Chemical engineering time series data early warning method and system based on trend variation analysis

Abstract

The embodiment of the invention provides a chemical engineering time series data early warning method and system based on trend variation analysis, and belongs to the technical field of early warning. The method comprises the following steps: collecting chemical engineering time sequence data, and preprocessing the chemical engineering time sequence data; on the basis of a pre-constructed trend prediction model, trend prediction is carried out on the preprocessed chemical engineering time series data, and predicted time series data of a future preset time length is obtained; performing correlation analysis on the predicted time sequence data and expected time sequence data of the corresponding time section through a trend variation analysis model to obtain a trend difference; and determining and executing an early warning scheme based on the trend difference. According to the scheme of the invention, the problems of poor early warning instantaneity and low early warning precision of the existing early warning scheme are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of early warning, and particularly to a chemical process data early warning method based on trend variation analysis and a chemical process data early warning system based on trend variation analysis. Background Art

[0002] In the field of chemical production, safety production accidents are prone to occur, threatening personal and property safety. The state stipulates that abnormal conditions occurring in the safety operation management of chemical installations should be monitored and early warned, and online monitoring means should be adopted to give early warning of abnormal conditions in advance to avoid accidents. Therefore, establishing a set of accurate and rapid monitoring and early warning is of great significance for improving the intrinsic safety of enterprises and the safety production supervision level of regulatory departments.

[0003] Currently, in the dynamic monitoring and early warning in the chemical field, the method of safety risk level assessment is mainly adopted. By constructing index contents such as safety production management indicators, the overall risk value of the enterprise is calculated, and based on this, the risk level of the enterprise is divided and key supervision is carried out. Similarly, the research group of Liaoning Chemical Industry also chose the method of constructing an early warning evaluation model by establishing an index system. They used an analysis method combining analytic hierarchy process (AHP) and fuzzy comprehensive evaluation in the research to grade the risks of enterprises. Such monitoring and early warning methods are based on enterprises, with a small change index of risks, poor early warning timeliness, and unable to locate the devices where accidents may occur, making it difficult to make a rapid response to possible accidents.

[0004] Aiming at the problems of poor early warning timeliness and low early warning accuracy existing in the existing early warning schemes, a new early warning scheme applied to the chemical field needs to be proposed. Summary of the Invention

[0005] The purpose of the embodiments of the present invention is to provide a chemical process data early warning method and system based on trend variation analysis to at least solve the problems of poor early warning timeliness and low early warning accuracy existing in the existing early warning schemes.

[0006] To achieve the above purpose, the first aspect of the present invention provides a chemical process data early warning method based on trend variation analysis. The method includes: collecting chemical process data and preprocessing the chemical process data; based on a pre-constructed trend prediction model, performing trend prediction on the preprocessed chemical process data to obtain predicted time series data for a future predetermined time length; performing correlation analysis on the predicted time series data and the expected time series data of the corresponding time period through a trend variation analysis model to obtain a trend difference; determining and executing an early warning scheme based on the trend difference.

[0007] Optionally, the preprocessing of the chemical process data includes: outlier cleaning processing and missing value processing.

[0008] Optionally, the trend prediction model is an improved ARIMA model or a pre-trained neural network model. Optionally, the method further includes: constructing an improved ARIMA model, including: determining an autoregressive algorithm, expressed as:

[0009]

[0010] where y t is the current value; μ is the constant term; p is the order; y i is the autocorrelation coefficient; ε t is the error; determining the moving average term of the moving average process and generating a q-order autoregressive process algorithm of the moving average model, expressed as:

[0011]

[0012] Generating an ARIMA model based on the autoregressive algorithm and the q-order autoregressive process algorithm of the moving average model.

[0013] Optionally, the ARIMA model is expressed as:

[0014]

[0015] Optionally, the method further includes: pre-training a neural network model, including: collecting chemical engineering historical parameters and generating training samples based on the chemical engineering historical parameters; performing model training in a pre-constructed neural network based on the training samples to obtain a trend prediction model; wherein, the pre-constructed neural network includes: a forget gate, an input gate, and an output gate.

