Spark flashover real-time detection method and detection system based on PSO-SVM model

Abstract

The invention discloses a spark flashover real-time detection method based on a PSO-SVM model, and the method comprises the steps: inputting a resonance current # imgabs0 # value of a high-voltage electrostatic dust collection power system collected every 0.1 s into a classification decision function f (x), if f (x) is greater than gt; if 0, the high-voltage electrostatic dust collection power supply system is in a normal working state, and if f (x) lt is continuously performed twice; and 0, the high-voltage electrostatic dust collection power supply system is in a spark flashover state. The invention also discloses a detection system based on the method and a control method of the high-voltage electrostatic dust collection power supply system. The spark flashover of the high-voltage electrostatic dust collection power supply system can be detected in real time, the total flashover frequency can be effectively reduced by controlling the spark flashover, and the purpose of improving the dust collection efficiency is achieved.

Description

Technical Field

[0001] The invention relates to the field of high-voltage electrostatic dust removal, and in particular to a spark flashover real-time detection method and a detection system based on a PSO-SVM model. Background Art

[0002] As one of the main new energy sources, solar energy has become the most potential energy source because of its endless nature. It is estimated that by 2030, solar energy will provide 10% of the world's electricity, most of which comes from sunny desert areas. At present, most domestic photovoltaic power generation bases are built in Tibet, Xinjiang, Inner Mongolia, Gansu and other places, where the air transparency is high, the sunshine time is long, and the solar energy resources are rich. However, the area is windy and sandy, which easily raises a lot of dust, causing dust accumulation on solar panels. Data shows that in just one month, the energy output of solar panels is reduced by 20% due to dust accumulation, so solar panel cleaning is very necessary.

[0003] The main ways to clean solar panels are: manual high-pressure water gun cleaning, robot cleaning, vehicle-mounted mobile cleaning, and solar panel self-cleaning technology. However, nearly 38 billion liters of water are used to clean solar panels each year, which is enough to supply drinking water for 2 million people. In addition, most of these places are short of water resources, and the cleaning effect is poor, the labor consumption is large, and water resources are wasted, making it difficult to be widely used in actual production. Therefore, the high-voltage electrostatic precipitator power supply designed for electrostatic precipitator can not only achieve solar panel dust removal and water resource waste, but also improve the cleaning efficiency and effect of solar panels.

[0004] The basic working principle of high-voltage electrostatic dust removal power supply is to charge and charge dust particles through the high-voltage static electricity in the power supply body. When the dust is charged, it will be absorbed by the dust collecting plate under the action of electrostatic force and then processed.

[0005] During the operation of the high-voltage electrostatic precipitator power supply, with the changes in wind speed, temperature, humidity, thickness of solar panel dust and other factors, spark flashover will occur between the cathode plate and the anode plate, and the spark flashover will cause the output voltage to decrease sharply, and frequent flashover will seriously reduce the average electrostatic force inside the dust collector, directly affecting the ability and efficiency of handling solar panel dust. For the high-voltage electrostatic precipitator power supply body, repeated spark flashover will cause the plates and poles inside the dust removal power supply to form electrical corrosion, which will seriously reduce the service life of the high-voltage electrostatic precipitator power supply body. The design of the high-voltage electrostatic precipitator power supply should reduce spark flashover to increase the service life of the high-voltage electrostatic precipitator power supply body. In addition, the spark flashover inside the dust collector will cause voltage and current shocks of the power supply. For the high-voltage electrostatic precipitator power supply, the spark breakdown will cause a sudden increase in the current at the rear end of the transformer and a sudden decrease in the current at the front end of the transformer. Frequent spark flashovers may reduce the reliability of the dust removal power supply. Summary of the invention

[0006] The purpose of the present invention is to provide a real-time detection method and a detection system capable of detecting spark flashover during the operation of an electrostatic precipitator power supply.

[0007] The present invention adopts the following technical solution: A spark flashover real-time detection method based on PSO-SVM model comprises the following steps: (a) The resonant current of the high-voltage electrostatic precipitator power supply system is collected every 0.1s The current value; (b) Input the collected current value into the classification decision function f ( x )=sign(w x +b) If f ( x )>0, the high-voltage electrostatic precipitator power supply system is in normal working condition. f ( x )<0, the high-voltage electrostatic precipitator power supply system is in a spark flashover state.

