Real-time fault monitoring IoT system for chemical production equipment clusters

Abstract

The invention discloses a real-time fault monitoring Internet of Things system for a chemical production equipment cluster, relates to the technical field of equipment fault monitoring and control, and is used to solve the problem of poor fault monitoring accuracy of a chemical production equipment cluster; the system comprises a data acquisition module, a comparative analysis module, an equipment abnormality determination module, and a fault trend prediction module; the invention collects data of different frequencies by re-layering, and completes and corrects abnormal values ​​to ensure data integrity, and establishes a baseline model, extracts multidimensional features of a normal state of the equipment by an autoencoder, and quantifies operating features by a Gaussian mixture model to achieve deviation judgment, and then combines isolation forest and principal component analysis to comprehensively judge the abnormal state by deviation, isolation score and principal component deviation, and generates an alarm signal, and at the same time, adopts ARIMA and LSTM models to fuse and predict short-term and long-term trends to form a comprehensive prediction value, and automatically executes control measures when the trend is abnormal to improve the safety of equipment operation.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault monitoring and control, and more specifically, to a real-time fault monitoring Internet of Things system for a chemical production equipment cluster. Background Art

[0002] The real-time fault monitoring IoT system for chemical production equipment clusters is committed to ensuring the safety and efficiency of the chemical production process through advanced IoT and intelligent control technologies. Chemical production equipment clusters usually include pumps, compressors, reactors, heat exchangers and other equipment. These equipment operate in complex environments with high temperature, high pressure or corrosive media. Equipment failure or even safety accidents may occur if you are not careful. Operation and maintenance personnel rely on regular manual inspections or post-fault repairs. This model often cannot meet the high safety requirements of modern chemical production. With the development of IoT, big data and intelligent control technologies, chemical companies can use real-time data collection and analysis systems to monitor the operating status of equipment online to achieve preventive maintenance and predictive fault management.

[0003] Deficiencies in existing technologies: With the expansion of the scale of chemical production equipment clusters and the continuous increase in the amount of monitoring data, the system is unable to process the data of all equipment in real time due to excessive resource load, resulting in control response lags and monitoring delays. This resource bottleneck problem makes it difficult for the system to prioritize responses when faced with critical fault signals, especially under high-load conditions. The priority of monitoring data is difficult to reasonably allocate, and low-priority data often squeezes out the processing resources for fault data, resulting in inaccurate fault monitoring and untimely response, affecting the overall operating efficiency and safety of the equipment cluster. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, there is a solution as follows to solve the problem of poor fault monitoring accuracy of the chemical production equipment cluster in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The real-time fault monitoring IoT system for chemical production equipment clusters includes a data acquisition module, a comparative analysis module, an equipment anomaly determination module, and a fault trend prediction module, and each module is connected by signals;

[0007] The data collection module is used to collect data from chemical production equipment clusters in layers according to the frequency and importance of different types of sensor data, and to filter, complete and correct abnormal values ​​of the data;

[0008] The comparative analysis module is used to use historical data as baseline data, use an autoencoder to extract baseline data features, quantify the multi-dimensional feature distribution of the device under normal conditions through a Gaussian mixture model and build a baseline model, map the real-time data to the feature space through the encoder, and calculate the deviation from the baseline model to determine the abnormality of the device;

[0009] The equipment anomaly determination module is used to combine the isolation forest and principal component analysis to secondary identify data anomalies, comprehensively judge the equipment anomalies based on the deviation, isolation score and principal component deviation, and generate an alarm signal when the comprehensive threshold is exceeded;

[0010] The fault trend prediction module is used to build a trend prediction model, perform short-term predictions through the ARIMA model and the LSTM model for long-term predictions, and generate a comprehensive prediction value by weighted fusion results. It analyzes future trends in combination with the baseline model and automatically executes equipment control measures when equipment abnormalities are detected.

[0011] In a preferred embodiment, the method is used to collect hierarchical data of a chemical production equipment cluster according to the frequency and importance of different types of sensor data, and to filter, complete and correct abnormal values ​​of the data. The specific steps are as follows:

[0012] The collected data is divided into high-frequency data, medium-frequency data, and low-frequency data;

[0013] Noise filtering is performed based on the characteristics of the sensor and the historical noise level, and the data mutation points in the filtered data are smoothed using the cubic spline interpolation method;

[0014] The smoothed data is subjected to outlier processing and data correction. The data points are compared with the medians of the data at adjacent time points using the distributed distance correction method to detect and correct outliers.

