Dynamic reactive power compensation device control system and control method thereof

Abstract

The invention provides a dynamic reactive power compensation device control system and a control method thereof. The system comprises a TSC reactive power compensation unit, an SVG reactive power compensation unit and a coordination control unit. Object attribute information and related parameter information of reactive power generated by a power grid within a certain time period are collected, an LSTM neural network training data set is constructed and trained, the reactive power of the power grid is predicted, the reactive power needing to be compensated is calculated, timely reactive power compensation is achieved, and the reactive power compensation efficiency is improved. The problem that real-time reactive power compensation response is not timely enough is avoided, meanwhile, a prediction value advanced detection strategy is set, the problem caused by neural network model prediction overfitting is avoided, and control reliability is guaranteed. Meanwhile, a strategy that the TSC and the SVG are mixed to carry out reactive compensation is adopted, the control strategies of the TSC and the SVG are accurately optimized, the advantages of the characteristics of the TSC and the advantages of the characteristics of the SVG are fully fused, and the efficiency and quality of reactive power compensation are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of power reactive power compensation control, and specifically to a dynamic reactive power compensation device control system and a control method thereof. Background Art

[0002] In modern power systems, the balance of reactive power plays a vital role in maintaining system voltage stability, improving power quality, and ensuring the safe and efficient operation of power equipment. With the continuous expansion of industrial production and social electricity consumption, the load changes in the power system are becoming increasingly complex and frequent. The traditional static reactive power compensation method can no longer meet the needs of real-time dynamic regulation of reactive power. Dynamic reactive power compensation devices have emerged as the times require. They can quickly respond to changes in system reactive power and perform reactive power compensation in a timely manner.

[0003] The existing reactive power compensation technologies mainly include thyristor switched capacitors (TSC), thyristor controlled reactors (TCR), static VAR generators (SVG), etc. Their main characteristics are:

[0004] 1. Thyristor switched capacitors (TSC) achieve reactive power compensation by switching capacitors of fixed capacity. When capacitors need to be switched in or out for reactive power compensation, the fast on and off characteristics of thyristors can complete the switching of capacitors in a relatively short time (generally within half a cycle to several cycles), thereby making a relatively timely response to reactive power changes and helping to quickly adjust the reactive state of the power grid. However, the compensation amount is discrete and can only be adjusted in steps based on the capacity of the capacitor bank, and continuous and precise reactive power regulation cannot be achieved. Therefore, it may be difficult to meet the needs in situations where high reactive power compensation accuracy is required.

[0005] 2. Thyristor-controlled reactors (TCRs) change the equivalent reactance value of the reactor by controlling the trigger angle of the thyristor, thereby achieving continuous regulation of reactive power. Compared with TSCs, they can perform reactive compensation more precisely according to the reactive power demand of the power grid, and are more advantageous in scenarios where the dynamic range of reactive power changes is large and there are certain requirements for compensation accuracy. However, in the process of adjusting the reactance value of the reactor by changing the trigger angle of the thyristor, harmonic currents will inevitably be injected into the power grid. These harmonic currents will pollute the power quality of the power grid and may cause problems such as overheating, vibration, and malfunction of electrical equipment. Therefore, additional harmonic filters are often required to control the harmonics, which increases the complexity and cost of the system.

[0006] 3. Static VAR generator (SVG response speed) is based on power electronic conversion technology, which can quickly and accurately output or absorb reactive power, and can realize continuous adjustment of reactive compensation. Whether it is compensating small-capacity reactive changes or responding to large-capacity, rapidly changing reactive demand scenarios, it can achieve high compensation accuracy and better meet the strict requirements of modern power systems for power quality. First of all, the cost is high: SVG uses advanced power electronic technology and complex control and conversion systems, and the research and development, manufacturing, and procurement costs of the equipment itself are relatively high. Secondly, the environmental requirements are relatively high: due to the relatively sophisticated internal electronic devices, there are certain requirements for the temperature, humidity, electromagnetic interference and other conditions of the working environment. For example, high temperature environment may affect the performance and life of electronic devices, and strong electromagnetic interference may cause errors in control signals, etc., so it is necessary to equip with better environmental adjustment and protection facilities, which increases the ancillary costs during use.

[0007] It can be seen that the above reactive power compensation technologies all have certain defects. Therefore, an improved dynamic reactive power compensation device control system and control method thereof are urgently needed to overcome the above shortcomings. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a dynamic reactive power compensation device control system and a control method thereof, so as to solve the deficiencies of the prior art described in the background technology.

[0009] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:

[0010] A dynamic reactive power compensation device control system includes: a TSC reactive power compensation unit, an SVG reactive power compensation unit, and a coordination control unit, wherein:

[0011] The TSC reactive power compensation unit comprises a first grid-side parameter detection circuit, a first controller, a TSC switching circuit, a TSC main circuit, and a first power supply circuit, wherein the input end of the first controller is connected to the first grid-side parameter detection circuit, the output end of the first controller is connected to the TSC switching circuit, the output end of the TSC switching circuit is connected to the TSC main circuit, and the first power supply circuit is electrically connected to the first controller, the first grid-side parameter detection circuit, and the TSC switching circuit;

[0012] The SVG reactive power compensation unit comprises a second grid-side parameter detection circuit, a second controller, an SVG drive circuit, an SVG inverter main circuit, and a second power supply circuit, wherein the input end of the second controller is connected to the second grid-side parameter detection circuit, the output end of the second controller is connected to the SVG drive circuit, the output end of the SVG drive circuit is connected to the SVG inverter main circuit, and the second power supply circuit is electrically connected to the second controller, the second grid-side parameter detection circuit, and the SVG drive circuit;

[0013] The coordination control unit includes a coordination server, a first compensation current detection module, a second compensation current detection module, and a load-side current feedback module, wherein the first compensation current detection module, the second compensation current detection module, and the load-side current feedback module are all electrically connected to the input end of the coordination server, and the coordination server is electrically connected to the first controller and the second controller;

[0014] Among them, the first grid-side parameter detection circuit and the second grid-side parameter detection circuit are used to collect object attribute information and related parameter information of reactive power generated by the grid within a certain period of time;

[0015] The coordination server is used to assign weights based on the frequency and size of reactive power generated by the object, and to construct an LSTM neural network training data set, in which object attribute information, operating parameter information, and weight assignments are used as neural network feature values, and reactive power prediction values ​​are used as neural network target values; at the same time, a reactive power prediction model of the LSTM neural network is constructed and trained to calculate the reactive power after time T. The reactive power that needs to be compensated during the period is based on the power grid Reactive power required to be compensated during the time period, and calculation of the allocation strategy of SVG and TSC reactive power compensation;

[0016] The first controller and the second controller are used to execute the reactive power compensation allocation strategy issued by the coordination server and perform reactive power compensation control on the SVG and the TSC.