[0008] In the detection method, the classification decision function f ( x ) is obtained through the trained PSO-SVM model.

[0009] In the detection method, the training method of the PSO-SVM model is: (1) Collect the resonant current when the high-voltage electrostatic precipitator power supply is working normally The current value and the resonant current during spark flashover The current value; (2) Set the labels of the current values ​​of normal operation and spark flashover to 1 and -1 respectively to construct data samples; (3) Construct training data samples and test data samples from the obtained data samples; (4) Input the training data samples into the PSO-SVM model for training to obtain a trained PSO-SVM model; (5) Use the test data samples to test the trained PSO-SVM model and verify whether the model can identify the normal working state and spark flashover state of the power supply.

[0010] A detection system using the above-mentioned spark flashover real-time detection method based on the PSO-SVM model includes: a current sampling circuit connected to the main circuit of the high-voltage electrostatic precipitator power supply system, a drive circuit and an ARM control circuit.

[0011] The main circuit of the high-voltage electrostatic precipitator power supply system is an LCC resonant converter, which is composed of an input voltage source, a single-phase full-bridge inverter circuit, an LCC resonant cavity, a high-frequency and high-voltage boost transformer, and a full-bridge rectifier circuit.

[0012] In the detection system, the current sampling circuit includes: diodes D1~D5, resistors R1~R6, capacitors C1~C5, high-frequency current transformer U2, and operational amplifier U1. The first pin of the current transformer U2 is connected to the positive pole of the diode D2 and the negative pole of D4, and the second pin of U2 is connected to the positive pole of the diode D3 and the negative pole of D5. One end of R6, C5, and R4 is connected to the negative pole of D2 and D5, and the other end of R6 and C5 is connected to the positive pole of D4 and D5 and grounded.

[0013] The first pin of the operational amplifier U1 is connected to one end of R1, R3, and C1, the other end of R3 is connected to the anode of D1, one end of C3, and one pin of ADC1_IN1 of STM32, the cathode of D1 is connected to 3.3V, the other end of C3 is grounded, the second pin of U1 is connected to one end of C1, R1, and R2, the other end of R2 is grounded, the third pin of U1 is connected to one end of R4, R5, and C4, the other ends of R5 and C4 are grounded, the fourth pin of U1 is grounded, the eighth pin of U1 is connected to +12V and one end of C2, and the other end of C2 is grounded.

[0014] In the detection system, the driving circuit includes: high-speed optocoupler TLP2362, U3, U5, U8, capacitors C6-C16, diodes D6-D11, driving chip EG2104s, U4, U7. Control chip STM32F103, U6, resistors R7-R17.

[0015] The first pin of U4 is connected to the positive electrode of diode D6, one end of capacitor C9 and +12V, the other end of C9 is grounded, the second pin of U4 is connected to the fifth pin of U3, the third pin of U4 is connected to the third pin of U7 and the fifth pin of U8, one end of C11 is connected to the third pin of U4, the other end is connected to the fourth pin of U4 and grounded, the fifth pin of U4 is connected to one end of R10 and the negative electrode of D8, the other end of R10, the positive electrode of D8 and one end of R11, Q2 g Connect the other end of R11 to Q2 s Ground, the sixth pin of U4 and one end of C10, one end of R9 and Q1 s Connect the seventh pin of U4 to the negative terminal of D7 and one end of R7, and the positive terminal of D7 and the other end of R7 to Q1. g , the other end of R9 is connected, the eighth pin of U4 is connected to the other end of C10 and the negative pole of D6.

[0016] The first pin of U3 is connected to the positive pole of C6, one end of C7, the first pins of U5 and U8, and +3.3, the third pin of U3 is connected to one end of R8, the other end of R8 is connected to PWM1 of U6, the fourth pin of U3 is connected to one end of C8 and grounded, and the sixth pin of U3 is connected to the other end of C8 and +5.