[0016] Optionally, the step of performing correlation analysis on the predicted time series data and the expected time series data of the corresponding time period through the trend variation analysis model to obtain a trend difference includes: obtaining expected time series data within a preset prediction time interval based on the expected operating conditions of the target chemical engineering equipment; calculating a correlation coefficient between the predicted time series data and the expected time series data based on the trend variation analysis model; using the correlation coefficient as the trend difference between the predicted time series data and the expected time series data.

[0017] Optionally, the correlation coefficient is a Pearson correlation coefficient, a Spearman correlation coefficient, or a Kendall correlation coefficient.

[0018] Optionally, when the correlation coefficient is a Pearson correlation coefficient, the calculation rule of the correlation coefficient between the predicted time series data and the expected time series data is:

[0019]

[0020] Wherein, X is the predicted time-series data; Y is the expected time-series data; ρ X,Y is the correlation coefficient between the predicted time-series data and the expected time-series data; cov(X, Y) is the covariance between the predicted time-series data and the expected time-series data; σ X is the standard deviation of the predicted time-series data; σ Y is the standard deviation of the expected time-series data.

[0021] Optionally, determining and executing an early warning plan based on the trend difference includes: when the trend difference is greater than a preset trend difference threshold, determining that the probability of an abnormality occurring in the target chemical equipment within a predetermined future time length reaches the early warning standard; then generating an early warning instruction and generating an alarm message based on the early warning instruction.

[0022] The second aspect of the present invention provides a chemical process time-series data early warning system based on trend variation analysis. The system includes: a collection unit for collecting chemical process time-series data and preprocessing the chemical process time-series data; a prediction unit for performing trend prediction on the preprocessed chemical process time-series data based on a pre-constructed trend prediction model to obtain predicted time-series data for a predetermined future time length; a comparison unit for performing correlation analysis on the predicted time-series data and the expected time-series data in the corresponding time period through a trend variation analysis model to obtain a trend difference; and an early warning unit for determining and executing an early warning plan based on the trend difference.

[0023] The third aspect of the present invention provides a computer-readable storage medium, on which instructions are stored, and when the instructions are run on a computer, the computer is caused to execute the above-mentioned chemical process time-series data early warning method based on trend variation analysis.

[0024] The fourth aspect of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned chemical process time-series data early warning method based on trend variation analysis is implemented.

[0025] Through the above technical solutions, the solution of the present invention processes the outlier of the real data of chemical parameters, and based on the real chemical data without outliers, uses the time series analysis method to train a time series prediction model to predict the parameter value fluctuation in a certain time window in the future. Based on the real chemical parameter data without outliers as the theoretical trend, the received new real-time trend of the parameter is used as the real trend. The real trend and the theoretical trend are analyzed for trend comparison by means such as correlation analysis to identify whether the fluctuation of the real parameter data has generated an abnormal trend. If it is determined to be an abnormal trend, an early warning is generated. The solution of the present invention automatically reports the warning information according to the warning algorithm, provides early fault warning to enterprises and the government, so that they have sufficient time to check and process, and avoid the occurrence of accidents as much as possible in advance.

[0026] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific embodiments section. Brief Description of the Drawings

[0027] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0028] Figure 1 is a flowchart of the steps of a method for warning chemical time series data based on trend variation analysis provided by an embodiment of the present invention;

[0029] Figure 2 is a system structure diagram of a system for warning chemical time series data based on trend variation analysis provided by an embodiment of the present invention. Detailed Description of the Embodiments

[0030] The following will describe the specific embodiments of the present invention in detail with reference to the drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0031] In the field of chemical production, safety production accidents are prone to occur, threatening personal and property safety. The state stipulates that abnormal working conditions in the safety operation management of chemical installations should be monitored and warned, and online monitoring means should be adopted to give early warning of abnormal working conditions to avoid accidents. Therefore, establishing a set of accurate and rapid monitoring and warning is of great significance for improving the intrinsic safety of enterprises and the safety production supervision level of regulatory departments.

