor and stabilizing it to above 0.99, effectively solving the problems of compensation blind spots, response lag, insufficient accuracy, rigid hardware configuration, and poor adaptability of traditional devices.

[0033] When the AI ​​decision-making unit, as the core control unit, generates erroneous instructions due to model uncertainty, data anomalies, or hardware failures, or completely loses its decision-making ability, the entire compensation system will fall into a state of loss of control or paralysis, losing its basic compensation function, and the system reliability cannot be guaranteed. In another technical solution, the FPGA high-speed processing unit is further configured to: automatically switch to the built-in backup control logic when the confidence level of the output instruction of the AI ​​decision-making unit is lower than a preset threshold or when the AI ​​decision-making unit malfunctions, and directly issue a backup switching instruction to the control drive module; the backup control logic is a PID control algorithm or fuzzy control logic. Specifically, during system operation, the FPGA high-speed processing unit continuously evaluates the output status of the AI ​​decision-making unit, firstly by evaluating the confidence level of the output instruction. The AI ​​model usually outputs a probability value or confidence score for each decision. If the score is lower than a preset threshold, it indicates that the AI ​​model lacks confidence in the current decision. Secondly, it directly... The system detects faults such as loss of communication heartbeat with the AI ​​decision-making unit, incorrect format of received command data, or data exceeding reasonable limits. Once any of the above abnormal conditions are met, the state machine inside the FPGA high-speed processing unit will automatically trigger a switch. The switch process is disturbance-free, that is, the FPGA high-speed processing unit immediately stops forwarding AI commands and instead activates its built-in backup control logic. This backup logic takes the real-time reactive power status data (such as reactive power deviation) calculated by the FPGA high-speed processing unit itself as input, and directly generates a set of conservative but robust backup switching commands through PID calculation or fuzzy inference. These commands are then directly sent to the control drive module to drive capacitor switching.

[0034] In one example, the FPGA high-speed processing unit extracts the confidence score (e.g., probability output based on the softmax function) from the decision results output by the AI ​​model. A preset threshold of 0.85 is used. When the confidence score of a switching instruction (e.g., switching on 3 25kvar branches) is only 0.72, it is judged as a low-confidence instruction. The FPGA high-speed processing unit performs hard validation on the instruction format and numerical range. If the AI ​​output switches on 10 4kvar branches (but the system only has 4 4kvar branches configured), or if the instruction contains illegal values ​​such as switching on -1 capacitor, an anomaly is directly triggered, for example, in a car... When the load changes abruptly, the AI ​​outputs instructions for two 16kvar and five 25kvar lines due to data noise. The hard verification module of the FPGA high-speed processing unit first determines that the five 25kvar lines in the instruction exceed the upper limit of the system's actual configured eight 25kvar branch resources, and directly judges it as an instruction abnormality. At the same time, the confidence level of the instruction is only 0.68, which is lower than the threshold. Under any one or both abnormal conditions, the FPGA immediately starts the backup logic. Based on the real-time 60kvar reactive power demand, it generates a conservative instruction to output two 25kvar lines and one 16kvar line through the PID algorithm to ensure a smooth switching.

[0035] In the above technical solution, through intelligent fault detection and switching, when the advanced AI control fails due to a fault, the system can automatically and seamlessly degrade to the mature and reliable traditional control mode, ensuring the continuous provision of basic reactive power compensation functions, which greatly enhances the applicability and long-term operational stability of the entire device in complex industrial environments.

[0036] During device operation, it is difficult to monitor and report in a timely and automatic manner whether critical internal components (such as capacitor branches) malfunction. Another technical solution also includes: The fault detection and alarm module is connected to the FPGA high-speed processing unit and the power input module. The fault detection and alarm module is configured to: monitor the harmonic distortion rate and power factor calculated by the FPGA high-speed processing unit in real time, and when the harmonic distortion rate is greater than 5% or the power factor is less than 0.99, it indicates that the current compensation effect has not achieved the expected target, output an alarm signal and send fault information including fault code, timestamp and specific data to the external monitoring system.

[0037] In the above technical solution, the fault detection and alarm module monitors the harmonic distortion rate and power factor in real time, ensuring that an alarm is triggered and fault information is uploaded in a timely manner when the compensation effect fails to meet the standard, so that the problem can be detected.

