A compressor control method and system

Abstract

The present invention provides a compressor control method and system. Belonging to the technical field of compressor control, the method includes: real-time collecting multi-dimensional data during the operation of the compressor through a multi-parameter sensor network; preprocessing the collected multi-dimensional data and constructing a data set; through a reinforcement learning model based on a deep neural network, learning and optimizing the compressor control strategy based on historical data and the current environmental state; continuously iterating and updating through simulation and online learning; based on an adaptive neural network model, real-time predicting the operation state and performance parameters of the compressor for a period of time in the future, and dynamically adjusting the network structure and parameters according to the current working conditions and the compressor historical data. Real-time collect multi-dimensional data during the operation of the compressor through a multi-parameter sensor network. These data are preprocessed, including denoising, filling missing values, outlier detection, etc., to ensure data quality.

Description

Technical Field

[0001] The present invention provides a compressor control method and system, belonging to the technical field of compressor control. Background Art

[0002] Most traditional compressor control technologies rely on preset algorithms and fixed control logics, making it difficult to cope with dynamically changing working conditions. Although attempts have been made to optimize with simple machine learning models in recent years, these methods are often limited by model complexity, data dependence, and generalization ability, and it is difficult to achieve in-depth optimization of compressor performance. In particular, when facing complex working conditions (such as extreme temperatures, pressure fluctuations, load changes, etc.), traditional methods often cannot quickly and accurately adjust the compressor operating parameters, resulting in low energy efficiency and frequent failures. Summary of the Invention

[0003] The present invention provides a compressor control method and system to solve the problems mentioned in the above background art:

[0004] A compressor control method proposed by the present invention, the method includes:

[0005] S1. Through a multi-parameter sensor network, multi-dimensional data during the operation of the compressor is collected in real time; the collected multi-dimensional data is preprocessed, and a data set is constructed;

[0006] S2. Through a reinforcement learning model based on a deep neural network, based on historical data and the current environmental state, the compressor control strategy is learned and optimized; through simulation and online learning, continuous iterative updates are performed;

[0007] S3. Based on an adaptive neural network model, the operating state and performance parameters of the compressor in the next period of time are predicted in real time, and according to the current working conditions and the compressor historical data, the network structure and parameters are dynamically adjusted;

[0008] S4. The outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module are fused to generate a comprehensive control strategy;

[0009] S5. According to the comprehensive control strategy, the compressor operating parameters are intelligently adjusted. At the same time, through a closed-loop feedback mechanism, the compressor operating state is continuously monitored, and the control strategy is adjusted in a timely manner.

[0010] A compressor control system proposed by the present invention, the system includes:

[0011] Data acquisition module: Through a multi-parameter sensor network, multi-dimensional data during the operation of the compressor is collected in real time, the collected multi-dimensional data is preprocessed, and a data set is constructed;

[0012] Optimization Learning Module: Based on a reinforcement learning model of a deep neural network, it learns and optimizes the compressor control strategy based on historical data and the current environmental state; through simulation and online learning, it continuously iterates and updates;

[0013] Real-time Prediction Module: Based on an adaptive neural network model, it makes real-time predictions on the operating state and performance parameters of the compressor in the next period of time, and dynamically adjusts the network structure and parameters according to the current working conditions and compressor historical data;

[0014] Strategy Generation Module: It fuses the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module to generate a comprehensive control strategy;

[0015] Closed-loop Control Module: According to the comprehensive control strategy, it intelligently adjusts the operating parameters of the compressor; at the same time, through a closed-loop feedback mechanism, it continuously monitors the operating state of the compressor and timely adjusts the control strategy.

[0016] Advantages of the present invention: Multidimensional data during the operation of the compressor is collected in real time through a multi-parameter sensor network. These data are preprocessed, including denoising, filling missing values, outlier detection, etc., to ensure data quality. Then, these data are organized into a time series dataset, providing a basis for subsequent model training and prediction; Using historical data and the current environmental state, a deep neural network reinforcement learning model is used to learn and optimize the control strategy of the compressor. The model is continuously iteratively updated through simulation and online learning to adapt to different working conditions and environmental changes; An adaptive neural network model is used to make real-time predictions on the operating state and performance parameters of the compressor in the future for a period of time. Dynamically adjust the network structure and parameters according to the current working condition and historical data to improve the accuracy of prediction; The outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module are fused to generate a comprehensive control strategy. This strategy takes into account immediate rewards and long-term returns, as well as stability and robustness under different working conditions; Intelligently adjust the operating parameters of the compressor according to the comprehensive control strategy, and continuously monitor the operating state and performance parameters of the compressor through a closed-loop feedback mechanism. When an anomaly or deviation from the expected value is detected, adjust the control strategy in a timely manner, and ensure the safe operation of the compressor through an anomaly handling process and a fault warning system; As new data continuously arrives, the model will be updated according to the importance of the data to maintain its accuracy and adaptability. At the same time, through the rolling time window technology and the online learning framework, the model can continuously receive new operating data and self-optimize; Build a simulation system that highly simulates the actual operating environment of the compressor, receive the operating data of the compressor in the simulation environment in real time through the online learning framework, and interact with the deep reinforcement learning model. This allows the model to learn and adapt in a changing environment, thereby improving the performance of the control strategy; Verify and optimize the comprehensive control strategy in the simulation environment, evaluate the performance of the strategy through comparative experiments, and adjust the parameters of the fusion algorithm and the control strategy according to the verification results; Through the fault warning system, when a potential fault is predicted, notify the operator in advance to take corresponding measures. At the same time, the anomaly handling process can immediately initiate emergency protection measures when an anomaly is detected, ensuring that the critical events and abnormal states of the compressor are recorded and processed. Description of the Drawings

[0017] Figure 1 It is a flowchart of the method steps described in the present invention;

[0018] Figure 2 It is a block diagram of the system described in the present invention. Detailed Embodiments

[0019] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0020] An embodiment of the present invention is as follows Figure 1 shown, a compressor control method, the method comprising:

[0021] S1. Through a multi-parameter sensor network, multi-dimensional data during the operation of the compressor is collected in real time, the multi-dimensional data including pressure, temperature, current, vibration, and flow rate; the collected multi-dimensional data is preprocessed, and a data set is constructed;

[0022] S2. Through a reinforcement learning model based on a deep neural network, based on historical data and the current environmental state (such as operating conditions, compressor state, etc.), the compressor control strategy is learned and optimized; through simulation and online learning, continuous iterative updates are performed to generate an optimal control instruction;

[0023] S3. Based on an adaptive neural network model, the operating state and performance parameters of the compressor for a period of time in the future are predicted in real time, and according to the current operating conditions and the compressor historical data, the network structure and parameters are dynamically adjusted;

[0024] S4. The outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module are fused to generate a comprehensive control strategy; this strategy not only considers the optimal control under the current operating conditions, but also incorporates anticipatory adjustments to future state changes;

[0025] S5. According to the comprehensive control strategy, the operating parameters of the compressor (such as speed, intake air volume, loading / unloading strategy, etc.) are intelligently adjusted; at the same time, through a closed-loop feedback mechanism, the operating state of the compressor is continuously monitored, and the control strategy is adjusted in a timely manner.

[0026] The working principle of the above technical solution is as follows: Use a multi-parameter sensor network to collect key data during the operation of the compressor in real time, including pressure, temperature, current, vibration, and flow rate, etc.; preprocess the collected multi-dimensional data, including data cleaning (removing noise, filling missing values, etc.), anomaly detection (identifying and processing abnormal data points), and feature engineering (extracting useful features, dimensionality reduction, etc.) to construct a high-quality data set; use a reinforcement learning model based on a deep neural network, combined with historical data and the current environmental state (such as operating conditions, compressor status, etc.), to learn and optimize the control strategy of the compressor; through simulation and online learning, the model is continuously iteratively updated, aiming to maximize long-term benefits (such as energy efficiency, stability, etc.), and generate optimal control instructions; based on an adaptive neural network model, real-time predict the operating state and performance parameters of the compressor for a period of time in the future; dynamically adjust the network structure and parameters according to the current operating conditions and compressor historical data to improve the accuracy of prediction. The prediction result is used as one of the inputs of the deep reinforcement learning model to assist it in making more accurate decisions; fuse the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module to generate a comprehensive control strategy; this strategy not only considers the optimal control under the current operating conditions but also incorporates anticipatory adjustments to future state changes, thus achieving a comprehensive optimization of the compressor's operating parameters; according to the comprehensive control strategy, intelligently adjust the operating parameters of the compressor (such as speed, intake air volume, loading / unloading strategy, etc.); through a closed-loop feedback mechanism, continuously monitor the operating state of the compressor and timely adjust the control strategy.