[0017] The present invention also provides a control method for a dynamic reactive power compensation device control system, comprising the following steps:

[0018] The first grid-side parameter detection circuit and the second grid-side parameter detection circuit collect object attribute information and related parameter information of reactive power generated in the grid within a certain period of time, and the coordination server assigns weights based on the frequency and size of reactive power generated by the object, and constructs an LSTM neural network training data set, in which the object attribute information, operating parameter information, and weight assignment are used as neural network feature values, and the reactive power prediction value is used as the neural network target value;

[0019] The coordination server builds a reactive power prediction model of the LSTM neural network, optimizes it based on the improved genetic algorithm, divides the training data set into a training set and a test set, inputs the training set into the LSTM model for model training, and tests the model through the test set;

[0020] The coordination server T collects the operating parameter information corresponding to the object in real time, inputs the trained LSTM model to the power grid T+ The reactive power at time T is predicted, and the load-side current feedback module collects the voltage and current of the grid at time T to calculate the reactive power of the grid, and then calculates the reactive power after time T. Reactive power that needs to be compensated during the period;

[0021] Coordinate servers according to the grid Reactive power required to be compensated during the time period, and calculation of the allocation strategy of SVG and TSC reactive power compensation;

[0022] The second controller optimizes the inverter main circuit drive strategy of the SVG through the PID algorithm according to the allocation strategy, if reactive power compensation is performed only through the SVG or through both the SVG and TSC at the same time; the first controller controls the switching operation of the TSC;

[0023] In T+ After the moment, the reactive power prediction and the allocation strategy of SVG and TSC reactive compensation are repeated continuously to keep the reactive power of the power grid infinitely close to the set minimum value.

[0024] Furthermore, the reactive power prediction model of the LSTM neural network is constructed, and optimized based on the improved genetic algorithm, the training data set is divided into a training set and a test set, and the training set is input into the LSTM model for model training, specifically:

[0025] The constructed LSTM model includes input layer, hidden layer and output layer. The input layer has at least 10 neurons, the hidden layer has three layers, the output layer has at least 5 neurons, and Softmax is used as the activation function. A fully connected layer is added, and a Dropout layer is added before the fully connected layer to prevent overfitting.

[0026] Construct an evaluation function for the improved genetic algorithm, use binary coding to initialize the population, and set various parameters; set the fitness function, calculate the fitness value of each individual, and retain the individual with the largest fitness; determine whether the algorithm termination conditions are met. If the convergence conditions are met, output the result, otherwise perform crossover, mutation, and selection operations; determine whether the result of the selection operation meets the algorithm termination conditions. If so, end the algorithm;

[0027] The constructed LSTM neural network model is used as the main architecture of the reactive power prediction model, the number of input layers is set to 3, and the output layer outputs the state matrix; RMSE is determined as the error evaluation standard, and RMSE is also used as the evaluation function of the improved genetic algorithm;

[0028] Use the improved genetic algorithm to dynamically calculate the fitness value of each LSTM training step and determine

[0029] The step size with the best fitness value is used as the training step size of the LSTM neural network;

[0030] The object attribute information, operating parameter information, and weight assignment in the training set are used as input values, the reactive power prediction value is used as the output value, and the cross entropy is used as the loss function of the model. The training is carried out according to the loss function and the improved genetic algorithm. The weights between the neurons are continuously adjusted according to the value of the loss function so that the value of the loss function is gradually reduced.

[0031] After each complete training data set traversal, the loss function value of the validation set is recorded and verified. When the loss function value meets the set threshold, the training ends and a trained LSTM neural network model is obtained.

[0032] Furthermore, the calculation after time T Before the period when the reactive power to be compensated is very small, that is, before Always verify your predictions:

[0033] If the deviation of the predicted reactive power value is less than the set deviation threshold, the calculated reactive power value after time T is calculated. Reactive power that needs to be compensated during the time period, and the allocation strategy of reactive power compensation using SVG and TSC;

[0034] If the deviation of the verified predicted reactive power value exceeds the set deviation threshold, the reactive power to be compensated is calculated according to the voltage and current of the power grid collected in real time at time T, and the allocation strategy of SVG and TSC for reactive compensation is executed.

[0035] Furthermore, the allocation strategy of the SVG and TSC for reactive power compensation is specifically as follows:

[0036] Divide the reactive power W of the power grid that needs to be compensated by the maximum power W1 that a single TSC can compensate, expressed as N=W / W1;

[0037] When N is less than 1, the coordination control unit sends a signal to the first controller and the second controller, and reactive power compensation is performed only through SVG;

[0038] When N is greater than 1 and is not a positive integer and the decimal point a is less than the set threshold TH, the coordination control unit sends a signal to the first controller and the second controller to perform reactive power compensation only through TSC;

[0039] When N is greater than 1 and is not a positive integer and the decimal point digit b is greater than the set threshold TH, the coordination control unit sends a signal to the first controller and the second controller, and performs reactive power compensation through SVG and TSC at the same time;

[0040] The threshold value TH is set to 0.001, and the maximum power that can be compensated by all TSCs and SVGs in the power grid is greater than the reactive power that needs to be compensated in general in the power grid.