[0017] The first pin of U7 is connected to the positive electrode of diode D9, capacitor C14 and 12V, the other end of C14 is grounded, the second pin of U7 is connected to the fifth pin of U5, the third pin of U7 is connected to the third pin of U4 and the fifth pin of U8, one end of C16 is connected to the third pin of U7, the other end is connected to the fourth pin of U7 and grounded, the fifth pin of U7 is connected to one end of R16 and the negative electrode of D11, the other end of R16, the positive electrode of D11 and one end of R17, Q4 g Connect the other end of R17 to Q4 s Ground, the sixth pin of U4 and one end of C15, one end of R14 and Q3 s Connect the seventh pin of U4 to the negative terminal of D10 and one end of R10, and the positive terminal of D10 and the other end of R10 to Q3. g , the other end of R14 is connected, the eighth pin of U7 is connected to the other end of C15 and the negative pole of D9.

[0018] The first pin of U5 is connected to the positive pole of C6, one end of C7, the first pins of U3 and U8, and +3.3, the third pin of U5 is connected to one end of R12, the other end of R12 is connected to PWM2 of U6, the fourth pin of U5 is connected to one end of C12 and grounded, and the sixth pin of U5 is connected to the other end of C12 and +5.

[0019] The first pin of U8 is connected to the positive pole of C6, one end of C7, the first pins of U3 and U5, and +3.3. The third pin of U8 is connected to one end of R15, and the other end of R15 is connected to the positive pole of U6. The fourth pin of U8 is connected to one end of C13 and grounded, and the sixth pin of U8 is connected to the other end of C13 and +5.

[0020] In the detection system, the spark flashover real-time detection method based on the PSO-SVM model is deployed in the STM32F103 control circuit.

[0021] A control method for an electrostatic precipitator power supply utilizes the above detection system.

[0022] In the control method, the detection system collects the resonant current of the high-voltage electrostatic precipitator power system every 0.1s. Value, the collected current value is input into the classification decision function f (x ), if f ( x )>0, the high-voltage electrostatic precipitator power supply system is in normal working condition; if it is f ( x )<0, the high-voltage electrostatic dust removal power system is in a spark flashover state, and the ARM main control chip outputs a shutdown signal to the locking terminal of the MOSFET switch tube driver chip EG2104s. , block the drive output, suspend system operation; adjust the output reference voltage V ref value, increase the system operating frequency f s , reducing the output voltage, the main control chip outputs an open signal until the system works in normal working state.

[0023] The beneficial effects of the present invention are as follows: the present invention discloses a spark flashover real-time detection method and detection system based on the PSO-SVM model, and obtains a classification decision function by training the PSO-SVM model. f ( x ), and write the spark flashover real-time detection strategy software by the classification decision function, control the detection system to perform circuit detection and control spark flashover, which can effectively reduce the total number of flashovers and achieve the purpose of improving dust removal efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flowchart constructed for the PSO-SVM model.

[0025] Figure 2 It is a real-time detection method for spark flashover.

[0026] Figure 3 It is the main circuit of the high-voltage electrostatic precipitator power supply system.

[0027] Figure 4 It is a current sampling circuit.

[0028] Figure 5 For the drive circuit.

[0029] Figure 6 The resonant current when the main circuit is working normally.

[0030] Figure 7 It is the resonant current when spark flashes over in the main circuit.

[0031] Figure 8 Here are some screenshots of data samples.

[0032] Fig. 9 This is the trained PSO-SVM model.

[0033] Fig.10 To test the accuracy of PSO-SVM model samples. DETAILED DESCRIPTION

[0034] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0035] 1. Training of PSO-SVM model like Figure 1 As shown, the training process of the PSO-SVM model includes the following steps.

[0036] 1) Collect the resonant current when the high-voltage electrostatic precipitator power supply system is working normally The current value and the resonant current during spark flashover The labels of the current values ​​of normal operation and spark flashover are set to 1 and -1 respectively, and data samples are constructed. The collected current signals of the two states are processed by Python software, and M groups of samples are taken for each state. M groups of samples are randomly selected as training samples, and the remaining s groups are used as test samples.

[0037] 2) Input the training samples into the particle swarm optimization support vector machine model (PSO-SVM) for training to obtain the trained particle swarm optimization support vector machine model.

[0038] The SVM model uses the linear kernel function sklearn. The classification accuracy of SVM on the training set is used as the fitness evaluation standard when designing the fitness function. For each particle's C value, a linear SVM model is trained to calculate its classification accuracy on the training set.

[0039] The specific process is: a) Initialize the basic parameters of the particle swarm and determine the fitness function in the optimization process.

