Adaptive filtering method and device for industrial predictive control based on time series decomposition

Abstract

The application relates to the technical field of industrial predictive control, in particular to an industrial predictive control adaptive filtering method and device based on time series decomposition. The method comprises the following steps: collecting historical data of a controlled variable in an industrial process in a preset time period to obtain a time series data set; decomposing the time series data set through a time series decomposition algorithm to obtain three types of independent components; performing feature extraction on the three types of independent components respectively, and matching a filtering algorithm and determining target filtering parameters of the filtering algorithm based on the feature extraction results. Compared with related technologies, the time series data set is decomposed into three types of independent components through the time series decomposition algorithm, the complex characteristics of the industrial process data are analyzed from three dimensions of change trend, period fluctuation and specific residual data, and the problem that a traditional filtering algorithm depends on experience is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial predictive control, and particularly relates to an industrial predictive control adaptive filtering method and device based on time series decomposition. BACKGROUND

[0002] Industrial predictive control refers to, in the related field of industrial production, processing complex fluctuation industrial process data in an application scenario by applying a predictive control software, and then improving the efficiency of industrial production, which can be widely applied to control systems in industrial production processes such as chemical industry and petroleum industry.

[0003] In industrial predictive control, a filtering module is used to suppress noise of collected industrial process data, while effective process information needs to be retained to provide high-quality data guarantee for subsequent data input into a predictive model. The filtering effect depends on the rationality of the filtering parameter values. In the current filtering parameter calculation method, the filtering parameter depends on the experience setting of a staff, and the actual production process is relatively complex. The manual parameter setting mode is difficult to adapt to the complex characteristics of the industrial process data, resulting in insufficient filtering precision, and then affecting the prediction precision and control effect of the predictive control model.

[0004] Therefore, there is an urgent need for a filtering method that can adapt to and analyze the complex characteristics of industrial process data, thereby improving the filtering precision. SUMMARY

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present application provides an industrial predictive control adaptive filtering method and device based on time series decomposition, and the main purpose is to solve the problem that the current manual parameter setting mode is difficult to adapt to the complex characteristics of industrial process data, resulting in insufficient filtering precision.

[0006] In order to achieve the above-mentioned purpose, the main technical scheme adopted by the present application includes:

[0007] In a first aspect, the present application provides an industrial predictive control adaptive filtering method based on time series decomposition, comprising:

[0008] Collecting historical data of a controlled variable in an industrial process in a preset time period to obtain a time series data set;

[0009] Decomposing the time series data set by a time series decomposition algorithm to obtain three types of independent components; wherein the three types of independent components include a trend component representing trend change, a periodic component representing fluctuation frequency in a period, and a residual component representing noise intensity, and the trend component, the periodic component and the residual component are independent of each other;

[0010] extract features from the three types of independent components respectively, and match a filtering algorithm based on the feature extraction results and determine target filtering parameters of the filtering algorithm.

[0011] In a second aspect, the embodiments of the present application provide an industrial predictive control adaptive filtering device based on time series decomposition, comprising:

[0012] A collection unit is configured to collect historical data of a controlled variable in an industrial process within a preset time period to obtain a time series data set;

[0013] A calculation unit is configured to decompose the time series data set by a time series decomposition algorithm to obtain three types of independent components; wherein the three types of independent components include a trend component representing trend changes, a periodic component representing fluctuation frequencies within a period, and a residual component representing noise intensities, and the trend component, the periodic component, and the residual component are mutually independent;

[0014] A processing unit is configured to extract features from the three types of independent components respectively, and match a filtering algorithm based on the feature extraction results and determine target filtering parameters of the filtering algorithm.

[0015] By the above technical solution, the present application provides an industrial predictive control adaptive filtering method and device based on time series decomposition. First, historical data of a controlled variable in an industrial process within a preset time period is collected to obtain a time series data set. Then, the time series data set is decomposed by a time series decomposition algorithm to obtain three types of independent components; wherein the three types of independent components include a trend component representing trend changes, a periodic component representing fluctuation frequencies within a period, and a residual component representing noise intensities, and the trend component, the periodic component, and the residual component are mutually independent. Then, features are extracted from the three types of independent components respectively, and a filtering algorithm is matched based on the feature extraction results and target filtering parameters of the filtering algorithm are determined. Compared with related technologies, the present application first arranges historical data in an industrial process in chronological order to form a time series data set, and then decomposes the time series data set into three types of independent components by a time series decomposition algorithm. The complex characteristics of industrial process data are analyzed from three different dimensions of change trend, fluctuation within a period, and specific residual data, solving the problem of "no clear rules for parameter adjustment and relying on experience" in traditional filtering. Finally, features are extracted based on the three types of independent components, and a filtering algorithm is matched based on the feature extraction results and target filtering parameters of the filtering algorithm are determined, thereby improving filtering precision and enhancing the adaptability of industrial predictive control. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of an industrial predictive control adaptive filtering method based on time series decomposition provided by the embodiments of the present application is shown in the figure;

[0017] Figure 2 A flowchart of a time series decomposition algorithm provided for an embodiment of the present application is shown in FIG. 1.

[0018] Figure 3 A flowchart of a target filtering parameter calculation method provided for an embodiment of the present application is shown in FIG. 2.

[0019] Figure 4 A module diagram of an industrial predictive control adaptive filtering method based on time series decomposition provided for an embodiment of the present application is shown in FIG. 3.

[0020] Figure 5 A diagram of an industrial predictive control adaptive filtering device based on time series decomposition provided for an embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0021] In order to better understand the above technical solutions, exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer, more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0022] As described above, industrial predictive control can improve the efficiency of industrial production through some industrial control software. In industrial predictive control, the filtering module is one of the core links: it needs to suppress noise on the collected industrial process data (such as temperature, pressure, hardness, flow, concentration, purity, etc.) while retaining effective process information to provide high-quality input data for subsequent prediction models. Filtering is a key technology in signal processing, which refers to the process of suppressing interference or extracting useful information by selecting specific frequency components, and its core role is to separate noise and effective information from the original data to improve data usability and accuracy. The effectiveness of filtering directly depends on the rationality of the filtering parameter values.

