Low-temperature plasma temperature curve prediction method combined with time sequence model

Abstract

The invention provides a low-temperature plasma temperature curve prediction method combined with a time sequence model, and the method comprises the steps: recognizing a temperature abrupt change abnormal value in discharge spot temperature data, carrying out the normalization processing of the temperature data through removing the temperature abrupt change abnormal value, and combining the size and density parameters of a discharge spot, and obtaining the temperature curve of the discharge spot. Obtaining discharge spot temperature space-time distribution data under a unified scale; acquiring a periodic fluctuation curve of the temperature of the arc discharge area along with time in the sliding arc discharge process, extracting period and amplitude characteristic parameters of temperature fluctuation, and analyzing relevance between the temperature fluctuation and arc discharge voltage and current electrical parameters to obtain arc discharge temperature fluctuation data; a temperature prediction result is used for guiding and adjusting power supply parameters of dielectric barrier discharge and sliding arc discharge, and active control over temperature distribution in the discharge process is achieved by optimizing discharge voltage and frequency, inhibiting local overheating of discharge spots and weakening arc discharge temperature fluctuation amplitude.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for predicting a low-temperature plasma temperature curve in combination with a time series model. Background Art

[0002] In actual low-temperature plasma wastewater treatment systems, due to the non-uniformity of fluid flow and changes in electrode surface conditions, there are significant temporal and spatial differences in plasma temperature distribution. This inhomogeneity in temperature distribution poses a challenge to accurately predicting and optimizing system performance. Specifically, during dielectric barrier discharge, the presence of local discharge spots leads to non-uniformity in temperature field distribution. The temperature at the discharge spot is significantly higher than that in other areas and changes dynamically over time. At the same time, in sliding arc discharge, the temperature exhibits periodic fluctuations. This complex temperature distribution pattern coupled in time and space is difficult to capture and describe for traditional prediction models. Therefore, how to introduce factors such as temporal and spatial correlation and non-stationarity, and dynamically correct and optimize the model in combination with experimental data, so as to achieve accurate prediction of temperature distribution under different discharge modes, is of great significance for a deep understanding of energy transfer and reaction mechanisms in low-temperature plasma wastewater treatment, as well as for optimizing reactor design and operating parameters. Summary of the invention

[0003] The present invention provides a method for predicting a low-temperature plasma temperature curve in combination with a time series model, which mainly includes:

[0004] Acquire instantaneous temperature data at different discharge spots during dielectric barrier discharge, collect temperature change data of the discharge spot within a preset time range, and spatial distribution data of the temperature near the discharge spot at a preset specific time, to form a data set reflecting the spatiotemporal evolution characteristics of the discharge spot temperature;

[0005] Identify the temperature mutation anomalies in the discharge spot temperature data, remove the temperature mutation anomalies, and normalize the temperature data in combination with the discharge spot size and density parameters to obtain the discharge spot temperature spatiotemporal distribution data under a unified scale;

[0006] The temperature spatiotemporal distribution data of the discharge spot temperature under a unified scale are analyzed by a density clustering algorithm to obtain characteristic parameters of the temperature gradient and temperature fluctuation amplitude of the temperature spatiotemporal distribution data of the discharge spot temperature, identify the pattern reflecting the uneven distribution of the discharge spot temperature, and count the occurrence frequency and duration of the pattern;

[0007] Obtain the periodic fluctuation curve of the temperature in the arc discharge area over time during the sliding arc discharge process, extract the period and amplitude characteristic parameters of the temperature fluctuation, analyze the correlation between the temperature fluctuation and the arc discharge voltage and current electrical parameters, and obtain the arc discharge temperature fluctuation data;

[0008] The pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data are used as inputs. Through the support vector machine algorithm, the mapping relationship between the temporal and spatial distribution of the temperature in the discharge process and the discharge parameters is established to predict the temperature of the discharge spot and arc discharge characteristics.

[0009] In the actual wastewater treatment process, the instantaneous temperature data in the discharge area is collected in real time. According to the identified pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data, the temperature of the discharge spot and arc discharge characteristics is predicted, and the real-time prediction of the temperature spatiotemporal distribution of the discharge process is realized;

[0010] The temperature prediction results are used to guide the adjustment of power supply parameters of dielectric barrier discharge and sliding arc discharge. By optimizing the discharge voltage and frequency, local overheating of the discharge spot is suppressed, the amplitude of arc discharge temperature fluctuation is weakened, and active control of the temperature distribution of the discharge process is achieved.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for predicting the temperature curve of low-temperature plasma combined with a time series model. The method collects temperature data of discharge spots and arc discharge areas, identifies temperature anomalies and performs normalization processing, uses a density clustering algorithm to analyze temperature distribution characteristics, and identifies typical uneven distribution patterns. At the same time, an arc discharge temperature fluctuation model is established to analyze the correlation between temperature fluctuations and electrical parameters. These characteristics are used as input to establish a temperature prediction model through a support vector machine algorithm. In the actual wastewater treatment process, temperature data is collected in real time and input into the prediction model, and parameters are dynamically adjusted to achieve real-time prediction of temperature distribution. Finally, the discharge voltage and frequency are optimized according to the prediction results, local overheating is suppressed, temperature fluctuations are weakened, and active control of the temperature distribution of the discharge process is achieved. The present invention realizes precise control of dielectric barrier discharge and sliding arc discharge processes through the analysis of temperature spatiotemporal distribution characteristics and the establishment of a prediction model, thereby improving wastewater treatment efficiency and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flow chart of a method for predicting a low-temperature plasma temperature curve in combination with a timing model.

[0014] Figure 2 It is a schematic diagram of a method for predicting a low-temperature plasma temperature curve in combination with a time series model according to the present invention.

[0015] Figure 3 It is another schematic diagram of a method for predicting a low-temperature plasma temperature curve in combination with a timing model according to the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1-3 In this embodiment, a method for predicting a low-temperature plasma temperature curve combined with a time series model may specifically include:

[0018] Step S101, obtaining instantaneous temperature data at different discharge spots during dielectric barrier discharge, collecting temperature change data of the discharge spot within a preset time range, and spatial distribution data of the temperature near the discharge spot at a preset specific moment, to form a data set reflecting the spatiotemporal evolution characteristics of the discharge spot temperature.