[0038] During device operation, it is difficult to monitor and report in a timely and automatic manner whether the compensation effect (such as harmonic distortion rate and power factor) meets the standards. Another technical solution also includes: The fault diagnosis and self-healing module is connected to the FPGA high-speed processing unit, the control drive module, and the power input module. The module is configured to monitor the switching status (by monitoring whether the voltage or current across each capacitor branch changes as expected to determine successful switching) and health status (by monitoring the effective value of the current flowing through the capacitor, harmonic content, or temperature to determine health status). When a specific branch fault is detected (e.g., a command to switch on but no current is detected, or an abnormal current waveform indicating internal component damage), this information is fed back to the AI ​​decision unit via the FPGA high-speed processing unit. The AI ​​decision unit dynamically adjusts subsequent switching combination strategies to bypass the faulty branch, recombining it with the remaining healthy branches to find a new combination scheme that best approximates the target compensation capacity, thereby achieving the compensation target.

[0039] In one instance, in a typical configuration adapted to a 500KVA transformer, only 16 of the 80 configurable switching ports may be enabled. Suppose that the capacitor in the 5th port (16kvar specification) fails. When the target compensation capacity is 48kvar, the original optimal strategy is to activate 3 16kvar branches. After the fault diagnosis and self-healing module identifies the fault, the AI ​​decision unit will dynamically reconstruct the strategy by utilizing the healthy capacitor branches connected to the remaining large number of available spare ports. For example, it may call an alternative combination of 2 16kvar and 2 4kvar to complete the compensation. Even in small-scale systems with low port utilization, the ample port resources provide flexible space for fault avoidance. In fully or high-configuration scenarios adapted to 3000KVA transformers, 80 configurable switching ports may be used extensively. At this time, the scale advantage of the port and capacitor branch library is more significant. Even if multiple branch failures occur, the system can still recalculate and execute a new combination strategy that is closest to the target compensation capacity based on the huge healthy branch resource pool, keeping the compensation error within the allowable range without immediate shutdown. This fully demonstrates the strong redundancy, fault tolerance and self-healing capabilities of the system designed with sufficient configurable ports.

[0040] In practical applications, this device can achieve precise compensation for three-phase power factor imbalance by using single-phase capacitor banks. For example, in a factory, a 20 kVA two-phase welding machine load has a phase voltage of 220V and a power factor of only 0.6, which needs to be increased to 0.99. Only a small number of single-phase capacitors of suitable capacity need to be configured in the compensation cabinet. The device's at least 80 configurable switching ports share resources with the three-phase capacitor branches. After the AI ​​decision unit identifies the low power factor and three-phase imbalance of a phase, it automatically generates a single-phase compensation command, controlling the multi-channel capacitor switch to selectively switch single-phase capacitors. This ensures that the power factor of each phase is stabilized above 0.99 without manual intervention. The configuration of this single-phase compensation function is entirely determined by the user during the manufacturing of the compensation cabinet according to actual needs. The core control device requires no modification, fully demonstrating the flexibility of the 80 configurable switching ports.

[0041] In the above technical solution, the fault diagnosis and self-healing module monitors the status of the capacitor branch, identifies faults and provides feedback, dynamically adjusts the switching strategy to bypass the faulty branch, and uses the remaining healthy branch to reorganize and compensate, so that the device has fault tolerance capability and autonomous maintenance potential, ensuring the long-term stable and reliable operation of the system.

[0042] Pre-trained AI models using static or transient analysis models struggle to capture periodic and trend-based load changes, leading to short-sighted switching strategies that cannot adapt to future operating conditions and exhibit reduced compensation effectiveness during dynamic load fluctuations. In another technical solution, the pre-trained AI model is a Long Short-Term Memory (LSTM) network, which possesses a strong ability to model long-term dependencies in time-series data. It effectively learns and memorizes historical patterns and time dependencies in load power changes. During model training, the input data is a continuous sequence of historical load data, including time-series values ​​of historical active power, reactive power, and load capacity. The training framework uses TensorFlow, with the dataset divided into training and validation sets in an 8:2 ratio. The loss function is mean squared error (MSE), the optimizer is Adam, the learning rate is 0.001, and the training run is 200 epochs. The validation set error is continuously 1. If the load does not decrease in round 0, the process stops. During the deployment and operation phase, when the FPGA high-speed processing unit sends real-time data, the LSTM in the AI ​​decision unit performs time series analysis on the historical load data. This involves combining the new data with recent historical data to create a new time series, capturing the dynamic characteristics of load changes. After receiving the real-time load data transmitted by the FPGA high-speed processing unit, the AI ​​decision unit combines it with the load change prediction results output by the LSTM model. When it is observed that the load is slowly increasing, the LSTM may add slightly more capacitors in advance. When it is predicted that the load is about to drop sharply, the LSTM may issue a command to cut off some capacitors in advance, rather than waiting until the reactive power has been compensated before taking action.