[0027] The effects of the above technical solutions are as follows: By collecting multi-dimensional data in real time and performing preprocessing, a high-quality data set is constructed, providing a solid foundation for subsequent model learning and optimization. It helps to more accurately reflect the actual operating state of the compressor, thereby formulating more efficient control strategies; The reinforcement learning model can continuously iterate and update the control strategy according to historical data and the current environmental state, aiming to maximize long-term benefits. It can not only improve the operating efficiency of the compressor, but also optimize energy use and reduce energy consumption costs while maintaining high efficiency; The adaptive neural network model can predict the operating state and performance parameters of the compressor in the future for a period of time in real time and make dynamic adjustments according to the current working conditions and historical data. This predictive adjustment helps to detect and address potential operating problems in advance, thereby enhancing the operating stability and reliability of the compressor; The comprehensive control strategy integrates the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module, considering both the optimal control under the current working conditions and the predictive adjustment of future state changes. This comprehensive control strategy helps to ensure that the compressor can maintain a stable operating state under various working conditions; By introducing advanced artificial intelligence technologies such as deep neural networks, reinforcement learning, and adaptive neural networks, the intelligent optimization and real-time adjustment of the compressor control strategy are realized. This greatly improves the intelligent level and automation degree of the compressor, reducing manual intervention and dependence; The closed-loop feedback mechanism can continuously monitor the operating state of the compressor and adjust the control strategy in a timely manner according to the actual situation. This automatic adjustment mechanism helps to ensure that the compressor always operates in the best state, improving the reliability and flexibility of the overall system; By real-time monitoring and predicting the operating state and performance parameters of the compressor, potential operating problems and faults can be detected in a timely manner. This helps to take maintenance measures in advance to avoid the occurrence or expansion of faults, thereby reducing maintenance costs and downtime; The intelligent and automatic control strategy helps to optimize the operating parameters and working modes of the compressor, reducing unnecessary wear and tear. This helps to extend the service life of the compressor and improve the economy and sustainability of the overall system.

[0028] In one embodiment of the present invention, S1 includes:

[0029] S11. Real-time collect multi-dimensional data during the operation of the compressor through a multi-parameter sensor network, and use a sliding window algorithm to smooth the collected multi-dimensional data to remove noise;

[0030] S12. Detect and fill in missing values through statistical methods, detect outliers using machine learning algorithms (such as isolation forest), and mark or eliminate them;

[0031] S13. Implement feature engineering to extract feature variables that have important impacts on the operating state of the compressor, such as pressure volatility and temperature change trend; organize the preprocessed data in a time series to construct a data set containing historical data and current data.

[0032] S14. Store the data set through a distributed storage scheme and back up the data set based on the set data backup mechanism.

[0033] The working principle of the above technical solution is as follows: A multi-parameter sensor network is deployed on the compressor to collect multi-dimensional data during the operation of the compressor in real time. These data include but are not limited to pressure, temperature, current, vibration, and flow. To remove noise in the data, a sliding window algorithm is used to smooth the collected data. The sliding window algorithm defines a fixed-size or variable-size window on the data sequence and slides the window on the sequence at a certain step length, so as to perform average or weighted average processing on the data within the window at each position to achieve the effect of smoothing the data; missing values are detected and filled through statistical methods (such as interpolation method, regression method, etc.) to ensure the integrity of the data. Then, a machine learning algorithm (such as Isolation Forest) is applied to detect outliers. The Isolation Forest algorithm is an outlier detection method based on decision trees. It distinguishes outliers from normal points by placing outliers in the shallower branches of the tree. The detected outliers can be marked or removed to improve the quality and reliability of the data; in the feature engineering stage, feature variables that have important impacts on the operating state of the compressor are extracted. These feature variables can reflect the operating conditions, performance change trends, and potential faults of the compressor. For example, feature variables such as pressure volatility and temperature change trend can be extracted. Then, the preprocessed data is organized in a time series to construct a data set containing historical data and current data. Such a data set is helpful for subsequent model learning and optimization; to efficiently store and manage large-scale data sets, a distributed storage scheme is adopted. The distributed storage scheme improves the reliability and scalability of the data by dispersing the data storage on multiple nodes. At the same time, the data set is backed up based on the set data backup mechanism to prevent data loss or damage. The data backup can be saved locally or on a backup server, or an online backup service such as cloud storage can be used to ensure the security and recoverability of the data.

[0034] The effects of the above technical solutions are as follows: By using a multi-parameter sensor network to collect multi-dimensional data during the operation of the compressor in real time, the timeliness and comprehensiveness of the data are ensured; the sliding window algorithm is used to smooth the collected multi-dimensional data, effectively removing noise and improving the accuracy and reliability of the data; statistical methods are used to detect and fill in missing values, avoiding analysis biases caused by data missing; machine learning algorithms (such as Isolation Forest) are applied to detect outliers and mark or remove them, further improving the purity and accuracy of the data set; feature engineering is implemented to extract feature variables that have important impacts on the operation state of the compressor, such as pressure volatility, temperature change trend, etc., which can more intuitively reflect the operation state and performance of the compressor; feature extraction helps subsequent data analysis and model establishment, improving the accuracy and generalization ability of the model; the preprocessed data is organized in time series to construct a data set containing historical data and current data, providing a basis for time series analysis and prediction; the data set is stored through a distributed storage solution, making full use of network resources and improving storage efficiency and scalability; distributed storage also helps to achieve high availability and fault tolerance of the data, enhancing the reliability and security of the data; based on setting up a data backup mechanism to back up the data set, ensuring the recoverability of the data in case of accidents; the data backup mechanism helps to reduce losses and risks caused by data loss or damage.

[0035] In one embodiment of the present invention, the S13 includes:

[0036] Use correlation analysis to evaluate the correlation between each dimension of data (such as pressure, temperature, current, vibration, etc.) and the operation state of the compressor (such as efficiency, fault tendency, etc.). Based on the analysis results, screen out the feature variables that are highly correlated with the operation state of the compressor.

[0037] Further quantify the importance of each feature variable through feature importance evaluation methods (such as the feature importance of Random Forest, the coefficients of Lasso regression, etc.); according to the evaluation results, retain the feature variables that have important impacts on the operation state of the compressor, such as pressure volatility, temperature change trend, etc.

[0038] Perform standardization or normalization processing on the screened feature variables, and derive new feature variables based on domain knowledge and data characteristics; for example, the ratio of pressure to temperature, the spectral characteristics of vibration signals, etc. can be calculated to enrich the information content of the data set.

[0039] Align all feature variables in time series, that is, there are corresponding data records at each timestamp; for missing timestamps, perform interpolation or filling according to the trends of the front and back data.

[0040] According to the operating cycle of the compressor and the data analysis requirements, the data is divided into multiple time windows; each time window contains a time series data of a certain length;

[0041] Organize the preprocessed historical data and current data in time series to construct a comprehensive data set containing rich information; among them, the historical data is used for model training, and the current data is used for model validation and prediction;

[0042] Divide the comprehensive data set into a training set, a validation set and a test set; the training set is used for model training, the validation set is used for model selection and parameter tuning, and the test set is used for evaluating the generalization ability of the model.

[0043] The working principle of the above technical solution is as follows: Use statistical methods (such as Pearson correlation coefficient, etc.) to evaluate the correlation between data in each dimension (such as pressure, temperature, current, vibration, etc.) and the operating state of the compressor (such as efficiency, fault tendency, etc.); Through the analysis results, identify the characteristic variables that are highly correlated with the operating state of the compressor. These variables can usually better reflect the operating state and performance changes of the compressor; Use feature importance evaluation methods (such as the feature importance of random forest, the coefficients of Lasso regression, etc.) to further quantify the importance of each characteristic variable; According to the evaluation results, screen out the characteristic variables that have an important impact on the operating state of the compressor, such as pressure volatility, temperature change trend, etc. These variables will play a key role in subsequent model training; Standardize or normalize the selected characteristic variables to eliminate the dimensional differences between different features and improve the convergence speed and prediction performance of the model; Based on domain knowledge and data characteristics, derive new characteristic variables. For example, calculate the ratio of pressure to temperature, the spectral characteristics of vibration signals, etc., to enrich the information content of the data set. These derived features help the model to more comprehensively understand the operating state of the compressor; Align all characteristic variables in the time series to ensure that there are corresponding data records at each timestamp. This helps to maintain the consistency and integrity of the data; For missing timestamps, interpolate or fill them according to the trends of the previous and subsequent data. This can ensure the continuity and reliability of the data set and avoid negative impacts on model training; According to the operating cycle of the compressor and the data analysis requirements, divide the data into multiple time windows. Each time window contains a certain length of time series data, which helps the model to capture the temporal dependence and periodic changes of the data; Organize the preprocessed historical data and current data in time series to construct a comprehensive data set containing rich information. Among them, the historical data is used for model training, and the current data is used for model validation and prediction; Divide the comprehensive data set into a training set, a validation set, and a test set. The training set is used for model training, providing enough data for the model to learn the internal laws and patterns of the data; The validation set is used for model selection and parameter tuning, helping to select the optimal model parameters and configurations; The test set is used to evaluate the generalization ability of the model to ensure that the model can also perform well on unseen data.