[0023] Currently, the filtering parameters of existing industrial predictive control are usually manually adjusted by engineers according to field experience. However, it is obvious that the feedback industrial process data is a complex result, for example, the poor purity in the data is composed of multiple factors and is not caused by only a single factor. Therefore, the method of manually setting filtering parameters is difficult to adapt to the complex characteristics of industrial process data, resulting in insufficient filtering precision, which further affects the prediction accuracy and control effect of the prediction control model.

[0024] In order to improve the current artificial parameter setting mode which is difficult to adapt to the complex characteristics of industrial process data, and to solve the problem of insufficient filtering precision, an industrial predictive control adaptive filtering method based on time series decomposition is proposed. The method can be applied to the adaptive filtering system of industrial predictive control. In the running, any one of the following industrial predictive control adaptive filtering methods based on time series decomposition can be executed. As shown in the following formula (1), the method comprises the following steps. Figure 1

[0025] S101, collecting historical data of a controlled variable in an industrial process in a preset time period to obtain a time series data set.

[0026] Firstly, the controlled variable (CV) in the industrial process refers to the industrial process data that needs to be analyzed, for example, temperature, pressure, hardness, flow, concentration, purity, etc. can be used as one of the controlled variables. Further, the time series is a set of observation data arranged in chronological order, and the core feature is that the data change is strongly related to time, which is mainly used to analyze historical rules, identify fluctuation reasons and predict future trends. The collection method can directly call the historical data in the preset time period through the control software used in the specific industrial field.

[0027] As a feasible implementation manner, S101 can collect the historical data of the controlled variable in the recent period through the data source interface of the industrial prediction software to obtain the time series data set.

[0028] S102, decomposing the time series data set by a time series decomposition algorithm to obtain three types of independent components.

[0029] The time series decomposition algorithm (Seasonal-Trend decomposition using Loess, STL) is an analysis method for decomposing time series data into multiple components (such as long-term trend, seasonal / periodic fluctuation and residual error). The core goal is to reveal the pattern relationship behind the data and support prediction. Among them, the three types of independent components include the trend component representing the trend change, the periodic component representing the fluctuation frequency within the period, and the residual error component representing the noise intensity, and the trend component, the periodic component and the residual error component are independent of each other.

[0030] ​Further, the trend component is also referred to as a trend term, and reflects a change trend of the industrial process data in a preset time period, such as a slow decrease in pressure data caused by equipment aging. The periodic component is also referred to as a periodic term, and reflects periodic fluctuations of the industrial process data, such as a periodic change in flow data caused by raw material replenishment every few hours; the residual component is also referred to as a residual term, and is only a direct reflection from the data level, and reflects random noise in the industrial process data, such as sensor measurement errors. The independent components in each component are independent of each other, that is, the components in three dimensions have no influence relationship with each other.

[0031] In S103, feature extraction is respectively performed on the three types of independent components, and a filtering algorithm is matched and target filtering parameters of the filtering algorithm are determined based on a result of the feature extraction.

[0032] In S103, feature extraction is respectively performed on the trend term, the periodic term, and the residual term, to obtain feature parameters. The feature parameters can specifically include an average value and a change rate of the trend term, a main period (i.e., a period with the highest fluctuation frequency) and a period amplitude (a difference between a maximum value and a minimum value of the periodic term) of the periodic term, and a standard deviation of the residual term. These features can better reflect changes of the respective components. For example, the greater the change rate of the trend term, the steeper the trend, and the greater the standard deviation of the residual term, the more serious the noise. After the feature extraction, a corresponding filtering algorithm is calculated and matched based on the result of the feature extraction, and target filtering parameters corresponding to the filtering algorithm are determined, so that the industrial prediction software performs filtering processing based on the target filtering parameters.

[0033] In this embodiment, first, historical data of a controlled variable in an industrial process in a preset time period is collected to obtain a time series data set. Then, the time series data set is decomposed by a time series decomposition algorithm to obtain three types of independent components. The three types of independent components include a trend component representing a trend change, a periodic component representing a fluctuation frequency in a period, and a residual component representing a noise intensity. The trend component, the periodic component, and the residual component are independent of each other. Feature extraction is respectively performed on the three types of independent components, and a filtering algorithm is matched and target filtering parameters of the filtering algorithm are determined based on a result of the feature extraction.

[0034] Compared with related technologies, in this embodiment, historical data in an industrial process is first arranged in time sequence to form a time series data set, and then the time series data set is decomposed into three types of independent components by a time series decomposition algorithm. The complex characteristics of the industrial process data are analyzed from three different dimensions of a change trend, a fluctuation in a period, and specific residual data, to solve the problem of "no clear rules for parameter adjustment and dependence on experience" in traditional filtering. Finally, feature extraction is performed based on the three types of independent components, and a filtering algorithm is matched and target filtering parameters of the filtering algorithm are determined based on a result of the feature extraction, to improve filtering precision and enhance adaptability of industrial prediction control.

[0035] Optionally, feature extraction is performed on the three types of independent components respectively, and a matching filter algorithm is determined based on the feature extraction result and target filter parameters of the filter algorithm are determined, including: feature extraction is performed on the three types of independent components respectively to obtain feature vectors corresponding to the time series data set; a noise-containing coefficient is calculated based on the feature vectors; the noise-containing coefficient is compared with a first preset threshold, and in the case that the noise-containing coefficient is greater than the first preset threshold, it is determined that the filter algorithm is a moving average filter algorithm; target filter parameters of the moving average filter algorithm are calculated based on the three types of independent components.

[0036] In the embodiment, feature extraction is first performed on the three types of independent components to form feature vectors and a noise-containing coefficient is calculated based on the feature vectors. Then, in the case that the noise-containing coefficient is greater than a first preset threshold, it is determined that the filter algorithm is a moving average filter algorithm, and target filter parameters of the moving average filter algorithm are calculated.

[0037] Firstly, the industrial predictive control algorithm itself has certain robustness and anti-interference ability. When the noise intensity of the controlled variable is not large, any filter strategy does not need to be adopted, and good control effect can be achieved. In order to judge whether the data needs to be filtered, the noise-containing coefficient is first created as a filter necessity judgment index in the embodiment, and a critical threshold (a first preset threshold) is set to avoid data distortion caused by "forced filtering regardless of noise intensity" in part of the related technology or control precision reduction caused by "missing necessary filtering", and the filter necessity and the robustness of the control algorithm are considered.