[0032] At present, in the dynamic monitoring and early warning in the chemical industry, the method of safety risk level assessment is mainly adopted. By constructing index contents such as safety production management indicators, the overall risk value of an enterprise is calculated, and based on this, the risk level of the enterprise is divided and key supervision is carried out. Similarly, the research group of Liaoning Chemical Industry also chose to build an early warning evaluation model by establishing an index system. They used an analysis method combining analytic hierarchy process (AHP) and fuzzy comprehensive evaluation in their research to classify the risks of enterprises. Such monitoring and early warning methods are based on enterprises, with a small change index of risks, poor timeliness of early warning, and unable to locate the device where an accident may occur, making it difficult to make a quick response to possible accidents.

[0033] Another type of strategy uses data for early warning analysis and conducts algorithm research based on the time series characteristics of historical data. Yao Yuman et al. reviewed the combined application of data-driven methods in chemical industry early warning research, focused on investigating data-driven fault diagnosis methods, and summarized the technical route for the development of early warning towards process fault prediction and diagnosis. Following the technical route of abnormal condition monitoring and early warning, Hu Jin proposed an ultra-early monitoring and early warning method for abnormal conditions combining improved particle swarm optimization (PSO) algorithm and least squares support vector machine (LSSVM). In the case analysis of ultra-early monitoring and early warning of overpressure abnormal conditions in a propane tower, it can accurately predict the process data within the next 500 s and issue an abnormal alarm 40 s earlier than the DCS system. One year later, Dong et al. used a method combining radial basis function and recursive algorithm for non-smooth and non-linear time series prediction in chemical processes, with high accuracy. In 2022, Wang Yingying et al. proposed to use LSTM (long short-term memory network) to warn of the temperature of underwater electronic modules. However, only the prediction of continuous time series data was achieved without studying the early warning model, and it does not have the ability of early warning. In the same year, Chen Liang first combined the deep learning time series prediction model with the fuzzy mathematics risk assessment model and proposed a set of online monitoring and dynamic risk early warning model for process parameters.

[0034] Aiming at the problems of poor warning timeliness and low warning accuracy existing in the existing warning schemes, the solution of the present invention proposes a warning method for chemical process time series data based on trend variation analysis. The solution of the present invention performs outlier processing on the real data of chemical parameters, and based on the chemical real data without outliers, uses time series analysis methods to train a time series prediction model to predict the parameter value fluctuation situation in a certain time window in the future. Based on the real chemical parameter data without outliers as the theoretical trend, the newly received real-time trend of the parameter is used as the real trend. The real trend and the theoretical trend are subjected to trend comparison analysis by means such as correlation analysis to identify whether the abnormal trend is generated by the fluctuation of the real data of the parameter. If it is determined as an abnormal trend, an early warning is generated. The solution of the present invention automatically reports warning information according to the warning algorithm, provides early fault warning to enterprises and the government, so that they have sufficient time to check and process, and avoid the occurrence of accidents as much as possible before the event.

[0035] Figure 1 It is the method flow chart of the warning method for chemical process time series data based on trend variation analysis provided by an embodiment of the present invention. As Figure 1 shown, the embodiment of the present invention provides a warning method for chemical process time series data based on trend variation analysis, and the method includes:

[0036] Step S10: Collect chemical process time series data and preprocess the chemical process time series data.

[0037] Specifically, the preprocessing of the chemical process time series data includes: outlier cleaning processing and missing value processing.

[0038] In the embodiment of the present invention, the solution of the present invention can automatically and accurately predict the change trend of the parameter value, and automatically report warning information according to the warning algorithm, provide early fault warning to enterprises and the government, so that they have sufficient time to check and process, and avoid the occurrence of accidents as much as possible before the event. To achieve the above purpose, the solution of the present invention adopts an overall idea based on big data for the real-time parameters in the chemical production field, proposes a trend prediction method including traditional machine learning methods and deep learning methods, proposes to use trend variation analysis methods with correlation analysis as the mainstream means, and innovatively proposes a time series trend prediction method for chemical parameters and a warning method based on trends, realizing an accident early warning technology based on artificial intelligence.