[0019] According to the temperature value collected by the thermal imaging sensor based on the coordinate information of the discharge spot, the spot area is scanned and sampled at multiple points by a high-frequency infrared detector within a preset scanning period to obtain a discharge spot temperature time series data set; for the discharge spot temperature time series data set, a dual-spectrum imaging detector is used to perform imaging scanning on the discharge spot area, and the temperature field distribution data of the discharge area is constructed by a cubic spline interpolation algorithm; according to the temperature field distribution data, a density-based spatial clustering algorithm is used to classify the discharge spot temperature evolution characteristics, and the temperature gradient matrix is ​​calculated by the finite difference method; for the temperature gradient matrix and the temperature field distribution data, a recursive neural network is used for feature fusion, and the spatiotemporal data sequence is aligned by a dynamic time warping algorithm to obtain a discharge spot temperature spatiotemporal evolution feature model.

[0020] Specifically, the temperature values ​​collected by the thermal imaging sensor are collected according to the coordinate information of the discharge spot. The spot area is scanned and sampled at multiple points within the preset scanning cycle by a high-frequency infrared detector. The collected temperature data are sorted in time series to generate a discharge spot temperature time series data set. The temperature time series data set is tested for data quality, and abnormal temperature fluctuations are removed by a median filter. The missing data points are supplemented by a linear interpolation method to generate a standardized temperature data sequence. The discharge spot area is imaged and scanned by a dual-spectrum imaging detector. The temperature distribution data of the discharge spot and its surrounding area are extracted from the acquired thermal imaging map. The temperature field distribution data of the discharge area is constructed by a cubic spline interpolation algorithm. According to the standardized temperature data sequence, the temperature evolution characteristics of the discharge spot are classified by a density-based spatial clustering algorithm, and the temperature gradient matrix is ​​calculated by the finite difference method to obtain the heat diffusion law of the discharge spot area. For the temperature gradient matrix, the least squares method is used to fit the temperature field contour line, and the temperature change rate at different positions of the discharge spot is obtained by interpolation calculation. The recursive neural network is used to fuse the temperature field distribution data and temperature change rate data of the discharge spot. The dynamic time warping algorithm is used to align the spatiotemporal data sequence to construct the spatiotemporal evolution characteristic model of the discharge spot temperature. The temperature field distribution characteristics during dielectric barrier discharge show dual spatiotemporal evolution characteristics. The temperature change of the discharge spot includes two dimensions: instantaneous change and spatial distribution. The positions of the discharge spots are randomly distributed in the dielectric barrier discharge area. The high-frequency infrared detector uses a scanning frequency of 1000Hz to scan the discharge area at high speed to obtain temperature time series data. For a typical discharge spot, its temperature rises rapidly from room temperature 25℃ to a peak value of 350℃, with a rise time of about 0.5ms, followed by exponential decay. After 10ms, the temperature drops below 100℃. For the collected temperature data, there are abnormal fluctuations caused by sensor noise. A median filter with a window length of 5 is used for noise reduction. The smoothness of the filtered temperature curve is improved by 30%. For the missing data points that appear during the sampling process, the linear interpolation method is used to reconstruct the missing interval data. The interpolation calculation formula is T(t)=T1+(T2-T1)(t-t1) / (t2-t1), where T1 and T2 are the temperature values ​​at both ends of the missing point, and t1 and t2 are the corresponding time points. The dual-spectrum imaging detector simultaneously acquires visible light and infrared images with a spatial resolution of 0.1mm, and performs imaging scanning on the discharge spot area. The cubic spline interpolation algorithm is used to spatially reconstruct the discrete temperature data, with an interpolation node spacing of 0.05mm, and a continuous temperature field distribution function is constructed. In the central area of ​​the discharge spot, the temperature presents an approximate Gaussian distribution feature, and the temperature decreases by 85% from the center to the outside within a radius of 0.5mm. The temperature evolution characteristics of the discharge spot are analyzed based on the density clustering algorithm. The clustering radius is set to 20℃, the minimum number of sampling points is 5, and the temperature change curve is divided into three stages: rapid heating, stable discharge, and residual heat dissipation.The temperature gradient is calculated using the second-order central difference format, with a spatial step of 0.1mm and a time step of 0.1ms. The heat diffusion law is obtained, and the maximum temperature gradient in the discharge center area can reach 450℃ / mm. The least squares method is used to fit the temperature field contours, the fitting polynomial order is 4, and the residual sum of squares is less than 1℃. The recursive neural network contains 64 hidden layer neurons, the input feature dimension is 128, and the dynamic time warping algorithm is used to align the time series data. The time window length is 20ms and the step size is 0.1ms. The model training uses a stochastic gradient descent optimizer with a learning rate of 0.001 and 1000 iterations. The root mean square error on the validation set is less than 2.5℃.

[0021] Step S102, identifying temperature mutation anomalies in the discharge spot temperature data, removing the temperature mutation anomalies, and normalizing the temperature data in combination with the discharge spot size and density parameters to obtain the discharge spot temperature spatiotemporal distribution data under a unified scale.

[0022] The temperature data is scanned by a sliding variance window, and temperature mutation abnormal points are identified according to a preset temperature fluctuation threshold. The temperature mutation abnormal points are corrected by local linear regression and bilateral exponential weighted filtering to obtain the temperature data after the first smoothing process; wavelet multi-scale decomposition is performed on the temperature data after the first smoothing process, and residual mutations are filtered out by a threshold denoising method to obtain the temperature data after the second smoothing process; the spatial normalization reference scale is calculated according to the discharge spot diameter size, and the maximum and minimum normalization mapping is performed on the temperature data after the second smoothing process. The spatial normalization coefficient is calculated in combination with the discharge spot density distribution function to obtain the temperature data after scale conversion; autoencoder feature extraction is performed on the temperature data after the scale conversion to obtain a temperature feature vector, and the temperature feature vector is interpolated and reconstructed by a radial basis function to obtain the discharge temperature spatiotemporal distribution data under a unified scale.