[0038] Optionally, target filter parameters of the moving average filter algorithm are calculated based on the three types of independent components, including: initial moving average filter parameters are calculated based on the three types of independent components; the time series data set is filtered by using the initial moving average filter parameters to obtain a first filter result; a filter evaluation index is calculated according to the first filter result, and it is judged whether the filter evaluation index meets a preset requirement; in the case that the filter evaluation index does not meet the preset requirement, the initial moving average filter parameters are iterated according to a first preset iteration formula; the time series data set is filtered by using the iterated moving average filter parameters, and a corresponding filter evaluation index is calculated, until the filter evaluation index meets the preset requirement, and the moving average filter parameters after this iteration are taken as the target filter parameters of the moving average filter algorithm.

[0039] In the embodiment, the process of how to calculate the target filtering parameter of the sliding average filtering algorithm is explained. Firstly, the initial sliding average filtering parameter is calculated based on the three types of independent components. In specific practical applications, the initial sliding average filtering parameter is also the initial sliding window length. Then, the time series data set is processed by sliding average filtering based on the initial sliding window length to obtain the first filtering result. The filtering result is essentially industrial process data, and the first filtering result refers to the result after filtering processing by the sliding average filtering algorithm, which is used to distinguish the second filtering result obtained after filtering processing by other filtering algorithms.

[0040] After filtering processing, the filtering evaluation index is calculated according to the first filtering result, and it is judged whether the filtering evaluation index meets the preset requirement. The filtering evaluation index specifically includes information retention degree and noise suppression degree. The role of the information retention degree is to judge whether there is excessive filtering, which leads to excessive denoising of the original data and causes distortion. The role of the noise suppression degree is to judge the suppression effect on noise.

[0041] Further, if the information retention degree and the noise suppression degree do not meet the preset requirement, the initial sliding window length is iterated according to the first iteration formula set in advance. The time series data set is processed by sliding average filtering based on the iterated sliding window length, and the corresponding filtering evaluation index is calculated until the filtering evaluation index meets the preset requirement. The sliding window length obtained by this iteration is taken as the target filtering parameter of the sliding average filtering algorithm, and thus the target filtering parameter for the control of the industrial prediction software is obtained. The specific formula will be explained in subsequent embodiments.

[0042] In the embodiment, the information retention degree and the noise suppression degree are constructed to quantify the filtering effect, avoiding the one-sidedness of the single-index evaluation in the prior art, such as only paying attention to noise suppression and ignoring the loss of original data information, so as to ensure that the filtered data neither has excessive noise remaining nor loses key industrial information.

[0043] Optionally, in the case that the iteration number reaches the maximum and the filtering evaluation index corresponding to the first filtering result after iteration still does not meet the preset requirement, the method further includes: clearing the iteration number, determining the filtering algorithm as a first-order exponential filtering algorithm, and calculating the target filtering parameter of the first-order exponential filtering algorithm based on the three types of independent components.

[0044] The target filtering parameter of the first-order exponential filtering algorithm is calculated based on the three types of independent components, including: calculating an initial first-order exponential filtering parameter based on the three types of independent components; filtering the time series dataset using the initial first-order exponential filtering parameter to obtain a second filtering result; calculating a filtering evaluation index according to the second filtering result, and determining whether the filtering evaluation meets a preset requirement; in the case that the filtering evaluation does not meet the preset requirement, iteratively calculating the initial first-order exponential filtering parameter according to a second preset iteration formula; performing first-order exponential filtering on the time series dataset based on the first-order exponential filtering parameter after iteration, and calculating the corresponding filtering evaluation index until the filtering evaluation index meets the preset requirement, and taking the first-order exponential filtering parameter after this iteration as the target filtering parameter of the first-order exponential filtering algorithm.

[0045] In the case that the iteration number of the moving average filtering algorithm reaches the maximum and the filtering evaluation index corresponding to the first filtering result after iteration still does not meet the preset requirement, a subsequent filtering scheme is further proposed in the embodiment. The specific steps are similar to the steps of calculating the target filtering parameter according to the moving average filtering algorithm, which will not be described here again, and only the technical effects will be described. It should be noted that the present embodiment is not for all filtering algorithms, but for two specific filtering algorithms, and a more comprehensive and more general adaptive filtering method based on time series decomposition for industrial predictive control is proposed. The moving average filtering algorithm is preferentially selected in the present embodiment because of its low distortion advantage, and the first-order exponential filtering algorithm is switched to only when the moving average filtering algorithm reaches the maximum iteration number but still does not meet the requirement, forming a strategy of "low distortion priority and high adaptability backup". Further, the "parameter adjustment formula based on information retention degree and noise suppression degree" is designed for the two types of filtering respectively, realizing dynamic and directional optimization of the parameters, thereby solving the problem of "no clear rules for parameter adjustment and relying on experience" in traditional filtering, and improving the applicability and generality.

[0046] Optionally, the filtering evaluation index includes an information retention degree and a noise suppression degree.

[0047] According to the filtering result, the filtering evaluation index is calculated, and it is determined whether the filtering evaluation index meets a preset requirement, including: calculating a correlation coefficient of the filtering result relative to the trend component and the periodic component to obtain an information retention degree; calculating a residual term of the filtering result and a standard deviation of the residual term, and calculating a noise suppression degree based on the standard deviation of the residual term of the filtering result and the standard deviation of the residual component; in the case that the information retention degree is greater than or equal to a second preset threshold and the noise suppression degree is less than or equal to a third preset threshold, it is determined that the preset requirement is met; wherein the filtering result includes the first filtering result or the second filtering result.

[0048] In the embodiment, how to calculate the filtering evaluation index, that is, how to calculate the information retention degree and the noise suppression degree is explained. Whether the filtering algorithm selects the sliding average filtering algorithm or the first-order exponential filtering algorithm, the way of calculating the filtering evaluation index is the same, and the information retention degree and the noise suppression degree are determined based on the three types of independent components. And it is limited that in the case that the information retention degree is greater than or equal to the second preset threshold, and the noise suppression degree is less than or equal to the third preset threshold, it is determined that the preset requirement is met, so that the obtained filtering parameter is more accurate, and the effect of avoiding both "excessive filtering leading to information loss" and "insufficient filtering leading to noise residue" is realized.