[0039] Embodiment 1:

[0040] Adopt the method of directly connecting to the time series database, use the real-time data collected from the chemical engineering scenario in the application, and directly access the dynamic database through the interface. Establish a time series of parameter values with the parameter collection time as the index. In the preprocessing of the data, clean the outliers according to the 3σ principle, and then use the Lagrange interpolation method or the linear interpolation method to fill in the missing values. The input of the data collection and processing process is the time series data of the parameters within a period of length l, and the output is the anomaly-free smooth time series data for this period of length l.

[0041] Step S20: Based on the pre-constructed trend prediction model, perform trend prediction on the preprocessed chemical engineering time series data to obtain the predicted time series data for a future predetermined time length.

[0042] Specifically, the trend prediction model is an improved ARIMA model or a pre-trained neural network model.

[0043] Specifically, the method further includes: constructing an improved ARIMA model, including: determining the autoregressive algorithm, expressed as:

[0044]

[0045] where y t is the current value; μ is the constant term; p is the order; y i is the autocorrelation coefficient; ε t is the error; determine the moving average term of the moving average process, and generate the q-order autoregressive process algorithm of the moving average model, expressed as:

[0046]

[0047] Based on the autoregressive algorithm and the q-order autoregressive process algorithm of the moving average model, generate the ARIMA model.

[0048] Example two:

[0049] It is realized by establishing a time series prediction model. There are two mainstream strategies for time series prediction models. One is the traditional time series prediction model represented by the ARIMA model, and the other is the neural network-based method represented by the RNN model and the LSTM model.

[0050] In machine learning methods, we define the parameter p as the autoregressive term of the autoregressive process, q as the moving average term of the moving average process, and the parameter d as the number of differences made for the time series to reach stationarity. Then the formula for the p-order autoregressive process of autoregression is:

[0051]

[0052] where y tis the current value, μ is the constant term, p is the order, and y i is the autocorrelation coefficient, and ε t is the error.

[0053] The formula for the q-order autoregressive process of the moving average model is:

[0054]

[0055] The autoregressive moving average model combines the above two models, and its formula is defined as:

[0056]

[0057] Before using the above model for modeling, first determine the differencing order d through the stationarity of the differenced sequence, and then obtain the optimal parameters p and q based on the analysis of the autocorrelation function ACF and partial autocorrelation PACF of the stationary time series after differencing. After determining the three key parameters, perform modeling, and model testing is required before the model is put into use to verify its effect.

[0058] Preferably, the method further includes: performing pre-training on the neural network model, including: collecting chemical engineering historical parameters and generating training samples based on the chemical engineering historical parameters; performing model training in a pre-constructed neural network based on the training samples to obtain a trend prediction model; wherein, the pre-constructed neural network includes: a forgetting gate, an input gate, and an output gate.

[0059] Example 3:

[0060] Another type of deep learning method has a long short-term memory network ( Figure 1 ), which is a special variant of the recurrent neural network (RNN), introducing the concept of "gates" and being controlled by a forgetting gate, an input gate, and an output gate. The control logic of the gate unit determines the update or discard of data, and the algorithm formula is:

[0061] Forgetting gate: f t = σ(W f [h t-1 + b f )

[0062] Input gate: i t = σ(W i [h t-1 , x t + b i )

[0063] Ct = tanh(Wc[ht-1,xt]+bc)

[0064] Output gate: O t = σ(W o [ht-1 , x t + b o )

[0065] The final output is:

[0066] ht = Ottanh(Ct)

[0067] As an improved RNN network, it has the characteristics of the RNN network, that is, it can "remember" past information and use it to process the current input. There is a problem of gradient disappearance in the RNN, which makes the RNN network unable to remember long-term dependencies. The introduction of "gates" in the improved model can control the flow and loss of features. Because it has the "memory" of the acquired long-term and short-term data, it often can achieve better results in time series prediction. When implementing this model, univariate time series can be used for prediction. If the influence characteristics of a certain parameter can be analyzed, multivariate time series can also be used for trend prediction.