[0041] Furthermore, when reactive power compensation is performed by SVG and TSC at the same time, reactive power compensation is performed by SVG first. When the reactive power compensation power of SVG reaches Ws=K1 at time t1, b The W2 coordination control unit sends a signal to control TSC to perform reactive power compensation, and the time when the coordination control unit sends the signal satisfies t=t1+t2-t3, where t2 is the SVG reactive power compensation handover power. W=K2 b W2 is the time taken, t3 is the time from TSC receiving the control signal to responding, and t2 is equal to or equivalent to t3, K1+K2=1, K1 is much larger than K2, and W2 is the maximum power that a single SVG can compensate.

[0042] Furthermore, the PID algorithm is used to optimize the inverter main circuit driving strategy of the SVG, specifically:

[0043] Set the desired reactive power compensation current value to , the reactive power compensation current value measured in real time As the feedback variable, the difference between the two is e = - As the error input signal of the PID algorithm, the commonly used position discrete PID algorithm formula is adopted:

[0044]

[0045] in, is the control quantity output at the kth sampling moment, , , are proportional coefficient, integral coefficient, and differential coefficient respectively. and are the error values ​​at the kth and k-1th sampling moments respectively.

[0046] Furthermore, the object attribute information and its related parameter information include: load and its characteristic parameters, grid elements and their parameters, power supply and its characteristic parameters, and grid operation parameters, wherein the grid elements include but are not limited to capacitors, inductors, and transformers, and the power supply includes generators and distributed power sources.

[0047] Furthermore, in the LSTM neural network training data set, the reactive power prediction value is calculated using the instantaneous reactive power method.

[0048] Furthermore, after receiving the reactive power instruction to be compensated according to the reactive power compensation allocation strategy, the SVG calculates the instruction current of the SVG through the three-phase circuit instantaneous reactive power algorithm.

[0049] The beneficial effects of the above technical solution of the present invention are as follows:

[0050] 1. The present invention collects the attribute information of objects that generate reactive power in the power grid within a certain period of time and their related parameter information, and assigns weights based on the frequency and size of reactive power generated by the objects, constructs and trains an LSTM neural network training data set, predicts the reactive power of the power grid, and calculates the reactive power that needs to be compensated, thereby achieving timely reactive power compensation and avoiding the problem of insufficient response of real-time reactive power compensation. At the same time, a prediction value advance detection strategy is set to avoid the problem caused by overfitting of the neural network model prediction, thereby ensuring control reliability.

[0051] 2. The present invention adopts a strategy of hybrid TSC and SVG for reactive power compensation, and accurately optimizes the control strategies of TSC and SVG, fully integrating the advantages of the respective characteristics of TSC and SVG, and ensuring the efficiency and quality of reactive power supplement.

[0052] 3. The present invention optimizes and controls the inverter main circuit driving strategy of SVG through PID algorithm, so that the reactive power quickly approaches the target value, and can further fine-tune the driving strategy to eliminate possible steady-state errors, ensuring that the reactive power output by SVG is long-term stable and accurately maintained near the set value, meeting the accuracy requirements of reactive power compensation of the power grid, and allowing SVG to respond quickly and accurately change the reactive output. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is the overall principle block diagram of the control system of the dynamic reactive power compensation device of the present invention;

[0054] Figure 2 The hierarchical control structure of the control system of the dynamic reactive power compensation device of the present invention;

[0055] Figure 3 It is a principle block diagram of a TSC reactive power compensation unit of a dynamic reactive power compensation device control system of the present invention;

[0056] Figure 4 It is a principle block diagram of the SVG reactive power compensation unit of the control system of the dynamic reactive power compensation device of the present invention;

[0057] Figure 5 It is a principle block diagram of a coordination control unit of a control system of a dynamic reactive power compensation device of the present invention;

[0058] Figure 6The schematic diagram of the PID control method for SVG of the present invention;

[0059] Figure 7 The three-phase voltage signal detection and conditioning circuit diagram of the present invention;

[0060] Figure 8 The three-phase current signal detection and conditioning circuit diagram of the present invention;

[0061] Fig. 9 This is a schematic diagram of the SVG driving circuit of the present invention;

[0062] Fig.10 This is the schematic diagram of the SVG inverter main circuit of the present invention;

[0063] Fig.11 This is a schematic diagram of the current tracking PWM control principle of the present invention. DETAILED DESCRIPTION

[0064] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0065] Example 1

[0066] like Figure 1 As shown, the present invention proposes a dynamic reactive power compensation device control system, including: a TSC reactive power compensation unit, an SVG reactive power compensation unit, and a coordination control unit, wherein:

[0067] like Figure 3 As shown, the TSC reactive power compensation unit includes a first grid-side parameter detection circuit, a first controller, a TSC switching circuit, a TSC main circuit, and a first power supply circuit. The input end of the first controller is connected to the first grid-side parameter detection circuit, the output end of the first controller is connected to the TSC switching circuit, the output end of the TSC switching circuit is connected to the TSC main circuit, and the first power supply circuit is electrically connected to the first controller, the first grid-side parameter detection circuit, and the TSC switching circuit.

[0068] like Figure 4 As shown, the SVG reactive power compensation unit includes a second grid-side parameter detection circuit, a second controller, an SVG drive circuit, an SVG inverter main circuit, and a second power supply circuit. The input end of the second controller is connected to the second grid-side parameter detection circuit, the output end of the second controller is connected to the SVG drive circuit, the output end of the SVG drive circuit is connected to the SVG inverter main circuit, and the second power supply circuit is electrically connected to the second controller, the second grid-side parameter detection circuit, and the SVG drive circuit;

[0069] like Figure 5The figure shows a block diagram of the coordination control unit. The coordination control unit includes a coordination server, a first compensation current detection module, a second compensation current detection module, and a load-side current feedback module. The first compensation current detection module, the second compensation current detection module, and the load-side current feedback module are all electrically connected to the input end of the coordination server, and the coordination server is electrically connected to the first controller and the second controller. Figure 1 It can be seen that the coordination control unit detects the reactive compensation current of the SVG reactive compensation unit and the TSC reactive compensation unit respectively through the first compensation current detection module and the second compensation current detection module. , , the load side current is detected by the load side current feedback module Based on the changes of these parameters, the first controller of the TSC reactive compensation unit and the second controller of the SVG reactive compensation unit are controlled, and then the first controller and the second controller send compensation current control signals to the TSC switching circuit and the SVG driving circuit respectively, so as to realize the load side current The reactive power balance is the goal.