[0040] The basic parameters include: number of particles, maximum number of iterations, learning factor (C1, C2), and inertia weight.

[0041] Number of particles: The number of particles determines the coverage of the search space, and the typical value is 20 to 100. The present invention selects 50.

[0042] Maximum number of iterations: controls the algorithm running time and accuracy, typical value: 100~500. The present invention uses 200.

[0043] Learning factors (C1, C2): balance individual experience and group experience. The present invention sets C1=2, C2=2.

[0044] Inertia weight: controls the balance between global search and local search. The present invention sets it to w_start=0.9 and w_end=0.4.

[0045] Fitness function: used to evaluate the quality of each particle position (i.e., penalty coefficient C). The present invention adopts 5-fold cross-validation accuracy.

[0046] b) Generate a particle population, take the parameter C that affects the classification performance of the support vector machine as the position of the particle, and randomly generate a certain number of influencing parameter combinations as the initial position of the particle.

[0047] In the present invention, the general range of parameter C is 10 -3 ~10 3 .

[0048] c) Train the support vector machine under different particle position conditions and calculate the fitness value of each particle.

[0049] When C approaches a large value, it means that the classification cannot be wrong. When C approaches a small value, it means that there can be a greater tolerance for errors. Therefore, the particle swarm optimization algorithm is used to find the optimal solution for C. The SVM model is trained by adjusting the C value. The goal of the fitness function is to guide the PSO algorithm to find the optimal parameter C by quantifying the classification performance of the SVM model.

[0050] The fitness value is a value calculated by the fitness function, which indicates the ability of the current particle position (i.e., the parameter C of SVM) to fit the training data.

[0051] Specific steps for calculating the fitness value: i) Traverse each particle in the particle swarm and evaluate the parameter value represented by each particle.

[0052] ii) Get the parameter value of the current particle: Extract the support vector machine penalty parameter C value represented by the particle currently being processed.

[0053] iii) Prepare training data for cross-validation: Prepare the training data set for subsequent 5-fold division.

[0054] iv) Perform 5-fold cross validation: Divide the training data set evenly and randomly into 5 subsets. Perform 5 iterations, select 4 subsets in each iteration and combine them as the training set, and the remaining subset as the validation set. For each iteration, use the parameter C of the current particle to build a linear support vector machine model. Use the training set of this iteration to train the constructed SVM model. Use the trained model to predict the validation set of this iteration and calculate the classification accuracy on the validation set.

[0055] v) Calculate the average classification accuracy: sum up the classification accuracy obtained by 5 cross-validations and take the average value. This average value is the fitness value corresponding to the particle.

[0056] vi) Store and return the fitness value.

[0057] d) By comparing the size of the fitness value, the individual optimal solution of the particle swarm and the global optimal solution are updated, and then the position coordinates of the next movement of the particle are determined.

[0058] By comparing the fitness values, the individual optimal solution is updated and the global optimal solution is updated. On this basis, the particle speed and updated position are calculated. The boundary condition set by the present invention is C∈[10 -3 ~10 3 ].

[0059] e) until the expected accuracy is reached or the predetermined number of cycles is completed, the particle position corresponding to the final global optimal solution is the optimal parameter of the support vector machine; otherwise, return to step c) to continue. In the present invention, the number of cycles is 200, and the expected accuracy is greater than 98%.

[0060] (3) The trained particle swarm optimization support vector machine model is used to test the test samples to verify whether the model can identify the normal working state and spark flashover state of the power supply.

[0061] (4) Finally, the classification decision function is obtained f ( x )=sign(w x +b).

[0062] 2. Spark flashover real-time detection method like Figure 2 As shown, the spark flashover real-time detection method specifically includes the following steps.

[0063] (1) The high-voltage electrostatic precipitator power supply system is started.

[0064] (2) The system delays for 2 seconds until the power system runs smoothly.

[0065] (3) Collect the resonant current of the high-voltage electrostatic precipitator power supply system every 0.1s value.

[0066] (4) Input the collected current value into the classification decision function f ( x ), if f ( x )>0, the high-voltage electrostatic precipitator power supply system is in normal working condition. f ( x )<0, the high-voltage electrostatic precipitator power supply system is in a spark flashover state and the system operation is suspended.