[0015] In a preferred embodiment, historical data is used as baseline data, an autoencoder is used to extract baseline data features, a Gaussian mixture model is used to quantify the multi-dimensional feature distribution of the device in a normal state and a baseline model is constructed, real-time data is mapped to the feature space through the encoder, and the deviation from the baseline model is calculated to determine the abnormality of the device. The specific steps include:

[0016] Collect and organize historical data of baseline data, and select historical data of equipment within a fixed period of time as baseline data samples. The baseline data samples include normal operating data of equipment under different working conditions and loads;

[0017] Use autoencoder neural networks as unsupervised feature extraction tools to identify nonlinear characteristics of baseline data;

[0018] In the feature space, a Gaussian mixture model is used to establish a baseline model for the feature vectors of the device in a normal state, quantify the data feature range in a normal state, and form a baseline model that adapts to the multi-dimensional feature distribution;

[0019] Based on the constructed baseline model, the collected real-time data is compared, the real-time data is mapped to the feature space through the encoder, and the deviation from the baseline model is calculated;

[0020] When the deviation is greater than the set threshold, it is preliminarily determined that there is an abnormality in the current equipment operation.

[0021] In a preferred embodiment, the isolation forest and principal component analysis are combined to identify data anomalies for the second time, and the device anomalies are comprehensively judged based on the deviation, isolation score and principal component deviation. When the comprehensive threshold is exceeded, an alarm signal is generated. The specific steps are as follows:

[0022] Use Isolation Forest to build a decision tree to separate outliers of different features and determine anomaly scores;

[0023] Use principal component analysis to identify deviations from the principal components of the data, determine the data variation pattern, and calculate the principal component deviation;

[0024] According to the results of deviation, isolation score and principal component deviation, the abnormality type is classified by multi-layer abnormality detection to determine the comprehensive abnormality of different equipment and equipment under different working conditions;

[0025] When the comprehensive abnormality exceeds the comprehensive threshold, it is determined that the equipment operation is abnormal, and the comprehensive threshold is adjusted according to the real-time operating condition changes of the equipment.

[0026] In a preferred embodiment, a trend prediction model is constructed, short-term prediction is performed by ARIMA model, long-term prediction is performed by LSTM model, and the weighted fusion results are used to generate a comprehensive prediction value, including the following steps:

[0027] Arrange the features of real-time data and historical data, re-encode the cleaned data that has passed the anomaly detection into feature vectors, construct feature sequences, and standardize the feature sequences;

[0028] Predict the short-term trend of the equipment and use the autoregressive integrated moving average model to determine the short-term forecast value of the equipment;

[0029] Use long short-term memory networks to predict long-term trends, determine long-term dependencies and nonlinear relationships of data, use feature sequences as input to long short-term memory networks, and determine long-term prediction values ​​of devices;

[0030] The prediction results of the ARIMA model and the LSTM model are fused using a weighted average method to form a comprehensive prediction value.

[0031] In a preferred embodiment, the future trend is analyzed in combination with the baseline model, and when an equipment abnormality is detected, equipment control measures are automatically executed, including the following steps:

[0032] After determining the comprehensive prediction value, a trend analysis is performed and corresponding warning information is generated. If the comprehensive prediction value exceeds the normal baseline model range, the system triggers a warning signal, notifies the operation and maintenance personnel, and optimizes the comprehensive threshold;

[0033] Calculate the deviation of the future operating state of the equipment, compare the future moment prediction value generated by the trend prediction model with the normal range of the baseline model, and determine the prediction deviation. The prediction deviation indicates the degree to which the future operating state of the equipment deviates from the normal state.

[0034] Adaptively adjust the comprehensive threshold according to the prediction deviation, and determine the dynamically adjusted comprehensive threshold;

[0035] Multi-level allocation of comprehensive thresholds is carried out according to different monitored parameters, including vibration, temperature, and pressure. Layered comprehensive thresholds are adjusted according to the parameters, and control measures for the equipment are automatically executed.

[0036] The technical effects and advantages of the present invention for the real-time fault monitoring Internet of Things system of chemical production equipment clusters are as follows:

[0037] The present invention ensures the integrity and accuracy of the data by collecting important data of different frequencies in layers, and adopts filtering, data completion and outlier correction. The system uses historical data to establish a baseline model, extracts multi-dimensional features of the equipment in normal state through the autoencoder, and quantifies the equipment operation characteristics in combination with the Gaussian mixture model, so as to judge the deviation of the equipment in real time. For possible false alarms or missed alarms, the state of the equipment is secondary judged through isolation forest and principal component analysis, and the abnormal situation of the equipment is comprehensively determined by the deviation degree, isolation score and principal component deviation degree, which significantly improves the detection accuracy.