[0070] Among them, the first grid-side parameter detection circuit and the second grid-side parameter detection circuit are used to collect object attribute information and related parameter information of reactive power generated by the grid within a certain period of time;

[0071] The coordination server is used to assign weights based on the frequency and size of reactive power generated by the object, and to construct an LSTM neural network training data set, in which object attribute information, operating parameter information, and weight assignments are used as neural network feature values, and reactive power prediction values ​​are used as neural network target values; at the same time, a reactive power prediction model of the LSTM neural network is constructed and trained to calculate the reactive power after time T. The reactive power that needs to be compensated during the period is based on the power grid Reactive power required to be compensated during the time period, and calculation of the allocation strategy of SVG and TSC reactive power compensation;

[0072] The first controller and the second controller are used to execute the reactive power compensation allocation strategy issued by the coordination server and perform reactive power compensation control on the SVG and the TSC.

[0073] The IGBT used in the SVG device is a new type of power electronic device, and its breaking time does not exceed 10 microseconds, while the TSC device uses a thyristor, and its breaking time is about 10 milliseconds, which is a thousand times different. Therefore, the response speed of SVG is much greater than that of TSC, and when SVG adopts closed-loop control, the response speed will become faster. When the hybrid dynamic reactive power compensation device is put into operation in the power grid, since the SVG has a very fast response speed, the SVG will first perform reactive power compensation and quickly reach its critical value, while the response speed of the TSC is very slow, which will cause the SVG to continue to work in the limit output state. Only after the TSC completes the switching, the SVG can exit the limit output state. In order to avoid this situation, it is necessary to reasonably allocate reactive power. The present invention fully integrates the advantages of the TSC reactive power compensation unit and the SVG reactive power compensation unit. The advantages of TSC are mature technology, simple and easy to control device, and low production cost. The disadvantages are that the TSC has a slow response speed and can only compensate in groups, and cannot perform continuous reactive power compensation. The advantages of SVG are fast response speed, wide adjustment range, and continuous reactive power compensation. However, when compensating large-capacity reactive power, the manufacturing cost of SVG is very high and the control is relatively complicated. First, use TSC to compensate large-capacity reactive power, and use SVG to compensate the remaining small-capacity reactive power difference.

[0074] The hybrid dynamic reactive power compensation device of the present invention is not a simple combination of SVG and TSC. Since the controllers of SVG and TSC will affect each other, it is necessary to coordinate the work between the SVG and TSC controllers to ensure the safe and efficient operation of the system. The overall control concept of the present invention is: using TSC grouping to compensate for most of the reactive power required by the system, the reactive power difference after TSC compensation is continuously compensated by SVG, and in addition, the system voltage fluctuation and voltage flicker can be suppressed. The hierarchical control structure is as follows: Figure 2 As shown, the hierarchical control method is mainly used to coordinate the control of the hybrid dynamic reactive power compensation device. The control is mainly divided into three parts: reactive power coordination layer, control layer and execution layer.

[0075] The overall control strategy is: first, in the reactive power coordination layer, the voltage at the public connection point of the power grid is and load current Detection is performed, and the reactive power that needs to be compensated is calculated, and then reasonably allocated to TSC and SVG. In the reactive control layer, several groups of TSC are determined according to the principle of reactive power distribution, and switching signals are sent. At the same time, the SVG direct current controller is controlled to make the SVG send the corresponding PWM control signal. In the reactive execution layer, the switching of TSC is controlled and the SVG is made to generate reactive current to complete the reactive power compensation.

[0076] like Figure 6As shown, it is a schematic diagram of the PID control method for SVG of the present invention. In the power system, the presence of inductive and capacitive loads will cause the generation and flow of reactive power, affecting the power factor and power quality of the system. The PID control of SVG can accurately control the SVG to output the corresponding reactive power according to the real-time reactive demand of the system, realize the dynamic compensation of reactive power, improve the power factor of the system, make the power factor close to 1, reduce the transmission of reactive current in the power grid, and reduce line loss. When the load in the power system changes, it will cause voltage fluctuations. SVG quickly adjusts the output reactive power through PID control, absorbs or emits reactive current, and thus regulates the grid voltage. For example, when the load increases and causes the voltage to drop, the SVG quickly increases the reactive output under the action of PID control to increase the voltage; conversely, when the load decreases, the reactive output is reduced to prevent the voltage from being too high and maintain the stability of the system voltage.

[0077] The first grid-side parameter detection circuit and the second grid-side parameter detection circuit both include a voltage detection circuit and a current detection circuit. Figure 7 As shown in the figure, it is a three-phase voltage signal detection and conditioning circuit. The TV1013 voltage transformer is selected as the initial voltage acquisition system, which has a good voltage acquisition effect. The LM358N operational amplifier and the parallel circuit composed of a 1000 ohm resistor and a 0.022uF capacitor are used to realize the acquisition of voltage signals. Figure 8 As shown in the figure, it is a current signal detection and conditioning circuit. The current transformer generally used for collecting current signals is not used here, but LA55P is used. Compared with the current transformer, this sensor has a better DC component suppression effect, the detection value is more in line with the actual situation, and it has good sampling characteristics.

[0078] like Fig. 9 As shown, it is a schematic diagram of the SVG driving circuit. Since the additional inductance of the driving circuit will have a greater impact on the static and dynamic characteristics of the element when the IGBT is turned off at a high frequency, the present invention adopts a pulse transformer method as the driving method of the driving circuit, because the influence of the additional inductance must be solved, and the IGBT is in a high-frequency operation state. When this method is used, the additional inductance loaded on the flexible element of the driving circuit will be negatively correlated with the frequency, which can reduce the influence of the additional parasitic inductance and achieve an ideal effect.

[0079] like Fig.10The figure shows the schematic diagram of the SVG inverter main circuit. Its characteristic is that there is a capacitor on the right, which is equivalent to a stable DC power supply. When the capacitor is regarded as infinite, 6 flexible IGBTs are used to switch on and off in different time periods to realize the inversion of DC current into AC, so as to achieve the purpose of connecting to the grid side and inject the stored electric energy into the grid. The inductor on the left is equivalent to playing the role of a filter, which can prevent the high-frequency harmonics generated by the full bridge on the right from interfering with the grid and improve the quality of the electric energy injected into the grid.