[0068] Step S30: Perform correlation analysis on the predicted time series data and the expected time series data of the corresponding time period through the trend variation analysis model to obtain the trend difference.

[0069] Specifically, based on the expected operating conditions of the target chemical equipment, obtain the expected time series data within the preset prediction time interval; based on the trend variation analysis model, calculate the correlation coefficient between the predicted time series data and the expected time series data; use the correlation coefficient as the trend difference between the predicted time series data and the expected time series data.

[0070] In the embodiment of the present invention, the correlation coefficient is the Pearson correlation coefficient, the Spearman correlation coefficient or the Kendall correlation coefficient.

[0071] Preferably, when the correlation coefficient is the Pearson correlation coefficient, the calculation rule of the correlation coefficient between the predicted time series data and the expected time series data is:

[0072]

[0073] where X is the predicted time series data; Y is the expected time series data; ρ X,Y is the correlation coefficient between the predicted time series data and the expected time series data; cov(X, Y) is the covariance between the predicted time series data and the expected time series data; σ X is the standard deviation of the predicted time series data; σ Y is the standard deviation of the expected time series data.

[0074] Example 4:

[0075] It is implemented through a trend variation analysis model, and the main content is to judge the correlation between two lines, that is, to conduct correlation analysis and significance difference analysis on the curves. The correlation analysis of curves can be analyzed and studied using algorithms such as Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient. The most commonly used is the Pearson correlation coefficient; the significance of curve differences can be analyzed using an independent samples t-test.

[0076] The formula for calculating the Pearson correlation coefficient of sequences X and Y is:

[0077]

[0078] where cov(X,Y) is the covariance of the two sequences, and σ X and σ Y are the standard deviations of the two sequences.

[0079] For different parameters, the degree of trend variation that may indicate abnormal situations in the future may also vary. Setting a default value that works well for the vast majority of parameters as a basis and making targeted adjustments when applied to different parameters is an effective, time-saving, and labor-saving strategy.

[0080] Step S40: Determine and execute an early warning plan based on the trend difference.

[0081] Specifically, when the trend difference is greater than a preset trend difference threshold, it is determined that the probability of an abnormality occurring in the target chemical equipment within a future predetermined time length reaches the early warning standard; then an early warning instruction is generated, and an alarm message is generated based on the early warning instruction.

[0082] In the embodiments of the present invention, through the method of correlation analysis, a comparative analysis of the ideal predicted trend and the actual observed trend is actually completed. That is, it is considered that when there is a large difference between the existing trend and the ideal trend, then at a certain future moment, an abnormal situation is very likely to occur. That is, using the current data, the trend for a period of time in the future is predicted, and then the trend for this period of time in the future is used to judge the future abnormal situation. Such a method solves the limitation of the existing trend prediction algorithm in terms of the prediction time length while maintaining accuracy, greatly advances the early warning time, and effectively improves the early warning ability.

[0083] The input of the early warning module is the future predicted time series data with a time length of s and the true time series data of s / w parameters with a time width of w, and the output is the trend variation analysis result represented by binary values, where 1 represents that the trend is abnormal and an alarm is required, and 0 represents that the trend is normal.

[0084] Figure 2This is the system structure diagram of a chemical process time series data warning system based on trend variation analysis provided by an embodiment of the present invention. As Figure 2 shown, an embodiment of the present invention provides a chemical process time series data warning system based on trend variation analysis, and the system includes:

[0085] A collection unit, configured to collect chemical process time series data and preprocess the chemical process time series data.

[0086] Specifically, the preprocessing of the chemical process time series data includes: outlier cleaning processing and missing value processing.

[0087] In an embodiment of the present invention, the solution of the present invention can automatically and accurately predict the change trend of parameter values, and automatically report warning information according to a warning algorithm, providing early fault warnings to enterprises and the government, so that they have sufficient time to conduct inspections and processing, and avoid accidents as much as possible before they occur. To achieve the above purpose, the solution of the present invention adopts an overall idea based on big data for real-time parameters in the chemical production field, proposes a trend prediction method including traditional machine learning methods and deep learning methods, proposes a trend variation analysis method with correlation analysis as the mainstream means, and innovatively proposes a chemical parameter time series trend prediction and a warning method based on trends, realizing an accident early warning technology based on artificial intelligence.

