Reactive compensation control system and method based on artificial intelligence autonomous learning

Abstract

The invention relates to the technical field of reactive compensation, in particular to a reactive compensation control system and method based on artificial intelligence autonomous learning, the reactive compensation control system comprises a data acquisition layer, an AI decision layer, a compensation execution layer and a feedback optimization layer, the data acquisition layer comprises a sensor module and a data acquisition module, the sensor module is used for collecting current and voltage data in the circuit. According to the method, firstly, a reactive power prediction model can be established through AI autonomous learning, and dynamic optimization of a compensation strategy is achieved; secondly, production load types are matched in real time, that is, products of the same kind are judged when the deviation is smaller than 0.05, the accuracy of a compensation strategy is ensured, in addition, instantaneous reactive power pre-judgment of the SVG and steady-state compensation cooperative control of the TBB / TSC can be achieved, and the response speed and accuracy of hybrid compensation are improved; and finally, through verification after compensation and a model updating mechanism, the adaptability and stability of long-term operation of the system are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of reactive power compensation, and particularly to a reactive power compensation control system and method based on artificial intelligence autonomous learning. Background Art

[0002] In industrial power systems, reactive power compensation is an important means to improve power quality, reduce line losses, and improve power factor. Traditional reactive power compensation control methods mainly rely on fixed capacitor banks (TBB / TSC) or static var generators (SVG). Their control strategies are usually based on preset thresholds or simple power factor feedback regulation, and have the following limitations: First, the static compensation strategy has poor adaptability: Traditional TBB / TSC compensation devices usually perform switching control based on average reactive power or power factor thresholds, and cannot respond to rapid load fluctuations in real time, resulting in insufficient compensation accuracy; Although SVG can dynamically compensate for instantaneous reactive power, existing control methods mostly rely on real-time feedback regulation, have response delays, and are difficult to achieve advanced prediction.

[0003] Second, lack of autonomous learning ability: Existing systems usually adopt fixed parameters or artificial experience to set compensation strategies, and cannot autonomously optimize models according to the periodic changes of production loads; When the load characteristics change due to different products produced by the factory, it is necessary to manually adjust the parameters again, resulting in high operation and maintenance costs and low efficiency.

[0004] Third, the hybrid compensation has insufficient coordination: In a high-voltage TBB+SVG (or low-voltage TSC+SVG) hybrid compensation system, the existing technology has not effectively combined the advantages of the two: TBB / TSC is suitable for steady-state compensation, and SVG is suitable for dynamic compensation. Most solutions only simply superimpose the functions of the two, lacking an intelligent distribution strategy based on load characteristics.

[0005] Fourth, the model update and verification are missing: After traditional methods of compensation, there is no closed-loop verification of the power factor and no model self-correction mechanism. Long-term operation is likely to result in a decline in compensation effect due to equipment aging or load changes. Summary of the Invention

[0006] The purpose of the present invention is to propose a reactive power compensation control system and method based on artificial intelligence autonomous learning to solve the problems of poor adaptability of the existing reactive power compensation control system, lack of autonomous learning ability, insufficient coordination of hybrid compensation, and missing model update and verification.

[0007] To achieve the above purpose, the present invention adopts the following technical solutions: Reactive power compensation control system based on artificial intelligence autonomous learning. The reactive power compensation control system includes a data acquisition layer, an AI decision-making layer, a compensation execution layer, and a feedback optimization layer. The data acquisition layer includes a sensor module and a data acquisition module. The sensor module is used to collect current and voltage data in the circuit, and the data acquisition module is used to collect relevant active and reactive power data of the PLC in the circuit; The AI decision-making layer includes a database model and an AI learning module. The database model is used to store historical data. The AI learning model is an LSTM neural network model, which is used to calculate the circuit current, voltage, active or reactive power and establish a prediction model; The compensation execution layer includes an SVG compensation module and a TBB / TSC compensation module. The SVG compensation module is used for dynamic reactive power compensation, and the TBB / TSC compensation module is used for static reactive power compensation; The feedback optimization layer includes a power factor detection module, which is used for the operator to set the set value of the power factor.

[0008] As a further description of the above technical solution: The output end of the sensor module is electrically connected to the input end of the data acquisition module. The output end of the data acquisition module is electrically connected to the input end of the database model. The output end of the database model is electrically connected to the input end of the AI learning module. The output end of the power factor detection module is electrically connected to the input end of the AI learning module. The SVG compensation module and the TBB / TSC compensation module are connected in parallel between the AI learning module and the power factor detection module.

[0009] As a further description of the above technical solution: The sensor module is a Hall sensor.