[0049] Optionally, the time series dataset is decomposed by a time series decomposition algorithm to obtain the three types of independent components, including: initializing a cycle window parameter, a trend window parameter and an inner loop iteration number; initializing the cycle component as zero, and calculating an initial residual based on the initialized cycle component and the time series dataset; performing inner loop processing on the initial residual to output an inner loop residual after inner loop processing; performing outer loop processing on the inner loop residual after inner loop processing to obtain an outer loop residual; judging whether the outer loop residual converges; if not, repeating the outer loop processing; if yes, outputting a final trend component and a cycle component, and calculating a residual component based on the time series dataset, the final trend component and the cycle component.

[0050] Further, the inner loop processing includes: based on the initial residual, smoothing processing is performed on the initial trend item by using the Loess algorithm and the trend window parameter to obtain an initial trend item; the inner loop residual is updated according to the initial trend item and the initial residual until the inner loop iteration number is reached, and the current trend component is output as the inner loop residual; the outer loop processing includes: smoothing processing is performed on the inner loop residual output by the inner loop processing by using the Loess algorithm and the cycle window parameter, and the cycle component is updated; and the outer loop residual is calculated according to the smoothed inner loop residual and the updated cycle component.

[0051] In the embodiment, how to decompose the time series dataset by the STL algorithm to obtain the three types of independent components is explained, which can be combined with the above-mentioned embodiments. Figure 2The process is divided into three steps of initial setting, inner loop and outer loop. The initial setting includes initializing the period window parameter, the trend window parameter and the number of inner loop iterations; initializing the period component to zero, and calculating the initial residual based on the initialized period component and the time series data set. Then the initial residual is subjected to inner loop and outer loop respectively, and three types of independent components are calculated. By using the STL algorithm, the analysis and decoupling of industrial process data are realized, and the problem of data distortion (overly strong filtering) or noise residue (overly weak filtering) caused by directly calculating filtering parameters based on original data without distinguishing the filtering needs of different components (such as retaining trend items, strongly suppressing noise items, and matching period characteristics) is solved.

[0052] Optionally, before the time series data set is decomposed by the time series decomposition algorithm to obtain three types of independent components, it further includes: preprocessing the time series data set; preprocessing the time series data set, including: removing outliers in the time series data set by the 3-sigma rule; and completing missing elements in the time series data set by linear interpolation.

[0053] In this embodiment, the step of preprocessing the time series data set is also proposed. Specifically, the 3-sigma rule is used to remove obvious outliers (such as extreme values beyond the normal distribution interval) in the time series data set, and the linear interpolation method is used to complete missing values due to communication interruption and other reasons, so as to obtain the preprocessed time series and improve the data quality.

[0054] Next, combined with S201-S206, a specific implementation is proposed to uniformly explain the formulas, preset requirements, threshold values and the like mentioned in the above technical solution of the industrial predictive control adaptive filtering method based on time series decomposition. S201-S206 is not shown in the attached drawings separately, and specifically includes:

[0055] S201: obtaining a time series data set

[0056] Through the data source interface (such as OPCUA, OPCDA, Modbus protocol) of the industrial predictive control software, the historical data of the expected controlled variable in the recent period is collected to obtain the original time series data set D={d1,d2,......,d k ,.....,d n}, where k is the sampling time, n is the total number of samples, d k is the industrial process data (such as temperature, liquid level, pressure, etc.) at time k.

[0057] Further, the dataset D is preprocessed: the 3-sigma criterion is used to eliminate obvious outliers (such as extreme values beyond the normal distribution interval), and the linear interpolation method is used to complete the missing values due to communication interruption and other reasons, thereby obtaining the preprocessed time series X = {x1, x2,..., x k ,.....,x n}.

[0058] S202: Decompose the time series dataset

[0059] The preprocessed time series X is decomposed by using the STL algorithm to obtain three types of independent components, specifically including:

[0060] Trend term T = {t1, t2,..., t k ,.....,t n}: reflecting the long-term trend of industrial process data;

[0061] Periodic term S = {s1, s2,..., s k ,.....,s n}: reflecting the periodic fluctuations of industrial process data;

[0062] Residual term R = {r1, r2,..., r k ,.....,r n}: reflecting random noise in industrial data;

[0063] Wherein, the three components satisfy X = T + S + R, and they are independent of each other.

[0064] The specific steps of the STL algorithm are as follows:

[0065] First, perform STL parameter initialization, which requires setting three core parameters: season window (also known as period window) S, for example, 12 sampling points, trend window T, for example, 25 sampling points, and iteration number L, for example, 15 times. After completing parameter setting, initialize the periodic term S0 to 0, where the initialized periodic term is the initialized periodic component, and calculate the initial residual R0 based on the initialized periodic term, and the calculation formula is R0 = X - S0.

[0066] Then enter the inner loop (trend smoothing) stage: first, use the Loess algorithm to smooth the initial residual R0 to obtain the initial trend term T0; then calculate the inner loop residual R1 by R1 = R0 - T0; then judge whether the number of inner loop times reaches the preset iteration number L = 15 times, if not, return to the starting point of the inner loop to continue smoothing and residual calculation, if yes, end the inner loop and enter the outer loop.

[0067] The outer loop stage takes the optimization of the periodic term as the core: first, the Loess algorithm is used to smooth the obtained inner loop residual R1, and the updated periodic term is S1; then the outer loop residual R2 is calculated according to R2=T-S1-T0. Then, it is judged whether the outer loop converges: the convergence can be judged by setting a preset convergence threshold, for example, if the residual fluctuation is greater than the preset convergence threshold, it means that it has not converged, and the periodic term and the residual need to be updated from the beginning of the outer loop; if the residual fluctuation is less than or equal to the preset convergence threshold, it means that it has converged, and the outer loop is ended. The preset convergence threshold here can be determined according to the actual application and the specific values of each independent component.

[0068] After the outer loop is ended, the final trend term T and the periodic term S are output, and the residual term R is calculated according to the residual term calculation formula R=X-T-S, and the residual term R is obtained and output, thus completing the execution process of the entire STL algorithm.