[0088] Adopt the method of directly connecting to the time series database, use real-time data collected from a real chemical scenario in the application, and directly access the dynamic database through an interface. Establish a time series of parameter values with the parameter collection time as the index. In the preprocessing of data, outlier cleaning is performed according to the 3σ principle, and then the Lagrange interpolation method or the linear interpolation method is used to fill in the missing values. The input of the data collection and processing process is the time series data of parameters within a period of time length l, and the output is the anomaly-free smooth time series data for this period of time with a length of l.

[0089] A prediction unit, configured to perform trend prediction on the preprocessed chemical process time series data based on a pre-constructed trend prediction model, and obtain prediction time series data for a future predetermined time length.

[0090] Specifically, the trend prediction model is an improved ARIMA model or a pre-trained neural network model.

[0091] Specifically, the method further includes: constructing an improved ARIMA model, including: determining an autoregressive algorithm, expressed as:

[0092]

[0093] where y t is the current value; μ is a constant term; p is the order; y iis the autocorrelation coefficient; ε t is the error; determine the moving average term of the moving average process, and generate the q-order autoregressive process algorithm of the moving average model, expressed as:

[0094]

[0095] Based on the autoregressive algorithm and the q-order autoregressive process algorithm of the moving average model, generate an ARIMA model.

[0096] It is realized by establishing a time series prediction model. There are two mainstream strategies for time series prediction models. One is the traditional time series prediction model represented by the ARIMA model, and the other is the neural network-based method represented by the RNN model and the LSTM model.

[0097] In machine learning methods, we define the parameter p as the autoregressive term of the autoregressive process, q as the moving average term of the moving average process, and the parameter d as the number of differences made for the time series to reach stationarity. Then the formula for the p-order autoregressive process of autoregression is:

[0098]

[0099] where, y t is the current value, μ is the constant term, p is the order, y i is the autocorrelation coefficient, ε t is the error.

[0100] The formula for the q-order autoregressive process of the moving average model is:

[0101]

[0102] The autoregressive moving average model combines the above two models, and the formula is defined as:

[0103]

[0104] Before using the above model for modeling, first determine the difference order d through the stationarity of the differenced sequence, and then obtain the optimal parameters p and q based on the analysis of the autocorrelation coefficient ACF and the partial autocorrelation PACF of the stationary time series after differencing. After determining the three key parameters, perform modeling, and model verification is required before the model is put into use to verify its effect.

[0105] Preferably, the method further includes: performing pre-training on the neural network model, including: collecting chemical engineering historical parameters, and generating training samples based on the chemical engineering historical parameters; performing model training in a pre-constructed neural network based on the training samples to obtain a trend prediction model; wherein, the pre-constructed neural network includes: a forgetting gate, an input gate, and an output gate.

[0106] Another type of deep learning method has a long short-term memory network ( Figure 1 ), which is a special variant of the recurrent neural network (RNN). The concept of "gates" is introduced and controlled through forget gates, input gates, and output gates. The control logic of the gate units determines the update or discard of data, and the algorithm formula is as follows:

[0107] Forget gate: f t = σ(W f [h t-1 + b f )

[0108] Input gate: i t = σ(W i [h t-1 , x t + b i )

[0109] C t = tanh(W c [h t-1 , x t + b c )

[0110] Output gate: O t = σ(W o [h t-1 , x t + b o )

[0111] The final output is:

[0112] h t = O t tanh(C t )

[0113] As an improved RNN network, it has the characteristics of the RNN network, that is, it can "remember" past information and use it to process the current input. The RNN has a problem of gradient disappearance, which makes the RNN network unable to remember long-term dependencies. The introduction of "gates" in the improved model can control the flow and loss of features. Because it has a "memory" of the acquired long-term and short-term data, it often achieves better results in time series prediction. When implementing this model, univariate time series can be used for prediction. If the influencing features of a certain parameter can be analyzed, multivariate time series can also be used for trend prediction.