[0067] (5) Adjust the output reference voltage V refvalue, increase the system operating frequency f s , reduce the output voltage and output the start signal until the system works in normal working state.

[0068] 3. Spark flashover real-time detection system The spark flashover real-time detection system includes: a current sampling circuit connected to the main circuit of the high-voltage electrostatic precipitator power supply system, a drive circuit and an STM32F103 control circuit.

[0069] like Figure 3 As shown, the main circuit of the high-voltage electrostatic precipitator power supply system is an LCC resonant converter, which is mainly composed of an input voltage source, a single-phase full-bridge inverter circuit, an LCC resonant cavity, a high-frequency high-voltage boost transformer, and a full-bridge rectifier circuit. It includes four MOSFET switches Q1~Q4 of the inverter circuit, resonant elements Lr, Cr, Cp of the resonant cavity, a high-frequency high-voltage boost transformer T1, D1~D4 of the rectifier circuit, a filter capacitor Co, and a load RL.

[0070] like Figure 4 As shown, the current sampling circuit includes: diodes D1-D5, resistors R1-R6, capacitors C1-C5, a high-frequency current transformer U2, and an operational amplifier U1.

[0071] The first pin of U2 of the current transformer is connected to the positive electrode of the diode D2 and the negative electrode of D4, and the second pin of U2 is connected to the positive electrode of the diode D3 and the negative electrode of D5.

[0072] One end of R6, C5 and R4 is connected to the negative pole of D2 and D5, and the other end of R6 and C5 is connected to the positive pole of D4 and D5 and grounded.

[0073] The first pin of the operational amplifier U1 is connected to one end of R1, R3, and C1, the other end of R3 is connected to the anode of D1, one end of C3, and one pin of ADC1_IN1 of STM32, the cathode of D1 is connected to 3.3V, the other end of C3 is grounded, the second pin of U1 is connected to one end of C1, R1, and R2, the other end of R2 is grounded, the third pin of U1 is connected to one end of R4, R5, and C4, the other ends of R5 and C4 are grounded, the fourth pin of U1 is grounded, the eighth pin of U1 is connected to +12V and one end of C2, and the other end of C2 is grounded.

[0074] like Figure 5 As shown, the driving circuit includes: high-speed optocoupler TLP2362, U3, U5, U8, capacitors C6~C16, diodes D6~D11, driving chip EG2104s, U4, U7. Control chip STM32F103, U6, resistors R7~R17.

[0075] The first pin of U4 is connected to the positive electrode of diode D6, one end of capacitor C9 and +12V, the other end of C9 is grounded, the second pin of U4 is connected to the fifth pin of U3, the third pin of U4 is connected to the third pin of U7 and the fifth pin of U8, one end of C11 is connected to the third pin of U4, the other end is connected to the fourth pin of U4 and grounded, the fifth pin of U4 is connected to one end of R10 and the negative electrode of D8, the other end of R10, the positive electrode of D8 and one end of R11, Q2 g Connect the other end of R11 to Q2 s Ground, the sixth pin of U4 and one end of C10, one end of R9 and Q1 s Connect the seventh pin of U4 to the negative terminal of D7 and one end of R7, and the positive terminal of D7 and the other end of R7 to Q1. g , the other end of R9 is connected, the eighth pin of U4 is connected to the other end of C10 and the negative pole of D6.

[0076] The first pin of U3 is connected to the positive pole of C6, one end of C7, the first pins of U5 and U8, and +3.3, the third pin of U3 is connected to one end of R8, the other end of R8 is connected to PWM1 of U6, the fourth pin of U3 is connected to one end of C8 and grounded, and the sixth pin of U3 is connected to the other end of C8 and +5.

[0077] The first pin of U7 is connected to the positive electrode of diode D9, capacitor C14 and 12V, the other end of C14 is grounded, the second pin of U7 is connected to the fifth pin of U5, the third pin of U7 is connected to the third pin of U4 and the fifth pin of U8, one end of C16 is connected to the third pin of U7, the other end is connected to the fourth pin of U7 and grounded, the fifth pin of U7 is connected to one end of R16 and the negative electrode of D11, the other end of R16, the positive electrode of D11 and one end of R17, Q4 g Connect the other end of R17 to Q4 s Ground, the sixth pin of U4 and one end of C15, one end of R14 and Q3 s Connect the seventh pin of U4 to the negative terminal of D10 and one end of R10, and the positive terminal of D10 and the other end of R10 to Q3. g , the other end of R14 is connected, the eighth pin of U7 is connected to the other end of C15 and the negative pole of D9.