[0043] In the above technical solution, the LSTM model is used as a pre-trained AI model, which can effectively analyze historical load data and capture the periodic and trend-based dynamic changes of the load. Compared with static or instantaneous models, the AI ​​decision of the LSTM model is not only based on the current state, but also can more accurately predict load changes, improve the long-term compensation effect under periodic load changes or slow drift conditions, and make the control strategy more forward-looking and adaptable.

[0044] In another technical solution, the capacitor bank module can employ three-phase self-healing parallel power capacitors and single-phase power capacitors. The specifications of the three-phase capacitor branches of the capacitor bank module include 4kvar, 16kvar, and 25kvar, and the specifications of the single-phase capacitor branches include 1kvar and 2kvar. Different specifications of three-phase and single-phase capacitor branches can be combined through the configurable switching ports of no less than 80 channels. Capacity expansion can be achieved by connecting multiple branches of the same specification in parallel, providing continuously adjustable compensation capacity from 4kvar upwards to adapt to the reactive power compensation requirements of transformer loads from 250KVA to 3000KVA. Users can determine the number of activated ports and the total number of capacitor branches of each specification connected according to the specific reactive power compensation requirements of the target transformer (250KVA to 3000KVA) without replacing the core control device, thereby achieving high configuration flexibility and application adaptability.

[0045] In another technical solution, the high-speed ADC module has a resolution of 16 bits. It samples the low-voltage safety analog signal at a rate of no less than 10kHz. Each sample yields a digital value that can extremely accurately reproduce the instantaneous amplitude of the original analog signal. Whether it's the fundamental amplitude of the grid voltage or the tiny harmonic components superimposed on it, both can be captured and quantified more accurately. After converting the low-voltage safety analog signal into a digital signal, the high-speed ADC module transmits it to the FPGA high-speed processing unit, which then calculates real-time reactive power status data, load type information, and load capacity information. Using 16-bit resolution ensures information quality from the source of the data link, providing a high-fidelity data foundation for subsequent power calculations, harmonic analysis, and load feature extraction performed by the FPGA, thus guaranteeing overall compensation accuracy.

