Chemical process intelligent monitoring method and system based on artificial intelligence

Abstract

The invention discloses a chemical process intelligent monitoring method and system based on artificial intelligence, and the method comprises the following steps: obtaining an original data flow of a chemical process through a distributed sensor network, and carrying out the preprocessing of the original data flow, and obtaining a preprocessing data set; according to the preprocessed data set, an initial prediction model is constructed by adopting a long-short-term memory network, and the initial prediction model is used for capturing time sequence characteristics of the chemical process; establishing a mechanism model according to the physical mechanism of the chemical process, and fusing the output of the mechanism model and the prediction result of the initial prediction model through a weighted average method to obtain mixed data; working condition change characteristics are obtained according to the mixed data, a working condition recognition model is established according to the working condition change characteristics, when the working condition change is detected through the working condition recognition model, a control strategy adjustment mechanism is triggered, and a self-adaptive control algorithm is adopted to update control parameters according to the optimal control parameter set. According to the invention, self-adaptive control can be carried out on the chemical process.

Description

Technical Field

[0001] The present invention belongs to the technical field of chemical production, and in particular relates to an intelligent monitoring method and system for a chemical process based on artificial intelligence. Background Art

[0002] The intelligent monitoring system of chemical processes faces challenges in data quality and model accuracy in practical applications. First, the data collection environment in chemical processes is complex, and sensors are susceptible to interference, resulting in noise and outliers in the collected data, which affects data quality. Secondly, the physical mechanism of chemical processes is complex, involving the transformation of multiple substances and energy transfer, and it is difficult to establish an accurate physical model. Although data-driven models can learn process characteristics from historical data, they face problems such as insufficient data and changes in operating conditions, resulting in insufficient model generalization capabilities.

[0003] In addition, the control objectives of chemical processes are diverse. It is necessary to ensure product quality while taking into account factors such as energy consumption and safety. The design of intelligent control algorithms requires a balance between multiple objectives. At the same time, the operating conditions of chemical processes change frequently, and the control algorithm needs to have a certain degree of adaptive ability to automatically adjust the control strategy according to the changes in operating conditions. Finally, the intelligent monitoring system needs to be integrated with existing control systems, MES systems, etc. to achieve seamless data docking and interaction, which poses a challenge to the compatibility and stability of the system. Therefore, how to build a high-quality data acquisition and processing mechanism in a complex chemical process, establish a process model with strong robustness and good generalization ability, design a multi-objective optimization intelligent control algorithm, and achieve effective integration with existing systems are key technical issues that need to be solved by the intelligent monitoring system of chemical processes. Summary of the invention

[0004] The present invention proposes an artificial intelligence-based intelligent monitoring method and system for a chemical process to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a chemical process intelligent monitoring method based on artificial intelligence, comprising the following steps:

[0006] Acquire the original data stream of the chemical process through a distributed sensor network, and preprocess the original data stream to obtain a preprocessed data set;

[0007] According to the preprocessed data set, an initial prediction model is constructed using a long short-term memory network, wherein the initial prediction model is used to capture the time series characteristics of the chemical process;

[0008] A mechanism model is established according to the physical mechanism of the chemical process, and the output of the mechanism model is merged with the prediction result of the initial prediction model through a weighted average method to obtain mixed data;

[0009] The operating condition change characteristics are obtained according to the mixed data, and an operating condition identification model is established according to the operating condition change characteristics. When the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the control parameters are updated according to the optimal control parameter set using an adaptive control algorithm.

[0010] Preferably, preprocessing the original data stream includes:

[0011] Use multi-source data fusion technology to perform time alignment processing to obtain a synchronized data set;

[0012] Based on the sliding window mechanism, the synchronous data set is processed in segments, and each data segment is subjected to denoising processing by using a wavelet transform method to obtain a denoised data sequence;

[0013] The denoised data sequence is subjected to outlier detection by using an isolation forest algorithm, and the detected outliers are cleaned to obtain a cleaned data set.

[0014] Preferably, the time alignment process using multi-source data fusion technology includes:

[0015] The Kalman filter algorithm is used to align the timestamps of data from different data sources; the time synchronization of multi-source heterogeneous data is achieved through data interpolation and data compensation technology.