[0044] The effects of the above technical solutions are as follows: Through correlation analysis, characteristic variables highly correlated with the operating state of the compressor can be accurately identified, avoiding the interference of irrelevant or redundant data; the selected characteristic variables are more focused on the key indicators of compressor operation, improving the efficiency and accuracy of data analysis and model training; using the feature importance evaluation method, the importance of each characteristic variable is further quantified, ensuring that the most valuable features can be fully utilized during model training; characteristic variables that have an important impact on the operating state of the compressor are retained, such as pressure volatility, temperature change trend, etc., and these characteristics can more accurately reflect the operating state and performance changes of the compressor; the selected characteristic variables are standardized or normalized to eliminate the dimensional differences between different characteristics, improving the convergence speed and prediction performance of the model; standardization or normalization helps reduce the risk of overfitting of the model, improving the stability and generalization ability of the model; based on domain knowledge and data characteristics, new characteristic variables are derived, such as the ratio of pressure to temperature, spectral characteristics of vibration signals, etc., and these characteristics can provide richer information, helping the model to more comprehensively understand the operating state of the compressor; feature derivation also helps to capture the potential relationships between data, improving the prediction accuracy and robustness of the model; all characteristic variables are aligned in the time series to ensure that there are corresponding data records at each timestamp, helping to maintain the consistency and integrity of the data; time series alignment provides a reliable data basis for subsequent model training and prediction; for missing timestamps, interpolation or filling is performed according to the trends of the previous and subsequent data to ensure the continuity and reliability of the data set; missing value processing avoids analysis biases and model performance degradation caused by missing data; according to the operating cycle of the compressor and the data analysis requirements, the data is divided into multiple time windows, and each time window contains a certain length of time series data; the time window setting helps the model capture the temporal dependence and periodic changes of the data, improving the prediction accuracy and adaptability of the model; the preprocessed historical data and current data are organized in time series to construct a comprehensive data set containing rich information; the comprehensive data set provides sufficient data support for model training and prediction, helping to improve the accuracy and generalization ability of the model; the comprehensive data set is divided into a training set, a validation set, and a test set to ensure the independence of model training, selection, and evaluation; the training set is used for model training, the validation set is used for model selection and parameter tuning, and the test set is used for evaluating the generalization ability of the model. This division method helps to avoid overfitting and underfitting problems, improving the reliability and practicality of the model.

[0045] In one embodiment of the present invention, the S2 includes:

[0046] S21. Construct a reinforcement learning model architecture based on a deep neural network. The model framework includes a policy network and a value network. The policy network is used to generate control instructions, and the value network is used to evaluate the pros and cons of the control instructions.

[0047] S22. Divide the constructed dataset into a training set and a validation set for model training and validation. Use the stochastic gradient descent algorithm to perform offline training on the model to learn the optimal control strategy of the compressor under different working conditions.

[0048] S23. Optimize the model performance by adjusting hyperparameters (such as the learning rate and batch size). Construct a compressor operation simulation environment to simulate the behavior of the compressor under different working conditions, including normal working conditions, abnormal working conditions, etc.

[0049] S24. Implement online learning in the simulation environment and collect new data for model update at the same time. Continuously update the data in the simulation environment through the rolling time window technology.

[0050] S25. Generate optimal control instructions based on the output of the deep reinforcement learning model, including adjustment suggestions for parameters such as the compressor speed, intake air volume, and loading / unloading strategy.

[0051] The working principle of the above technical solution is as follows: A reinforcement learning model architecture based on a deep neural network is constructed, which includes a policy network and a value network; the policy network is responsible for generating control instructions that will directly act on the compressor to adjust its operating state; the value network is used to evaluate the quality of the control instructions generated by the policy network, so as to provide feedback for the optimization of the policy network; the constructed data set is divided into a training set and a validation set, which are used for model training and validation respectively; the model is trained offline using the stochastic gradient descent algorithm. In the offline training stage, the model gradually optimizes the parameters of the policy network and the value network by repeatedly learning the data in the training set to learn to optimally control the compressor under different working conditions; the performance of the model is optimized by adjusting hyperparameters (such as the learning rate and batch size). The adjustment of these hyperparameters will directly affect the training speed and convergence performance of the model; a compressor operation simulation environment is constructed to simulate the behavior of the compressor under different working conditions. These working conditions include normal working conditions and abnormal working conditions, etc., to comprehensively test the adaptability and robustness of the model; online learning is implemented in the simulation environment. In the online learning stage, the model will continuously receive new data inputs and adjust its parameters and strategies in real time according to these data to adapt to more complex and changeable working condition conditions; through the rolling time window technology, the data in the simulation environment is continuously updated. This helps to maintain the timeliness and accuracy of the model data, thereby improving the prediction and control performance of the model; based on the output of the deep reinforcement learning model, optimal control instructions are generated. These instructions include adjustment suggestions for parameters such as compressor speed, intake air volume, loading / unloading strategy, etc.; these control instructions will directly guide the operation of the compressor to achieve a more efficient and stable operating state. At the same time, these instructions can also be dynamically adjusted according to the actual situation to adapt to changing working condition conditions.

[0052] The effects of the above technical solution are as follows: Deep neural networks have powerful learning capabilities and can automatically learn complex and high-order feature representations, making them suitable for processing large-scale and high-dimensional data. In a reinforcement learning model, deep neural networks are used to construct a policy network and a value network, which are responsible for generating control instructions and evaluating the pros and cons of control instructions respectively. This structure enables the model to more accurately learn the optimal control strategy of the compressor under different operating conditions; Reinforcement learning learns by the interaction between the agent and the environment to maximize the long-term cumulative reward. In the compressor control scenario, the agent can select actions (such as adjusting the speed, intake air volume, etc.) according to the state of the environment (such as the compressor operating condition), and adjust its behavior strategy through the feedback of the environment (such as compressor performance, energy consumption, etc.). This trial-and-error method enables the model to adapt to different operating conditions and improve robustness; Dividing the constructed data set into a training set and a validation set helps the training and validation of the model. The training set is used for the learning process of the model, while the validation set is used to evaluate the performance of the model to ensure that the model can also show good generalization ability on unseen data; Using the stochastic gradient descent algorithm to perform offline training on the model can efficiently update the model parameters and accelerate the training process. At the same time, by continuously adjusting hyperparameters such as the learning rate, the model performance can be further optimized; Constructing a compressor operation simulation environment can simulate the behavior of the compressor under different operating conditions, including normal operating conditions and abnormal operating conditions, etc. This provides the possibility for the online learning and adaptation of the model, enabling the model to show good performance under more complex and variable operating conditions; Implementing online learning in the simulation environment enables the model to adapt to new operating conditions and collect new data for model update. By continuously updating the data in the simulation environment through the rolling time window technology, it is ensured that the model can always reflect the latest operating conditions; Based on the output of the deep reinforcement learning model, optimal control instructions can be generated, including adjustment suggestions for parameters such as compressor speed, intake air volume, loading / unloading strategy, etc. These suggestions can guide the optimal control of the compressor during actual operation and improve the system performance and stability; This technical solution has broad application prospects in the field of compressor control and can bring significant economic and social benefits to enterprises. For example, by optimizing the compressor control strategy, energy consumption can be reduced, production efficiency can be improved, and the equipment life can be extended, etc.

[0053] In one embodiment of the present invention, the S24 includes:

[0054] S241. Construct a simulation system that highly simulates the actual operating environment of the compressor, receive the compressor operation data in the simulation environment in real time through an online learning framework, and interact the operation data with the deep reinforcement learning model;

[0055] S242. Simulate various complex and changeable operating conditions in the simulation environment. The complex and changeable operating conditions include different load demands, environmental temperature changes, and power supply voltage fluctuations. During the simulation process, collect the compressor operation data in real time, including key parameters such as pressure, temperature, current, and vibration.

[0056] S243. Through the rolling time window technology, continuously update the data in the simulation environment, and adjust the size of the time window according to the change speed of the compressor operation characteristics and the update requirements of the model.

[0057] S244. Conduct quality assessment and preprocessing on the newly collected data, online verify and evaluate the model with the new data, and evaluate the accuracy and performance of the model by comparing the model prediction results with the actual simulation results.

[0058] S245. According to the results of the online verification, adjust the parameters of the deep reinforcement learning model in real time, including the learning rate, batch size, and network weights. Based on the adaptive learning rate adjustment mechanism, dynamically adjust the learning rate according to the learning progress and error change of the model.

[0059] S246. Update the model parameters and policies obtained from online learning to the actual operating compressor control system, and iteratively optimize the control policy through the policy optimization algorithm, combining historical data and real-time data.

[0060] The working principle of the above technical solution is as follows: The simulation system can accurately replicate the physical characteristics and behavior patterns of the compressor, including stable operation under normal working conditions and response mechanisms under abnormal working conditions. Through the online learning framework, it receives the compressor operation data in the simulation environment in real time and interacts these data with the deep reinforcement learning model to provide a basis for the online learning and update of the model; the simulation system can simulate various complex and changeable working conditions, such as different load demands, ambient temperature changes, power supply voltage fluctuations, etc.; during the simulation process, it collects the operation data of the compressor in real time, including key parameters such as pressure, temperature, current, vibration, etc., and these data will be used for the online learning and verification of the model; using the rolling time window technology, it continuously updates the data in the simulation environment to ensure that the model can learn the latest compressor operation characteristics and trends; according to the change speed of the compressor operation characteristics and the update requirements of the model, it dynamically adjusts the size of the time window to improve the adaptability and accuracy of the model; it conducts quality assessment on the newly collected data to ensure the accuracy and reliability of the data; it preprocesses the data to meet the requirements of model input; by comparing the model prediction results with the actual simulation results, it evaluates the accuracy and performance of the model to provide a basis for parameter adjustment and optimization of the model; according to the results of online verification, it adjusts the parameters of the model in real time, including learning rate, batch size, network weights, etc.; it adopts an adaptive learning rate adjustment mechanism to dynamically adjust the learning rate according to the learning progress and error changes of the model to improve the convergence speed and accuracy of the model; it updates the model parameters and strategies obtained from online learning to the actual operating compressor control system to achieve the real-time application and optimization of the model; by combining historical data and real-time data, it iteratively optimizes the control strategy through a strategy optimization algorithm to find a better control instruction and parameter adjustment scheme to improve the operation efficiency and stability of the compressor.