[0010] Reactive power compensation control method based on artificial intelligence autonomous learning, which includes the following steps: S100. Collect the current and voltage data in the circuit through the sensor module, then collect the relevant active and reactive power data of the PLC in the circuit through the data acquisition module, and then centrally transmit the collected data to the database model for calculation, comparison and analysis; S200. The AI compares and analyzes the data calculated in step S100 with the historical data in the database model. When the data deviation amount ≤ 0.05, directly proceed to the next step. When the data deviation amount > 0.05, calculate the compensation amount output control through the AI, and transmit the data collected this time to the AI learning module for recording and calculation, then update and calibrate the data stored in the database model, and finally continue to compare and analyze the deviation amount. When the data deviation amount ≤ 0.05, the next step can be carried out; S300. Based on the data deviation calculated in step S200, the AI learning module selectively performs dynamic reactive power compensation on the reactive power compensation system through the SVG compensation module, or performs static reactive power compensation on the reactive power compensation system through the TBB / TSC compensation module. Then, it analyzes the data of the set value of the power factor and the actual working power of the circuit. When the ratio of the set power factor to the actual working power of the circuit ≥ 0.95, the current reactive power compensation control is completed. When the ratio of the set power factor to the actual working power of the circuit < 0.95, it outputs control by calculating the compensation amount through AI, and transmits the data collected this time to the AI learning module for recording and calculation. Then, it updates and calibrates the data stored in the database model, and then repeats step S200 and step S300 until the ratio of the set power factor to the actual working power of the circuit ≥ 0.95.

[0011] As a further description of the above technical solution: The AI learning modules used in step S200 and step S300 perform comprehensive calculation, comparison, and analysis through a three-layer 128-unit LSTM neural network model. Then, based on the calculated output instantaneous reactive power prediction and average reactive power prediction data, it controls the SVG compensation module and the TBB / TSC compensation module to perform instantaneous reactive power output and average reactive power output respectively.

[0012] As a further description of the above technical solution: The LSTM neural network model includes two layers of LSTM and a fully connected output layer. The specific formula is as follows: =LSTM1(X norm , ) ; =LSTM2( ) ; y = W d + b d ; In the formula, ∈R 128 , ∈R 128 , W d ∈R 2×128 , b d ∈R 2 , y ∈ R 2 contains [avg_q, instant_q].

[0013] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: In the present invention, first, a reactive power prediction model can be established through AI autonomous learning to achieve dynamic optimization of the compensation strategy. Second, the production load types are matched in real time, that is, a deviation < 0.05 is determined as the same type of product to ensure the accuracy of the compensation strategy. In addition, the instantaneous reactive power prediction of SVG and the steady-state compensation coordination control of TBB / TSC can be realized to improve the response speed and accuracy of hybrid compensation. Finally, through the compensation verification and model update mechanism, the adaptability and stability of the long-term operation of the system are ensured. Brief Description of the Drawings

[0014] Figure 1 It is a schematic diagram of the hierarchical architecture of the reactive power compensation control system proposed by the present invention; Figure 2 It is a flowchart of the operation logic of the reactive power compensation control method in the present invention; Figure 3 It is a schematic diagram of the process of the AI training model; Figure 4 It is a prediction model chart generated by AI in the present invention. Detailed Embodiments

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0016] Please refer to Figure 1 , the present invention provides a technical solution: a reactive power compensation control system based on artificial intelligence autonomous learning. The reactive power compensation control system includes a data acquisition layer, an AI decision layer, a compensation execution layer, and a feedback optimization layer. The data acquisition layer includes a sensor module and a data acquisition module. The sensor module is used to collect current and voltage data in the circuit, and the data acquisition module is used to collect relevant active and reactive power data of the PLC in the circuit; The AI decision layer includes a database model and an AI learning module. The database model is used to store historical data, and the AI learning model is an LSTM neural network model, which is used to perform operations on circuit current, voltage, active or reactive power and establish a prediction model; The compensation execution layer includes an SVG compensation module and a TBB / TSC compensation module. The SVG compensation module is used for dynamic reactive power compensation, and the TBB / TSC compensation module is used for static reactive power compensation; The feedback optimization layer includes a power factor detection module, and the power factor detection module is used for the operator to set the set value of the power factor.

[0017] Among them, the sensor module is a Hall sensor, which can dynamically collect the current and voltage change data of the circuit in real time according to the slight change of the magnetic field of the circuit. The output end of the sensor module is electrically connected to the input end of the data acquisition module, and the collected current and voltage data can be transmitted to the data acquisition module for summarization.

[0018] The output end of the data acquisition module is electrically connected to the input end of the database model, and the collected current, voltage, reactive power and actual working power data of the circuit can be transmitted to the database model in real time.

[0019] The output end of the database model is electrically connected to the input end of the AI learning module. The AI learning module can calculate, compare and analyze the collected current, voltage, reactive power and actual working power data of the circuit with the historical data stored in the database model, and then send a control signal to the SVG compensation module and the TBB / TSC compensation module. According to the comparison of the deviation between the actual working power of the actual circuit and the collected data, the switching or collaborative work of the automatic dynamic reactive power compensation or static reactive power compensation working mode can be realized.