[0078] The first pin of U5 is connected to the positive pole of C6, one end of C7, the first pins of U3 and U8, and +3.3, the third pin of U5 is connected to one end of R12, the other end of R12 is connected to PWM2 of U6, the fourth pin of U5 is connected to one end of C12 and grounded, and the sixth pin of U5 is connected to the other end of C12 and +5.

[0079] The first pin of U8 is connected to the positive pole of C6, one end of C7, the first pins of U3 and U5, and +3.3. The third pin of U8 is connected to one end of R15, and the other end of R15 is connected to the positive pole of U6. The fourth pin of U8 is connected to one end of C13 and grounded, and the sixth pin of U8 is connected to the other end of C13 and +5.

[0080] The above power supplies +5, +3.3, and +12 are provided by the auxiliary power supply of the high-voltage electrostatic precipitator power supply.

[0081] 4. Spark flashover real-time detection system samples the resonant current of the high-voltage electrostatic precipitator power supply system The process of value 1) Obtain proportional main circuit resonant current through high-frequency current transformer U2 Signal.

[0082] 2) Resonant current The signal is rectified by the full-bridge rectifier circuit composed of D2~D5 to obtain a DC current signal, and then converted into a DC voltage signal by the sampling resistor R6 and the filter capacitor C5.

[0083] 3) Figure 4 The ARM control chip uses STM32F103, and the differential amplifier circuit composed of the operational amplifier U1 and its peripheral circuits amplifies the DC voltage signal to about 1.65V. At this time, the ADC linearity of the control chip STM32F103 is relatively good and the error is relatively small.

[0084] 4) The control chip STM32F103 collects the resonant current of the high-voltage electrostatic precipitator power supply system every 0.1s value.

[0085] 5. Working process of spark flashover real-time detection system 1) The ADC sampling frequency of the control chip STM32F103 is 12MHz, and the shortest conversion time is 1.17us. In order to ensure the stability of the sampled data, the maximum and minimum values ​​of the current values ​​sampled every 40 times are removed and then the average value is calculated to remove some interference items.

[0086] 2) Input the processed current data into the classification decision function f(X) in the particle swarm optimization support vector machine model.

[0087] 3) If f(X)>0, the whole system works normally, the ADC of the STM32F103 main control chip continues to sample, and the above process is repeated.

[0088] 4) If f(X)<0 twice in a row, the system determines that a spark flashover has occurred in the high-voltage electrostatic precipitator power supply.

[0089] 5) The STM32F103 main control chip outputs a shutdown signal to the lock terminal of the MOSFET switch driver chip EG2104s , block the drive output and suspend system operation.

[0090] 6) Adjust the output reference voltage V ref value, increase the system operating frequency f s , reducing the output voltage, the main control chip outputs an open signal until the system works in normal working state.

[0091] 6. Effect Examples The high-voltage electrostatic precipitator power supply system in this experiment has a maximum output power of 80W, an input voltage of 36V, an output voltage of about 0~12kV, and an operating frequency of about 37kHz~50KHz. The goal is to identify the two states of normal operation and spark flashover of the high-voltage electrostatic precipitator power supply system, so as to achieve the control of spark flashover.

[0092] like Figure 6 and Figure 7 The figure shows the resonant current of the high-voltage electrostatic precipitator power supply system during normal operation and spark flashover. Waveforms of two states. Figure 6 and Figure 7 The main circuit resonant current Normal working current value and spark flashover current value, and then set the labels of normal working and spark flashover current values ​​to 1 and -1 respectively to construct data samples. Some data samples are as follows Figure 8 As shown, the first column is the number of the sampled current data, the second column is the size of the sampled current data, and the last column is the label value.

[0093] The above experimental data include two states of the current on the primary side of the main circuit transformer during normal operation and spark flashover. Each current signal has 100 groups, totaling 200 groups of samples. Now, 80 groups of samples are randomly selected from the 100 groups of samples in each signal as training samples, and the remaining 20 groups of samples are used as test samples, totaling 160 groups of training samples and 40 groups of test samples.