[0061] The effects of the above technical solutions are as follows: By constructing a simulation system that highly simulates the actual operating environment of the compressor, the physical characteristics and behavior patterns of the compressor can be accurately replicated, including responses under normal and abnormal operating conditions. This helps the model to learn and optimize under conditions close to the real environment, improving the practicality and accuracy of the model; The simulation system receives compressor operation data in real time through an online learning framework and interacts with the deep reinforcement learning model. This real-time data-driven learning method enables the model to quickly adapt to changes in the compressor operation state, improving the model's response speed and adaptability; The simulation environment can simulate various complex and variable operating conditions, such as different load demands, ambient temperature changes, power supply voltage fluctuations, etc. This helps the model to learn the optimal control strategies under different operating conditions and improve the model's generalization ability; During the simulation process, key operating parameters of the compressor, such as pressure, temperature, current, vibration, etc., are collected in real time. These data provide rich information for the training and verification of the model, helping the model to more accurately predict and control the operating state of the compressor; By using the rolling time window technology to continuously update the data in the simulation environment, it is ensured that the model can learn the latest operating characteristics and trends of the compressor. This helps the model to maintain real-time tracking and adaptability to the compressor operation state; Adjust the size of the time window according to the change speed of the compressor operation characteristics and the update requirements of the model to improve the update efficiency and accuracy of the model. This flexibility enables the model to better adapt to changes in the compressor operation state; Conduct quality assessment and preprocessing on the newly collected data to ensure the accuracy and reliability of the data. This helps to avoid misleading learning results of the model due to data quality problems; Conduct online verification and evaluation of the model with new data, compare the model prediction results with the actual simulation results, and evaluate the accuracy and performance of the model. This verification method helps to timely discover problems existing in the model and make adjustments, improving the stability and reliability of the model; According to the results of online verification, adjust the parameters of the deep reinforcement learning model in real time, including learning rate, batch size, network weights, etc. This real-time adjustment method helps the model to converge to the optimal solution faster and improve the learning efficiency of the model; Based on the adaptive learning rate adjustment mechanism, dynamically adjust the learning rate according to the learning progress and error change of the model. This adaptive adjustment method helps the model to maintain a stable convergence speed during training and avoid problems of overfitting or underfitting; Update the model parameters and strategies obtained from online learning to the actual operating compressor control system to achieve real-time application and optimization of the model. This helps to transform the learning results of the model into actual production benefits and improve the operating efficiency and stability of the compressor; Through the policy optimization algorithm, combine historical data and real-time data to iteratively optimize the control strategy to find a better control instruction and parameter adjustment scheme. This iterative optimization method helps to continuously improve the performance and intelligent level of the control system.

[0062] In one embodiment of the present invention, the S243 includes:

[0063] Evaluate the timeliness of data by comparing the timestamps, rates of change, and correlations with other parameters of the data to determine whether the data still represents the current operating state of the compressor;

[0064] Design a dynamically adjusted time window size based on the change speed of the compressor's operating characteristics and the update requirements of the model; initially, set a larger time window to capture the operating trends over a longer period; as the model's adaptability to the data increases, gradually narrow the time window to focus on recent data changes;

[0065] Set a sliding strategy for the time window based on a fixed time interval or changes in the data volume, and based on the data automatic update mechanism, when new data enters the time window, automatically replace the oldest data;

[0066] Through the model feedback loop, re-enter the updated data into the deep reinforcement learning model for a new round of training and verification; evaluate the adaptability and accuracy of the model by comparing the model prediction results of the old and new data.

[0067] Identify and filter out abnormal data in the simulation environment through an anomaly detection algorithm; among them, abnormal data may be caused by sensor failures, data transmission errors, or abnormal behaviors of the compressor itself. According to the results of the anomaly detection, formulate a data filtering strategy; for example, for data determined to be abnormal, you can choose to directly delete it, replace it with an average value, or perform other reasonable processing;

[0068] Quantify the model's adaptability to new data according to the adaptability evaluation index, and dynamically adjust the model's optimization strategy according to the results of the adaptability evaluation. For example, when the model's adaptability to new data is weak, the learning rate can be increased or the network structure can be adjusted; when the model already shows strong adaptability, the learning rate can be decreased to avoid overfitting.

[0069] The working principle of the above technical solution is as follows: By comparing the timestamps, rates of change, and correlations with other parameters of the data, the timeliness of the data is evaluated. The timestamp is used to determine the data collection time, the rate of change reflects the trend of data change over time, and the correlation with other parameters helps to identify potential connections between data. These data characteristics together form the basis for evaluating whether the data still represents the current operating state of the compressor; According to the change speed of the compressor operating characteristics and the update requirements of the model, a dynamically adjusted time window size is designed. Initially, to capture the operating trend over a longer period, a larger time window is set. As the model's adaptability to the data gradually increases, the time window gradually shrinks to focus on recent data changes, thus more accurately reflecting the current state of the compressor; Based on a fixed time interval or changes in the data volume, a sliding strategy for the time window is set. When new data enters the time window, the oldest data is automatically replaced according to the sliding strategy to maintain the timeliness and representativeness of the data within the time window; Realize the automatic update and replacement of the data to ensure that the time window always contains the latest and most accurate data. This mechanism provides stable data support for the continuous learning and optimization of the model; The updated data is re-input into the deep reinforcement learning model for a new round of training and verification. By comparing the model prediction results of the new and old data, the adaptability and accuracy of the model are evaluated. This feedback loop helps the model to continuously learn and optimize to adapt to changes in the compressor operating state; Identify and filter out abnormal data in the simulation environment through anomaly detection algorithms. Abnormal data may be caused by sensor failures, data transmission errors, or abnormal behaviors of the compressor itself. According to the results of anomaly detection, a data filtering strategy is formulated, such as directly deleting abnormal data, replacing it with an average value, or performing other reasonable processing, to ensure the accuracy and reliability of the data input into the model; According to the adaptability evaluation index, quantify the adaptability of the model to new data. When the model's adaptability to new data is weak, strategies such as increasing the learning rate and adjusting the network structure are used to enhance the model's adaptability. When the model already shows strong adaptability, the learning rate is reduced to avoid overfitting and ensure the stability and accuracy of the model.

[0070] The effects of the above technical solutions are as follows: By comparing the timestamps, change rates, and correlations with other parameters of the data, the timeliness of the data is accurately evaluated to ensure that the data input into the model can truly reflect the current operating state of the compressor. This helps the model make more accurate predictions and decisions, improving the response speed and accuracy of the control system; According to the change speed of the compressor operating characteristics and the update requirements of the model, the size of the time window is dynamically adjusted. A larger time window is initially set to capture the operating trends over a longer period. As the model's adaptability to the data increases, the time window is gradually narrowed to focus on recent data changes. This dynamic adjustment strategy helps the model better adapt to changes in the compressor operating state, improving the generalization ability and robustness of the model; A sliding strategy for setting the time window is based on fixed time intervals or changes in the data volume, and automatic data update is achieved. When new data enters the time window, the oldest data is automatically replaced to ensure that the time window always contains the latest and most accurate data. This helps the model continuously learn the latest compressor operating characteristics, improving the real-time performance and accuracy of the model; Through the model feedback loop, the updated data is re-input into the deep reinforcement learning model for a new round of training and verification. By comparing the model prediction results of the new and old data, the adaptability and accuracy of the model are evaluated. This feedback loop mechanism helps the model continuously learn and optimize, improving the performance and stability of the model; Abnormal data in the simulation environment is identified and filtered out through an anomaly detection algorithm. Abnormal data may be caused by sensor failures, data transmission errors, or abnormal behaviors of the compressor itself. By detecting and filtering abnormal data in a timely manner, the model can be prevented from being misled, improving the prediction accuracy and reliability of the model; According to the results of the anomaly detection, a reasonable data filtering strategy is formulated. For data determined to be abnormal, options such as direct deletion, replacement with an average value, or other reasonable processing can be selected. This helps ensure the accuracy and consistency of the data input into the model, improving the stability and robustness of the model; According to the adaptability evaluation index, the adaptability of the model to new data is quantified. This helps to timely discover problems and deficiencies in the model, providing strong support for model optimization; According to the results of the adaptability evaluation, the optimization strategy of the model is dynamically adjusted. When the model's adaptability to new data is weak, the learning rate can be increased or the network structure can be adjusted to enhance the model's adaptability; when the model already shows strong adaptability, the learning rate can be decreased to avoid overfitting. This dynamic adjustment strategy helps the model maintain the best learning state and performance at different stages.