[0114] A comparison unit is used to perform a correlation analysis on the predicted time series data and the expected time series data in the corresponding time period through a trend variation analysis model to obtain a trend difference.

[0115] Specifically, based on the expected operating conditions of the target chemical equipment, obtain the expected time-series data within a preset prediction time interval; based on the trend variation analysis model, calculate the correlation coefficient between the predicted time-series data and the expected time-series data; use the correlation coefficient as the trend difference between the predicted time-series data and the expected time-series data.

[0116] In the embodiments of the present invention, the correlation coefficient is the Pearson correlation coefficient, the Spearman correlation coefficient, or the Kendall correlation coefficient.

[0117] Preferably, when the correlation coefficient is the Pearson correlation coefficient, the calculation rule of the correlation coefficient between the predicted time-series data and the expected time-series data is:

[0118]

[0119] where X is the predicted time-series data; Y is the expected time-series data; ρ X,Y is the correlation coefficient between the predicted time-series data and the expected time-series data; cov(X,Y) is the covariance between the predicted time-series data and the expected time-series data; σ X is the standard deviation of the predicted time-series data; σ Y is the standard deviation of the expected time-series data.

[0120] Example 4:

[0121] It is implemented through a trend variation analysis model. The main content is to judge the correlation degree of two lines, that is, to perform correlation analysis and significance difference analysis on the curves. The correlation analysis of the curves can be analyzed and studied using algorithms such as the Pearson correlation coefficient, the Spearman correlation coefficient, and the Kendall correlation coefficient. The most commonly used is the Pearson correlation coefficient; the significance of the curve difference can be analyzed using an independent samples t-test.

[0122] The formula for calculating the Pearson correlation coefficient of sequences X and Y is:

[0123]

[0124] where cov(X,Y) is the covariance of the two sequences, σ X and σ Y are the standard deviations of the two sequences.

[0125] For different parameters, the degree of trend variation that may indicate abnormal situations in the future may also vary. Setting a default value that can achieve good results for the vast majority of parameters as a basis and making targeted adjustments when applied to different parameters is an effective, time-saving, and labor-saving strategy.

[0126] An early warning unit, configured to determine and execute an early warning plan based on the trend difference.

[0127] Specifically, when the trend difference is greater than a preset trend difference threshold, it is determined that the probability of an abnormality occurring in the target chemical equipment within a predetermined future time period reaches the early warning standard; then an early warning instruction is generated, and an alarm message is generated based on the early warning instruction.

[0128] In the embodiment of the present invention, through the method of correlation analysis, the comparative analysis of the predicted ideal trend and the observed actual trend is actually completed. That is, it is considered that when there is a large difference between the existing trend and the ideal trend, then at a certain future moment, an abnormal situation is very likely to occur. That is, with the current data, the trend for a period of time in the future is predicted, and then the trend for this period of time in the future is used to judge the abnormal situation in the future. Such a method solves the limitation of the existing trend prediction algorithm in the prediction time length while maintaining accuracy, greatly advances the early warning time, and effectively improves the early warning ability.

[0129] The input of the early warning module for advance is the future prediction time series data with a time length of s and the parameter real time series data with a time width of w for s / w, and the output is the trend variation analysis result represented by binary values, where 1 represents that the trend is abnormal and an alarm is required, and 0 represents that the trend is normal.

[0130] The embodiment of the present invention also provides a computer-readable storage medium, on which instructions are stored, and when the instructions are run on a computer, the computer is made to execute the above-mentioned chemical process time series data early warning method based on trend variation analysis.

[0131] Those skilled in the art can understand that all or part of the steps in the method for implementing the above embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions to make a single-chip microcomputer, a chip or a processor execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks or optical disks and other various media that can store program codes.

[0132] The optional embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. In addition, it should be noted that, in the above specific embodiments, the various specific technical features described can be combined in any suitable manner without conflict. To avoid unnecessary repetition, the embodiments of the present invention will not separately describe various possible combination methods.