[0080] Example 2

[0081] The present invention provides a control method for a dynamic reactive power compensation device control system, comprising the following steps:

[0082] S1, the first grid-side parameter detection circuit and the second grid-side parameter detection circuit collect the object attribute information and related parameter information of the reactive power generated by the grid within a certain period of time, and the coordination server assigns weights based on the frequency and size of the reactive power generated by the object, and constructs an LSTM neural network training data set, in which the object attribute information, operating parameter information, and weight assignment are used as neural network feature values, and the reactive power prediction value is used as the neural network target value; in the LSTM neural network training data set, the reactive power prediction value is calculated using the instantaneous reactive power method;

[0083] S2. The coordination server builds a reactive power prediction model of the LSTM neural network, optimizes it based on the improved genetic algorithm, divides the training data set into a training set and a test set, inputs the training set into the LSTM model for model training, and tests the model through the test set;

[0084] S3, the coordination server T collects the operating parameter information corresponding to the object in real time, inputs the trained LSTM model to the power grid T+ The reactive power at time T is predicted, and the voltage and current of the power grid at time T are collected to calculate the reactive power of the power grid, and then the reactive power after time T is calculated. Reactive power that needs to be compensated during the period;

[0085] S4, coordination server according to the power grid Reactive power required to be compensated during the time period, and calculation of the allocation strategy of SVG and TSC reactive power compensation;

[0086] S5, the second controller optimizes and controls the inverter main circuit driving strategy of the SVG through the PID algorithm according to the allocation strategy, if reactive power compensation is performed only through the SVG or through both the SVG and the TSC at the same time; the first controller controls the switching operation of the TSC;

[0087] S6, at T+ After the moment, the reactive power prediction and the allocation strategy of SVG and TSC reactive compensation are repeated continuously to keep the reactive power of the power grid infinitely close to the set minimum value.

[0088] In this embodiment, a reactive power prediction model of an LSTM neural network is constructed and optimized based on an improved genetic algorithm. The training data set is divided into a training set and a test set. The training set is input into the LSTM model for model training, specifically:

[0089] The constructed LSTM model includes input layer, hidden layer and output layer. The input layer has at least 10 neurons, the hidden layer has three layers, the output layer has at least 5 neurons, and Softmax is used as the activation function. A fully connected layer is added, and a Dropout layer is added before the fully connected layer to prevent overfitting.

[0090] Construct an evaluation function for the improved genetic algorithm, use binary coding to initialize the population, and set various parameters; set the fitness function, calculate the fitness value of each individual, and retain the individual with the largest fitness; determine whether the algorithm termination conditions are met. If the convergence conditions are met, output the result, otherwise perform crossover, mutation, and selection operations; determine whether the result of the selection operation meets the algorithm termination conditions. If so, end the algorithm;

[0091] The constructed LSTM neural network model is used as the main architecture of the reactive power prediction model, the number of input layers is set to 3, and the output layer outputs the state matrix; RMSE is determined as the error evaluation standard, and RMSE is also used as the evaluation function of the improved genetic algorithm;

[0092] Use the improved genetic algorithm to dynamically calculate the fitness value of each LSTM training step and determine

[0093] The step size with the best fitness value is used as the training step size of the LSTM neural network;

[0094] The object attribute information, operating parameter information, and weight assignment in the training set are used as input values, the reactive power prediction value is used as the output value, and the cross entropy is used as the loss function of the model. The training is carried out according to the loss function and the improved genetic algorithm. The weights between the neurons are continuously adjusted according to the value of the loss function so that the value of the loss function is gradually reduced.

[0095] After each complete training data set traversal, the loss function value of the validation set is recorded and verified. When the loss function value meets the set threshold, the training ends and a trained LSTM neural network model is obtained.

[0096] Due to its characteristics, the LSTM neural network model is more adept at the field of time series prediction, and the reactive power prediction of the power grid is time series data. Through past power load data, seasonal changes, time cycles and related influencing factors (such as the impact of temperature on cooling or heating electricity demand), LSTM can predict the size of power load in a certain period of time in the future, realize the reasonable scheduling of the power grid, and optimize the power generation plan.

[0097] When reactive power compensation is performed during the operation of the power grid, the reactive power compensation device needs to be able to respond to the changes in the reactive power demand of the power grid in a timely and effective manner and provide corresponding reactive power to maintain a good voltage level of the power grid, reduce power loss, improve power quality, and ensure the stable operation of the power grid. Timely reactive power compensation is very important. For example, during peak hours of electricity consumption, a large number of inductive loads are connected, which causes a sharp increase in the reactive power demand of the power grid. If the reactive power compensation cannot keep up in time, the grid voltage will drop significantly, resulting in poor power quality and affecting the normal operation of various electrical equipment. During the trough period when the load decreases rapidly, if the reactive power compensation device cannot be adjusted in time, there may be excess reactive power, causing overvoltage and other problems, which also endangers the safety of the power grid. Through the LSTM neural network model, early prediction and backup reactive power compensation control strategy can be used to achieve timely reactive power compensation.

[0098] In this embodiment, the calculation after time T is Before the period when the reactive power to be compensated is very small, that is, before T+ - Always verify your predictions:

[0099] If the deviation of the predicted reactive power value is less than the set deviation threshold, the calculated reactive power value after time T is calculated. Reactive power that needs to be compensated during the time period, and the allocation strategy of reactive power compensation using SVG and TSC;

[0100] If the deviation of the verified predicted reactive power value exceeds the set deviation threshold, the reactive power to be compensated is calculated according to the voltage and current of the power grid collected in real time at time T, and the allocation strategy of SVG and TSC for reactive compensation is executed.

[0101] This step is to make up for the inaccurate prediction of some situations caused by overfitting or other factors in the neural network operation. It is verified very shortly before the prediction time value. If the verification is successful, the predicted strategy is executed. If the verification is inaccurate, a control strategy based on real-time monitoring data is adopted.