[0046] The intelligent automatic reactive power compensation control method, applying the aforementioned intelligent automatic reactive power compensation control device, includes the following steps: The data acquisition steps are completed through the sampling and signal conditioning module: the PT and CT are used to acquire the voltage analog signal, current analog signal and current waveform signal of the power grid in real time, respectively, and the current waveform signal of the transformer load end. The voltage-current phase difference signal is acquired synchronously through the phase detection circuit. The module is equipped with an optical isolator to achieve physical isolation between the high voltage signal and the low voltage circuit. The FIR low-pass filter circuit is configured to filter out high frequency interference in the signal. After the above processing, the conditioned 0~5V low voltage safety analog signal is output. The analog-to-digital conversion process is completed through a high-speed ADC module: After receiving the low-voltage safety analog signal output from the sampling and signal conditioning module, the high-speed ADC module stabilizes the signal through an internal sample-and-hold circuit, and then converts it into a digital signal through an analog-to-digital converter. This digital signal is a discrete digital sequence representing the original power grid waveform, which completely preserves the characteristic information of the original signal. The calculation and feature extraction steps are completed through the FPGA high-speed processing unit: The FPGA high-speed processing unit receives digital signals, performs instantaneous reactive power calculation and harmonic analysis, and generates real-time reactive power status data, load type information, and load capacity information of the system based on current waveform signals and voltage-current phase difference signals. Specifically, the following operations are performed: ① Signal input: The FPGA receives digital signals transmitted by the high-speed ADC module through a high-speed parallel bus to ensure real-time data transmission; ② Parallel operation: The built-in reactive power calculation unit, based on the pq instantaneous reactive power theory, synchronously calculates instantaneous active power, instantaneous reactive power, and apparent reactive power using instantaneous voltage and current values. Power and power factor are used to distinguish between inductive and capacitive load types based on the positive and negative voltage-current phase difference; the harmonic analysis unit decomposes the current signal using the FFT algorithm to identify the main harmonic frequencies and the content of each harmonic in the power grid; the load identification unit calculates the capacity of the load carried by the transformer based on the sinusoidal nature and distortion degree of the current waveform (reflecting the nonlinear characteristics of the load) and the voltage-current phase difference, combined with the effective values ​​of voltage and current; ③ Data output: The above calculation results are integrated into a data packet including real-time reactive power status data, load type information and load capacity information, and transmitted to the AI ​​decision unit through a high-speed bus, with a single calculation and feature extraction cycle of no more than 10ms; The decision-making process is completed through an AI decision-making unit: receiving real-time reactive power status data, load type information, and load capacity information; calling pre-stored capacitor switching combination strategies; and outputting matrix capacitor switching instructions through a pre-trained AI model. Specifically, the following operations are performed: ① Data interaction: The AI ​​decision-making unit receives feature data packets transmitted by the FPGA high-speed processing unit and simultaneously calls various predefined capacitor switching combination strategies in the data storage module through the SPI interface. This strategy set can be preset by engineers based on typical load scenarios or obtained by the system through learning and optimization during long-term operation; ② Strategy matching: determining the compensation direction based on the received load type information, calculating the required total reactive power compensation capacity based on the load capacity information, and calling a benchmark strategy set that matches the current load type and capacity range from the pre-stored strategies; ③ Instruction generation: optimizing the benchmark strategy through a pre-trained AI model (which has learned a large number of mapping relationships between grid states and optimal compensation actions), and outputting matrix capacitor switching instructions. These instructions clearly indicate the number and quantity of capacitor branches to be put into or cut off, with a single decision cycle not exceeding 2ms. The drive generation steps are completed by controlling the drive module: Multiple drive signals are generated based on the matrix capacitor switching instructions, specifically performing the following operations: ① Instruction priority determination: The module has a built-in instruction priority determination unit, setting the priority of instructions output by the AI ​​decision unit to be higher than the backup instructions of the FPGA high-speed processing unit; ② Instruction reception and switching: The matrix capacitor switching instructions issued by the AI ​​decision unit are received first. When an abnormal situation such as communication interruption or illegal output value is detected in the AI ​​decision unit, the module automatically switches to receiving the backup switching instructions issued by the FPGA high-speed processing unit. These backup instructions are generated by the FPGA's built-in basic compensation logic based on the real-time reactive power value, directly calculating the minimum adaptive compensation capacity; ③ Drive signal generation: The received switching instructions are converted into multiple drive signals that can directly drive semiconductor switching devices. A high level represents the switching signal, and a low level represents the disconnect signal. The drive signals are then transmitted to the multiple capacitor switch. The execution and compensation steps are completed through a multi-channel capacitor switch: The multi-channel capacitor switch includes multiple sets of solid-state relays or thyristors, each controlled by a corresponding drive signal output by the control drive module. Upon receiving the signal, it controls the on / off state of each branch in the capacitor bank module. The capacitor bank module includes multiple independently controlled capacitor branches, each with at least three different capacity levels. Each branch is connected in series with a fuse and a status sensor. The multi-channel capacitor switch dynamically switches and combines the branches according to the drive signal. For example, under low load, only the small capacity branch is switched on; under medium load, the small and medium capacity branches are switched on in combination; and under heavy load, the medium and large capacity branches are switched on in combination, forming a matrix switching structure to achieve instantaneous and accurate compensation of reactive power capacity.

[0047] To achieve real-time monitoring and convenient operation and maintenance, the intelligent automatic reactive power compensation device, in addition to its multi-channel capacitor switching function, should also include a human-machine interface panel. The panel is arranged in the order of status display, fault alarm, and operation control. For example, the power supply is indicated by a red LED, which illuminates when the power supply is normal; voltage, current, and power factor are displayed in real-time via digital tubes, intuitively presenting the core parameters of the power grid and load; overvoltage and undervoltage are indicated by yellow LEDs, which illuminate to trigger an early warning; faults 1 to 3 are indicated by red LEDs, corresponding to open circuit, short circuit, and communication abnormality of the main control unit, respectively; and a high-temperature alarm is indicated by a red LED, suggesting abnormal capacitor bank temperature. The operation buttons include an alarm clearing button and a manual / automatic cycle button. The alarm clearing button clears non-fault alarms, while the manual / automatic cycle button switches control modes. An illuminated light indicates manual mode, allowing manual switching of branches; an off light indicates automatic mode, where the system makes autonomous decisions. The human-machine interface panel can simultaneously display the single-phase compensation status and related fault alarms. The panel functions are implemented using the existing FPGA data acquisition and fault diagnosis module, requiring no additional core hardware.