[0038] When the abnormal signal of the equipment exceeds the comprehensive threshold, the system generates an alarm signal and performs trend forecasting, using the ARIMA model and LSTM model to integrate and forecast short-term and long-term trends to form a comprehensive forecast value and accurately predict the future status of the equipment. At the same time, the system can dynamically adjust the threshold based on the real-time analysis results and automatically execute control measures when a fault trend is detected to ensure the safe and stable operation of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The figure is a schematic diagram of the structure of the Internet of Things system for real-time fault monitoring of chemical production equipment clusters according to the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] In order to achieve the above objectives, Figure 1 The present invention provides a structural schematic diagram of a real-time fault monitoring Internet of Things system for a chemical production equipment cluster, which specifically includes a data acquisition module, a comparison and analysis module, an equipment anomaly determination module, and a fault trend prediction module, and each module is connected by a signal;

[0042] The data collection module is used to collect data from chemical production equipment clusters in layers according to the frequency and importance of different types of sensor data, and to filter, complete and correct abnormal values ​​of the data;

[0043] The comparative analysis module is used to use historical data as baseline data, use an autoencoder to extract baseline data features, quantify the multi-dimensional feature distribution of the device under normal conditions through a Gaussian mixture model and build a baseline model, map the real-time data to the feature space through the encoder, and calculate the deviation from the baseline model to determine the abnormality of the device;

[0044] The equipment anomaly determination module is used to combine the isolation forest and principal component analysis to secondary identify data anomalies, comprehensively judge the equipment anomalies based on the deviation, isolation score and principal component deviation, and generate an alarm signal when the comprehensive threshold is exceeded;

[0045] The fault trend prediction module is used to build a trend prediction model, perform short-term predictions through the ARIMA model and the LSTM model for long-term predictions, and generate a comprehensive prediction value by weighted fusion results. It analyzes future trends in combination with the baseline model and automatically executes equipment control measures when equipment abnormalities are detected.

[0046] Step 1: In the real-time fault monitoring IoT system of chemical production equipment clusters, multi-layer data collection and cleaning are key basic steps to ensure data quality and improve monitoring accuracy. Due to the complexity of chemical production equipment clusters and the variability of data collection environments, the data processing process includes data collection, noise filtering, data smoothing, and anomaly correction. When processing equipment operation status data, each link needs to be designed in combination with the characteristics of the scenario to ensure that the collected data can accurately and stably reflect the true status of the equipment. The specific steps are as follows:

[0047] Data collection and hierarchical integration: In a chemical equipment cluster, each device is equipped with multiple types of sensors to collect key operating data. Typical sensors include vibration sensors, pressure sensors, temperature sensors, etc. The data dimensions and frequencies collected by each type of sensor are different. To meet the monitoring needs of different devices and different types of data, the data collection process integrates data in layers according to importance and sampling frequency.

[0048] The hierarchical structure prioritizes the collection and transmission of high-frequency data (such as vibration data) to ensure that the system can detect key state changes in a timely manner; medium-frequency data (such as pressure data) and low-frequency data (such as equipment status signals) are processed step by step according to demand to avoid unnecessary data transmission load. For the data set of the i-th device, ,in, Vibration data, which represents high-frequency data related to equipment vibration; is pressure data, which indicates important medium-frequency parameters during equipment operation and monitors the pressure status inside or outside the equipment; T is temperature data, which indicates the temperature information of key areas of the equipment and belongs to medium-frequency or low-frequency data. For example, the temperature of a reactor or the temperature of some mechanical parts; It is the equipment status signal, which is low-frequency data and reflects the overall operating status of the equipment. Unified collection and synchronous integration of data streams from different devices into a total data set D. This design ensures that key information is not lost during data transmission, providing a reliable time series data foundation for subsequent monitoring and analysis;

[0049] Noise filtering and outlier correction are performed on the collected data. In the chemical production environment, the complexity of equipment operation leads to interference noise in the collected data, such as electromagnetic interference, offset caused by mechanical vibration, etc. These noises will interfere with the judgment of the true state of the equipment. Therefore, noise filtering must be performed before entering the analysis data. Adaptive filtering method is used for the collected data. The filtering parameters are determined according to the characteristics of the sensor and the historical noise level. and time period , filter out short-term interference and retain effective equipment operation signals. The specific noise filtering formula is: ,in, is the jth sensor at time The collected raw data, n represents the size of the smoothing window, is the filter coefficient, which is related to the sensor type and environmental factors. is the time period, indicating the sampling period of the data;