[0016] Preferably, obtaining the denoised data sequence includes:

[0017] The sliding window mechanism is used to segment the data and divide the data set into multiple data segments. For each data segment, the wavelet transform method is used to denoise it and the noise data is separated from the original data through wavelet transform. The interpolation algorithm is used to correct the noise data and the denoised data segments are recombined to obtain the denoised data sequence.

[0018] Preferably, the cleaned data set includes:

[0019] The isolation forest algorithm is used to build an anomaly detection model; according to the preset anomaly threshold, each data point is judged to be an outlier; if the data point matches the anomaly threshold, it is marked as an outlier; all data that are not marked as outliers are obtained to form a cleaned data subset; by performing cluster analysis on the cleaned data subset, data sets of different categories are obtained.

[0020] Preferably, using an adaptive control algorithm to update the control parameters according to the optimal control parameter set includes:

[0021] The changing trend of mixed data is determined by a time series analysis algorithm to obtain the characteristics of operating condition changes; based on the operating condition change characteristics, a support vector machine algorithm is used to establish an operating condition identification model; the established operating condition identification model is used to monitor operating condition changes in real time, and if the model determines that the operating condition has changed significantly, a control strategy adjustment mechanism is triggered; when the control strategy adjustment is triggered, a reinforcement learning algorithm is used to search for an optimal control parameter set, and the algorithm is guided to optimize the control effect through a reward function; the optimal control parameter set obtained by the search is input into an adaptive control algorithm, and the parameters of the PID controller are dynamically adjusted according to the characteristics of the operating condition changes.

[0022] The present invention also discloses an artificial intelligence-based intelligent monitoring system for a chemical process, comprising:

[0023] A data acquisition and preprocessing module is used to acquire the original data stream of the chemical process through a distributed sensor network, and preprocess the original data stream to obtain a preprocessed data set;

[0024] An initial prediction model building module, used to build an initial prediction model using a long short-term memory network according to the preprocessed data set, wherein the initial prediction model is used to capture the time series characteristics of the chemical process;

[0025] A fusion module is used to establish a mechanism model according to the physical mechanism of the chemical process, and fuse the output of the mechanism model with the prediction result of the initial prediction model through a weighted average method to obtain mixed data;

[0026] The operating condition identification and control adjustment module is used to obtain the operating condition change characteristics based on the mixed data, establish an operating condition identification model based on the operating condition change characteristics, and when the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the adaptive control algorithm is used to update the control parameters according to the optimal control parameter set.

[0027] The present invention also discloses a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0028] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[0029] The present invention also discloses a computer program product, comprising a computer program, which implements the steps of the method when executed by a processor.

[0030] Compared with the prior art, the present invention has the following advantages and technical effects:

[0031] The present invention discloses an intelligent monitoring method and system for a chemical process based on artificial intelligence. First, the raw data of a distributed sensor network is subjected to multi-source fusion and preprocessing, including time alignment, sliding window segmentation and wavelet denoising. Then, an isolation forest algorithm is used to detect outliers and clean the data set. Next, a data-driven model constructed by a long short-term memory network and a mechanism model based on a physical mechanism are combined to form a hybrid prediction model through weighted fusion to improve the ability to capture the timing characteristics of the chemical process and the prediction accuracy. Finally, a working condition identification model is established to monitor the changes in key parameters in real time. Once a working condition change is detected, an adaptive control algorithm is triggered to dynamically adjust the control strategy. The present invention realizes accurate modeling and intelligent control of complex and multi-changing process flows through a hybrid modeling method combining data-driven and mechanism models, as well as an intelligent control strategy for working condition adaptation, thereby improving production efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0033] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0035] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] Embodiment 1

[0037] like Figure 1 As shown, this embodiment provides a chemical process intelligent monitoring method based on artificial intelligence, comprising the following steps:

[0038] Acquire the original data stream of the chemical process through a distributed sensor network, and preprocess the original data stream to obtain a preprocessed data set;

[0039] According to the preprocessed data set, an initial prediction model is constructed using a long short-term memory network, wherein the initial prediction model is used to capture the time series characteristics of the chemical process;

[0040] A mechanism model is established according to the physical mechanism of the chemical process, and the output of the mechanism model is merged with the prediction result of the initial prediction model through a weighted average method to obtain mixed data;

[0041] The operating condition change characteristics are obtained according to the mixed data, and an operating condition identification model is established according to the operating condition change characteristics. When the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the control parameters are updated according to the optimal control parameter set using an adaptive control algorithm.