[0023] Specifically, according to the temperature data of the discharge spot, a sliding variance window with a length of n is used to scan the temperature data. By calculating the change rate of adjacent temperature points and comparing them with the preset temperature fluctuation threshold, the temperature mutation abnormal points are identified from the temperature data. For the identified temperature mutation abnormal points, the local linear regression method is used to calculate the temperature trend function of the interval before and after the abnormal point, and the abnormal point is corrected by bilateral exponential weighted filtering to obtain the temperature data after the first smoothing process. The temperature data after the first smoothing process is multi-scale decomposed by wavelet transform, and the residual mutation is filtered out by the threshold denoising method to obtain the temperature data after the second smoothing process. The spatial normalization reference scale is calculated according to the diameter size of the discharge spot, and the maximum and minimum normalization method is used to map and transform the temperature data after the second smoothing process to obtain the normalized temperature data. For the normalized temperature data, the spatial normalization coefficient is calculated in combination with the discharge spot density distribution function, and the temperature data is scaled by the spatial normalization coefficient. The autoencoder is used to extract the features of the scaled temperature data to obtain the discharge spot temperature feature vector. According to the temperature feature vector, the radial basis function is used to interpolate and reconstruct the spatial distribution of the temperature field, and the spatiotemporal distribution data of the discharge temperature under a unified scale are obtained. The characteristics of the discharge spot temperature data are mutation and spatial distribution unevenness. The sliding variance window is used to detect the abnormality of the temperature data. The window length is set to 25 data points. When the change rate of adjacent temperature points exceeds the preset threshold of 0.8, it is determined to be a mutation abnormal point. In a typical discharge process, the normal temperature change rate is maintained between 0.3 and 0.5, while the temperature change rate of the mutation abnormal point often exceeds 1.2, showing a significant jump feature. For the identified temperature mutation abnormal points, the local linear regression method is used for correction processing. 10 data points before and after the abnormal point are selected to construct the regression interval. The regression function form is T(t)=at+b, where T is the temperature value, t is the time point, and a and b are the regression coefficients. The weight function of the bilateral exponential weighted filter adopts w(t)=exp(-|t| / τ), the time constant τ is set to 5 data point intervals, and t is the time, to achieve a smooth transition of the abnormal point. The wavelet transform uses the db4 wavelet basis function to perform a three-layer decomposition of the temperature data to obtain the approximate coefficients and detail coefficients at different scales. The threshold denoising uses the soft threshold method, and the threshold value is set to 2.5 times the standard deviation of the detail coefficient to effectively remove the residual mutation components. The discharge spot diameter distribution is between 0.5mm and 2mm. When calculating the normalized reference scale, the maximum diameter of 2mm is taken as the benchmark, and other sizes are mapped proportionally. The temperature data is normalized using the maximum and minimum standardization method, and the mapping interval is [0,1]. The discharge spot density distribution shows Gaussian characteristics, and the density function is ρ(r)=exp(-r 2 / 2σ 2), where r is the radial distance from the center point and σ is the standard deviation parameter. The spatial normalization coefficient is calculated based on the density distribution function, ranging from 0.6 to 1.2, and the coefficient is larger in areas with higher density. The autoencoder network contains 3 encoding layers and 3 decoding layers, with 64, 32, and 16 hidden neurons, respectively, and uses the ReLU activation function. The radial basis function uses a Gaussian kernel function, where β is a shape parameter with a value of 0.1. During interpolation and reconstruction, 50 uniformly distributed control points are selected to construct an interpolation matrix to solve the interpolation coefficients. The reconstructed temperature field distribution data achieves a continuous and smooth spatial distribution description on a 0.1mm×0.1mm grid. The spatial resolution of the temperature field is increased to 4 times that of the original data, and the local discontinuities in the original data are eliminated.

[0024] Step S103, analyzing the spatiotemporal distribution data of the discharge spot temperature under a unified scale by a density clustering algorithm, obtaining characteristic parameters of the temperature gradient and temperature fluctuation amplitude of the spatiotemporal distribution data of the discharge spot temperature, identifying a pattern reflecting the non-uniformity of the discharge spot temperature distribution, and counting the occurrence frequency and duration of the pattern.

[0025] According to the discharge spot temperature field data, the maximum and minimum standardization method is used to obtain standardized temperature field data, and the standardized temperature field data is used for density clustering calculation; for the standardized temperature field data, the local density value is calculated by the density clustering algorithm, and the local density value is used to obtain the temperature field distribution cluster center through Euclidean distance calculation; for the temperature field distribution cluster center, the temperature field spatial gradient matrix is ​​calculated by the second-order central difference format, and the temperature field spatial gradient matrix is ​​used to obtain the temperature distribution non-uniformity characteristics through Laplace operator operation; for the standardized temperature field data, the temperature field fluctuation characteristics are obtained by Fourier transform, and the temperature field fluctuation frequency components of the discharge spot temperature field are obtained through wavelet packet analysis.

[0026] Specifically, according to the spatiotemporal distribution data of the discharge spot temperature, the temperature field data is normalized by the maximum and minimum normalization method to obtain the standardized temperature field data. According to the standardized temperature field data, the density clustering algorithm is used to calculate the local density value of the spot temperature field, and the similarity between the data points is calculated by the Euclidean distance to obtain the cluster center of the discharge spot temperature field distribution. For the temperature field data of the cluster center, the second-order central difference format is used to calculate the temperature field spatial gradient matrix, and the second-order derivative of the temperature field is calculated by the five-point Laplace operator to obtain the temperature distribution inhomogeneity characteristics. According to the temperature field data, the three-dimensional Fourier transform is used to decompose the temperature field fluctuation characteristics, and the fluctuation amplitude spectrum is extracted by wavelet packet analysis to obtain the discharge spot temperature field fluctuation frequency component. For the temperature field fluctuation frequency component, the adaptive threshold method is used to extract the main frequency characteristics, and the power spectrum density analysis is used to determine the main frequency distribution of the temperature fluctuation. According to the temperature distribution inhomogeneity characteristics and the main frequency distribution, the characteristic vector matching method is used to identify the typical distribution mode of the temperature field, and the pattern recognition result is verified by the cross-validation method. The histogram statistical method is used to count the frequency of occurrence of each typical mode, and the duration of each typical mode is recorded through a sliding time window to obtain the temporal characteristics of the temperature distribution mode. The discharge spot temperature field data shows significant temporal and spatial distribution inhomogeneity. The original temperature data ranges from 25°C to 350°C. The data is mapped to the [0,1] interval through maximum and minimum normalization processing. The standardized temperature field data eliminates the dimension effect, which is convenient for subsequent feature extraction. In the density clustering algorithm, the local density is calculated using the Gaussian kernel function ρi=Σexp(-(dij / dc) 2 ), where dij is the Euclidean distance between data points i and j, and dc is the cutoff distance parameter, which is set to the median of all distances. For typical discharge spots, the local density value is distributed between 0.2 and 0.8, and the high-density area corresponds to the local aggregation characteristics of the temperature distribution. The spatial gradient of the temperature field is calculated using the second-order central difference format: in is the value of the temperature field at different grid points, h is the spatial step length, and its value is 0.1 mm. The five-point Laplace operator expression is Used to extract the second-order spatial derivative characteristics of the temperature field. The calculation results show that the temperature gradient value reaches the maximum at the edge of the discharge spot, which is about 450℃ / mm. The three-dimensional Fourier transform performs time-space spectrum decomposition on the temperature field, and the complex spectrum F(u,v,w) is obtained after transformation, where u and v are spatial frequency components and w is the time frequency component. The wavelet packet analysis uses the db4 wavelet basis function to decompose the spectrum into 6 layers to obtain frequency characteristics at different scales. The main frequency components are concentrated in the range of 10Hz to 100Hz, corresponding to the characteristic oscillation frequency of the discharge process. The power spectral density analysis uses the Welch method, the time window length is 256 points, the overlap rate is 50%, and the Hanning window is weighted. The analysis results show that there are three main frequency components in the temperature field fluctuation, 25Hz, 45Hz and 75Hz, which correspond to the characteristic frequencies of different discharge modes. The feature vector matching uses the cosine similarity metric, and the similarity threshold is set to 0.85, and 4 typical temperature distribution modes are identified. The time series statistics of typical patterns use a sliding time window of 1 second in length, with a window sliding step of 0.1 second. The statistical results show that pattern 1 has the highest frequency, accounting for 45% of the total time, and the average duration is 0.3 seconds; pattern 2 and pattern 3 each account for 25% and last for 0.2 seconds; pattern 4 has the lowest frequency, accounting for only 5%, but the longest duration is 0.5 seconds. The pattern recognition results were verified by the cross-validation method, and the recognition accuracy on the validation set reached 92%, indicating that the feature extraction and pattern recognition methods have good stability.