[0071] In one embodiment of the present invention, the S245 includes:

[0072] Analyze the error between the model prediction results and the actual simulation results; based on the error analysis, identify the performance bottlenecks of the model under different working conditions; for example, the model may perform poorly under specific load demands or environmental temperatures, which may be related to the physical characteristics or the complexity of the behavior pattern of the compressor;

[0073] Through the adaptive learning rate adjustment mechanism, dynamically adjust the learning rate according to the learning progress and error changes of the model, and analyze the impact of batch size on the model training efficiency and stability; by comparing the model performance under different batch sizes, select the optimal batch size that can ensure both training efficiency and model stability;

[0074] Implement parameter adjustment experiments in the simulation environment, select the optimal parameter settings by comparing the model performance under different parameter combinations, continuously monitor the performance changes of the model during the parameter adjustment process, and provide real-time feedback based on the monitoring results;

[0075] Evaluate the performance of the adjusted model on the validation set, apply the adjusted model parameters and strategies to the compressor control system in actual operation, and iteratively optimize the control strategy through the combination of real-time data and historical data.

[0076] The working principle of the above technical solution is as follows: Analyze the errors between the model prediction results and the actual simulation results, and calculate statistical indicators such as absolute error, relative error, and mean square error (MSE); These error indicators can quantify the accuracy of model prediction and provide a basis for subsequent model optimization; Based on the results of error analysis, identify the performance bottlenecks of the model under different working conditions. For example, the model may perform poorly under specific load demands or environmental temperatures, which may be related to the physical characteristics of the compressor or the complexity of the behavior pattern; By identifying the performance bottlenecks, the model can be optimized targeted to improve its prediction accuracy under various working conditions; Through the adaptive learning rate adjustment mechanism, dynamically adjust the learning rate according to the learning progress and error changes of the model; Monitor the change of the loss function on the validation set, and reduce the learning rate when the loss function tends to be stable to promote the convergence of the model; Increase the learning rate when the loss function suddenly increases to avoid falling into local optimal solutions; This way of dynamically adjusting the learning rate helps to improve the training efficiency and accuracy of the model; Analyze the impact of batch size on the training efficiency and stability of the model; By comparing the model performance under different batch sizes, select the optimal batch size that can ensure both training efficiency and model stability; The optimization of batch size helps to balance the training speed and stability of the model; Implement parameter adjustment experiments in the simulation environment, and by comparing the model performance under different parameter combinations, select the optimal parameter settings; Parameter adjustment experiments can systematically explore the impact of model parameters on performance and provide the possibility to find the best parameter combination; During the parameter adjustment process, continuously monitor the performance changes of the model and provide real-time feedback based on the monitoring results; Real-time feedback helps to promptly detect the changes in model performance and adjust parameters to optimize model performance; Evaluate the performance of the adjusted model on the validation set to ensure the accuracy and reliability of the model in practical applications; The performance evaluation on the validation set provides strong support for the practical application of the model; Apply the adjusted model parameters and strategies to the compressor control system in actual operation; Combine real-time data and historical data to iteratively optimize the control strategy to improve the performance and stability of the compressor control system; Iterative optimization is a continuous process that helps to continuously improve the intelligent level of the compressor control system.

[0077] The effects of the above technical solution are as follows: by calculating statistical indicators such as absolute error, relative error, mean square error (MSE), etc., the difference between the model prediction results and the actual simulation results can be quantified; based on error analysis, the performance bottleneck of the model under different working conditions can be identified, such as poor performance under specific load requirements or ambient temperature; this helps to optimize the model in a targeted manner and improve its prediction accuracy; implement parameter adjustment experiments in a simulation environment, and select the optimal parameter settings by comparing the model performance under different parameter combinations; during the parameter adjustment process, continuously monitor the performance changes of the model and provide real-time feedback based on the monitoring results; this helps to dynamically adjust the model parameters to adapt to the compressor characteristics under different working conditions, thereby improving the prediction accuracy; through the adaptive learning rate adjustment mechanism, the learning rate is dynamically adjusted according to the learning progress and error changes of the model; when the loss function tends to be stable, the learning rate is reduced to promote the convergence of the model; when the loss function suddenly increases, the learning rate is increased to avoid falling into the local optimal solution; this helps to improve the training efficiency of the model while maintaining the stability of the model; through Compare the model performance under different batch sizes, and select the optimal batch size that can ensure both training efficiency and model stability; this helps to balance the training speed and stability of the model and improve the overall performance; evaluate the adjusted model performance on the validation set to ensure the accuracy and reliability of the model in practical applications; this helps to verify the generalization ability of the model and ensure its applicability under different working conditions; apply the adjusted model parameters and strategies to the actual operating compressor control system; through the combination of real-time data and historical data, iteratively optimize the control strategy; this helps the model to continuously learn and adapt in practical applications and improve the overall performance of the compressor control system; this technical solution is based on a data-driven approach, and continuously optimizes the model through error analysis, parameter adjustment and other means; it helps to promote the intelligent development of compressor control systems and improve their automation level and operating efficiency; this technical solution provides a framework for continuous optimization, which can continuously optimize models and control strategies as data accumulates and technology advances; it helps to maintain the advancement and competitiveness of compressor control systems.

[0078] In one embodiment of the present invention, S3 includes:

[0079] S31, selecting an adaptive neural network model, and designing an input layer and an output layer of the model, wherein the input layer is used to receive the pre-processed compressor operation data, and the output layer is used to predict the operation status and performance parameters of the compressor in the future;

[0080] S32. The model is trained offline through the back propagation algorithm and the Adam optimizer, so that the model can predict the operating status and performance parameters of the compressor in the future (for example, one week);

[0081] S33. Dynamically adjust the structure and parameters of the adaptive neural network according to the current working conditions and the compressor historical data, such as adding or reducing hidden layers, adjusting the number of neurons, etc., and prevent model overfitting through regularization;

[0082] S34. Based on the preset model update strategy, when new data arrives, update the model parameters according to the importance of the data; use the prediction result of the adaptive neural network as one of the inputs of the deep reinforcement learning model to assist it in making decisions.

[0083] The working principle of the above technical solution is as follows: In the solution, an adaptive neural network model such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU) is selected. These models are particularly suitable for processing data with time series characteristics, such as compressor operation data, including key parameters such as intake pressure, intake temperature, exhaust pressure, exhaust temperature, rotational speed, and flow rate. The output layer is used to predict the operation state and performance parameters of the compressor in the future for a period of time, such as exhaust pressure, efficiency, etc. The adaptive neural network model is offline trained using the backpropagation algorithm and the Adam optimizer. The backpropagation algorithm is used to calculate the error between the model prediction result and the actual value, and reduce the error by adjusting the model parameters. The Adam optimizer is an efficient optimization algorithm that can dynamically adjust the learning rate of each parameter according to the first-order moment estimate and the second-order moment estimate of the gradient. The training objective is to enable the model to accurately predict the operation state and performance parameters of the compressor in the future for a period of time (such as one week). Dynamically adjust the structure and parameters of the adaptive neural network according to the current working conditions and the compressor historical data. This includes measures such as adding or reducing hidden layers, adjusting the number of neurons, etc., to adapt to the data characteristics and prediction requirements under different working conditions. The purpose of dynamic adjustment is to improve the adaptability and prediction accuracy of the model. During the model training process, regularization techniques are used to prevent model overfitting. Regularization restricts the complexity of the model parameters by adding a penalty term to the loss function, thereby avoiding the problem that the model performs too well on the training data but has poor generalization ability on the test data. Based on the preset model update strategy, when new data arrives, update the model parameters according to the importance of the data. This can ensure that the model can continuously learn and adapt to new data characteristics. The model update strategy may include methods such as incremental learning and online learning to adapt to different application scenarios and data update frequencies. Use the prediction result of the adaptive neural network as one of the inputs of the deep reinforcement learning model. The deep reinforcement learning model can make more accurate decisions based on the prediction result and other relevant information to optimize the operation strategy and energy efficiency of the compressor. This combination can give full play to the advantages of the adaptive neural network in prediction and the ability of the deep reinforcement learning in decision-making, and jointly improve the operation efficiency and reliability of the compressor.