[0050] Through weighted and periodic adjustments, the filtering formula can adapt to the complex environmental characteristics of chemical equipment, filter out short-term random interference, retain key signal features, and effectively remove sudden noise. The resulting data It will be passed to the data smoothing step to ensure the stability of the data for subsequent analysis;

[0051] After noise filtering, in order to ensure the continuity and smoothness of the data, the data smoothing and completion method is applied. Since sensor signal mutations or data loss may occur during the operation of chemical equipment, it is necessary to smooth the data mutation points through cubic spline interpolation. The data smoothing result Calculated based on the following formula: , where the coefficient is adjusted dynamically , , , , so that the smoothing results can adapt to the actual state changes of chemical equipment; For the original data In time The first derivative of ;

[0052] The cubic spline interpolation method can not only eliminate mutations in the data, but also fill in the missing data in the short term, providing a smooth and continuous data sequence for the subsequent baseline model establishment;

[0053] The smoothed data is processed for outliers and corrected. In the monitoring data of chemical equipment, due to occasional sensor failures or external interference, outliers that deviate significantly from the true value may be generated. The system uses a distributed distance correction method to detect and correct outliers by comparing the data point with the median of the data at the adjacent time point. The correction formula is: ,in, for The median of the data within n time steps before and after, is the dispersion coefficient of data j, which is calculated by the range of historical data;

[0054] According to the relative median deviation of the data, exponential adjustment is performed to reduce the impact of extreme outliers. That is, the correction formula corrects the outliers through exponential decay to reduce their impact on the overall data. After correction, the data can more accurately reflect the actual operating status of the equipment. The data processed by outliers It will serve as the input for the baseline model in the next step to ensure that the model is built on reliable basic data;

[0055] After data collection and stratification, the noise-filtered results are passed as input to the data smoothing step. The smoothed data are then processed and corrected for outliers, and the final data is provided to the baseline model construction and subsequent anomaly detection.

[0056] Step 2: Establish a baseline model for data comparison. After completing multi-layer data collection and cleaning, establish a baseline model to provide a reference standard for real-time fault monitoring of chemical production equipment clusters. The baseline model is used to define the characteristic patterns of various data of the equipment under normal operating conditions, so that subsequent real-time data can be compared with the baseline model, and then identify abnormal situations that deviate from the normal state. The specific steps are as follows:

[0057] Collect and organize historical data of baseline data, select historical data of equipment within a certain period of time (for example, in the production process, every hour) as the baseline data sample. The baseline data sample should cover the normal operation data of the equipment under different working conditions and loads to reflect the full range of normal behavior of the equipment. For the jth sensor in the chemical equipment cluster, its historical collection data set is expressed as ,in, The time range for baseline data collection, the system extracts data features from the cleaned historical data set , used for subsequent modeling;

[0058] Extract key features from baseline data and use autoencoder neural network as an unsupervised feature extraction tool to capture the nonlinear characteristics of multi-dimensional data. As the baseline data input vector, the encoder of Autoencoder will Mapping to feature space : in, is an activation function, such as ReLU (rectified linear unit), and are the weight and bias parameters of the encoder;

[0059] It should be noted that the vectors in the feature space represent the multidimensional features of the device in the normal state. Such multidimensional features are used to define the normal state range of the device in the baseline model;

[0060] In feature space In this paper, a baseline model is established for the feature vector of the device in normal state to quantify the normal fluctuation range of the device. The Gaussian mixture model (GMM) is used for modeling. is the probability distribution of the feature space, and the formula is: , where K represents the number of Gaussian distributions (i.e., the number of components in GMM), is the weight of the kth Gaussian component, and are the mean and covariance matrices of the kth component respectively;

[0061] Through the GMM model, the data feature range under normal conditions can be quantified and a baseline model that adapts to the multi-dimensional feature distribution can be formed, which can provide an accurate probabilistic reference standard for subsequent comparisons;

[0062] Based on the constructed baseline model, the real-time data collected Compare to calculate the degree of deviation, that is, map the real-time data to the feature space through the encoder , calculate its deviation from the baseline model : ,in, It indicates the probability value of the feature vector in the baseline probability distribution. The larger the deviation, the more significant the deviation of the data from the normal state. When the deviation exceeds the set threshold, it is preliminarily determined that the current equipment operation is abnormal.