[0048] The above technical solution transforms advanced hardware capabilities into executable control logic, clarifies the intelligent processing steps of the entire chain from signal acquisition to capacitor switching, and provides clear methodological guidance for understanding and implementing the technical solution.

[0049] To reduce power factor fluctuations and enable compensation control to have a certain degree of predictability, in another technical solution, the FPGA high-speed processing unit also performs trend extrapolation on load characteristic information to generate load change prediction data. Specifically, based on the data of the current and past instants, it calculates the real-time type and capacity information of the load, and performs mathematical trend extrapolation analysis on this information (especially time-series data such as load capacity, active / reactive power, etc.). That is, through simple linear prediction and extrapolation using moving average filters, it predicts how the load may change in a very short period of time in the future, whether it will continue to rise, remain stable, or begin to decline, and describes the predicted direction and magnitude of change. When the AI ​​decision-making unit makes decisions, the input information received includes current real-time status data and prediction data. The AI ​​decision-making unit combines the load change prediction data to output matrix capacitor switching instructions to achieve more accurate and dynamic compensation with rapid response capabilities. For example, if the prediction data indicates that the load is about to increase rapidly, in the current decision, in addition to compensating for the current reactive power deficit, a small additional capacitor is added to prepare for the upcoming increase in reactive power demand. Conversely, if the predicted load is about to decrease, the amount of capacitor added is slightly reduced at the current time to avoid subsequent overcompensation.

[0050] In one example, with a 50ms time window, the most recent 10 sets of load capacity data (e.g., 52kvar, 55kvar, 58kvar, and 61kvar respectively) are cached. After removing impulse interference by moving average, a linear trend with a slope of 3kvar / 10ms is fitted based on the least squares method. It is predicted that the load capacity will rise to 67kvar within the next 20ms. The AI ​​decision unit combines the current reactive power demand of 58kvar with the predicted data and directly outputs a predictive instruction to put in 2 25kvar + 1 16kvar + 1 4kvar (total capacity 69kvar). When the load rises to 67kvar after 20ms, the compensation capacity has been matched in advance, and the power factor fluctuation has decreased from ±0.03 to ±0.005. If the predicted load drops suddenly from 80kvar to 40kvar (e.g., motor shutdown), the 2 25kvar branches are disconnected 15ms in advance to avoid overcompensation causing the power factor to drop below 0.95.

[0051] In the above technical solution, traditional feedback control is combined with predictive feedforward control. The compensation action leads the actual load change to a certain extent, thereby effectively offsetting the total delay caused by the inherent sampling, calculation and execution links of the system. This allows the power factor to maintain a smaller fluctuation amplitude and a faster stabilization speed when the load changes abruptly, achieving a smoother and more stable dynamic compensation effect and improving the overall performance of the system.

[0052] Switching reactive power compensation capacitors can alter the impedance characteristics of the power grid. If their capacitive reactance resonates with inductive components in the grid at a specific harmonic frequency, it will significantly amplify that harmonic, causing harmonic pollution and even damaging other electrical equipment. In another technical solution, pre-stored capacitor switching strategies include harmonic-sensitive strategies. This involves incorporating harmonic safety as a constraint or optimization objective into the reactive power compensation decision-making process. Under the premise of meeting basic reactive power compensation requirements, priority is given to capacitor combinations that will not resonate in parallel with the grid background impedance at the main harmonic frequencies after connection. Furthermore, it may even prioritize capacitor branch combinations that have a certain filtering effect on specific harmonics. The FPGA high-speed processing unit performs spectral decomposition on the acquired current signal using the FFT algorithm, calculates the total harmonic distortion rate (THD), accurately separates the fundamental component from each harmonic component, locates the dominant harmonic frequency with the highest content, and determines whether the harmonic is an inherent background harmonic of the power grid. When the FPGA high-speed processing unit analyzes and finds that the harmonic distortion rate exceeds a preset threshold, it sends a trigger signal containing the harmonic distortion rate value plus the dominant harmonic frequency to the AI ​​decision unit. This triggers the AI ​​decision unit to invoke a harmonic-sensitive strategy, optimize the theoretical compensation capacity based on the current reactive power demand, and generate a switching command that can suppress the resonance of the dominant harmonic frequency. In other words, the capacitor combination ultimately selected for switching will not form a dangerous resonance point between its capacitive reactance and the system inductive reactance at the main harmonic frequency, thereby avoiding the amplification of harmonics due to compensation.