[0094] Then the training samples are input into the particle swarm optimization support vector machine model for training to obtain the trained particle swarm optimization support vector machine model (such as Fig. 9 ).

[0095] The trained particle swarm optimization support vector machine model is used to test the test samples to verify whether the model can identify the normal working state and spark flashover state of the power system. Fig.10It can be seen that the diagnostic results obtained by using the test samples as the model input of the trained PSO-SVM have a diagnostic accuracy of 98.33% for the normal operation and spark flashover of the primary side of the step-up transformer. Finally, the classification decision function is obtained. f ( x )=sign(1.93 x -1.12), of which, x is the sampled current value.

[0096] Write the spark flashover real-time detection strategy software and deploy it to the main control chip STM32F103. Start the high-voltage power supply, delay 2s to avoid the system power-on shock, wait for the system to run smoothly, and then detect the spark flashover. The main control chip collects the resonant current of the high-voltage electrostatic precipitator power supply system every 0.1s. Value, the collected current value is input into the classification decision function f ( x ), if f ( x )>0, the system is in normal working state. f ( x )<0, it is a spark flashover state, and the main control chip outputs a shutdown signal to the lock terminal of the MOSFET driver chip IR2104S. , block the drive output and suspend system operation. Adjust the reference voltage V according to the output voltage ref value, increase the system operating frequency f s (37kHz~50KHz), reduce the output voltage, and the main control chip outputs an open signal until the system works in normal working state.

Claims

1. A real-time detection method for spark flashover based on PSO-SVM model, characterized by: 1) Collect the resonant current of the high-voltage electrostatic precipitator power supply system every 0.1s value; 2) Input the collected current value into the classification decision function f ( x )=sign(w x +b) If f ( x )>0, the high-voltage electrostatic precipitator power supply system is in normal working condition. f ( x )<0, the high-voltage electrostatic precipitator power supply system is in a spark flashover state.

2. The spark flashover real-time detection method based on the PSO-SVM model according to claim 1 is characterized in that: The classification decision function f ( x ) is obtained through the trained PSO-SVM model.

3. The spark flashover real-time detection method based on PSO-SVM model according to claim 2 is characterized in that: The training method of the PSO-SVM model is: 1) Collect the resonant current when the high-voltage electrostatic precipitator power supply system is working normally The current value and the resonant current during spark flashover The current value; 2) Set the labels of the current values ​​of normal operation and spark flashover to 1 and -1 respectively to construct data samples; 3) Use the acquired data samples to construct training data samples and test data samples; 4) Input the training data samples into the PSO-SVM model for training to obtain the trained PSO-SVM model; 5) Use the test data samples to test the trained PSO-SVM model to verify whether the model can identify the normal working state and spark flashover state of the power supply.

4. The detection system of the spark flashover real-time detection method based on the PSO-SVM model according to claim 1, characterized in that: The main circuit of the high-voltage electrostatic precipitator power supply system is connected to the current sampling circuit, the drive circuit and the STM32F103 control circuit.

5. The detection system according to claim 4, characterized in that: The main circuit of the high-voltage electrostatic precipitator power supply system is an LCC resonant converter, which is composed of an input voltage source, a single-phase full-bridge inverter circuit, an LCC resonant cavity, a high-frequency and high-voltage boost transformer, and a full-bridge rectifier circuit.

6. The detection system according to claim 5, characterized in that: The current sampling circuit includes: diodes D1-D5, resistors R1-R6, capacitors C1-C5, high-frequency current transformer U2, and operational amplifier U1; The first pin of U2 is connected to the positive electrode of D2 and the negative electrode of D4, and the second pin of U2 is connected to the positive electrode of D3 and the negative electrode of D5; One end of R6, C5, and R4 is connected to the negative pole of D2 and D5, and the other end of R6 and C5 is connected to the positive pole of D4 and D5 and grounded; The first pin of the operational amplifier U1 is connected to one end of R1, R3, and C1, the other end of R3 is connected to the anode of D1, one end of C3, and one pin of ADC1_IN1 of STM32, the cathode of D1 is connected to 3.3V, the other end of C3 is grounded, the second pin of U1 is connected to one end of C1, R1, and R2, the other end of R2 is grounded, the third pin of U1 is connected to one end of R4, R5, and C4, the other ends of R5 and C4 are grounded, the fourth pin of U1 is grounded, the eighth pin of U1 is connected to +12V and one end of C2, and the other end of C2 is grounded.