[0042] The specific steps are as follows:

[0043] Step S101, obtaining the original data stream of the distributed sensor network, and using multi-source data fusion technology to perform time alignment processing on data with different sampling frequencies to obtain a synchronized data set.

[0044] Specifically, the original data stream collected by each sensor node in the distributed sensor network is obtained; the data source that needs to be time synchronized is determined according to the sampling frequency of each sensor; the Kalman filter algorithm is used to align the timestamps of data from different data sources; and the time synchronization of multi-source heterogeneous data is achieved through data interpolation and data compensation technology.

[0045] In this embodiment, the data processing flow in the distributed sensor network involves multiple key steps, each of which is crucial to the final data quality. Taking the oil refining process as an example, common key parameters include reactor temperature, raw material feed pressure, catalyst flow rate, etc. By deploying devices such as temperature sensors, pressure transmitters, and flow meters, these parameter data can be collected in real time. They collect data at different frequencies. Time synchronization is the key to ensuring data comparability. Since the sampling frequencies of different sensors may be different, it is necessary to identify which data sources need to be synchronized. For example, the temperature sensor samples once a minute, while the pressure sensor samples once every five minutes, which requires time alignment. The Kalman filter algorithm plays an important role in timestamp alignment. It can predict and estimate the values ​​between data points and effectively handle noise and uncertainty in sensor data. In this way, data of different frequencies can be adjusted to a unified time scale.

[0046] Step S102, for the synchronous data set, segment processing is performed based on a sliding window mechanism, and each data segment is denoised using a wavelet transform method to obtain a denoised data sequence.

[0047] Specifically, according to the characteristics of the synchronous data set, the sliding window mechanism is used to segment the data and divide the data set into multiple data segments. For each data segment, the wavelet transform method is used to denoise it, and the noise data is separated from the original data through wavelet transform. According to the results of the wavelet transform, the characteristics of the noise data are determined, and each data point is judged to be noise based on the preset threshold. For the data points judged to be noise, the interpolation algorithm is used to correct them, and the adjacent non-noise data points are used to estimate and replace the noise data points. The denoised data segments are recombined to obtain a complete denoised data sequence to ensure the continuity and integrity of the data.

[0048] In this embodiment, the sliding window mechanism is a commonly used data processing method, which can divide long-time series data into multiple shorter data segments. For example, for a synchronous data set containing 24-hour temperature data, a 1-hour sliding window can be used, moving forward 30 minutes each time, so that 48 overlapping data segments can be obtained. This method not only ensures the continuity of the data, but also facilitates subsequent processing and analysis. Wavelet transform is a powerful signal processing tool that can effectively separate noise from raw data. Taking temperature data as an example, high-frequency noise can be separated from low-frequency temperature change trends through wavelet transform. Selecting appropriate wavelet basis functions (such as Daubechies wavelets) and decomposition layers can better capture the characteristics of the data. The feature judgment of noise data is usually based on statistical characteristics and threshold settings. For example, the mean and standard deviation of each data segment can be calculated, and data points that deviate from the mean by more than 3 times the standard deviation are regarded as potential noise. This method can effectively identify sudden outliers, but may ignore slowly changing systematic errors. For data points determined to be noise, interpolation algorithms can be used for correction. Linear interpolation is a simple and effective method suitable for data correction in a short time. For data missing for a long time, you can consider using spline interpolation or polynomial interpolation, which can better maintain the continuity and smoothness of the data. Data smoothing is an important step after denoising, which can further eliminate the influence of residual noise. The moving average method is simple and easy to use, and is suitable for processing stable time series data. For example, using a 5-point moving average for temperature data can effectively smooth short-term fluctuations and highlight long-term trends. Kalman filtering is more suitable for processing dynamically changing data, such as vehicle trajectory data, which can dynamically adjust the estimation results based on historical data and current measurements. The data series after denoising and smoothing lays the foundation for subsequent analysis. In the feature extraction stage, statistical features (such as mean, variance, kurtosis, etc.) can be calculated or time domain and frequency domain features can be extracted. For example, for temperature data, features such as daily variation amplitude and time of maximum temperature occurrence can be extracted. In terms of pattern recognition, the processed data can be used for tasks such as anomaly detection and trend prediction. For example, using the processed temperature data, a time series model can be established to predict future temperature changes; for multi-source sensor data, machine learning algorithms can be used to identify different working states or failure modes to improve the reliability and efficiency of the system. This series of data processing steps not only improves data quality, but also provides a reliable foundation for subsequent in-depth analysis and application. By removing noise and smoothing data, the real information and change trends in the data can be captured more accurately, thus supporting more accurate decision-making and prediction.