[0027] Step S104, obtaining a periodic fluctuation curve of the temperature in the arc discharge region over time during the sliding arc discharge process, extracting characteristic parameters of the period and amplitude of the temperature fluctuation, analyzing the correlation between the temperature fluctuation and the arc discharge voltage and current electrical parameters, and obtaining arc discharge temperature fluctuation data.

[0028] The original temperature data collected by the temperature sensor in the sliding arc discharge area is obtained, and the pre-processed temperature data is obtained by using a median filter and a linear interpolation method according to the original temperature data; for the pre-processed temperature data, the temperature fluctuation curve is multi-scale decomposed by wavelet transform, and the periodic amplitude characteristic data of the temperature fluctuation is obtained by using a peak detection method; according to the periodic amplitude characteristic data and the original data of the discharge voltage and current, the instantaneous frequency characteristics of the voltage and current are obtained by using a bandpass filter and a Hilbert transform; for the instantaneous frequency characteristics and the periodic amplitude characteristic data of the temperature fluctuation, a long short-term memory network is used to perform time series feature learning, and the functional relationship between the temperature fluctuation and the electrical parameters is obtained by using a nonlinear least squares method.

[0029] Specifically, according to the original data collected by the temperature sensor in the sliding arc discharge area, the median filter is used to reduce the noise of the temperature data, and the missing data points are supplemented by the linear interpolation method to obtain the preprocessed temperature time series data. For the preprocessed temperature data, the wavelet transform is used to perform multi-scale decomposition on the temperature fluctuation curve, the fluctuation period value is calculated by the autocorrelation function, and the amplitude characteristics of the temperature fluctuation are extracted by the peak detection method. For the original data of discharge voltage and current, a bandpass filter is used to filter out high-frequency noise and DC components, and the instantaneous phase and instantaneous frequency characteristics are extracted by Hilbert transform. According to the extracted instantaneous frequency characteristics, the logarithmic spectrum analysis method is used to obtain the main frequency components of the voltage and current waveforms, and the energy distribution of each frequency component is obtained by spectrum decomposition. According to the periodic amplitude characteristics of temperature fluctuations and the spectrum characteristics of electrical parameters, the Pearson correlation coefficient is used to calculate the correlation strength between temperature fluctuations and electrical parameters. According to the calculated correlation strength, the cross-correlation analysis is used to determine the phase relationship between temperature fluctuations and electrical parameter waveforms, and the temperature response characteristics are obtained by time delay analysis. The long short-term memory network is used to learn the time series features of temperature fluctuation data and voltage and current data, and the mapping relationship between temperature fluctuation characteristics and electrical parameters is verified by cross-validation method. According to the results of time series feature learning, the nonlinear least squares method is used to fit the functional relationship between temperature fluctuation and electrical parameters to obtain the description of arc discharge temperature fluctuation characteristics. The temperature fluctuation characteristics of the sliding arc discharge area show obvious periodicity and instability. The sampling frequency of the temperature sensor is set to 1000Hz, and there is significant high-frequency noise interference in the original temperature data. A median filter with a window length of 5 is used to reduce the noise of the temperature data. The signal-to-noise ratio after filtering is improved by 8dB. The missing data points that appear in the sampling process are repaired by the cubic spline interpolation method. The wavelet transform of the temperature fluctuation curve uses the db4 wavelet basis function to perform 5-layer wavelet decomposition. The frequency band ranges corresponding to the decomposition scale are 0-15.6Hz, 15.6-31.2Hz, 31.2-62.5Hz, 62.5-125Hz, and 125-250Hz respectively. The main period of temperature fluctuation calculated by autocorrelation function is 0.02 seconds, and the fluctuation amplitude varies between 25℃ and 45℃, showing quasi-periodic characteristics. The sampling frequency of the discharge voltage and current signal is 10kHz, and the passband range of the bandpass filter is 10Hz to 1000Hz, which filters out power frequency interference and high-frequency noise. The instantaneous phase extracted by Hilbert transform presents sawtooth wave characteristics, and the phase change rate reflects the frequency modulation characteristics of the discharge process. The logarithmic spectrum analysis of the voltage and current waveforms shows that the main frequency components are distributed at 50Hz, 150Hz and 250Hz, with energy accounts for 60%, 25% and 15% respectively. The Pearson correlation coefficient calculation shows that the correlation coefficient between temperature fluctuation and current waveform is 0.82, and the correlation coefficient with voltage waveform is 0.65, indicating that the correlation between temperature fluctuation and current change is higher.The cross-correlation analysis shows that the temperature fluctuation has a time delay of 2ms relative to the current change, which is consistent with the thermal inertia characteristics of the discharge plasma. The long short-term memory network contains 3 hidden layers, each layer contains 64 neurons, and the input feature dimension is 128, including the time series characteristics of temperature and electrical parameters. The training data set contains 1000 sets of discharge cycle data, and the validation set accounts for 20%. The training uses the Adam optimizer, the learning rate is 0.001, the batch size is 32, and the training cycle is 100. The root mean square error on the validation set is 1.8℃, R. 2 The coefficient is 0.92. The nonlinear least squares method uses the Levenberg-Marquardt algorithm, and the fitting function is T(t) = A1sin(ω1t+φ1) + A2sin(ω2t+φ2), where T is temperature, t is time, A is the amplitude coefficient, ω is the angular frequency, and φ1 and φ2 are the phases of the main temperature fluctuation component and the secondary component, respectively. The fitting results show that the amplitude coefficient of the main temperature fluctuation component is A1 = 35°C, the frequency f1 = 50Hz, the amplitude coefficient of the secondary component is A2 = 15°C, the frequency f2 = 150Hz, and the correlation coefficient between the fitting curve and the measured data reaches 0.95.