[0102] In this embodiment, the allocation strategy of reactive power compensation between SVG and TSC is as follows:

[0103] Divide the reactive power W of the power grid that needs to be compensated by the maximum power W1 that a single TSC can compensate, expressed as N=W / W1;

[0104] When N is less than 1, the coordination control unit sends a signal to the first controller and the second controller, and reactive power compensation is performed only through SVG; because when N is less than 1, it means that the reactive power to be compensated is less than the maximum power W1 that a single TSC can compensate, so continuous compensation through SVG is more accurate.

[0105] When N is greater than 1 and is not a positive integer and the digit a after the decimal point is less than the set threshold TH, the coordination control unit sends a signal to the first controller and the second controller, and reactive compensation is performed only through TSC; because when N is greater than 1 and is not a positive integer and the digit a after the decimal point is less than the set threshold TH, it means that the reactive power to be compensated is basically an integer multiple of the maximum power W1 that a single TSC can compensate, and the reactive power after the decimal point can be ignored, which facilitates operation and control.

[0106] When N is greater than 1 and is not a positive integer and the digit b after the decimal point is greater than the set threshold value TH, the coordination control unit sends a signal to the first controller and the second controller, and reactive compensation is performed through SVG and TSC at the same time; because when N is greater than 1 and is not a positive integer and the digit a after the decimal point is greater than the set threshold value TH, it means that the reactive power to be compensated still has a large "residual reactive power" after passing through several TSC discrete reactive compensations, but the reactive power cannot be accurately removed through a single TSC reactive compensation.

[0107] The threshold TH is set to 0.001, and the maximum power that can be compensated by all TSCs and SVGs in the power grid is greater than the reactive power that needs to be compensated in general in the power grid, ensuring that reactive power compensation fully covers the power grid.

[0108] In this embodiment, when reactive power compensation is performed by SVG and TSC at the same time, reactive power compensation is performed by SVG first. When the reactive power compensation power of SVG is calculated to reach Ws=K1 at time t1, b The W2 coordination control unit sends a signal to control TSC to perform reactive power compensation, and the time when the coordination control unit sends the signal satisfies t=t1+t2-t3, where t2 is the SVG reactive power compensation handover power. W=K2 b W2 is the time taken, t3 is the time from TSC receiving the control signal to responding, and t2 is equal to or equivalent to t3, K1+K2=1, K1 is much larger than K2, and W2 is the maximum power that a single SVG can compensate.

[0109] The response speed of SVG is faster than that of TSC, so SVG will start reactive power compensation first. When TSC has not started to respond, SVG will be in the limit output state. In order to prevent reactive power compensation oscillation in the system, the present invention sets a threshold for controlling the action of TSC, that is, the reactive power compensation power of SVG reaches Ws=K1. b The W2 coordination control unit sends a signal to control TSC for reactive power compensation. At the same time, considering the seamless transition of SVG and TSC in time handover, the coordination control unit sends a signal at t=t1+t2-t3, and starts t2 in advance when preparing to start TSC, and t is the time from TSC receiving the control signal to responding, and t2 is equal to or equivalent to t3, achieving seamless time connection and ensuring the stability of reactive power compensation.

[0110] In this embodiment, the PID algorithm is used to optimize the inverter main circuit driving strategy of the SVG, specifically:

[0111] Set the desired reactive power compensation current value to , the reactive power compensation current value measured in real time As the feedback variable, the difference between the two is e = - As the error input signal of the PID algorithm, the commonly used position discrete PID algorithm formula is adopted:

[0112]

[0113] in, is the control quantity output at the kth sampling moment, , , are proportional coefficient, integral coefficient, and differential coefficient respectively. and are the error values ​​at the kth and k-1th sampling moments respectively.

[0114] PID, or proportional, integral, and derivative control algorithm, is a classic feedback control strategy.

[0115] Proportional link (P): Its output is proportional to the input deviation, and it can respond quickly to the deviation. The larger the deviation, the stronger the regulation effect. However, using it alone may cause steady-state errors in the system and cannot completely eliminate the deviation. For example, when the reactive power output by the SVG deviates from the set value, the proportional link will output a corresponding regulation signal according to the deviation to change the drive of the inverter main circuit to make the reactive power approach the set value.

[0116] Integral link (I): Mainly used to eliminate steady-state errors, it will integrate the deviation. As long as the deviation exists, the integral effect will continue to accumulate and continuously adjust the output until the deviation is zero, so that the system can run stably at the set value. For example, during the long-term operation of SVG, if there is a small reactive power deviation that has not been eliminated, the integral link will play a role in gradually eliminating this steady-state error, allowing the inverter main circuit to output more accurate reactive power.

[0117] Differential link (D): It is adjusted according to the rate of change of the deviation, and can predict the trend of the deviation change, make adjustments in advance, improve the dynamic response speed of the system, and suppress the oscillation of the system. For example, when the grid is disturbed and the SVG needs to quickly change the reactive power output, the differential link can quickly adjust the inverter main circuit drive signal according to the reactive power deviation change rate, so that the SVG can respond to the change quickly.

[0118] The PID algorithm is used to optimize the inverter main circuit drive strategy of the SVG. First, the reactive power actually output by the SVG is detected and compared with the set target reactive power value to obtain a deviation signal. The proportional link in the PID algorithm can quickly adjust the drive pulse parameters (such as pulse width, frequency, etc.) of the inverter main circuit according to this deviation, so that the reactive power quickly approaches the target value. The integral link then continues to integrate the deviation, further fine-tunes the drive strategy, eliminates possible steady-state errors, and ensures that the reactive power output by the SVG is long-term stable and accurately maintained near the set value to meet the accuracy requirements of the reactive power compensation of the power grid. The differential link predicts the impact of changes in the power grid operating conditions or load fluctuations on the reactive power in advance based on the rate of change of the reactive power deviation, and adjusts the drive mode of the inverter main circuit in time. For example, when the load suddenly increases and the reactive demand changes rapidly, the SVG can respond quickly and accurately change the reactive output.