[0053] In one example, the equivalent inductance of the factory power grid is L = 0.01mH. When the FPGA detects a total harmonic distortion (THD) of 6.2% and a dominant harmonic of the 5th order (250Hz), the resonance formula is used. By reverse deduction, we arrive at the dangerous capacitor capacitance C = 1 / (4π) 2 f 2 L)=1 / (4×9.87×62500×0.01×10 -3=40.5μF (corresponding to a capacity of approximately 16kvar). When meeting the current reactive power demand of 50kvar, the AI ​​decision-making unit will actively avoid the combination of one 16kvar circuit and instead select two 25kvar circuits (total capacity 50kvar, corresponding to a 20μF capacitor). The resonant frequency of this combination is 356Hz, which deviates significantly from the 5th harmonic frequency, effectively avoiding resonance. If there is a 16kvar filter branch with a 6% series reactor, its tuning frequency matches the 5th harmonic. The AI ​​will prioritize selecting one 16kvar filter branch + one 25kvar conventional branch + two 4kvar circuits, compensating for the 5th harmonic current while absorbing some of the 5th harmonic current.

[0054] In the above technical solution, when harmonic exceedance is detected, priority is given to selecting capacitor combinations that are less likely to cause resonance or have a suppressive effect on specific harmonics. This allows the device to actively avoid negative impacts on the power grid harmonic environment while pursuing a high power factor, thus achieving comprehensive power quality management and improving application safety.

[0055] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0056] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. An intelligent automatic reactive power compensation control device, characterized in that, include: The power input module is used to connect to an AC power source and provide operating power. The sampling and signal conditioning module is connected to the power input module and is used to collect voltage analog signals, current analog signals, current waveform signals and voltage-current phase difference signals at the transformer load end in real time from the power grid, and to isolate and filter the signals to output a conditioned low-voltage safety analog signal. A high-speed ADC module, which is connected to the sampling and signal conditioning module and the power input module, is used to convert low-voltage safety analog signals into digital signals; The FPGA high-speed processing unit is connected to the high-speed ADC module and the power input module. It is used to receive digital signals, perform instantaneous reactive power calculation and harmonic analysis, and calculate the type and capacity of the load carried by the transformer based on the current waveform signal and the voltage-current phase difference signal, and generate real-time reactive power status data, load type information and load capacity information. A data storage module, which is connected to the power input module, is used to store a variety of predefined capacitor switching combination strategies; The AI ​​decision unit is connected to the FPGA high-speed processing unit, the data storage module, and the power input module, respectively. It receives real-time reactive power status data, load type information, and load capacity information. It then calls the switching combination strategy in the data storage module, calculates the required capacitor capacity based on the load type and capacity information, performs the calculation using a pre-trained AI model, and outputs a matrix-style capacitor switching command. The AI ​​decision unit is also configured to: identify three-phase power factor imbalance based on the three-phase power factor data provided by the FPGA high-speed processing unit; and generate a single-phase compensation switching command for a specific phase when an imbalance is detected. The control drive module is connected to the AI ​​decision unit, the FPGA high-speed processing unit and the power input module. It is used to prioritize receiving the matrix capacitor switching command issued by the AI ​​decision unit, and to receive the backup switching command issued by the FPGA high-speed processing unit when the AI ​​decision unit is abnormal, thereby generating the corresponding multi-channel drive signal. A multi-channel capacitor switch, which is connected to the control drive module and the power input module; and The capacitor bank module is connected to the multi-channel capacitor switch and the power input module. The capacitor bank module includes multiple independently controlled three-phase capacitor branches and at least one single-phase capacitor branch. The capacity specifications of each three-phase branch include at least three different levels. The intelligent automatic reactive power compensation control device is provided with no less than 80 configurable switching ports. The multi-channel capacitor switch, through the configurable switching ports, controls the three-phase and single-phase multi-channel capacitor branches to dynamically switch and combine them according to the multi-channel drive signals, forming a matrix switching structure to achieve instantaneous and accurate compensation of reactive power capacity and balance of three-phase power factor.