[0084] The effects of the above technical solution are as follows: By selecting adaptive neural network models such as long short-term memory network (LSTM) or gated recurrent unit (GRU), this solution can fully capture the temporal characteristics of compressor operation data; these models have excellent performance in processing time series data and can accurately predict the operation status and performance parameters of the compressor in the future for a period of time; through offline training using the backpropagation algorithm and Adam optimizer, the model can learn the internal laws and patterns of compressor operation data; during online prediction, the model can, based on the current working conditions and historical data, predict the operation status and performance parameters of the compressor in real time, improving the accuracy and real-time performance of the prediction; according to the current working conditions and compressor historical data, this solution can dynamically adjust the structure and parameters of the adaptive neural network; for example, increasing or decreasing the hidden layer, adjusting the number of neurons, etc., to adapt to different prediction requirements and changes in working conditions; through regularization techniques, this solution can effectively prevent the model from overfitting and improve the generalization ability of the model; this enables the model to maintain stable prediction performance when facing new data; this solution adopts a preset model update strategy to update the model parameters according to the importance of new data; this ensures that the model can timely learn the new data features and improve the accuracy and reliability of the prediction; taking the prediction results of the adaptive neural network as one of the inputs of the deep reinforcement learning model can assist it in making more accurate decisions; the deep reinforcement learning model can, based on the prediction results and other relevant information, optimize the operation parameters and strategies of the compressor, reducing energy consumption and maintenance costs; by accurately predicting the operation status and performance parameters of the compressor, this solution can timely detect potential problems and carry out optimization and adjustment; this helps to maintain the stable operation of the compressor and improve the operation efficiency; by predicting the failure risk and performance degradation trend of the compressor, this solution can formulate a maintenance plan in advance and take corresponding measures; this helps to reduce the maintenance cost and downtime of the compressor; the application of this solution can improve the production efficiency and product quality of the enterprise, reduce the operation cost; it helps to enhance the competitiveness of the enterprise in the market and achieve sustainable development.

[0085] In an embodiment of the present invention, the S33 includes:

[0086] Real-time collect the operation data of the compressor through the sensor network, including key parameters such as pressure, temperature, current, vibration, etc., and use principal component analysis (PCA) to perform feature selection and dimensionality reduction on the collected operation data, and extract the feature vector that can best represent the working conditions of the compressor;

[0087] Clean, denoise, and standardize the historical operation data of the compressor, and use clustering algorithms to perform pattern recognition on the historical data to identify different operation patterns and failure types;

[0088] Dynamically adjust the number of hidden layers and the number of neurons in each layer of the adaptive neural network according to the results of working condition analysis and historical data mining; for example, when it is recognized that the compressor is in a high-load or abnormal working condition, the number of hidden layers and neurons can be increased to improve the prediction accuracy of the model;

[0089] Based on network pruning and layer merging techniques, optimize the network topology according to model performance and computing resource limitations, use the Xavier or He initialization method, combine with the Adam or RMSprop optimizer to initialize the weights and biases of the network, and make dynamic adjustments during the training process;

[0090] Introduce L1 and L2 regularization terms and monitor the performance of the model on the validation set; when the performance of the validation set starts to decline, use the early stopping method to stop training.

[0091] Use the cross-validation method to evaluate the performance of the adjusted model, iterate and optimize the structure and parameters of the model according to the evaluation results, and apply the optimized model to the actual compressor control system for verification.

[0092] The working principle of the above technical solution is as follows: With the help of a sensor network, multiple operating data of the compressor are captured in real time, covering key indicators such as pressure, temperature, current, and vibration; the principal component analysis (PCA) technique is used to screen out the most representative feature vectors from the massive data, effectively reducing the data dimension and improving the processing efficiency; the historical operating data is deeply cleaned to remove noise and standardized; the clustering algorithm is applied to identify different operating modes and potential fault types, providing a strong basis for subsequent model adjustment; according to the working condition analysis and the results of historical data mining, the number of hidden layers and the number of neurons in each layer of the adaptive neural network are flexibly adjusted; for example, when the compressor is under high load or abnormal working conditions, the number of hidden layers and neurons is appropriately increased to enhance the prediction accuracy and generalization ability of the model; combined with network pruning and layer merging techniques, the topology of the network is refined according to the model performance and computational resource limitations; the Xavier or He initialization method is adopted to set reasonable initial values for the weights and biases of the network to ensure the stability and efficiency of the training process; efficient optimizers such as Adam or RMSprop are selected to dynamically adjust the weights and biases of the network, accelerating the training process and improving the model performance; L1 and L2 regularization terms are introduced to effectively prevent model overfitting and improve the generalization ability of the model; the cross-validation method is used to comprehensively evaluate the performance of the adjusted model, including key indicators such as accuracy, recall rate, and F1 score; by comparing the evaluation results of different models, the optimal model structure and parameter combination are selected; according to the evaluation results, the structure and parameters of the model are iteratively optimized to continuously improve the prediction accuracy and stability of the model; the optimized model is applied to the actual compressor control system for verification to ensure its reliability and effectiveness in the actual operating environment.

[0093] The effects of the above technical solutions are as follows: By collecting key parameters in real time through the sensor network and using principal component analysis (PCA) for feature selection and dimensionality reduction, the redundancy of data is effectively reduced, and the efficiency and accuracy of data processing are improved. This helps to quickly identify the feature vectors that best represent the compressor operating conditions and provides a high-quality data basis for subsequent model training; Deep cleaning, denoising, and standardization processing are performed on historical operation data, and clustering algorithms are used to identify different operating modes and fault types. This not only helps to understand the operating characteristics of the compressor but also provides an important basis for the dynamic adjustment of the model, improving the adaptability and prediction accuracy of the model under different operating conditions; According to the results of operating condition analysis and historical data mining, the number of hidden layers and the number of neurons in each layer of the adaptive neural network are dynamically adjusted. This flexibility enables the model to better adapt to the operating characteristics of the compressor under different operating conditions, improving the prediction accuracy and robustness; Based on network pruning and layer merging techniques, the network structure is optimized according to the model performance and computational resource limitations. This helps to reduce the complexity and computational cost of the model while ensuring the prediction performance, improving the practicality and scalability of the model; The weights and biases of the network are initialized using the Xavier or He initialization method combined with the Adam or RMSprop optimizer and dynamically adjusted during the training process. This helps to accelerate the convergence speed of the model and improve the training efficiency; L1 and L2 regularization terms are introduced to prevent the model from overfitting, and the model performance is monitored on the validation set. When the performance of the validation set begins to decline, the early stopping method is used to stop the training. This helps to ensure the generalization ability of the model and avoid performance degradation caused by overfitting; The performance of the adjusted model is evaluated using the cross-validation method, including metrics such as accuracy, recall, and F1 score. This helps to comprehensively and objectively evaluate the performance of the model and provides a reliable basis for iterative optimization; The structure and parameters of the model are iteratively optimized according to the evaluation results, and the optimized model is applied to the actual compressor control system for verification. This systematic optimization process helps to continuously improve the prediction accuracy and practicality of the model and provides strong support for the intelligent control and maintenance of the compressor.

[0094] In one embodiment of the present invention, the S4 includes:

[0095] S41. Based on the fusion algorithm, fuse the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module to generate a comprehensive control strategy;

[0096] S42. Verify and optimize the comprehensive control strategy in the simulation environment, and evaluate the performance of the strategy through comparative experiments; Adjust the parameters of the fusion algorithm and the control strategy according to the verification results;

[0097] S43. Based on the preset strategy update mechanism, when new data or operating conditions change, update the comprehensive control strategy in a timely manner.

[0098] The working principle of the above technical solution is as follows: Based on the fusion algorithm, the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module are effectively fused; a comprehensive control strategy is generated that not only considers the optimal control under the current working conditions but also anticipates future state changes and makes corresponding adjustments; the fusion algorithm needs to fully consider the output characteristics of the two modules to ensure the accuracy and effectiveness of the control strategy. The deep reinforcement learning decision-making module is responsible for making immediate decisions according to the current working conditions, while the adaptive neural network prediction module is responsible for predicting future states and adjusting the control strategy in advance; the generated comprehensive control strategy is verified and optimized in the simulation environment; the performance of the strategy is evaluated through comparative experiments, including comparing the control effects, stability, etc. under different parameters; according to the verification results, the parameters of the fusion algorithm and the control strategy are finely adjusted to further improve the performance of the strategy; this process may require multiple iterations until the optimal parameter combination and control strategy are found; based on the preset strategy update mechanism, when new data or working condition changes occur, the comprehensive control strategy is updated in a timely manner; the new data may come from the real-time collection of the sensor network or from the further mining and analysis of historical data; the changes in working conditions may include changes in compressor load, ambient temperature, etc., and these changes require the strategy to be able to respond and make adjustments in a timely manner; the update process should be as automated as possible to reduce manual intervention and improve the update efficiency.

[0099] The effects of the above technical solution are as follows: By fusing the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module, it is possible to comprehensively consider the optimal control under the current working conditions and the predictive adjustment of future state changes; this fusion not only takes into account the need for immediate decision-making but also the predictability of long-term planning, thus ensuring the accuracy and effectiveness of the control strategy; the generated comprehensive control strategy combines the advantages of the two advanced algorithms and can more comprehensively handle complex and variable working conditions; through verification and optimization in the simulation environment, the performance of the strategy can be further improved to ensure its stability and reliability in practical applications; verifying the comprehensive control strategy in the simulation environment can avoid the risks and losses that may be brought in practical applications; by evaluating the performance of the strategy through comparative experiments, it is possible to intuitively understand the performance of the strategy under different working conditions and provide a basis for subsequent parameter adjustment; according to the results of simulation verification, the parameters of the fusion algorithm and the control strategy can be finely adjusted to optimize its performance; the preset strategy update mechanism can ensure that when new data or working conditions change, the system can timely update the comprehensive control strategy, thus maintaining its adaptability and flexibility; deep reinforcement learning is good at dealing with complex decision-making problems, while the adaptive neural network is good at predicting future states; the combination of the two enables the system to have both decision-making ability and predictability, thus improving the intelligent level of the system; the preset strategy update mechanism can achieve automatic update, reducing the need for manual intervention; through continuous learning and optimization, the system can gradually improve its control performance and adaptability, achieving an improvement in the intelligent level; an accurate and effective control strategy can reduce the failure rate of equipment and extend the service life of the equipment; this helps to reduce the downtime and maintenance costs caused by equipment failures; the optimized control strategy can improve the operating efficiency of the equipment, thus enhancing the production efficiency; this helps enterprises reduce costs, increase revenues, and enhance market competitiveness.