[0119] In this embodiment, the object attribute information and its related parameter information include: load and its characteristic parameters, grid elements and their parameters, power supply and its characteristic parameters, and grid operation parameters, where the grid elements include but are not limited to capacitors, inductors, and transformers, and the power supply includes generators and distributed power sources. For example, a generator can not only output active power, but also provide reactive power. The magnitude of its excitation current directly affects the reactive power output. When the excitation current increases, the reactive power output of the generator usually increases; conversely, when the excitation current decreases, the reactive power output decreases accordingly. Moreover, the output reactive power of the generator is very different in the leading and lagging phase operation states. For example, when the leading phase is running, it will absorb the reactive power of the grid, and when the lagging phase is running, it will output reactive power. For example, a large number of devices such as asynchronous motors need to absorb reactive power from the grid to establish and maintain the magnetic field during operation. The size of its load rate will also affect the amount of reactive power absorbed. The lower the load, the lower the power factor, and the greater the proportion of reactive power absorbed. There are also transformers, even when running at no load, which will consume a certain amount of reactive power to establish a magnetic field. After loading, the reactive power consumption will change with the load.

[0120] In this embodiment, after receiving the reactive power instruction to be compensated according to the reactive power compensation allocation strategy, the SVG calculates the instruction current of the SVG through the three-phase circuit instantaneous reactive power algorithm, and then the SVG dynamically and accurately compensates the reactive power required by the system. The control method adopts the following method: Fig.11 The triangle wave comparison method shown in the figure The difference between i and i passes through the PI comparator, and the output result is compared with the triangular wave to obtain a rectangular pulse as a control signal to control the on or off of the switch tube. The current control loop of this control method is a first-order system, so its stability is better. The pulse width modulation technology (PWM) is used to reduce the harmonic content in the SVG output current.

[0121] In summary, the present invention collects the attribute information of objects and related parameter information of reactive power generated in the power grid within a certain period of time, and assigns weights based on the frequency and size of reactive power generated by the objects, constructs and trains an LSTM neural network training data set, predicts the reactive power of the power grid, and calculates the reactive power that needs to be compensated, thereby realizing timely reactive power compensation and avoiding the problem of insufficient response of real-time reactive power compensation. At the same time, a prediction value advance detection strategy is set to avoid the problem caused by overfitting of the neural network model prediction, thereby ensuring control reliability.

[0122] At the same time, a strategy of hybrid TSC and SVG is adopted for reactive power compensation, and the control strategies of TSC and SVG are accurately optimized, fully integrating the advantages of their respective characteristics to ensure the efficiency and quality of reactive power supplement.

[0123] The PID algorithm is used to optimize the inverter main circuit drive strategy of the SVG, so that the reactive power quickly approaches the target value. The drive strategy can be further fine-tuned to eliminate possible steady-state errors, ensuring that the reactive power output by the SVG is maintained near the set value in a long-term stable and accurate manner, meeting the accuracy requirements of the reactive power compensation of the power grid, allowing the SVG to respond quickly and accurately change the reactive output.

[0124] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A dynamic reactive power compensation device control system, characterized in that: include: TSC reactive power compensation unit, SVG reactive power compensation unit, coordination control unit, including: The TSC reactive power compensation unit comprises a first grid-side parameter detection circuit, a first controller, a TSC switching circuit, a TSC main circuit, and a first power supply circuit, wherein the input end of the first controller is connected to the first grid-side parameter detection circuit, the output end of the first controller is connected to the TSC switching circuit, the output end of the TSC switching circuit is connected to the TSC main circuit, and the first power supply circuit is electrically connected to the first controller, the first grid-side parameter detection circuit, and the TSC switching circuit; The SVG reactive power compensation unit comprises a second grid-side parameter detection circuit, a second controller, an SVG drive circuit, an SVG inverter main circuit, and a second power supply circuit, wherein the input end of the second controller is connected to the second grid-side parameter detection circuit, the output end of the second controller is connected to the SVG drive circuit, the output end of the SVG drive circuit is connected to the SVG inverter main circuit, and the second power supply circuit is electrically connected to the second controller, the second grid-side parameter detection circuit, and the SVG drive circuit; The coordination control unit includes a coordination server, a first compensation current detection module, a second compensation current detection module, and a load-side current feedback module, wherein the first compensation current detection module, the second compensation current detection module, and the load-side current feedback module are all electrically connected to the input end of the coordination server, and the coordination server is electrically connected to the first controller and the second controller; Among them, the first grid-side parameter detection circuit and the second grid-side parameter detection circuit are used to collect object attribute information and related parameter information of reactive power generated by the grid within a certain period of time; The coordination server is used to assign weights based on the frequency and size of reactive power generated by the object, and to construct an LSTM neural network training data set, in which object attribute information, operating parameter information, and weight assignments are used as neural network feature values, and reactive power prediction values ​​are used as neural network target values; at the same time, a reactive power prediction model of the LSTM neural network is constructed and trained to calculate the reactive power after time T. The reactive power that needs to be compensated during the period is based on the power grid Reactive power required to be compensated during the time period, and calculation of the allocation strategy of SVG and TSC reactive power compensation; The first controller and the second controller are used to execute the reactive power compensation allocation strategy issued by the coordination server and perform reactive power compensation control on the SVG and the TSC.

2. A control method for a dynamic reactive power compensation device control system, characterized in that: The following steps are involved: The first grid-side parameter detection circuit and the second grid-side parameter detection circuit collect object attribute information and related parameter information of reactive power generated in the grid within a certain period of time, and the coordination server assigns weights based on the frequency and size of reactive power generated by the object, and constructs an LSTM neural network training data set, in which the object attribute information, operating parameter information, and weight assignment are used as neural network feature values, and the reactive power prediction value is used as the neural network target value; The coordination server builds a reactive power prediction model of the LSTM neural network, optimizes it based on the improved genetic algorithm, divides the training data set into a training set and a test set, inputs the training set into the LSTM model for model training, and tests the model through the test set; The coordination server T collects the operating parameter information corresponding to the object in real time, inputs the trained LSTM model to the power grid T+ The reactive power at time T is predicted, and the load-side current feedback module collects the voltage and current of the grid at time T to calculate the reactive power of the grid, and then calculates the reactive power after time T. Reactive power that needs to be compensated during the period; Coordinate servers according to the grid Reactive power required to be compensated during the time period, and calculation of the allocation strategy of SVG and TSC reactive power compensation; The second controller optimizes the inverter main circuit drive strategy of the SVG through the PID algorithm according to the allocation strategy, if reactive power compensation is performed only through the SVG or through both the SVG and TSC at the same time; the first controller controls the switching operation of the TSC; In T+ After the moment, the reactive power prediction and the allocation strategy of SVG and TSC reactive compensation are repeated continuously to keep the reactive power of the power grid infinitely close to the set minimum value.