7. The detection system according to claim 6, characterized in that: The driving circuit includes: high-speed optocoupler TLP2362, U3, U5, U8, capacitors C6~C16, diodes D6~D11, driving chip EG2104s, U4, U7, ARM control chip, U6, resistors R7~R17; The first pin of U4 is connected to the positive electrode of diode D6, one end of capacitor C9 and +12V, the other end of C9 is grounded, the second pin of U4 is connected to the fifth pin of U3, the third pin of U4 is connected to the third pin of U7 and the fifth pin of U8, one end of C11 is connected to the third pin of U4, the other end is connected to the fourth pin of U4 and grounded, the fifth pin of U4 is connected to one end of R10 and the negative electrode of D8, the other end of R10, the positive electrode of D8 and one end of R11, Q2 g Connect the other end of R11 to Q2 s Ground, the sixth pin of U4 and one end of C10, one end of R9 and Q1 s Connect the seventh pin of U4 to the negative terminal of D7 and one end of R7, and the positive terminal of D7 and the other end of R7 to Q1. g , the other end of R9 is connected, the eighth pin of U4 is connected to the other end of C10 and the negative pole of D6; The first pin of U3 is connected to the positive electrode of C6, one end of C7, the first pins of U5 and U8, and +3.

3. The third pin of U3 is connected to one end of R8, the other end of R8 is connected to PWM1 of U6, the fourth pin of U3 is connected to one end of C8 and grounded, and the sixth pin of U3 is connected to the other end of C8 and +5. The first pin of U7 is connected to the positive electrode of diode D9, capacitor C14 and 12V, the other end of C14 is grounded, the second pin of U7 is connected to the fifth pin of U5, the third pin of U7 is connected to the third pin of U4 and the fifth pin of U8, one end of C16 is connected to the third pin of U7, the other end is connected to the fourth pin of U7 and grounded, the fifth pin of U7 is connected to one end of R16 and the negative electrode of D11, the other end of R16, the positive electrode of D11 and one end of R17, Q4 g Connect the other end of R17 to Q4 s Ground, the sixth pin of U4 and one end of C15, one end of R14 and Q3 s Connect the seventh pin of U4 to the negative terminal of D10 and one end of R10, and the positive terminal of D10 and the other end of R10 to Q3. g , the other end of R14 is connected, the eighth pin of U7 is connected to the other end of C15 and the negative pole of D9; The first pin of U5 is connected to the positive electrode of C6, one end of C7, the first pins of U3 and U8, and +3.

3. The third pin of U5 is connected to one end of R12. The other end of R12 is connected to PWM2 of U6. The fourth pin of U5 is connected to one end of C12 and grounded. The sixth pin of U5 is connected to the other end of C12 and +5. The first pin of U8 is connected to the positive pole of C6, one end of C7, the first pins of U3 and U5, and +3.

3. The third pin of U8 is connected to one end of R15, and the other end of R15 is connected to the positive pole of U6. The fourth pin of U8 is connected to one end of C13 and grounded, and the sixth pin of U8 is connected to the other end of C13 and +5.

8. The detection system according to claim 7, characterized in that: The spark flashover real-time detection method based on the PSO-SVM model described in claim 1 is deployed in an ARM control circuit.

9. A control method for a high-voltage electrostatic precipitator power supply system, characterized in that: The state detection is realized by using the detection system described in claims 4 to 8.

10. The control method according to claim 9, characterized in that: The detection system collects the resonant current of the high-voltage electrostatic precipitator power system every 0.1s. Value, the collected current value is input into the classification decision function f ( x ), if f ( x )>0, the system is in normal working state; if f ( x )<0, the system is in the spark flashover state, and the ARM main control chip outputs a shutdown signal to the lock terminal of the MOSFET switch driver chip EG2104s. , block the drive output, suspend system operation; adjust the output reference voltage V ref value, increase the system operating frequency f s , reducing the output voltage, the ARM main control chip outputs an open signal until the system works in normal working state.

Images