[0100] In one embodiment of the present invention, S41 includes:

[0101] Extract from the deep reinforcement learning module the control strategy optimized based on immediate rewards and long-term returns under the current working conditions; including key information such as action selection probability and state value estimation.

[0102] Conduct an adaptability evaluation on the extracted strategy, and analyze the stability and robustness of the extracted measurement under different working conditions; for example, by simulating conditions such as different loads, ambient temperatures, and power supply voltage fluctuations.

[0103] Use the adaptive neural network to predict the future operating state of the compressor, including the change trends of key parameters such as pressure, temperature, and current, and verify the accuracy of the prediction model by comparing the prediction results with the actual operating data.

[0104] Using the fusion algorithm architecture, the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module are taken as inputs, and a comprehensive control strategy is generated through non-linear transformation, etc.

[0105] The weights in the fusion algorithm are optimized using machine learning algorithms (such as genetic algorithms, particle swarm optimization, etc.), and the fusion strategy is preliminarily verified in a simulation environment.

[0106] According to the operating characteristics of the compressor under different working conditions, an adaptability analysis of the fusion strategy is carried out; including analyzing the performance of the strategy under conditions such as different loads, ambient temperatures, and power supply voltage fluctuations. According to the results of the working condition adaptability analysis, the parameters of the fusion algorithm and the control strategy are adjusted and optimized; including adjusting the weights, optimizing the neural network structure, etc.

[0107] The working principle of the above technical solution is as follows: From the deep reinforcement learning module, the system can extract control strategies optimized based on immediate rewards and long-term returns under the current working conditions; these strategies contain key information such as action selection probabilities and state value estimations, which together constitute the optimal behavior guidelines for the agent (i.e., the compressor control system) in a specific environment; the extracted strategies need to be adaptively evaluated to verify their stability and robustness under different working conditions; by simulating conditions such as different loads, ambient temperatures, and power supply voltage fluctuations, the system can evaluate the performance of the strategies under these conditions, thus ensuring their reliability in practical applications; using an adaptive neural network, the system can predict the future operating state of the compressor; the prediction content includes the change trends of key parameters such as pressure, temperature, and current, which are crucial for evaluating the compressor performance and formulating control strategies; by comparing the prediction results with the actual operating data, the system can verify the accuracy of the prediction model; this helps to ensure the reliability of the prediction model in practical applications and provides strong support for subsequent control strategy formulation; the fusion algorithm architecture takes the outputs of the deep reinforcement learning decision module and the adaptive neural network prediction module as inputs; through methods such as non-linear transformation, the fusion algorithm can generate a comprehensive control strategy that takes into account both the optimal control under the current working conditions and anticipates changes in future states; using machine learning algorithms (such as genetic algorithms, particle swarm optimization, etc.), the system can optimize the weights in the fusion algorithm; by optimizing the weights, the system can ensure that the comprehensive control strategy can fully consider various factors and is preliminarily verified in a simulation environment; the purpose of the preliminary verification is to evaluate the performance and stability of the fusion strategy in practical applications and provide a basis for subsequent optimization and adjustment; according to the operating characteristics of the compressor under different working conditions, the system conducts an adaptive analysis of the fusion strategy. The analysis content includes the performance of the strategy under conditions such as different loads, ambient temperatures, and power supply voltage fluctuations; based on the results of the adaptive analysis, the system can adjust and optimize the parameters of the fusion algorithm and the control strategy; the adjustment content may include weight adjustment, neural network structure optimization, etc., to ensure that the fusion strategy can exhibit the best performance under different working conditions.

[0108] The effects of the above technical solutions are as follows: The extracted control strategy not only considers immediate rewards but also takes into account long-term returns, thus ensuring the optimality and forward-looking nature of the strategy; the key information such as action selection probability and state value estimation it contains provides a clear direction and basis for the execution of the strategy; the stability and robustness analysis of the extracted strategy under different working conditions ensures the reliability and applicability of the strategy in practical applications; by simulating various conditions (such as different loads, ambient temperatures, and power supply voltage fluctuations), the adaptability of the strategy can be comprehensively evaluated, providing a basis for subsequent optimization; using an adaptive neural network to predict the future operating state of the compressor can accurately capture the change trends of key parameters (such as pressure, temperature, current); by comparing the prediction results with the actual operating data, the accuracy of the prediction model can be verified, providing strong support for the formulation of subsequent control strategies; fusing the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module to generate a comprehensive control strategy through non-linear transformation, etc.; this fusion method can make full use of the advantages of the two modules to achieve the comprehensive optimization of the control strategy; using machine learning algorithms such as genetic algorithms and particle swarm optimization to optimize the weights in the fusion algorithm further improves the performance of the comprehensive control strategy; by optimizing the weights, it can be ensured that the strategy can not only consider the optimal control under the current working conditions but also foresee the changes in future states and make adjustments; preliminarily verifying the fusion strategy in a simulation environment can evaluate its performance and stability in practical applications; this verification method can reduce the risks in practical applications and improve the reliability and safety of the strategy; according to the operating characteristics of the compressor under different working conditions, conduct an adaptability analysis of the fusion strategy; adjust and optimize the parameters of the fusion algorithm and control strategy according to the analysis results, such as adjusting the weights and optimizing the neural network structure, etc., to ensure that the strategy can exhibit the best performance under different working conditions.

[0109] In one embodiment of the present invention, step S5 includes:

[0110] S51. According to the comprehensive control strategy, intelligently adjust the operating parameters of the compressor, such as rotational speed, intake air volume, loading / unloading strategy, etc., and record the key data during the adjustment process, such as the operating state and energy efficiency before and after the adjustment, for subsequent analysis and optimization;

[0111] S52. Based on a closed-loop feedback mechanism, continuously monitor the operating state and performance parameters of the compressor, such as pressure, temperature, current, etc. When abnormalities or deviations from expectations are detected, adjust the control strategy in a timely manner;

[0112] S53. Through an exception handling process, when a serious abnormality is detected, immediately initiate emergency protection measures to prevent damage to the compressor, and record the key events and abnormal states during the operation of the compressor, including adjustment records, fault records, etc.;

[0113] S54. Through the fault warning system, when a potential fault is predicted, the operator is notified in advance to take corresponding measures to avoid the occurrence or expansion of the fault.

[0114] The working principle of the above technical solution is as follows: According to the comprehensive control strategy generated in the previous steps (such as S4), the system intelligently adjusts the operating parameters of the compressor, such as speed, intake air volume, loading / unloading strategy, etc.; these adjustments are aimed at optimizing the operating efficiency and energy efficiency of the compressor while meeting production requirements; during the adjustment process, the system records key data, including the operating status (such as pressure, temperature, current, etc.) and energy efficiency before and after the adjustment; this data is used for subsequent analysis and optimization to help the system better understand the operating characteristics and requirements of the compressor; based on a closed-loop feedback mechanism, the system continuously monitors the operating status and performance parameters of the compressor; the monitored parameters include but are not limited to pressure, temperature, current, etc., and these parameters can reflect the health status and operating efficiency of the compressor; when abnormal or deviated parameters are detected, the system adjusts the control strategy in a timely manner according to preset rules or algorithms; this adjustment is aimed at quickly responding to changes in the compressor operation and maintaining its stability and efficiency; through the abnormal handling process, when serious abnormalities are detected, such as too high temperature, abnormal pressure, etc., the system immediately activates emergency protection measures; these measures are aimed at preventing compressor damage and protecting equipment and production safety; the system records key events and abnormal states during the compressor operation, including adjustment records, fault records, etc.; these records provide valuable data support for subsequent fault analysis and handling; through the fault warning system, the system uses historical data and prediction models to predict potential faults; these predictions are based on the operating characteristics and historical fault patterns of the compressor and are aimed at detecting potential fault hazards in advance; when a potential fault is predicted, the system notifies the operator in advance; the operator can take corresponding preventive measures according to the notification, such as shutdown inspection, maintenance, etc., to avoid the occurrence or expansion of the fault.