[0030] Step S105, taking the pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data as input, establishing the mapping relationship between the spatiotemporal distribution of the temperature of the discharge process and the discharge parameters through the support vector machine algorithm, and performing temperature prediction on the discharge spot and arc discharge characteristics.

[0031] According to the temperature distribution data set of the discharge spot, singular value decomposition is used to reduce the dimension of the temperature distribution data, and a standardized feature data set is obtained by the maximum and minimum normalization method; for the standardized feature data set, a five-fold cross-validation method is used for evaluation, and abnormal data points are eliminated by a box plot detection method to obtain a training data set; for the training data set, a grid search method is used to optimize the bandwidth parameters of the Gaussian kernel function, and an optimal parameter combination is obtained by a cross-validation method; according to the optimal parameter combination, a support vector machine regression algorithm is used to construct a temperature distribution mapping function, and a temperature prediction function is obtained by a sequence minimum optimization method.

[0032] Specifically, based on the typical pattern data set of the temperature distribution inhomogeneity of the discharge spot and the arc discharge temperature fluctuation data set, the singular value decomposition is used to reduce the dimension of the feature data, and the discharge electrical parameter data is normalized by the maximum and minimum normalization method to obtain a standardized feature data set. For the standardized feature data set, the five-fold cross-validation method is used to evaluate the data quality, and the abnormal data points are removed by the box plot detection method to obtain the cleaned training data set. According to the cleaned training data set, the grid search method is used to optimize the bandwidth parameter of the Gaussian kernel function and the penalty factor of the support vector machine, and the optimal parameter combination is determined by the cross-validation method. For the optimal parameter combination, the support vector machine regression algorithm is used to construct the temperature spatiotemporal distribution mapping function, and the mapping function is solved by the sequence minimum optimization method to obtain the initial temperature prediction function. According to the initial temperature prediction function, the bootstrapping method is used to generate multiple groups of verification data sets, and the performance of the prediction function is evaluated by the holdout method. According to the evaluation results, the ensemble learning method is used to optimize the combination of multiple prediction functions, and the prediction results are fused by the weighted average method. The root mean square error and correlation coefficient are used to evaluate the performance of the fused prediction results, and the stability of the prediction results is verified by the chaos analysis method. According to the verified prediction function, the prediction function is dynamically updated by the online learning method to obtain the final prediction model of the spatiotemporal distribution of temperature in the discharge process. The prediction of the spatiotemporal distribution of temperature in the discharge process involves the processing of high-dimensional feature data. The original feature dimension is 128 dimensions, including 96 dimensions of temperature distribution pattern features and 32 dimensions of electrical parameter features. Singular value decomposition is used for dimensionality reduction, and the principal component with a cumulative contribution rate of 95% is selected to reduce the feature dimension to 36 dimensions. Standardization processing maps the eigenvalues ​​to the [-1,1] interval to eliminate the dimension effect. The data quality assessment adopts the five-fold cross validation method, and each fold contains 200 groups of samples. The box plot detection sets 1.5 times the interquartile range as the outlier judgment threshold, and identifies 15 outliers in the temperature data and 8 outliers in the electrical parameter data. The cleaned training data set contains 950 groups of valid samples. The support vector machine uses a Gaussian kernel function, and the parameter space of the grid search is set to take 20 logarithmically equally spaced points in the range of [0.1,10] for the bandwidth parameter σ, and take 20 logarithmically equally spaced points in the range of [1,1000] for the penalty factor C. The cross-validation results show that the optimal parameter combination is σ=1.5, C=100. The sequential minimum optimization algorithm solves the dual problem of the support vector machine, and the iteration termination condition is set to the objective function value change of 5 consecutive iterations is less than 0.001. The root mean square error of the initial temperature prediction function on the validation set is 2.5℃, and the correlation coefficient is 0.91. The bootstrapping method randomly selects 800 groups of samples from the training set to construct 50 sub-models. Ensemble learning uses the weighted average method to fuse multiple prediction functions, and the weight coefficient is proportional to the prediction accuracy of each sub-model on the validation set.The root mean square error of the prediction results after fusion was reduced to 1.8℃, and the correlation coefficient was increased to 0.95. The maximum Lyapunov exponent method was used for chaos analysis, and the calculation result was 0.15, indicating that the prediction results have good stability. During the online learning process, the model was updated every 50 new samples received, and the sliding window method was used to maintain the timeliness of the training samples, and the window length was set to 1000. The temperature prediction deviation of the updated prediction model under different working conditions was kept within the range of ±2℃, and the response time to the sudden change working condition was less than 0.1 second, realizing the real-time and accurate prediction of the temperature distribution of the discharge process. The spatial resolution of the prediction model for the temperature field reached 0.1mm, and the temporal resolution reached 1ms, which met the monitoring requirements of the dynamic evolution characteristics of the temperature field.

[0033] Step S106, in the actual wastewater treatment process, instantaneous temperature data in the discharge area is collected in real time, and the temperature of the discharge spot and arc discharge characteristics is predicted based on the identified pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data, so as to realize the real-time prediction of the temperature spatiotemporal distribution of the discharge process.

[0034] A sliding average filter is used to perform denoising on the temperature data collected by the temperature sensor in the discharge area to obtain a first-in-first-out cache sequence of the temperature data; based on the first-in-first-out cache sequence, the temperature data is decomposed at multiple scales through wavelet transform to obtain the spatial distribution characteristics of the temperature field; based on the spatial distribution characteristics of the temperature field, a Kalman filter is used to construct a state space equation, and the prediction model parameters are obtained through an adaptive gain matrix; based on the prediction model parameters, a random forest regressor is used to construct a temperature field mapping function, and the optimized prediction result is obtained through an online learning method.