[0049] Step S103, performing outlier detection on the denoised data sequence using an isolation forest algorithm, and if a data point matches a preset outlier threshold, marking it as an outlier to obtain a cleaned data set.

[0050] Specifically, for the denoised data sequence, the isolation forest algorithm is used to build an anomaly detection model; according to the preset anomaly threshold, each data point is judged to be an outlier; if the data point matches the anomaly threshold, it is marked as an outlier; all data that are not marked as outliers are obtained to form a cleaned data subset; by performing cluster analysis on the cleaned data subset, data sets of different categories are obtained; according to the characteristic distribution of the data set, the representative data points of each category are determined as input for subsequent analysis.

[0051] In this embodiment, the isolation forest algorithm is an efficient anomaly detection method. In the factory scenario, it can identify equipment failures, raw material anomalies, and so on. The algorithm calculates the isolation degree of each data point by constructing multiple decision trees. For example, if a sensor has an abnormally high temperature in a short period of time, it will be marked as an anomaly. This method can quickly discover potential problems and improve production efficiency and safety. The setting of abnormal thresholds needs to be combined with specific business scenarios. In a factory environment, a reasonable range can be determined based on historical data and expert experience. For example, the normal range of a temperature sensor is 20-30°C, and anything beyond this range is considered abnormal. By dynamically adjusting the threshold, it can adapt to the needs of different production stages and improve the accuracy of anomaly detection. The cleaned data subset can better reflect the normal production status. Taking the energy consumption data of the production line as an example, after removing the outliers, the energy utilization efficiency can be analyzed more accurately. This helps to optimize the production process and reduce operating costs.

[0052] Step S104, based on the cleaned data set, a data-driven model is constructed using a long short-term memory network, and an initial prediction model is obtained through historical data training. The initial prediction model is used to capture the time series characteristics of the chemical process.

[0053] Specifically, according to the cleaned data set, a data-driven model is constructed using a long short-term memory network, and an initial prediction model is obtained by training historical data to capture the time series characteristics in the chemical process. For the above business scenario, the following technical solution is designed: historical data of the chemical process is obtained, and the data is cleaned and preprocessed to obtain a data set that meets the requirements. According to the cleaned data set, a data-driven model based on a long short-term memory network is constructed, and the time series characteristics of the chemical process are effectively captured by setting appropriate network structures and parameters. The historical data is divided into a training set and a validation set, and the constructed long short-term memory network model is trained using the training set. After multiple rounds of iterative optimization, an initial prediction model is obtained. During the training process, the hyperparameters of the long short-term memory network, including the number of hidden layers, the number of hidden units, the learning rate, etc., are dynamically adjusted to improve the prediction performance of the model. For the validation set data, the initial prediction model obtained by training is used to make predictions, and the generalization ability of the model on unknown data is evaluated. If the prediction effect is not ideal, parameter tuning is performed. When the prediction performance on the validation set reaches the preset threshold, the initial prediction model obtained by training is saved as the basic model for subsequent chemical process prediction and optimization. In practical applications, the constructed data-driven model is used to predict new chemical process data. Combined with expert knowledge and prior experience, the prediction results are analyzed and interpreted to provide decision support for process optimization.

[0054] Furthermore, during the model training process, the network weights are continuously updated through the back-propagation algorithm and gradient descent optimization method, so that the model can effectively capture the time series characteristics of the chemical process. The trained long short-term memory network model can predict the future time series data of the chemical process and determine whether there are any abnormalities in the process operation. If the deviation between the predicted result and the actual data exceeds the preset threshold, the early warning mechanism is triggered to notify relevant personnel to check and adjust the process. The trained model is deployed to the online system to obtain chemical process data in real time, continuously monitor and warn the process operation status, and ensure the safe and stable operation of the process.