[0133] In addition, any combination can be made between various different embodiments of the present invention, as long as it does not violate the idea of the embodiments of the present invention, and it should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A warning method for chemical process time series data based on trend variation analysis, characterized in that, The method includes: Collecting chemical process time series data and preprocessing the chemical process time series data; Based on a pre-constructed trend prediction model, performing trend prediction on the preprocessed chemical process time series data to obtain predicted time series data for a future predetermined time length; Performing correlation analysis on the predicted time series data and the expected time series data in the corresponding time period through a trend variation analysis model to obtain a trend difference; Determining and executing an early warning plan based on the trend difference.

2. The method according to claim 1, wherein The preprocessing of the chemical process time series data includes: Outlier cleaning processing and missing value processing.

3. The method according to claim 1, wherein The trend prediction model is an improved ARIMA model or a pre-trained neural network model.

4. The method according to claim 3, wherein The method further includes: Constructing an improved ARIMA model, including: Determining an autoregressive algorithm, expressed as: where y t is the current value; μ is a constant term; p is the order; y i is the autocorrelation coefficient; ε t is the error; Determining the moving average term of the moving average process and generating a q-order autoregressive process algorithm of the moving average model, expressed as: Based on the autoregressive algorithm and the q-order autoregressive process algorithm of the moving average model, generating an ARIMA model.

5. The method according to claim 4, characterized in that, The ARIMA model is expressed as:

6. The method according to claim 3, characterized in that, The method further includes: Performing pre-training of a neural network model, including: Collecting chemical historical parameters and generating training samples based on the chemical historical parameters; Based on the training samples, performing model training in a pre-constructed neural network to obtain a trend prediction model; where The pre-constructed neural network includes: A forget gate, an input gate, and an output gate.

7. The method according to claim 1, characterized in that The performing correlation analysis on the predicted time series data and the expected time series data in the corresponding time period through a trend variation analysis model to obtain a trend difference includes: Based on the expected operating conditions of the target chemical equipment, obtaining the expected time series data within a preset prediction time interval; Based on the trend variation analysis model, calculating the correlation coefficient between the predicted time series data and the expected time series data; Taking the correlation coefficient as the trend difference between the predicted time series data and the expected time series data.

8. The method according to claim 7, wherein The correlation coefficient is a Pearson correlation coefficient, a Spearman correlation coefficient, or a Kendall correlation coefficient.

9. The method according to claim 8, wherein When the correlation coefficient is a Pearson correlation coefficient, the calculation rule of the correlation coefficient between the predicted time series data and the expected time series data is: Where X is the predicted time series data; Y is the expected time series data; ρ X,Y is the correlation coefficient between the predicted time series data and the expected time series data; cov(X,Y) is the covariance between the predicted time series data and the expected time series data; σ X is the standard deviation for predicting time series data; σ Y is the standard deviation of the expected timing data.

10. The method according to claim 1, wherein The determining and executing an early warning plan based on the trend difference includes: When the trend difference is greater than a preset trend difference threshold, determining that the probability of the target chemical equipment having an abnormality within a future predetermined time length reaches the early warning standard; Then generating an early warning instruction and generating an alarm message based on the early warning instruction.

11. A chemical process time series data warning system based on trend variation analysis, characterized in that, The system includes: A collection unit for collecting chemical process time series data and preprocessing the chemical process time series data; A prediction unit for performing trend prediction on the preprocessed chemical process time series data based on a pre-constructed trend prediction model to obtain predicted time series data for a future predetermined time length; A comparison unit, configured to perform correlation analysis on the predicted time-series data and the expected time-series data in a corresponding time period through a trend variation analysis model to obtain a trend difference; An early warning unit, configured to determine and execute an early warning plan based on the trend difference.

12. A computer-readable storage medium, characterized in that, Instructions are stored on the computer-readable storage medium, and when running on a computer, the computer is caused to execute the chemical process time-series data early warning method based on trend variation analysis according to any one of claims 1-10.

13. An electronic device, the electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the chemical process time-series data early warning method based on trend variation analysis according to any one of claims 1-10 is implemented.

Images