[0035] Specifically, according to the temperature data stream collected by the temperature sensor in the discharge area, a sliding average filter with a length of n is used to perform real-time noise reduction on the temperature data. The processed temperature data is stored in a first-in-first-out data cache queue to obtain the instantaneous temperature data sequence in the discharge area. For the temperature data sequence, the wavelet transform is used to perform multi-scale decomposition on the temperature data, and the spatial distribution characteristics of the temperature field are extracted by the soft threshold denoising method, and the similarity between the current temperature distribution and the pre-stored typical pattern is calculated. According to the spatial distribution characteristics of the temperature field and the temperature fluctuation characteristics, the Kalman filter is used to construct the state space equation, and the prediction model parameters are updated in real time through the adaptive gain matrix. For the updated prediction model parameters, the exponentially weighted moving average method is used to calculate the prediction residual sequence, and the noise covariance matrix of the Kalman filter is corrected by the residual analysis method. According to the corrected prediction model, the random forest regressor is used to construct the temperature field mapping function, and the prediction parameters are continuously optimized by the online learning method. For the optimized prediction results, the sliding time window is used to calculate the prediction error statistics, and the validity of the prediction results is judged by the adaptive threshold method. According to the prediction validity judgment results, the weighted combination method is used to fuse the results of multiple prediction cycles, and the combination weight is dynamically adjusted by the recursive least squares method. The cache update mechanism is used to maintain the parameter database of the prediction model, and the data is regularly cleaned up through the timestamp to maintain the real-time update capability of the prediction model. The temperature data acquisition frequency of the discharge area is set to 1000Hz. The original temperature data is significantly interfered by environmental noise, and the signal-to-noise ratio is about 15dB. A sliding average filter with a length of 25 is used for real-time noise reduction, and the signal-to-noise ratio is increased to 28dB after filtering. The data cache queue length is set to 1000 data points, corresponding to 1 second of temperature data, and the queue content is updated by a cyclic coverage method. The wavelet transform of the temperature data uses the db4 wavelet basis function, and performs a 4-layer decomposition to obtain wavelet coefficients of different scales. The soft threshold denoising uses an adaptive threshold λ=σ√(2logN), where σ is the noise standard deviation and N is the data length. The typical temperature distribution pattern contains 4 basic forms, and the similarity calculation uses the cosine distance, and the threshold is set to 0.85. The state variables of the Kalman filter include the spatial distribution parameters and time evolution parameters of the temperature field. The state equation is x(k+1)=Ax(k)+w(k), and the observation equation is y(k)=Hx(k)+v(k). Among them, the state transfer matrix A is estimated online by the least squares method, the observation matrix H is the unit matrix, and w(k) and v(k) are Gaussian white noise. The gain matrix K is iteratively solved by the Lyapunov equation, and the convergence criterion is that the norm change of 5 consecutive iterations is less than 0.001. The prediction residual sequence is processed by the exponential weighted moving average method, and the smoothing coefficient α=0.3. The residual mean and standard deviation are calculated. The noise covariance matrices Q and R are dynamically adjusted according to the statistical characteristics of the residuals, and the adjustment step size is 0.1.The random forest contains 50 decision trees with a maximum depth of 8 and a feature sampling ratio of 0.8. The prediction error statistics use a 60-second sliding time window to calculate the root mean square error RMSE and the mean absolute percentage error MAPE. The validity judgment threshold is set to RMSE < 5 ° C and MAPE < 10%. The fusion weight of the multi-period prediction results is calculated by recursive least squares, with a forgetting factor λ = 0.95 and an initial covariance matrix P0 = 100I. The parameter database uses a key-value pair storage structure, with the key being the timestamp and the value being the corresponding model parameter. The data cleaning cycle is 1 hour, and the parameter records of the last 24 hours are retained. In practical applications, the temperature field of the wastewater treatment process shows obvious periodicity and spatial inhomogeneity. The temperature prediction model shows good adaptability under different working conditions, and the prediction error is maintained at an average of.

[0036] Within the range of ±3°C, the response time to sudden temperature changes is less than 0.5 seconds, achieving accurate real-time prediction of temperature field distribution.

[0037] Step S107, using the temperature prediction result to guide the adjustment of the power supply parameters of the dielectric barrier discharge and the sliding arc discharge, by optimizing the discharge voltage and frequency, suppressing the local overheating of the discharge spot, weakening the arc discharge temperature fluctuation amplitude, and realizing active control of the temperature distribution of the discharge process.

[0038] According to the temperature field distribution data, the temperature field distribution characteristics are decomposed by multi-dimensional Fourier transform, and the temperature distribution evaluation parameters are obtained by calculating the spatial variation coefficient of the temperature field; for the temperature distribution evaluation parameters, the temperature fluctuation amplitude is monitored online by an adaptive threshold method, and the temperature anomaly discrimination result is obtained by calculating the time series fluctuation variance of the temperature field; according to the temperature anomaly discrimination result, the discharge voltage parameter and the frequency parameter are optimized by the gradient descent method, and the power supply parameter adjustment scheme is obtained by the constrained optimization solver; for the power supply parameter adjustment scheme, a linear quadratic regulator is used to construct a control function, and a voltage and frequency adjustment signal is obtained through a power modulation circuit.

[0039] Specifically, based on the discharge temperature prediction results, the temperature field distribution characteristics are decomposed by multidimensional Fourier transform, and the evaluation parameters of temperature distribution are obtained by calculating the spatial variation coefficient of the temperature field and the time series fluctuation variance. For the temperature distribution evaluation parameters, the adaptive threshold method is used to monitor the temperature fluctuation amplitude online, and the overheating index of the discharge spot is calculated by the sliding window method to obtain the temperature anomaly discrimination result. According to the temperature anomaly discrimination result, the gradient descent method is used to optimize the discharge voltage and frequency parameters, and the initial power supply parameter adjustment scheme is generated by the constrained optimization solver. For the initial power supply parameter adjustment scheme, the linear quadratic regulator is used to construct the control function, and the dynamic characteristics of the adjustment instruction are optimized by the feedforward compensation method. According to the optimized adjustment instruction, the control signal of the discharge power supply is generated by the digital signal processor, and the real-time adjustment of the voltage and frequency is realized by the power modulation circuit. Using the temperature field feedback data, the controller parameters are identified online by the recursive least squares method to obtain the update amount of the controller parameters.