[0069] S203: Feature extraction

[0070] For the trend term T, the periodic term S and the residual term R obtained in S202, feature parameters are extracted respectively, and are combined to form a feature vector F. The feature parameters specifically include:

[0071] Trend term feature F T : The average value μ of the trend term data T is used to represent the average value and the change rate δ. T T The average value can be obtained by simply taking the average value, and δ T is calculated according to formula one:

[0072] (Formula one)

[0073] In formula one, t k is the trend term of the industrial process data at time k, which is taken from the trend term T={t1,t2,......,t k ,.....,t n}, and n is the total number of samples. Among them, the change rate reflects the speed of trend change, and the greater δ T , the steeper the trend; the smaller δ T , the more gentle the trend.

[0074] Periodic term feature F S : The main period P S of the periodic term S (i.e. the period with the highest fluctuation frequency) and the period amplitude A S (the difference between the maximum value and the minimum value of the periodic term) are calculated by fast Fourier transform.

[0075] Residual term feature F R : The standard deviation σ R of the residual term R is calculated. The standard deviation reflects the noise intensity. σ​R The greater, the more serious the noise; σ R The smaller, the less the noise.

[0076] The feature vector F of the time series data is constructed as F = [μ T , δ T , P S , A S , σ R ].

[0077] S204: Filtering necessity judgment of data

[0078] The noise coefficient Y is calculated by Formula Two:

[0079] Y = 3 + log( ) (Formula Two)

[0080] In this embodiment, the first preset threshold is 0. If Y≤0, it indicates that the proportion of the residual term in the data is very small, and the noise intensity can be accepted by the predictive control algorithm, which will not significantly affect the performance of the controller, so that no filtering strategy needs to be adopted.

[0081] At this time, the output result is: filtering method = "no filtering", sliding window length N = 0, first-order exponential filtering coefficient α = 0, and the algorithm process directly ends. If Y>0, it indicates that the residual term cannot be ignored, and the data needs to be filtered, entering S205. At the same time, the process of calculating the target filtering parameters can be combined with Figure 3 , and the processes of S205 and S206 below are referred to.

[0082] S205: For a specific filtering algorithm, the initial value of the filtering parameter is calculated by analyzing the formula

[0083] The calculation formulas of the initial sliding average filtering parameter N 初始 (also referred to as the initial value of the sliding window length of the sliding average filtering in actual application) and the initial first-order exponential filtering parameter α 初始 (also referred to as the initial value of the first-order exponential filtering in actual application) are Formula Three and Formula Four, respectively:

[0084] N = max(8 + ceil(log( )), 3) (Formula Three)

[0085] In Formula Three, the max function is a function of taking the maximum of the two, and the ceil function is a function of rounding up. The specific logic is: the stronger the noise, the larger σ R , the larger the sliding window length N to enhance the filtering effect; the weaker the noise, the smaller σ R , the smaller the sliding window length N to weaken the filtering effect.

[0086] (Formula Four)

[0087] The specific logic of Formula Four is: the stronger the noise, the larger the σ R , and the larger the first-order exponential filter coefficient α to enhance the filtering effect; the more obvious the long-term trend or periodic trend, the larger the δ T or A S , and the smaller the first-order exponential filter coefficient α to weaken the filtering effect, thereby preserving data information.

[0088] S206: Filter effect evaluation and parameter iterative optimization

[0089] S2061: Application of moving average filtering strategy

[0090] In industrial predictive control software, the moving average filtering strategy is preferred because this method causes less data distortion. The initial sliding window length N 初始 is substituted into the filter module of the industrial predictive control software to filter the original time series X after removing outliers, and the filtered data X 滤波 is obtained.

[0091] S2062: Evaluate the filtering effect through two indicators

[0092] Information retention I: Calculate the correlation coefficient of X 滤波 , trend term T, and periodic term S through weighted average. Specifically, as shown in Formula Five:

[0093] (Formula Five)

[0094] In Formula Five, the corr function is a statistical measure of the strength and direction of the linear relationship between two variables. Its return value is a value between 0 and 1. The closer I is to 1, the better the information retention. The initial value of the weighting coefficient w is 0.7, which can be dynamically adjusted. It is recommended to take a value in the range of 0.6-0.8.

[0095] Noise suppression degree V: First, calculate the residual term R 滤波 of X 滤波 , and then calculate the ratio of its standard deviation to the standard deviation of the original data residual term R. Specifically, as shown in Formula Six:

[0096] (Formula Six)

[0097] In Formula Six, the closer V is to 0, the better the noise suppression.

[0098] The threshold for the filtered data to meet the standard is I≥0.875 and V≤0.25, that is, in the embodiment, the second preset threshold is 0.875, and the third preset threshold is 0.25.

[0099] S2063: Iterative optimization

[0100] If X 滤波 meets the qualified standard, N 初始 is the final filtering parameter, that is, the target filtering parameter N 最终 ; if not, the filtering parameter needs to be adjusted according to the following criteria:

[0101] If I < 0.875 (too much data distortion): the sliding window length N needs to be reduced, and the iterative formula is N 新 =N-1;

[0102] If V > 0.25 (too much noise residue): the sliding window length N needs to be increased, and the iterative formula is N 新 =N+1;

[0103] Then, using the new parameter N 新 , repeat S2061 and S2062 until the threshold is met. If the maximum number of iterations C max is not reached when the condition is met, output the sliding window length N=N 最终 after this iteration, the first-order exponential filtering coefficient a=0, and the algorithm flow ends.

[0104] S2064: First-order exponential filtering strategy application

[0105] If the sliding average filtering strategy still cannot meet the threshold when the maximum number of iterations C max is reached, the first-order exponential filtering strategy needs to be used. At this time, the current iteration number is cleared, and the initial first-order exponential filtering coefficient a 初始 is substituted into the filtering module of the industrial prediction control software to filter the original time series X after removing outliers, and filtered data X 滤波 is obtained.

[0106] S2065: Iterative optimization

[0107] The evaluation index after applying filtering is calculated through S2062. If X 滤波 meets the qualified standard, a 初始 is the final filtering parameter a 最终 ; if not, the filtering parameter needs to be adjusted according to the following criteria:

[0108] If I < 0.875 (too much data distortion): reduce the filtering coefficient a, and the iterative formula is ;

[0109] If V > 0.25 (too much noise remains): increase the filter coefficient a, the iteration formula is ;

[0110] Then use the new parameter a new to calculate the filter evaluation index until the threshold value is met or the maximum iteration number C is reached max At this time, output: filter method = "first-order exponential filter", sliding window length N = 0, first-order exponential filter coefficient a = a 最终 , the algorithm flow ends.