[0063] Furthermore, in order to ensure the long-term applicability of the baseline model, the baseline model can also be set to be dynamically updated over time to ensure that it reflects the long-term status changes of the current equipment. After each time period, the system re-evaluates the feature distribution based on the newly collected normal operation data and updates the GMM parameters in a step-by-step weighted manner, so that the model adapts to the long-term status trend changes of the equipment.

[0064] Step 3: Perform anomaly detection on the data. By comparing the current data with the baseline model in real time, identify abnormal situations that deviate from the normal state, discover equipment failures or potential problems, and issue timely warnings when the data deviates from the baseline model. The specific steps are as follows:

[0065] Multi-layer anomaly detection and classification is performed, and Isolation Forest and Principal Component Analysis (PCA) are introduced as auxiliary detection algorithms. Isolation Forest is used to separate outliers with different characteristics by constructing multiple decision trees; PCA is used to identify the deviation of the principal components of the data, to capture the main change patterns of the data, and to convert the characteristics determined by the real-time data into Input to the Isolation Forest model and calculate the anomaly score: , where IsolationScore is an isolation scoring function used to quantify the degree of abnormality of each data point;

[0066] At the same time, the principal component deviation is calculated by PCA: ,in, is the projection of the data in the principal component space, , and The results are used to classify anomaly types in a multi-layer anomaly detection manner;

[0067] In the complex environment of chemical equipment, the abnormal characteristics of different equipment and different working conditions are different. Therefore, it is necessary to set a specific threshold for each abnormal type and define the comprehensive abnormality: ,in, and is the weight coefficient, which is adjusted according to the device type and operating environment;

[0068] When the comprehensive abnormality exceeds the comprehensive threshold, the system determines that the equipment is abnormal, dynamically adjusts the comprehensive threshold to adapt to the real-time working condition changes of the equipment, and improves the flexibility of abnormality detection; the specific dynamic adjustment can be based on the fluctuation range of the real-time data of the equipment as the basis for adjusting the comprehensive threshold. For example, the comprehensive threshold is updated according to the fluctuation of the data within a time window. When the fluctuation range is large, the comprehensive threshold is appropriately reduced, otherwise it is increased;

[0069] It should be noted that after detecting that the comprehensive abnormality exceeds the threshold and determining that the equipment is abnormal, the purpose of dynamically adjusting the comprehensive threshold is to provide adaptability for future monitoring, rather than immediately using it for current abnormality judgment. Since the operating conditions of chemical production equipment are complex and often change, such as seasonal factors, process adjustments or production load changes, these changes may affect the normal fluctuation range of equipment operation. If the system threshold is updated after the abnormality is detected, it can more accurately reflect the "normal" state of the equipment under the current working conditions in subsequent monitoring, thereby avoiding false alarms due to excessive sensitivity;

[0070] For example, in some working conditions, the operation of the equipment may be on the verge of abnormality. If the threshold is not adjusted in time, it may cause the equipment to fluctuate repeatedly between abnormal and normal, causing a "frequent alarm" problem. Dynamically adjusting the threshold can help prevent this situation and improve the accuracy and importance of the alarm.

[0071] Furthermore, after an anomaly is detected, an alarm signal is sent in real time to notify the operation and maintenance personnel to conduct an inspection. To ensure the accuracy of the feedback, the system records the anomaly characteristics, type and comprehensive anomaly degree, and generates an anomaly trend chart. The trend chart displays the evolution of the anomaly through time series backtracking analysis, providing a reference for subsequent decision-making. The abnormal alarm information will also be fed back to the baseline model update to adjust the baseline standard in the long term and improve the detection accuracy of the system.

[0072] Step 4: After completing the anomaly detection in step 3, a trend prediction model is constructed. The trend prediction model is used to predict the future change trends of various key parameters in the chemical production equipment cluster, so as to provide early warning before the equipment anomaly is about to occur. This step uses time series analysis and deep learning algorithms to predict the short-term and long-term trends of the equipment's operating status, further improving the intelligence of the system. The following are the specific processing steps:

[0073] For real-time data The cleaned data that has passed the anomaly detection is re-encoded into a feature vector , and construct feature sequences The characteristic sequence Indicates from to Historical data is used to capture the dynamic changes in the equipment's operating status, standardize the feature sequence to fit the model input, and ensure the comparability of data at different time points;