3. The control method of the dynamic reactive power compensation device control system according to claim 2, characterized in that: The reactive power prediction model of the LSTM neural network is constructed, and is optimized based on an improved genetic algorithm, the training data set is divided into a training set and a test set, and the training set is input into the LSTM model for model training, specifically: The constructed LSTM model includes input layer, hidden layer and output layer. The input layer has at least 10 neurons, the hidden layer has three layers, the output layer has at least 5 neurons, and Softmax is used as the activation function. A fully connected layer is added, and a Dropout layer is added before the fully connected layer to prevent overfitting. Construct an evaluation function for the improved genetic algorithm, use binary coding to initialize the population, and set various parameters; set the fitness function, calculate the fitness value of each individual, and retain the individual with the largest fitness; determine whether the algorithm termination conditions are met. If the convergence conditions are met, output the result, otherwise perform crossover, mutation, and selection operations; determine whether the result of the selection operation meets the algorithm termination conditions. If so, end the algorithm; The constructed LSTM neural network model is used as the main architecture of the reactive power prediction model, the number of input layers is set to 3, and the output layer outputs the state matrix; RMSE is determined as the error evaluation standard, and RMSE is also used as the evaluation function of the improved genetic algorithm; Use the improved genetic algorithm to dynamically calculate the fitness value of each LSTM training step and determine The step size with the best fitness value is used as the training step size of the LSTM neural network; The object attribute information, operating parameter information, and weight assignment in the training set are used as input values, the reactive power prediction value is used as the output value, and the cross entropy is used as the loss function of the model. The training is carried out according to the loss function and the improved genetic algorithm. The weights between the neurons are continuously adjusted according to the value of the loss function so that the value of the loss function is gradually reduced. After each complete training data set traversal, the loss function value of the validation set is recorded and verified. When the loss function value meets the set threshold, the training ends and a trained LSTM neural network model is obtained.

4. The control method of the dynamic reactive power compensation device control system according to claim 2, characterized in that: The calculation after time T Before the period when the reactive power to be compensated is very small, that is, before Always verify your predictions: If the deviation of the predicted reactive power value is less than the set deviation threshold, the calculated reactive power value after time T is calculated. Reactive power that needs to be compensated during the time period, and the allocation strategy of reactive power compensation using SVG and TSC; If the deviation of the verified predicted reactive power value exceeds the set deviation threshold, the reactive power to be compensated is calculated according to the voltage and current of the power grid collected in real time at time T, and the allocation strategy of SVG and TSC for reactive compensation is executed.

5. The control method of the dynamic reactive power compensation device control system according to claim 4, characterized in that: The allocation strategy of the SVG and TSC for reactive power compensation is specifically as follows: Divide the reactive power W of the power grid that needs to be compensated by the maximum power W1 that a single TSC can compensate, expressed as N=W / W1; When N is less than 1, the coordination control unit sends a signal to the first controller and the second controller, and reactive power compensation is performed only through SVG; When N is greater than 1 and is not a positive integer and the decimal point a is less than the set threshold TH, the coordination control unit sends a signal to the first controller and the second controller to perform reactive power compensation only through TSC; When N is greater than 1 and is not a positive integer and the decimal point digit b is greater than the set threshold TH, the coordination control unit sends a signal to the first controller and the second controller, and performs reactive power compensation through SVG and TSC at the same time; The threshold value TH is set to 0.001, and the maximum power that can be compensated by all TSCs and SVGs in the power grid is greater than the reactive power that needs to be compensated in general in the power grid.

6. The control method of the dynamic reactive power compensation device control system according to claim 5, characterized in that: When reactive power compensation is performed by SVG and TSC at the same time, reactive power compensation is performed by SVG first. When the reactive power compensation power of SVG is calculated to reach Ws=K1 at time t1, b The W2 coordination control unit sends a signal to control TSC to perform reactive power compensation, and the time when the coordination control unit sends the signal satisfies t=t1+t2-t3, where t2 is the SVG reactive power compensation handover power. W=K2 b W2 is the time taken, t3 is the time from TSC receiving the control signal to responding, and t2 is equal to or equivalent to t3, K1+K2=1, K1 is much larger than K2, and W2 is the maximum power that a single SVG can compensate.

7. The control method of the dynamic reactive power compensation device control system according to claim 2, characterized in that: The PID algorithm is used to optimize the driving strategy of the inverter main circuit of the SVG, specifically: Set the desired reactive power compensation current value to , the reactive power compensation current value measured in real time As the feedback variable, the difference between the two is e = - As the error input signal of the PID algorithm, the commonly used position discrete PID algorithm formula is adopted: ； in, is the control quantity output at the kth sampling moment, , , are proportional coefficient, integral coefficient, and differential coefficient respectively. and are the error values ​​at the kth and k-1th sampling moments respectively.

8. The control method of the dynamic reactive power compensation device control system according to claim 2, characterized in that: The object attribute information and its related parameter information include: load and its characteristic parameters, grid elements and their parameters, power supply and its characteristic parameters, and grid operation parameters, where the grid elements include but are not limited to capacitors, inductors, and transformers, and the power supply includes generators and distributed power sources.

9. The control method of the dynamic reactive power compensation device control system according to claim 2, characterized in that: In the LSTM neural network training data set, the reactive power prediction value is calculated using the instantaneous reactive power method.

10. The control method of the dynamic reactive power compensation device control system according to claim 2, characterized in that: After receiving the reactive power instruction to be compensated according to the reactive power compensation allocation strategy, the SVG calculates the instruction current of the SVG through the three-phase circuit instantaneous reactive power algorithm.

Images