[0055] Step S105, establishing a mechanism model according to the physical mechanism equation of the chemical process, fusing the output of the mechanism model with the prediction result of the initial prediction model through a weighted average method to obtain a hybrid model, and the hybrid model is used to improve the prediction accuracy and generalization ability.

[0056] Specifically, relevant physical mechanism equations are obtained according to the chemical process, and the corresponding mechanism model is established; relevant historical data are collected for the chemical process, and the data-driven model is trained based on the machine learning algorithm to obtain the initial prediction model; the output result is obtained from the mechanism model, and the prediction result is obtained from the initial prediction model; the output of the mechanism model and the prediction result of the initial prediction model are integrated through the weighted average algorithm to obtain the hybrid model output; according to the preset threshold, it is judged whether the prediction accuracy of the hybrid model meets the requirements, if it does, the hybrid model is determined, otherwise the weight coefficient of the weighted average algorithm is adjusted; the cross-validation method is used to evaluate the generalization ability of the hybrid model, and the hybrid model is optimized through multiple iterations; the established hybrid model is applied to the prediction and optimization of the chemical process to improve production efficiency and product quality.

[0057] In this embodiment, in the optimization of chemical processes, it is an effective strategy to build a hybrid model by combining physical mechanisms and data-driven methods. Taking the petroleum refining process as an example, the mechanism model of the distillation tower is first established, including equations such as material balance, heat balance and phase balance. At the same time, historical operation data such as feed composition, tower top temperature, tower bottom temperature, etc. are collected, and a data-driven model is constructed using a long short-term memory network (LSTM). The mechanism model can calculate the product yield and quality based on the feed conditions, while the LSTM model can predict the change in tower top temperature in the future. The outputs of the two models are fused by a weighted average algorithm. For example, a weight of 0.6 can be given to the mechanism model and a weight of 0.4 to the LSTM model to obtain the final hybrid model prediction result. To evaluate the performance of the hybrid model, a threshold value of no more than 3% of the prediction error can be set. If the prediction accuracy of the hybrid model does not meet the requirements, the weight coefficient needs to be adjusted. For example, the weight of the mechanism model is increased to 0.7, the weight of the LSTM model is reduced to 0.3, and the output of the hybrid model is recalculated. The cross-validation method is used to evaluate the generalization ability of the hybrid model. The data set can be divided into 5 parts, 4 parts are used as training sets each time, and 1 part is used as test set. The performance of the model on different data subsets is obtained by repeating the cycle 5 times. Through multiple iterations, the model structure and parameters are continuously optimized to improve the prediction accuracy. The optimized hybrid model can be applied to actual production to more accurately predict product quality and output. For example, in the ethylene production process, the hybrid model can simultaneously consider the theoretical conversion rate and actual operation data of the thermal cracking furnace to provide operators with more reliable decision support. This can not only improve product yield, but also reduce energy consumption, achieving a dual improvement in economic and environmental benefits. The advantage of the hybrid model is that it retains the interpretability of the mechanism model and incorporates the data-driven model's ability to capture complex nonlinear relationships. In practical applications, the weights of the two models can be dynamically adjusted according to the specific situation to meet the prediction needs under different working conditions. Through continuous optimization and updating, the hybrid model can continuously improve its prediction accuracy and adaptability, providing strong support for the intelligent control and optimization of chemical processes.

[0058] Step S106, obtaining the operating condition change characteristics, which include the change trends of key parameters such as temperature, pressure, and flow, and establishing an operating condition identification model based on the operating condition change characteristics. If the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the control parameters are updated according to the optimal control parameter set using an adaptive control algorithm.

[0059] Specifically, the changing trend of the mixed data is determined by the time series analysis algorithm. According to the obtained working condition change characteristics, the support vector machine algorithm is used to establish a working condition identification model, and the optimal parameters of the model are obtained by training the historical working condition data. The established working condition identification model is used to monitor the working condition changes in real time. If the model determines that the working condition has changed significantly, the control strategy adjustment mechanism is triggered. When the control strategy adjustment is triggered, the reinforcement learning algorithm is used to search for the optimal control parameter set, and the algorithm is guided by the reward function to optimize the control effect. The optimal control parameter set obtained by the search is input into the adaptive control algorithm, and the parameters of the PID controller are dynamically adjusted according to the working condition change characteristics. After the controller parameters are updated, new key parameter data such as temperature, pressure, and flow are obtained to determine whether the working condition has returned to normal. If the working condition has not returned to normal, return to step 4 to continue searching for the optimal control parameters until the working condition is stable or the preset control target is reached.