[0020] The output end of the power factor detection module is electrically connected to the input end of the AI learning module. The staff can set the set value of the power factor according to the requirements of the actual factory working scenario, which is convenient for the AI to perform automatic judgment and control operations of reactive power compensation. The SVG compensation module and the TBB / TSC compensation module are connected in parallel between the AI learning module and the power factor detection module. According to the ratio of the set power factor to the actual working power of the circuit, the automatic dynamic reactive power compensation, static reactive power compensation operation or collaborative compensation operation of the circuit can be realized.

[0021] Specifically, as Figure 2 shown, the reactive power compensation control method based on artificial intelligence autonomous learning includes the following steps: Step 1: Collect the current and voltage data in the circuit through the sensor module, then collect the relevant active and reactive power data of the PLC in the circuit through the data acquisition module, and then centrally transmit the collected data to the database model for calculation, comparison and analysis; Step 2: The AI compares and analyzes the data calculated in Step 1 with the historical data in the database model. When the data deviation ≤ 0.05, directly proceed to the next step. When the data deviation > 0.05, calculate the compensation amount through the AI to output control, transmit the data collected this time to the AI learning module for recording and calculation, then update and calibrate the data stored in the database model, and finally continue to compare and analyze the deviation. When the data deviation ≤ 0.05, the next step can be carried out; Step 3: Based on the data deviation calculated in Step 2, the AI learning module selectively performs dynamic reactive power compensation on the reactive power compensation system through the SVG compensation module, or performs static reactive power compensation on the reactive power compensation system through the TBB / TSC compensation module. Then, it analyzes the set value of the power factor and the actual working power of the circuit. When the ratio of the set power factor to the actual working power of the circuit ≥ 0.95, the current reactive power compensation control is completed. When the ratio of the set power factor to the actual working power of the circuit < 0.95, it outputs control by calculating the compensation amount through AI, and transmits the data collected this time to the AI learning module for recording and calculation. Then, it updates and calibrates the data stored in the database model, and then repeats Step 2 and Step 3 until the ratio of the set power factor to the actual working power of the circuit ≥ 0.95.

[0022] Specifically, as Figure 3 shown, the AI learning modules used in Step 2 and Step 3 both perform comprehensive calculation, comparison, and analysis through a three-layer 128-unit LSTM neural network model. Then, based on the calculated output instantaneous reactive power prediction and average reactive power prediction data, it controls the SVG compensation module and the TBB / TSC compensation module to perform instantaneous reactive power output and average reactive power output respectively. The LSTM neural network model includes two layers of LSTM and a fully connected output layer. The input layer can input historical current, voltage, and reactive power into LSTM layer 1 for preliminary calculation, and the calculated data features are transmitted to LSTM layer 2 for further calculation. Finally, the data features after the secondary calculation are transmitted to LSTM layer 3 for the final calculation, so as to accurately predict the instantaneous reactive power and average reactive power data. The specific formula of this LSTM neural network model is as follows: =LSTM1(X norm , ); =LSTM2( ); y=W d +b d ; In the formula, ∈R 128 , ∈R 128 ,W d ∈R 2×128 ,b d ∈R 2 ,y∈R 2 contains [avg_q, instant_q].

[0023] Specifically, as Figure 4As shown, according to the operation results of the above LSTM neural network model, a schematic diagram of the prediction model is obtained. By analyzing the prediction model in Figure 4 it can be seen that: through the reactive power compensation control system and method of artificial intelligence autonomous learning in this application, the power quality can be effectively improved, the power factor is stable ≥ 0.95, the power grid harmonics are reduced by 30%, and in terms of energy efficiency, compared with the previous single-mode reactive power compensation method with manual participation, the line loss is reduced by 15% - 20%, and the annual electricity cost is saved by 10% - 25%. In addition, compared with the previous reactive power compensation method with manual participation, the degree of intelligence supports automatic switching of multi-product production modes without manual reset. The reactive power compensation control system and method of this artificial intelligence autonomous learning establish a reactive power prediction model through AI autonomous learning, dynamically optimize the compensation strategy, achieve real-time load matching (deviation < 0.05 determines the same type of production), improve the compensation accuracy, design TBB / TSC + SVG intelligent collaborative control, take into account the steady-state and dynamic compensation requirements, introduce closed-loop verification and model update, and ensure long-term operation stability.