[0040] For the controller parameter update amount, the Lyapunov stability criterion is used to constrain the parameter adjustment process, and the convergence of the control parameters is ensured by robustness analysis. According to the real-time data of the temperature field, the statistical process control chart is used to monitor the discharge temperature distribution online, and the effectiveness of the control effect is judged by the control limit calculation. The temperature field distribution characteristics of the discharge area are decomposed by multidimensional Fourier transform. The spatial coefficient of variation of the temperature field CV=σ / μ, where σ is the standard deviation and μ is the mean. Under typical working conditions, the CV value varies between 0.15 and 0.35. The time series fluctuation variance is calculated by exponential weighting method, and the weight factor α=0.3, which reflects the dynamic stability of the temperature field. The temperature fluctuation amplitude monitoring adopts the adaptive threshold method, the threshold λ=μ+kσ, where k is the adjustment coefficient, and the initial value is set to 2.5. The discharge spot overheating index POI is calculated by sliding window, POI = (Tmax-Tavg) / Tavg × 100%, the window length is 1 second, when POI exceeds 35%, it is judged as abnormal working condition, Tavg is the average temperature value in the window, Tmax is the maximum temperature value in the window. The voltage and frequency parameter optimization adopts gradient descent method, and the objective function J(V,f) includes temperature uniformity term and stability term, J = w1 × CV + w2 × Var(T), where w1 = 0.6, w2 = 0.4 are weight coefficients, and Var(T) is temperature variance. Constraints include voltage range 15-30kV, frequency range 20-100kHz, and optimization step size adaptive adjustment. The control function of the linear quadratic regulator adopts state feedback form, u(k) = -Kx(k), where the state vector x contains voltage deviation, frequency deviation and its rate of change, u(k) is the control input, which represents the control signal applied at time k, used to adjust the system state, and K is the state feedback gain matrix. The feedforward compensation adopts the temperature prediction model, and the compensation function is in the form of Gc(s)=α(1+Tds), where α is the preset proportional coefficient and the time constant Td=0.01s. The control signal is generated using a 16-bit digital signal processor with a sampling frequency of 10kHz and a PWM modulation frequency of 200kHz. The power modulation adopts the SPWM method with a carrier ratio of 20 and a dead time of 2μs to achieve a voltage accuracy of 0.1kV and a frequency accuracy of 0.1kHz. The recursive least squares method with a forgetting factor of 0.95 is used for the online identification of the controller parameters, and the state equation is

[0041] x(k+1)=Ax(k)+Bu(k), the observation equation is y(k)=Cx(k), A is the state transfer matrix, B is the input matrix, which represents the influence of the control input u(k) on the system state, and C is the output matrix, which represents how the system state x(k) is mapped to the observation value y(k). The projection algorithm is used for parameter update to ensure that the eigenvalues ​​of the parameter matrices A, B, and C are within the unit circle. The statistical process control uses an exponentially weighted moving average control chart, and the upper and lower control limits are the process mean plus or minus three times the standard deviation. The center line uses the target temperature value, and the control limit is updated every 60 seconds. When the temperature value exceeds the control limit for 3 consecutive points or 7 consecutive points are on the same side of the center line, the control parameter correction mechanism is triggered. In this way, closed-loop control of the discharge temperature field distribution is achieved, and the temperature uniformity and stability are significantly improved.

[0042] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting a low-temperature plasma temperature curve in combination with a time series model, characterized in that: The method comprises: Acquire instantaneous temperature data at different discharge spots during dielectric barrier discharge, collect temperature change data of the discharge spot within a preset time range, and spatial distribution data of the temperature near the discharge spot at a preset specific time, to form a data set reflecting the spatiotemporal evolution characteristics of the discharge spot temperature; Identify the temperature mutation anomalies in the discharge spot temperature data, remove the temperature mutation anomalies, and normalize the temperature data in combination with the discharge spot size and density parameters to obtain the discharge spot temperature spatiotemporal distribution data under a unified scale; The temperature spatiotemporal distribution data of the discharge spot temperature under a unified scale are analyzed by a density clustering algorithm to obtain characteristic parameters of the temperature gradient and temperature fluctuation amplitude of the temperature spatiotemporal distribution data of the discharge spot temperature, identify the pattern reflecting the uneven distribution of the discharge spot temperature, and count the occurrence frequency and duration of the pattern; Obtain the periodic fluctuation curve of the temperature in the arc discharge area over time during the sliding arc discharge process, extract the period and amplitude characteristic parameters of the temperature fluctuation, analyze the correlation between the temperature fluctuation and the arc discharge voltage and current electrical parameters, and obtain the arc discharge temperature fluctuation data; The pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data are used as inputs. The mapping relationship between the spatiotemporal distribution of the temperature in the discharge process and the discharge parameters is established through the support vector machine algorithm to predict the temperature of the discharge spot and arc discharge characteristics. In the actual wastewater treatment process, the instantaneous temperature data in the discharge area is collected in real time. According to the identified pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data, the temperature of the discharge spot and arc discharge characteristics is predicted, and the real-time prediction of the temperature spatiotemporal distribution of the discharge process is realized; The temperature prediction results are used to guide the adjustment of power supply parameters of dielectric barrier discharge and sliding arc discharge. By optimizing the discharge voltage and frequency, local overheating of the discharge spot is suppressed, the amplitude of arc discharge temperature fluctuation is weakened, and active control of the temperature distribution of the discharge process is achieved.

2. The method according to claim 1, characterized in that The instantaneous temperature data at different discharge spots during the dielectric barrier discharge process are obtained, and the temperature change data of the discharge spot within a preset time range and the spatial distribution data of the temperature near the discharge spot at a preset specific time are respectively collected to form a data set reflecting the spatiotemporal evolution characteristics of the discharge spot temperature, including: The temperature value collected by the thermal imaging sensor is collected according to the coordinate information of the discharge spot, and the spot area is scanned and sampled at multiple points within a preset scanning period by a high-frequency infrared detector to obtain a discharge spot temperature time series data set; For the discharge spot temperature time series data set, a dual-spectrum imaging detector is used to perform imaging scanning on the discharge spot area, and the temperature field distribution data of the discharge area is constructed by a cubic spline interpolation algorithm; According to the temperature field distribution data, a density-based spatial clustering algorithm is used to classify the discharge spot temperature evolution characteristics, and a temperature gradient matrix is ​​calculated by a finite difference method; A recursive neural network is used to perform feature fusion on the temperature gradient matrix and the temperature field distribution data, and a dynamic time warping algorithm is used to align the spatiotemporal data sequence to obtain a discharge spot temperature spatiotemporal evolution feature model.