2. The intelligent automatic reactive power compensation control device as described in claim 1, characterized in that, The FPGA high-speed processing unit is also configured to automatically switch to the built-in backup control logic when the confidence level of the output instruction of the AI ​​decision unit is lower than a preset threshold or when the AI ​​decision unit malfunctions, and directly issue a backup switching instruction to the control drive module; the backup control logic is a PID control algorithm or fuzzy control logic.

3. The intelligent automatic reactive power compensation control device as described in claim 1, characterized in that, Also includes: The fault detection and alarm module is connected to the FPGA high-speed processing unit and the power input module. The fault detection and alarm module is configured to: monitor the system harmonic distortion rate and power factor in real time, and output an alarm signal and send fault information to the external monitoring system when the harmonic distortion rate is greater than 5% or the power factor is lower than 0.

99.

4. The intelligent automatic reactive power compensation control device as described in claim 3, characterized in that, Also includes: The fault diagnosis and self-healing module is connected to the FPGA high-speed processing unit, the control drive module, and the power input module. The fault diagnosis and self-healing module is configured to monitor the switching status and health status of each branch in the capacitor bank module. When a fault is detected in a specific branch, the information is fed back to the AI ​​decision unit. The AI ​​decision unit dynamically adjusts the subsequent switching combination strategy to bypass the faulty branch and recombines the remaining healthy branches to achieve the compensation target.

5. The intelligent automatic reactive power compensation control device as described in claim 1, characterized in that, The pre-trained AI model is a long short-term memory network model that performs time series analysis on historical load data to capture the dynamic characteristics of load changes.

6. The intelligent automatic reactive power compensation control device as described in claim 1, characterized in that, The specifications of the three-phase capacitor branches of the capacitor bank module include 4kvar, 16kvar, and 25kvar, and the specifications of the single-phase capacitor branches include 1kvar and 2kvar. The three-phase and single-phase capacitor branches of different specifications can be combined through the no less than 80 configurable switching ports, and the capacity can be expanded by connecting multiple branches of the same specification in parallel. It provides a continuously adjustable compensation capacity starting from 4kvar to adapt to the reactive power compensation requirements of transformer loads from 250KVA to 3000KVA.

7. The intelligent automatic reactive power compensation control device as described in claim 1, characterized in that, The high-speed ADC module has a resolution of 16 bits. After converting the low-voltage safety analog signal into a digital signal, the high-speed ADC module transmits it to the FPGA high-speed processing unit for the FPGA high-speed processing unit to calculate real-time reactive power status data, load type information and load capacity information.

8. An intelligent automatic reactive power compensation control method, characterized in that, The application of the intelligent reactive power automatic compensation control device according to any one of claims 1-7 includes the following steps: The sampling and signal conditioning module acquires voltage analog signals, current analog signals, current waveform signals and voltage-current phase difference signals from the power grid in real time, as well as transformer load terminals, and isolates and filters the signals to output a conditioned low-voltage safety analog signal. Low-voltage safety analog signals are converted into digital signals using a high-speed ADC module; The system receives digital signals through the FPGA high-speed processing unit, performs instantaneous reactive power calculation and harmonic analysis, and generates real-time reactive power status data, load type information and load capacity information of the system based on current waveform signals and voltage-current phase difference signals. The AI ​​decision unit receives real-time reactive power status data, load type information, and load capacity information, calls pre-stored capacitor switching combination strategies, and outputs matrix capacitor switching instructions through a pre-trained AI model. The control drive module generates multiple drive signals based on the matrix capacitor switching command. The multi-channel capacitor switcher controls the dynamic switching combination of multiple capacitor branches, including at least three different capacity specifications, through at least 80 configurable switching ports based on multiple drive signals to form a matrix switching structure, thereby achieving instantaneous and accurate compensation of reactive power capacity.

9. The intelligent automatic reactive power compensation control method as described in claim 8, characterized in that, The FPGA high-speed processing unit also performs trend extrapolation on the load characteristic information to generate load change prediction data. The AI ​​decision-making unit combines load change prediction data to output matrix capacitor switching commands to achieve precise compensation.

10. The intelligent automatic reactive power compensation control method as described in claim 8, characterized in that, Pre-stored capacitor switching combination strategies include harmonic-sensitive strategies; When the FPGA high-speed processing unit analyzes and finds that the harmonic distortion rate exceeds a preset threshold, it triggers the AI ​​decision unit to call the harmonic sensitive strategy to generate a switching command that can suppress the resonance of the dominant harmonic frequency.

Images