[0074] In order to predict the short-term trend of the equipment, the autoregressive integrated moving average (ARIMA) model is used. The ARIMA model is suitable for time series with linear trends and can capture the volatility of the equipment in a short period of time. is the predicted value of the jth device, and the ARIMA model formula is:

[0075] ;

[0076] in, and is the model coefficient, obtained through historical data training, is the error term, used to capture random disturbances;

[0077] The prediction results of the ARIMA model provide the fluctuation trend of equipment operation in the short term and have high sensitivity to abnormal changes in a short period;

[0078] Use deep learning models to predict long-term trends, using the long short-term memory network (LSTM) model. LSTM can determine the long-term dependencies and nonlinear relationships of data through memory cells and gating mechanisms, and convert feature sequences into As input to the LSTM model, define the predicted value of the device at a future time point: ,in, and is the weight matrix of LSTM, obtained after model training. represents the activation function, h is the predicted time step, is the bias term of the forget gate;

[0079] The LSTM model can make accurate predictions of complex nonlinear trends and is suitable for the operation prediction of chemical equipment over a long period of time, making up for the limitations of the ARIMA model in long-term trends.

[0080] The prediction results of the ARIMA model and the LSTM model are combined to form a comprehensive prediction value. , the outputs of the two models are fused using weighted average: ,in, and is the weight coefficient, which is determined by historical data verification, so that the prediction contribution of the two models is optimal, and , to ensure the balance of the fusion results;

[0081] The integrated prediction value after fusion has both short-term sensitivity and long-term stability, and can more accurately reflect the trend changes of the equipment;

[0082] After determining the future trend forecast value, the system performs trend analysis on the forecast results and generates warning information. If the comprehensive forecast value exceeds the normal baseline model range, the system triggers a warning signal to notify the operation and maintenance personnel and optimize the future comprehensive threshold. Through trend warning, the system can identify potential risks before a failure occurs, providing proactive and forward-looking management methods for equipment operation and maintenance.

[0083] Step 5 is to dynamically adjust the comprehensive threshold of the equipment in real time based on the prediction results. The purpose of dynamically adjusting the threshold is to ensure that the system has higher sensitivity when potential anomalies or mutations are about to occur based on the future operation trend of the equipment, so as to respond to potential failures in a timely manner. The specific steps are as follows:

[0084] Before dynamically adjusting the comprehensive threshold, calculate the deviation of the future operating state of the equipment, and generate the future moment prediction value through the trend prediction model It is compared with the normal range of the baseline model to assess its degree of deviation. The central value of the baseline model is set to , then the prediction deviation is: ,in, It is a measure of uncertainty in the trend forecasting model and can be estimated from the statistics of forecast errors in historical data;

[0085] The predicted deviation is used to indicate the degree to which the future equipment operation status deviates from the normal status and provides a basis for subsequent dynamic threshold adjustment;

[0086] The comprehensive threshold of the monitoring system is adaptively adjusted according to the predicted deviation. The original static threshold is set as , define the dynamically adjusted threshold as: ,in, is the sensitivity coefficient, which is used to control the response intensity of the threshold to the deviation. When the predicted deviation is large, the dynamic threshold It will be significantly reduced, making the system more sensitive when potential anomalies occur; when the deviation is small or there is no obvious trend change, the threshold value will not change much to avoid false alarms;

[0087] Perform multi-level allocation of dynamic comprehensive thresholds, that is, according to different monitoring parameters (such as vibration, temperature, pressure), the system can adjust the comprehensive thresholds in layers according to the importance of the parameters. is the dynamic threshold of the jth parameter, then the dynamic adjustment formula is: ,in, It represents the threshold sensitivity coefficient of the jth parameter. Different parameters can be set differently according to the sensitivity to the fault. , used to ensure that important parameters have high sensitivity, while minor parameters remain stable to avoid frequent alarms;

[0088] For example, real-time monitoring and control of vibration anomalies. In chemical production, the vibration level of key equipment (such as pumps, compressors, etc.) is usually a key indicator of equipment health. When the vibration is detected to exceed the dynamically adjusted threshold, it indicates that the equipment may have problems such as mechanical failure or imbalance.