[0060] In this embodiment, the acquisition of operating condition change characteristics is a key link in chemical process control. Taking the petroleum refining process as an example, common key parameters include reactor temperature, raw material feed pressure, catalyst flow rate, etc. By deploying equipment such as temperature sensors, pressure transmitters and flow meters, these parameter data can be collected in real time. Using time series analysis algorithms such as autoregressive integral moving average model (ARIMA), the changing trend of each parameter can be effectively judged. For example, a rapid rise in reactor temperature in a short period of time may indicate the occurrence of an abnormal exothermic reaction. The establishment of a working condition identification model helps to detect production anomalies in a timely manner. The support vector machine (SVM) algorithm is suitable for working condition identification because of its advantages in small samples and nonlinear problems. Taking polymer production as an example, a variety of historical working condition data such as normal working conditions, catalyst deactivation, and reactor fouling can be collected, and the kernel function type and parameters of SVM can be determined by cross-validation to obtain the optimal working condition identification model. Real-time working condition monitoring is an important means to ensure production safety. When the working condition identification model determines that an abnormality occurs, such as a sudden increase in the pressure of the polymerization reactor, the control strategy adjustment mechanism will be triggered. This mechanism is designed to respond quickly to changes in operating conditions and prevent safety accidents. Reinforcement learning algorithms perform well in finding optimal control parameters. Taking the continuous stirred tank reactor (CSTR) as an example, the temperature control problem of the reactor can be modeled as a Markov decision process. By designing a suitable reward function, such as taking into account factors such as product quality, energy consumption, and safety, the reinforcement learning algorithm is guided to search for the optimal heating power and cooling water flow. The adaptive control algorithm can dynamically adjust the controller parameters according to changes in operating conditions. Taking the PID controller as an example, different parameter adjustment strategies can be selected according to the operating condition identification results. When the reaction intensity is detected to increase, the proportional coefficient and differential coefficient can be appropriately increased to improve the controller's response speed and ability to suppress oscillation. The evaluation and iterative optimization of the control effect are important links to ensure stable production. After the parameters are updated, key indicators such as temperature fluctuation amplitude and product quality need to be continuously monitored. If it is found that the operating conditions have not returned to normal, such as the temperature is still fluctuating, it is necessary to return to the reinforcement learning algorithm to search for the control parameters again. This process may require multiple iterations until the preset control target is achieved, such as controlling the temperature fluctuation within the range of ±0.5℃. Through the above method, intelligent control of chemical processes can be achieved to improve production efficiency and product quality. For example, in the ethylene production process, timely detection and adjustment of abnormal temperature distribution in the cracking furnace can significantly improve the selectivity of ethylene and reduce the generation of by-products. In polymer production, precise control of the pressure and temperature of the reactor can effectively regulate the molecular weight distribution of the product to meet the needs of different application scenarios. This method based on operating condition identification and adaptive control can not only cope with various disturbances in the production process, but also flexibly adjust production parameters according to market demand, bringing significant economic benefits to chemical companies.

[0061] This embodiment also discloses an artificial intelligence-based intelligent monitoring system for a chemical process, comprising:

[0062] A data acquisition and preprocessing module is used to acquire the original data stream of the chemical process through a distributed sensor network, and preprocess the original data stream to obtain a preprocessed data set;

[0063] An initial prediction model building module, used to build an initial prediction model using a long short-term memory network according to the preprocessed data set, wherein the initial prediction model is used to capture the time series characteristics of the chemical process;

[0064] A fusion module is used to establish a mechanism model according to the physical mechanism of the chemical process, and fuse the output of the mechanism model with the prediction result of the initial prediction model through a weighted average method to obtain mixed data;

[0065] The operating condition identification and control adjustment module is used to obtain the operating condition change characteristics based on the mixed data, establish an operating condition identification model based on the operating condition change characteristics, and when the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the adaptive control algorithm is used to update the control parameters according to the optimal control parameter set.