[0111] Further, on the basis of the above embodiment, a more specific implementation is given in combination with a specific application scenario. Specifically, Figure 4 Some specific execution modules are also exemplarily given to facilitate understanding of the overall processing flow. Taking the application scenario of "rectifying column still temperature data filtering" as an example, the normal fluctuation range of the rectifying column still temperature is 237±4℃. The sampling frequency of the historical data is 30s, the data coverage period is 2 hours, and there are 240 data points in total. The goal is to determine the appropriate filtering method (no filtering / moving average filtering / first-order exponential filtering) and give the corresponding filtering parameters (sliding window length N or first-order exponential filter coefficient a).

[0112] S301: Data acquisition and preprocessing

[0113] The still temperature data is collected through the OPC UA protocol to obtain the original data set D. Then, 12 outliers are removed by the three-sigma criterion, and 4 missing values are completed by linear interpolation to obtain the preprocessed time series X.

[0114] S302: Time series decomposition

[0115] After STL decomposition of X, we get:

[0116] Trend term T: the temperature slowly rises from 235℃ to 241℃ (average value μ T = 237.6, change rate δ T = 0.05 / min);

[0117] Periodic term S: there is a small wave every 30 minutes (main period P S = 30min, amplitude A S = 2℃);

[0118] Residual term R: standard deviation σ R = 1.4℃.

[0119] S303: Extract feature vector

[0120] F = [μ T= 237.6, δ T = 0.05, P S = 30, A S = 2, σ R = 1.4]

[0121] S304: Filtering necessity judgment of data

[0122] The noise-containing coefficient of the data of the kettle temperature is calculated: Y = 3 + log(1.4 / 237.6) ≈ 0.77 > 0

[0123] This shows that the noise of the data cannot be ignored, and therefore the subsequent filtering step needs to be performed.

[0124] S305: Initial value calculation of filtering parameters

[0125] The initial value of the sliding window length of the sliding average filtering is calculated:

[0126] N 初始 = max(8 + ceil(log(1.4 / 237.6)), 3) = 6

[0127] The initial value of the filtering coefficient of the first-order exponential filtering is calculated:

[0128]

[0129] S306: Filtering effect evaluation and parameter iterative optimization

[0130] First, the sliding average filtering strategy is selected, and the maximum number of iterations C max is set to 5:

[0131] After filtering with the initial sliding window length N 初始 = 6, the information retention degree I = 0.815 (not meeting the threshold value) and the noise suppression degree V = 0.231 (meeting the threshold value) are obtained;

[0132] In the second iteration, the sliding window length needs to be reduced due to the low information retention degree. Taking N = 5, the information retention degree I = 0.843 (not meeting the threshold value) and the noise suppression degree V = 0.235 (meeting the threshold value) are obtained;

[0133] In the third iteration, the sliding window length is further reduced. Taking N = 4, the information retention degree I = 0.882 (meeting the threshold value) and the noise suppression degree V = 0.242 (meeting the threshold value) are obtained. At this time, the maximum number of iterations has not been exceeded, and therefore the final algorithm calculation result is output as: filtering method = "sliding average filtering", sliding window length N 最终 = 4, and first-order exponential filtering coefficient a = 0.

[0134] By means of the industrial prediction control adaptive filtering method based on time series decomposition provided by any of the above embodiments, at least the following technical effects are achieved:

[0135] (1) Industrial data preprocessing technology: The combination scheme of "three-sigma criterion to remove outliers + linear interpolation to complete missing values" is adopted to solve the problems of extreme values and data missing in industrial scenarios, so as to ensure the integrity and reliability of the time series after preprocessing, and provide a high-quality data basis for subsequent decomposition.

[0136] (2) Time series decomposition based on STL algorithm: The STL algorithm is used to accurately decompose the time series into three independent components: trend item (long-term change), periodic item (periodic fluctuation) and residual item (random noise). The core parameters of the STL algorithm (seasonal window, trend window, iteration number) and the execution logic of "inner loop trend smoothing (Loess algorithm) - outer loop periodic item optimization (residual fluctuation convergence judgment)" are determined, so as to effectively separate the data components.

[0137] (3) Feature extraction method of decomposition components: Differentiated feature indicators are designed for the three types of decomposition components: "average value + change rate" is used to represent the long-term change characteristics of the trend item, "main period + amplitude calculated by fast Fourier transform" is used to represent the fluctuation law of the periodic item, and "standard deviation" is used to represent the noise intensity of the residual item. Finally, a feature vector is formed by combining the above indicators, which provides accurate basis for filter parameter calculation.

[0138] (4) Quantitative judgment of filtering necessity: The "noise coefficient" is defined as a judgment index, and a critical threshold is set to avoid data distortion caused by "forced filtering regardless of noise intensity" and control accuracy decline caused by "missing necessary filtering" in the prior art, which balances the filtering necessity and the robustness of the control algorithm.

[0139] (5) A double-filter priority strategy and iterative optimization logic is proposed: "sliding average priority, first-order exponential backup", which prioritizes sliding average filtering (low distortion advantage) and switches to first-order exponential filtering only when it fails to meet the requirements after the maximum number of iterations. This strategy prioritizes low distortion and provides high adaptability as a backup. For the two types of filters, "parameter adjustment formula based on information retention and noise suppression" is designed to achieve dynamic and directional optimization of parameters, solving the problem of "no clear rules for parameter adjustment, relying on experience" in traditional filtering.

[0140] Further, as a specific implementation of the method shown in Figures 1 to 4 , the embodiment provides an industrial prediction control adaptive filtering device based on time series decomposition, as shown in Figure 5 , the device comprises an acquisition unit 501, a calculation unit 502 and a processing unit 503.

[0141] The collection unit 501 is configured to collect historical data of the controlled variable in the industrial process in a preset time period to obtain a time series data set;

[0142] The calculation unit 502 is configured to decompose the time series data set by a time series decomposition algorithm to obtain three types of independent components; wherein the three types of independent components include a trend component representing a trend change, a periodic component representing a fluctuation frequency in a period, and a residual component representing a noise intensity, and the trend component, the periodic component, and the residual component are independent of each other;

[0143] The processing unit 503 is configured to respectively extract features from the three types of independent components, and match a filtering algorithm and determine a target filtering parameter of the filtering algorithm based on the feature extraction result.