[0024] At the same time, the reactive power compensation method of the present invention has the characteristics of fast compensation response speed, strong decision-making adaptability, high hybrid compensation efficiency, and good long-term stability compared with the reactive power compensation methods in the prior art. First, the SVG of the original compensation response speed depends on feedback, with a delay ≥ 50ms. The SVG of the present invention predicts compensation, with a delay ≤ 10ms, and the dynamic load adaptability is improved by 80%. Secondly, the original reactive power compensation requires manual adjustment of parameters. The present invention automatically learns the load trend through AI, and matches when the deviation < 5%, effectively reducing manual intervention and reducing the operation and maintenance cost by 60%. In addition, the original reactive power compensation TBB + SVG is independently controlled, and over-compensation / under-compensation occur frequently. The present invention's AI intelligently allocates TBB (steady state) + SVG (dynamic) to operate collaboratively, and the compensation accuracy is improved to within ±1% error. Finally, the original reactive power compensation has no model update, and the compensation effect gradually decreases. The present invention continuously optimizes the model through closed-loop verification + online learning, and the power factor still remains ≥ 0.95 after 1 year of operation.

[0025] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A reactive power compensation control system based on artificial intelligence autonomous learning, characterized in that, The reactive power compensation control system includes a data acquisition layer, an AI decision-making layer, a compensation execution layer, and a feedback optimization layer. The data acquisition layer includes a sensor module and a data acquisition module. The sensor module is used to collect current and voltage data in the circuit, and the data acquisition module is used to collect relevant active and reactive power data of the PLC in the circuit; The AI decision-making layer includes a database model and an AI learning module. The database model is used to store historical data. The AI learning model is an LSTM neural network model, which is used to calculate the circuit current, voltage, active or reactive power and establish a prediction model; The compensation execution layer includes an SVG compensation module and a TBB / TSC compensation module. The SVG compensation module is used for dynamic reactive power compensation, and the TBB / TSC compensation module is used for static reactive power compensation; The feedback optimization layer includes a power factor detection module, which is used for the operator to set the set value of the power factor.

2. The reactive power compensation control system based on artificial intelligence autonomous learning according to claim 1, wherein The output end of the sensor module is electrically connected to the input end of the data acquisition module. The output end of the data acquisition module is electrically connected to the input end of the database model. The output end of the database model is electrically connected to the input end of the AI learning module. The output end of the power factor detection module is electrically connected to the input end of the AI learning module. The SVG compensation module and the TBB / TSC compensation module are connected in parallel between the AI learning module and the power factor detection module.

3. The reactive power compensation control system based on artificial intelligence autonomous learning according to claim 1, wherein The sensor module is a Hall sensor.

4. The reactive power compensation control method based on artificial intelligence autonomous learning according to claim 1, characterized in that, It includes the following steps: S100. Collect the current and voltage data in the circuit through the sensor module, then collect the relevant active and reactive power data of the PLC in the circuit through the data acquisition module, and then centrally transmit the collected data to the database model for calculation, comparison and analysis; S200. The AI compares and analyzes the data calculated in step S100 with the historical data in the database model. When the data deviation ≤ 0.05, directly proceed to the next step. When the data deviation > 0.05, calculate the compensation amount output control through the AI, and transmit the data collected this time to the AI learning module for recording and calculation, then update and calibrate the data stored in the database model, and finally continue to compare and analyze the deviation. When the data deviation ≤ 0.05, the next step can be carried out; S300. Based on the data deviation calculated in step S200, the AI learning module selectively performs dynamic reactive power compensation on the reactive power compensation system through the SVG compensation module, or performs static reactive power compensation on the reactive power compensation system through the TBB / TSC compensation module. Then, it analyzes the data of the set value of the power factor and the actual working power of the circuit. When the ratio of the set power factor to the actual working power of the circuit is ≥ 0.95, the current reactive power compensation control is completed. When the ratio of the set power factor to the actual working power of the circuit is < 0.95, it outputs control by calculating the compensation amount through AI, transmits the data collected this time to the AI learning module for recording and calculation, then updates and calibrates the data stored in the database model, and then repeats steps S200 and S300 until the ratio of the set power factor to the actual working power of the circuit is ≥ 0.

95.

5. The reactive power compensation control method based on artificial intelligence autonomous learning according to claim 4, wherein, The AI learning modules used in step S200 and step S300 perform comprehensive calculation, comparison, and analysis through a three-layer LSTM neural network model with 128 units. Then, based on the calculated output instantaneous reactive power prediction and average reactive power prediction data, it controls the SVG compensation module and the TBB / TSC compensation module to perform instantaneous reactive power output and average reactive power output respectively.

6. The reactive power compensation control method based on artificial intelligence autonomous learning according to claim 5, characterized in that The LSTM neural network model includes two layers of LSTM and a fully connected output layer, and the specific formula is as follows: =LSTM1(X norm , ); =LSTM2( ); y = W d + b d ; wherein, ∈R 128 , ∈R 128 , W d ∈R 2×128 , b d ∈R 2 , y ∈ R 2 includes [avg_q, instant_q].

Images