3. The method according to claim 1, characterized in that The method of identifying temperature mutation anomalies in the discharge spot temperature data, removing temperature mutation anomalies, and normalizing the temperature data in combination with the discharge spot size and density parameters to obtain the discharge spot temperature spatiotemporal distribution data under a unified scale includes: The temperature data is scanned using a sliding variance window, and temperature mutation abnormal points are identified according to a preset temperature fluctuation threshold. The temperature mutation abnormal points are corrected through local linear regression and bilateral exponential weighted filtering to obtain the temperature data after the first smoothing process; Performing wavelet multi-scale decomposition on the temperature data after the first smoothing process, filtering out residual mutations by a threshold denoising method, and obtaining the temperature data after the second smoothing process; Calculating a spatial normalization reference scale according to the discharge spot diameter, performing maximum and minimum normalization mapping on the temperature data after the second smoothing process, calculating a spatial normalization coefficient in combination with the discharge spot density distribution function, and obtaining the temperature data after scale conversion; The temperature data after the scale conversion is subjected to autoencoder feature extraction to obtain a temperature feature vector, and the temperature feature vector is interpolated and reconstructed by a radial basis function to obtain the spatiotemporal distribution data of the discharge temperature under a unified scale.

4. The method according to claim 1, characterized in that The density clustering algorithm is used to analyze the spatiotemporal distribution data of the discharge spot temperature under a unified scale, obtain the temperature gradient and temperature fluctuation amplitude characteristic parameters of the spatiotemporal distribution data of the discharge spot temperature, identify the pattern reflecting the uneven distribution of the discharge spot temperature, and count the occurrence frequency and duration of the pattern, including: According to the discharge spot temperature field data, a maximum and minimum standardization method is used to obtain standardized temperature field data, and the standardized temperature field data is used for density clustering calculation; For the standardized temperature field data, local density values ​​are calculated by density clustering algorithm, and the local density values ​​are used to obtain the temperature field distribution cluster center through Euclidean distance calculation; For the cluster center of the temperature field distribution, a second-order central difference format is used to calculate the temperature field spatial gradient matrix, and the temperature field spatial gradient matrix is ​​operated by the Laplace operator to obtain the temperature distribution inhomogeneity characteristics; For the standardized temperature field data, the temperature field fluctuation characteristics are obtained through Fourier transformation, and the temperature field fluctuation frequency components of the discharge spot temperature field are obtained through wavelet packet analysis of the temperature field fluctuation characteristics.

5. The method according to claim 1, characterized in that The method of obtaining the periodic fluctuation curve of the temperature in the arc discharge area over time during the sliding arc discharge process, extracting the period and amplitude characteristic parameters of the temperature fluctuation, analyzing the correlation between the temperature fluctuation and the arc discharge voltage and current electrical parameters, and obtaining the arc discharge temperature fluctuation data includes: Acquire the original temperature data collected by the temperature sensor in the sliding arc discharge area, and obtain the pre-processed temperature data by using the median filter and linear interpolation method according to the original temperature data; For the pre-processed temperature data, multi-scale decomposition of the temperature fluctuation curve is performed by wavelet transform, and the periodic amplitude characteristic data of the temperature fluctuation is obtained by using a peak detection method; According to the periodic amplitude characteristic data and the original data of the discharge voltage and current, a bandpass filter and Hilbert transform are used to obtain the instantaneous frequency characteristics of the voltage and current; Aiming at the instantaneous frequency characteristics and the periodic amplitude characteristic data of temperature fluctuation, a long short-term memory network is used to perform time series feature learning, and the functional relationship between temperature fluctuation and electrical parameters is obtained by nonlinear least squares method.

6. The method according to claim 1, characterized in that The method uses the pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data as input, establishes the mapping relationship between the spatiotemporal distribution of the temperature of the discharge process and the discharge parameters through the support vector machine algorithm, and performs temperature prediction on the discharge spot and arc discharge characteristics, including: According to the temperature distribution data set of the discharge spot, singular value decomposition is used to perform dimension reduction processing on the temperature distribution data, and a standardized feature data set is obtained by a maximum and minimum standardization method; For the standardized feature data set, a five-fold cross validation method is used for evaluation, and abnormal data points are eliminated through a box plot detection method to obtain a training data set; For the training data set, a grid search method is used to optimize the bandwidth parameters of the Gaussian kernel function, and an optimal parameter combination is obtained by a cross-validation method; According to the optimal parameter combination, a support vector machine regression algorithm is used to construct a temperature distribution mapping function, and a temperature prediction function is obtained through a sequence minimum optimization method.

7. The method according to claim 1, characterized in that In the actual wastewater treatment process, instantaneous temperature data in the discharge area is collected in real time, and the temperature of the discharge spot and arc discharge characteristics is predicted according to the identified pattern reflecting the uneven temperature distribution of the discharge spot and the arc discharge temperature fluctuation data, so as to realize the real-time prediction of the temperature spatiotemporal distribution of the discharge process, including: A sliding average filter is used to perform noise reduction processing on the temperature data collected by the discharge area temperature sensor to obtain a first-in-first-out cache sequence of the temperature data; According to the first-in-first-out cache sequence, the temperature data is decomposed at multiple scales by wavelet transform to obtain the spatial distribution characteristics of the temperature field; According to the spatial distribution characteristics of the temperature field, a Kalman filter is used to construct a state space equation, and the prediction model parameters are obtained through an adaptive gain matrix; According to the prediction model parameters, a random forest regressor is used to construct a temperature field mapping function, and an optimized prediction result is obtained through an online learning method.

8. The method according to claim 1, characterized in that The temperature prediction results are used to guide the adjustment of power supply parameters of dielectric barrier discharge and sliding arc discharge, and the local overheating of the discharge spot is suppressed by optimizing the discharge voltage and frequency, and the arc discharge temperature fluctuation amplitude is weakened to realize active control of the temperature distribution of the discharge process, including: According to the temperature field distribution data, the multi-dimensional Fourier transform is used to decompose the temperature field distribution characteristics, and the temperature distribution evaluation parameters are obtained by calculating the spatial variation coefficient of the temperature field. According to the temperature distribution evaluation parameters, the temperature fluctuation amplitude is monitored online by using the adaptive threshold method, and the temperature anomaly discrimination result is obtained by calculating the time series fluctuation variance of the temperature field; According to the temperature anomaly discrimination result, the discharge voltage parameter and the frequency parameter are optimized and calculated by using the gradient descent method, and the power supply parameter adjustment scheme is obtained by using the constraint optimization solver; According to the power supply parameter adjustment scheme, a linear quadratic regulator is used to construct a control function, and a voltage-frequency adjustment signal is obtained through a power modulation circuit.

Images