[0144] In a specific application scenario, the processing unit 503 is specifically further configured to respectively extract features from the three types of independent components to obtain a feature vector corresponding to the time series data set; calculate a noise-containing coefficient based on the feature vector; compare the noise-containing coefficient with a first preset threshold, and in a case where the noise-containing coefficient is greater than the first preset threshold, determine that the filtering algorithm is a moving average filtering algorithm; and calculate a target filtering parameter of the moving average filtering algorithm based on the three types of independent components.

[0145] In a specific application scenario, the processing unit 503 is specifically further configured to calculate an initial moving average filtering parameter based on the three types of independent components; perform filtering processing on the time series data set by using the initial moving average filtering parameter to obtain a first filtering result; calculate a filtering evaluation index according to the first filtering result, and determine whether the filtering evaluation index meets a preset requirement; in a case where the filtering evaluation index does not meet the preset requirement, iteratively update the initial moving average filtering parameter according to a first preset iteration formula; perform moving average filtering processing on the time series data set based on the iteratively updated moving average filtering parameter, and calculate a corresponding filtering evaluation index until the filtering evaluation index meets the preset requirement, and take the iteratively updated moving average filtering parameter as the target filtering parameter of the moving average filtering algorithm.

[0146] In a specific application scenario, the processing unit 503 is specifically further configured to clear the iteration number, determine that the filtering algorithm is a first-order exponential filtering algorithm, and calculate a target filtering parameter of the first-order exponential filtering algorithm based on the three types of independent components; wherein, based on the three types of independent components, calculating the target filtering parameter of the first-order exponential filtering algorithm includes: calculating an initial first-order exponential filtering parameter based on the three types of independent components; performing filtering processing on the time series data set using the initial first-order exponential filtering parameter to obtain a second filtering result; calculating a filtering evaluation index according to the second filtering result, and determining whether the filtering evaluation meets the preset requirement; in the case that the filtering evaluation does not meet the preset requirement, iterating the initial first-order exponential filtering parameter according to a second preset iteration formula; performing first-order exponential filtering processing on the time series data set based on the first-order exponential filtering parameter after iteration, and calculating the corresponding filtering evaluation index until the filtering evaluation index meets the preset requirement, and taking the first-order exponential filtering parameter after this iteration as the target filtering parameter of the first-order exponential filtering algorithm.

[0147] In a specific application scenario, the processing unit 503 is specifically further configured to calculate a correlation coefficient of the filtering result with respect to the trend component and the periodic component to obtain an information retention degree; calculate a residual term of the filtering result and a standard deviation of the residual term, and calculate a noise suppression degree based on the standard deviation of the residual term of the filtering result and the standard deviation of the residual component; in the case that the information retention degree is greater than or equal to a second preset threshold and the noise suppression degree is less than or equal to a third preset threshold, it is determined that the preset requirement is met; wherein, the filtering result includes the first filtering result or the second filtering result.

[0148] In a specific application scenario, the calculation unit 502 is specifically further configured to initialize a periodic window parameter, a trend window parameter and an inner loop iteration number; initialize the periodic component to zero, and calculate an initial residual based on the initialized periodic component and the time series data set; perform inner loop processing on the initial residual to output an inner loop residual after inner loop processing; perform outer loop processing on the inner loop residual after inner loop processing to obtain an outer loop residual; determine whether the outer loop residual converges; if not, repeat the outer loop processing; if it has converged, output the final trend component and the periodic component, and calculate the residual component based on the time series data set, the final trend component and the periodic component.

[0149] The inner loop includes smoothing based on the initial residual, using the Loess algorithm and a trend window parameter to obtain an initial trend item; updating the inner loop residual according to the initial trend item and the initial residual, until the number of inner loop iterations is reached, and outputting the current trend component and the inner loop residual; and the outer loop processing includes: smoothing the inner loop residual output by the inner loop processing using the Loess algorithm and a cycle window parameter, and updating the cycle component; and calculating the outer loop residual according to the smoothed inner loop residual and the updated cycle component.

[0150] In a specific application scenario, the calculation unit 502 is specifically further configured to pre-process the time series data set; the pre-processing of the time series data set includes: removing outliers in the time series data set through the 3-sigma criterion; and completing missing elements in the time series data set through linear interpolation.

[0151] It should be noted that other corresponding descriptions of the various functional units involved in the industrial prediction control adaptive filtering device based on time series decomposition provided in this embodiment can be referred to the corresponding descriptions in Figures 1 to 4 , which will not be described here.

[0152] Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.), and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0153] Through the above description of the implementation mode, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware platform, or through hardware. By applying the scheme of the present embodiment, compared with related technologies, the historical data in the industrial process is first arranged in time sequence to form a time series data set, and then the time series data set is decomposed into three types of independent components through a time series decomposition algorithm, the complex characteristics of the industrial process data are analyzed from three different dimensions of change trend, period internal fluctuation and specific residual data, and the problem of "no clear rules for parameter adjustment, relying on experience" in traditional filtering is solved. Finally, feature extraction is performed based on the three types of independent components, and a filtering algorithm is matched based on the feature extraction result and the target filtering parameter of the filtering algorithm is determined, thereby improving the filtering precision and enhancing the adaptability of the industrial prediction control.

[0154] In the description of the application, it should be understood that the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0155] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the skilled person in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0156] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can modify, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An industrial predictive control adaptive filtering method based on time series decomposition, characterized in that, The method comprises the following steps: Collecting historical data of a controlled variable in an industrial process within a preset time period to obtain a time series data set; The controlled variable includes temperature, pressure, hardness, flow, concentration or purity; Decomposing the time series data set by a time series decomposition algorithm to obtain three types of independent components; wherein the three types of independent components include a trend component representing trend changes, a periodic component representing fluctuation frequencies within a period, and a residual component representing noise intensities, and the trend component, the periodic component and the residual component are independent of each other; Respectively extracting features from the three types of independent components, and matching a filtering algorithm based on the feature extraction results and determining target filtering parameters of the filtering algorithm; The step of respectively extracting features from the three types of independent components and matching a filtering algorithm based on the feature extraction results and determining target filtering parameters of the filtering algorithm comprises: Respectively extracting features from the three types of independent components to obtain a feature vector corresponding to the time series data set; Calculating a noise-containing coefficient based on the feature vector; Comparing the noise-containing coefficient with a first preset threshold, and in the case that the noise-containing coefficient is greater than the first preset threshold, determining that the filtering algorithm is a moving average filtering algorithm; Based on the three types of independent components, calculating target filtering parameters of the moving average filtering algorithm; Wherein the noise-containing coefficient is calculated based on the feature vector by the following formula: Y = 3 + log( ) where Y is a noise factor, is a standard deviation of residual terms in the feature vector, is an average of trend term data in the feature vector.