[0066] This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0067] This embodiment also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[0068] This embodiment also discloses a computer program product, including a computer program, which implements the steps of the method when executed by a processor.

[0069] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A chemical process intelligent monitoring method based on artificial intelligence, characterized in that: The following steps are involved: Acquire the original data stream of the chemical process through a distributed sensor network, and preprocess the original data stream to obtain a preprocessed data set; According to the preprocessed data set, an initial prediction model is constructed using a long short-term memory network, wherein the initial prediction model is used to capture the time series characteristics of the chemical process; A mechanism model is established according to the physical mechanism of the chemical process, and the output of the mechanism model is merged with the prediction result of the initial prediction model through a weighted average method to obtain mixed data; The operating condition change characteristics are obtained according to the mixed data, and an operating condition identification model is established according to the operating condition change characteristics. When the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the control parameters are updated according to the optimal control parameter set using an adaptive control algorithm.

2. The method according to claim 1, characterized in that Preprocessing the original data stream includes: Use multi-source data fusion technology to perform time alignment processing to obtain a synchronized data set; Based on the sliding window mechanism, the synchronous data set is processed in segments, and each data segment is subjected to denoising processing by using a wavelet transform method to obtain a denoised data sequence; The denoised data sequence is subjected to outlier detection by using an isolation forest algorithm, and the detected outliers are cleaned to obtain a cleaned data set.

3. The method according to claim 2, characterized in that The time alignment processing using multi-source data fusion technology includes: The Kalman filter algorithm is used to align the timestamps of data from different data sources; the time synchronization of multi-source heterogeneous data is achieved through data interpolation and data compensation technology.

4. The method according to claim 2, characterized in that: The denoised data sequence includes: The sliding window mechanism is used to segment the data and divide the data set into multiple data segments. For each data segment, the wavelet transform method is used to denoise it and the noise data is separated from the original data through wavelet transform. The interpolation algorithm is used to correct the noise data and the denoised data segments are recombined to obtain the denoised data sequence.

5. The method according to claim 2, characterized in that: The cleaned data set includes: The isolation forest algorithm is used to build an anomaly detection model; according to the preset anomaly threshold, each data point is judged to be an outlier; if the data point matches the anomaly threshold, it is marked as an outlier; all data that are not marked as outliers are obtained to form a cleaned data subset; by performing cluster analysis on the cleaned data subset, data sets of different categories are obtained.

6. The method according to claim 1, characterized in that Adopting adaptive control algorithm to update control parameters according to the optimal control parameter set includes: The changing trend of mixed data is determined by a time series analysis algorithm to obtain the characteristics of operating condition changes; based on the operating condition change characteristics, a support vector machine algorithm is used to establish an operating condition identification model; the established operating condition identification model is used to monitor operating condition changes in real time, and if the model determines that the operating condition has changed significantly, a control strategy adjustment mechanism is triggered; when the control strategy adjustment is triggered, a reinforcement learning algorithm is used to search for an optimal control parameter set, and the algorithm is guided to optimize the control effect through a reward function; the optimal control parameter set obtained by the search is input into an adaptive control algorithm, and the parameters of the PID controller are dynamically adjusted according to the characteristics of the operating condition changes.

7. An artificial intelligence-based intelligent monitoring system for chemical processes, characterized in that: include: A data acquisition and preprocessing module is used to acquire the original data stream of the chemical process through a distributed sensor network, and preprocess the original data stream to obtain a preprocessed data set; An initial prediction model building module, used to build an initial prediction model using a long short-term memory network according to the preprocessed data set, wherein the initial prediction model is used to capture the time series characteristics of the chemical process; A fusion module is used to establish a mechanism model according to the physical mechanism of the chemical process, and fuse the output of the mechanism model with the prediction result of the initial prediction model through a weighted average method to obtain mixed data; The operating condition identification and control adjustment module is used to obtain the operating condition change characteristics based on the mixed data, establish an operating condition identification model based on the operating condition change characteristics, and when the operating condition change is detected by the operating condition identification model, the control strategy adjustment mechanism is triggered, and the adaptive control algorithm is used to update the control parameters according to the optimal control parameter set.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images