[0089] After the system detects abnormal vibration, it can perform the following automatic control measures:

[0090] Automatic deceleration: If the vibration level exceeds the warning value but does not reach the shutdown requirement, the control system will automatically reduce the speed of the equipment to a safe level to reduce vibration shock and avoid further damage to the equipment;

[0091] Balance adjustment: For equipment with balance adjustment function (such as multi-stage pumps), the system can start the automatic balancing module to reduce vibration by adjusting the balance position of pump blades or other components;

[0092] Automatic alarm and pre-shutdown: If the vibration level continues to rise and exceeds the safety limit threshold, the control system will send an alarm signal and trigger the equipment's pre-shutdown control program to ensure the safety of personnel and equipment;

[0093] Through the above control measures, the system can respond to vibration faults at an early stage, thereby reducing equipment losses and safety hazards;

[0094] In terms of temperature anomaly monitoring and cooling control, in a reactor or heat exchange equipment, when it is detected that the equipment temperature exceeds the dynamic adjustment threshold, it may mean thermal runaway, cooling system failure or reaction abnormality;

[0095] When the temperature deviates from the normal range, the system automatically executes the following control strategies:

[0096] Increase cooling flow: When the temperature reaches the warning range, the system adjusts the flow valve of cooling water or coolant to increase the coolant flow to quickly reduce the temperature;

[0097] Reduce heating power: If the temperature continues to rise and approaches the set critical value, the system will reduce or suspend the power of the heating device to avoid the impact of excessive temperature on equipment and processes;

[0098] Start emergency cooling: When the temperature exceeds the safety limit, the system can start the backup cooling device and send an alarm signal to notify the operation and maintenance personnel to intervene;

[0099] The above-mentioned automatic temperature control measures can effectively avoid equipment failure due to overheating and improve the safety and stability of the production process.

[0100] When a fault or abnormality is detected in real time, automated equipment adjustment measures can be executed to ensure the safe and efficient operation of the chemical production equipment cluster. Through automatic deceleration, cooling adjustment and other control measures, the system can effectively respond to different types of equipment failures and achieve a full range of automated fault response.

[0101] It should be noted that the threshold information in this embodiment is pre-set by professionals and will not be explained in detail here. Some parameter English letters in the embodiments have the same situation, but different meanings are explained when used, which will not be explained one by one here.

[0102] The present invention ensures the integrity and accuracy of the data by collecting important data of different frequencies in layers, and adopts filtering, data completion and outlier correction. The system uses historical data to establish a baseline model, extracts multi-dimensional features of the equipment in normal state through the autoencoder, and quantifies the equipment operation characteristics in combination with the Gaussian mixture model, so as to judge the deviation of the equipment in real time. For possible false alarms or missed alarms, the state of the equipment is secondary judged through isolation forest and principal component analysis, and the abnormal situation of the equipment is comprehensively determined by the deviation degree, isolation score and principal component deviation degree, which significantly improves the detection accuracy.

[0103] When the abnormal signal of the equipment exceeds the comprehensive threshold, the system generates an alarm signal and performs trend forecasting, using the ARIMA model and LSTM model to integrate and forecast short-term and long-term trends to form a comprehensive forecast value and accurately predict the future status of the equipment. At the same time, the system can dynamically adjust the threshold based on the real-time analysis results and automatically execute control measures when a fault trend is detected to ensure the safe and stable operation of the equipment.

[0104] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0105] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0106] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0107] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0108] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which 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.

[0109] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time fault monitoring IoT system for chemical production equipment clusters, characterized by: It includes a data acquisition module, a comparison and analysis module, an equipment abnormality determination module, and a fault trend prediction module, and each module is connected through signals; The data collection module is used to collect data from chemical production equipment clusters in layers according to the frequency and importance of different types of sensor data, and to filter, complete and correct abnormal values ​​of the data; The comparative analysis module is used to use historical data as baseline data, use an autoencoder to extract baseline data features, quantify the multi-dimensional feature distribution of the device under normal conditions through a Gaussian mixture model and build a baseline model, map the real-time data to the feature space through the encoder, and calculate the deviation from the baseline model to determine the abnormality of the device; The equipment anomaly determination module is used to combine the isolation forest and principal component analysis to secondary identify data anomalies, comprehensively judge the equipment anomalies based on the deviation, isolation score and principal component deviation, and generate an alarm signal when the comprehensive threshold is exceeded; The fault trend prediction module is used to build a trend prediction model, perform short-term predictions through the ARIMA model and long-term predictions through the LSTM model, and generate a comprehensive prediction value by weighted fusion results. It analyzes future trends in combination with the baseline model and automatically executes equipment control measures when equipment abnormalities are detected. It is used to combine the isolation forest and principal component analysis to identify data anomalies for the second time. It comprehensively judges the equipment anomalies based on the deviation, isolation score and principal component deviation. When the comprehensive threshold is exceeded, an alarm signal is generated. The specific steps are as follows: Use Isolation Forest to build a decision tree to separate outliers of different features and determine anomaly scores; Use principal component analysis to identify deviations from the principal components of the data, determine the data variation pattern, and calculate the principal component deviation; According to the results of deviation, isolation score and principal component deviation, the abnormality type is classified by multi-layer abnormality detection to determine the comprehensive abnormality of different equipment and equipment under different working conditions; When the comprehensive abnormality exceeds the comprehensive threshold, it is determined that the equipment operation is abnormal, and the comprehensive threshold is adjusted according to the real-time operating condition changes of the equipment.