2. The method of claim 1, wherein, The step of calculating target filtering parameters of the moving average filtering algorithm based on the three types of independent components comprises: Based on the three types of independent components, calculating an initial moving average filtering parameter; Using the initial moving average filtering parameter to perform filtering processing on the time series data set to obtain a first filtering result; According to the first filtering result, calculating a filtering evaluation index and determining whether the filtering evaluation index meets a preset requirement; In the case that the filtering evaluation index does not meet the preset requirement, iterating the initial moving average filtering parameter according to a first preset iteration formula; Based on the iterated moving average filtering parameter, performing moving average filtering processing on the time series data set and calculating a corresponding filtering evaluation index until the filtering evaluation index meets the preset requirement, and taking the iterated moving average filtering parameter as the target filtering parameter of the moving average filtering algorithm.

3. The method of claim 2, wherein, In the case that the number of iterations reaches a maximum and the filtering evaluation index corresponding to the first filtering result after iteration still does not meet the preset requirement, the method further comprises: Clearing the number of iterations, determining that the filtering algorithm is a first-order exponential filtering algorithm, and calculating target filtering parameters of the first-order exponential filtering algorithm based on the three types of independent components; Wherein the step of calculating target filtering parameters of the first-order exponential filtering algorithm based on the three types of independent components comprises: Based on the three types of independent components, calculating an initial first-order exponential filtering parameter; Using the initial first-order exponential filtering parameter to perform filtering processing on the time series data set to obtain a second filtering result; According to the second filtering result, calculating a filtering evaluation index and determining whether the filtering evaluation meets the preset requirement; In a case where the filtering evaluation does not satisfy the preset requirement, the initial first-order exponential filtering parameter is iterated according to a preset second iteration formula; The time series data set is processed by first-order exponential filtering based on the first-order exponential filtering parameter after iteration, and a corresponding filtering evaluation index is calculated until the filtering evaluation index satisfies the preset requirement, and the first-order exponential filtering parameter after this iteration is taken as a target filtering parameter of the first-order exponential filtering algorithm.

4. The method according to claim 2 or 3, characterized in that, The filtering evaluation index includes information retention and noise suppression; According to the filtering result, a filtering evaluation index is calculated, and it is judged whether the filtering evaluation index satisfies a preset requirement, including: A correlation coefficient of the filtering result relative to the trend component and the periodic component is calculated to obtain the information retention; A residual term of the filtering result and a standard deviation of the residual term are calculated, and the noise suppression is calculated based on the standard deviation of the residual term of the filtering result and the standard deviation of the residual component; In a case where the information retention is greater than or equal to a second preset threshold and the noise suppression is less than or equal to a third preset threshold, it is determined that the preset requirement is satisfied; wherein the filtering result includes a first filtering result or a second filtering result.

5. The method of claim 1, wherein, The time series data set is decomposed by a time series decomposition algorithm to obtain three types of independent components, including: Periodic window parameters, trend window parameters and inner loop iteration numbers are initialized; The periodic component is initialized to zero, and an initial residual is calculated based on the initialized periodic component and the time series data set; The initial residual is processed by an inner loop to output an inner loop residual after the inner loop processing; The inner loop residual after the inner loop processing is processed by an outer loop to obtain an outer loop residual; It is judged whether the outer loop residual converges; If not, the outer loop processing is repeated; If it has converged, the final trend component and periodic component are output, and the residual component is calculated based on the time series data set, the final trend component and the periodic component.

6. The method of claim 5, wherein, The inner loop processing includes: An initial trend term is obtained by smoothing the initial residual using a Loess algorithm and the trend window parameters; The inner loop residual is updated according to the initial trend term and the initial residual until the inner loop iteration number is reached, and the current trend component and the inner loop residual are output; The outer loop processing includes: The inner loop residual output by the inner loop processing is smoothed using a Loess algorithm and the periodic window parameters, and the periodic component is updated; The outer loop residual is calculated based on the smoothed inner loop residual and the updated periodic component.

7. The method of claim 1, wherein, Before the time series data set is decomposed by the time series decomposition algorithm to obtain three types of independent components, it further includes: The time series data set is preprocessed; The preprocessing of the time series data set includes: Abnormal values in the time series data set are removed by 3-sigma criterion; Missing elements in the time series data set are completed by linear interpolation.

8. An industrial predictive control adaptive filtering device based on time series decomposition, characterized by, The method comprises the following steps: a collection unit configured to collect historical data of a controlled variable in an industrial process within a preset time period to obtain a time series data set; the controlled variable includes temperature, pressure, hardness, flow, concentration or purity; a calculation unit configured to decompose the time series data set by a time series decomposition algorithm to obtain three types of independent components; wherein the three types of independent components include a trend component representing a trend change, a periodic component representing a fluctuation frequency within a period, and a residual component representing a noise intensity, and the trend component, the periodic component and the residual component are independent of each other; a processing unit configured to respectively extract features from the three types of independent components, and match a filtering algorithm and determine target filtering parameters of the filtering algorithm based on the feature extraction results; the processing unit respectively extracts features from the three types of independent components, and matches a filtering algorithm and determines target filtering parameters of the filtering algorithm based on the feature extraction results, which comprises: extracting features from the three types of independent components to obtain a feature vector corresponding to the time series data set; calculating a noise-containing coefficient based on the feature vector; comparing the noise-containing coefficient with a first preset threshold, and determining that the filtering algorithm is a sliding average filtering algorithm if the noise-containing coefficient is greater than the first preset threshold; calculating target filtering parameters of the sliding average filtering algorithm based on the three types of independent components; wherein the noise-containing coefficient is calculated based on the feature vector by the following formula: Y = 3 + log( ) where Y is a noise factor, is the mean of the trend term data in the feature vector. R , is the mean of the trend term data in the feature vector.

Images