2. The real-time fault monitoring Internet of Things system for chemical production equipment cluster according to claim 1 is characterized in that: It is used to collect hierarchical data of chemical production equipment clusters according to the frequency and importance of different types of sensor data, and to filter, complete and correct abnormal values ​​of data. The specific steps are as follows: The collected data is divided into high-frequency data, medium-frequency data, and low-frequency data; Noise filtering is performed based on the characteristics of the sensor and the historical noise level, and the data mutation points in the filtered data are smoothed using the cubic spline interpolation method; The smoothed data is subjected to outlier processing and data correction. The data points are compared with the medians of the data at adjacent time points using the distributed distance correction method to detect and correct outliers.

3. The real-time fault monitoring Internet of Things system for chemical production equipment cluster according to claim 2 is characterized in that: It is used to use historical data as baseline data, use an autoencoder to extract baseline data features, quantify the multi-dimensional feature distribution of the device in a normal state through a Gaussian mixture model and build a baseline model, map real-time data to the feature space through the encoder, calculate the deviation from the baseline model to determine the abnormality of the device, and the specific steps include: Collect and organize historical data of baseline data, and select historical data of equipment within a fixed period of time as baseline data samples. The baseline data samples include normal operating data of equipment under different working conditions and loads; Using autoencoder neural networks as unsupervised feature extraction tools to identify nonlinear characteristics of baseline data; In the feature space, a Gaussian mixture model is used to establish a baseline model for the feature vectors of the device in a normal state, quantify the data feature range in a normal state, and form a baseline model that adapts to the multi-dimensional feature distribution; Based on the constructed baseline model, the collected real-time data is compared, the real-time data is mapped to the feature space through the encoder, and the deviation from the baseline model is calculated; When the deviation is greater than the set threshold, it is preliminarily determined that there is an abnormality in the current equipment operation.

4. The real-time fault monitoring Internet of Things system for chemical production equipment cluster according to claim 3 is characterized in that: It is used to build a trend prediction model, use the ARIMA model for short-term prediction and the LSTM model for long-term prediction, and weight the fusion results to generate a comprehensive prediction value, including the following steps: Arrange the features of real-time data and historical data, re-encode the cleaned data that has passed the anomaly detection into feature vectors, construct feature sequences, and standardize the feature sequences; Predict the short-term trend of the equipment and use the autoregressive integrated moving average model to determine the short-term forecast value of the equipment; Use long short-term memory networks to predict long-term trends, determine long-term dependencies and nonlinear relationships of data, use feature sequences as input to long short-term memory networks, and determine long-term prediction values ​​of devices; The prediction results of the ARIMA model and the LSTM model are fused using a weighted average method to form a comprehensive prediction value.

5. The real-time fault monitoring Internet of Things system for chemical production equipment cluster according to claim 4 is characterized in that: Combined with the baseline model, future trends are analyzed and when device anomalies are detected, device control measures are automatically executed, including the following steps: After determining the comprehensive prediction value, a trend analysis is performed and corresponding warning information is generated. If the comprehensive prediction value exceeds the normal baseline model range, the system triggers a warning signal, notifies the operation and maintenance personnel, and optimizes the comprehensive threshold; Calculate the deviation of the future operating state of the equipment, compare the future moment prediction value generated by the trend prediction model with the normal range of the baseline model, and determine the prediction deviation. The prediction deviation indicates the degree to which the future operating state of the equipment deviates from the normal state. Adaptively adjust the comprehensive threshold according to the prediction deviation, and determine the dynamically adjusted comprehensive threshold; Multi-level allocation of comprehensive thresholds is carried out according to different monitored parameters, including vibration, temperature, and pressure. Layered comprehensive thresholds are adjusted according to the parameters, and control measures for the equipment are automatically executed.

Images