[0115] The effects of the above technical solution are as follows: According to the comprehensive control strategy, the key operating parameters such as the rotation speed, intake air volume, and loading / unloading strategy of the compressor are intelligently adjusted; this adjustment can ensure that the compressor maintains the best operating state under different working conditions, thereby improving the operating efficiency and energy efficiency; during the adjustment process, key data such as the operating state and energy efficiency before and after the adjustment are recorded; these data provide a basis for subsequent analysis and optimization, helping the system continuously learn and improve the control strategy, and further enhancing the operating efficiency of the compressor; based on the closed-loop feedback mechanism, the operating state and performance parameters (such as pressure, temperature, current, etc.) of the compressor are continuously monitored; when abnormalities or deviations from the expected values are detected, the system can promptly adjust the control strategy to maintain the stable operation of the compressor; when serious abnormalities are detected, emergency protection measures are immediately initiated to prevent damage to the compressor; this instant response mechanism can greatly reduce the risk of the compressor stopping due to failures, ensuring the continuity and stability of the production line; the key events and abnormal states during the operation of the compressor are recorded, including adjustment records, fault records, etc.; these records provide valuable data support for subsequent fault analysis and handling, helping the system continuously accumulate experience and improve the fault response ability; through the fault warning system, potential faults are predicted using historical data and prediction models; this prediction ability enables the system to notify the operator in advance before the fault occurs, thus avoiding the occurrence or expansion of the fault; the operator can take corresponding preventive measures (such as shutting down for inspection, maintenance, etc.) according to the notification of the fault warning system; this preventive maintenance strategy can significantly reduce the failure rate and maintenance cost of the compressor, and improve the reliability and service life of the equipment.

[0116] An embodiment of the present invention, as Figure 2 shown, a compressor control system, the system includes:

[0117] Data acquisition module: Through a multi-parameter sensor network, multi-dimensional data during the operation of the compressor is collected in real time, and the multi-dimensional data includes pressure, temperature, current, vibration, and flow; the collected multi-dimensional data is preprocessed, and the preprocessing includes data cleaning, anomaly detection, and feature engineering, and a high-quality data set is constructed;

[0118] Optimization learning module: Through a reinforcement learning model based on a deep neural network, based on historical data and the current environmental state (such as working conditions, compressor state, etc.), the compressor control strategy is learned and optimized; through simulation and online learning, continuous iterative updates are performed;

[0119] Real-time prediction module: Based on an adaptive neural network model, the operating state and performance parameters of the compressor in the future for a period of time are predicted in real time, and the network structure and parameters are dynamically adjusted according to the current working conditions and the compressor historical data;

[0120] Policy Generation Module: Fuses the outputs of the deep reinforcement learning decision module and the adaptive neural network prediction module to generate a comprehensive control policy;

[0121] Closed-loop Control Module: Intelligently adjusts the operating parameters of the compressor (such as speed, intake air volume, loading / unloading strategy, etc.) according to the comprehensive control policy; meanwhile, through the closed-loop feedback mechanism, continuously monitors the operating state of the compressor and timely adjusts the control policy.

[0122] The working principle of the above technical solution is as follows: Use a multi-parameter sensor network to collect key data during the operation of the compressor in real time, including pressure, temperature, current, vibration, and flow rate, etc.; preprocess the collected multi-dimensional data, including data cleaning (removing noise, filling missing values, etc.), anomaly detection (identifying and processing abnormal data points), and feature engineering (extracting useful features, dimensionality reduction, etc.) to construct a high-quality data set; use a reinforcement learning model based on a deep neural network, combined with historical data and the current environmental state (such as operating conditions, compressor state, etc.), to learn and optimize the control policy of the compressor; through simulation and online learning, the model is continuously iteratively updated to maximize the long-term benefits (such as energy efficiency, stability, etc.) as the goal, and generate optimal control instructions; based on the adaptive neural network model, real-time predict the operating state and performance parameters of the compressor in the next period of time; dynamically adjust the network structure and parameters according to the current operating conditions and compressor historical data to improve the accuracy of the prediction. The prediction result is used as one of the inputs of the deep reinforcement learning model to assist it in making more accurate decisions; fuse the outputs of the deep reinforcement learning decision module and the adaptive neural network prediction module to generate a comprehensive control policy; this policy not only considers the optimal control under the current operating conditions, but also incorporates anticipatory adjustments to future state changes, thus realizing the comprehensive optimization of the operating parameters of the compressor; according to the comprehensive control policy, intelligently adjust the operating parameters of the compressor (such as speed, intake air volume, loading / unloading strategy, etc.); through the closed-loop feedback mechanism, continuously monitor the operating state of the compressor and timely adjust the control policy.

[0123] The effects of the above technical solution are as follows: By collecting multi-dimensional data in real time and performing preprocessing, a high-quality data set is constructed, providing a solid foundation for subsequent model learning and optimization. It helps to more accurately reflect the actual operating state of the compressor, thereby formulating a more efficient control strategy; The reinforcement learning model can continuously iterate and update the control strategy based on historical data and the current environmental state, aiming to maximize the long-term benefit. It can not only improve the operating efficiency of the compressor, but also optimize energy use and reduce energy consumption costs while maintaining high-efficiency operation; The adaptive neural network model can predict the operating state and performance parameters of the compressor in the future for a period of time in real time and make dynamic adjustments according to the current working conditions and historical data. This predictive adjustment helps to detect and address potential operating problems in advance, thereby enhancing the operating stability and reliability of the compressor; The comprehensive control strategy integrates the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module, taking into account both the optimal control under the current working conditions and the predictive adjustment of future state changes. This comprehensive control strategy helps to ensure that the compressor can maintain a stable operating state under various working conditions; By introducing advanced artificial intelligence technologies such as deep neural networks, reinforcement learning, and adaptive neural networks, the intelligent optimization and real-time adjustment of the compressor control strategy are realized. This greatly improves the intelligent level and automation degree of the compressor, reducing manual intervention and dependence; The closed-loop feedback mechanism can continuously monitor the operating state of the compressor and adjust the control strategy in a timely manner according to the actual situation. This automatic adjustment mechanism helps to ensure that the compressor always operates in the best state, improving the reliability and flexibility of the overall system; By real-time monitoring and predicting the operating state and performance parameters of the compressor, potential operating problems and faults can be detected in a timely manner. This helps to take maintenance measures in advance to avoid the occurrence or expansion of faults, thereby reducing maintenance costs and downtime; The intelligent and automatic control strategy helps to optimize the operating parameters and working modes of the compressor, reducing unnecessary wear and tear. This helps to extend the service life of the compressor and improve the economy and sustainability of the overall system.

[0124] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A compressor control method, characterized in that, The method includes: S1. Collect multi-dimensional data during the operation of the compressor in real time through a multi-parameter sensor network; preprocess the collected multi-dimensional data and construct a data set; S2. Through a reinforcement learning model based on a deep neural network, learn and optimize the compressor control strategy based on historical data and the current environmental state; continuously iterate and update through simulation and online learning; S3. Based on an adaptive neural network model, real-time predict the operating state and performance parameters of the compressor in the future for a period of time, and dynamically adjust the network structure and parameters according to the current working conditions and the compressor historical data; S4. Integrate the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module to generate a comprehensive control strategy; S5. Intelligently adjust the operating parameters of the compressor according to the comprehensive control strategy. At the same time, through a closed-loop feedback mechanism, continuously monitor the operating state of the compressor and timely adjust the control strategy; The S4 includes: S41. Based on a fusion algorithm, integrate the outputs of the deep reinforcement learning decision-making module and the adaptive neural network prediction module to generate a comprehensive control strategy; S42. Verify and optimize the comprehensive control strategy in a simulation environment, evaluate the performance of the strategy through comparative experiments; adjust the parameters of the fusion algorithm and the control strategy according to the verification results; S43. Based on a preset strategy update mechanism, when new data or working conditions change, update the comprehensive control strategy in a timely manner.

2. The compressor control method according to claim 1, characterized in that The S1 includes: S11. Collect multi-dimensional data during the operation of the compressor in real time through a multi-parameter sensor network, and use a sliding window algorithm to smooth the collected multi-dimensional data to remove noise; S12. Detect and fill in missing values through statistical methods, apply machine learning algorithms to detect outliers, and mark or eliminate them; S13. Implement feature engineering, extract feature variables that affect the operating state of the compressor, organize the preprocessed data in a time series, and construct a data set containing historical data and current data; S14. Store the data set through a distributed storage scheme, and back up the data set based on a set data backup mechanism.

3. The compressor control method according to claim 1, wherein, The S2 includes: S21. Construct a reinforcement learning model architecture based on a deep neural network, and the model architecture includes a policy network and a value network; the policy network is used to generate control instructions, and the value network is used to evaluate the advantages and disadvantages of the control instructions; S22. Divide the constructed data set into a training set and a validation set for model training and validation, and use the stochastic gradient descent algorithm to perform offline training on the model to learn the optimal control strategy of the compressor under different working conditions; S23. Optimize the model performance by adjusting hyperparameters, construct a compressor operation simulation environment, and simulate the behavior of the compressor under different working conditions; S24. Implement online learning in the simulation environment, and at the same time collect new data for model update; continuously update the data in the simulation environment through the rolling time window technology; S25. Generate control instructions based on the output of the deep reinforcement learning model.

4. The compressor control method according to claim 3, wherein The S24 includes: S241. Construct a simulation system that highly simulates the actual operating environment of the compressor, receive the compressor operation data in the simulation environment in real time through an online learning framework, and interact the operation data with a deep reinforcement learning model; S242. Simulate various complex and changeable working conditions in the simulation environment; during the simulation process, collect the compressor operation data in real time; S243. Through the rolling time window technology, continuously update the data in the simulation environment, and adjust the size of the time window according to the change speed of the compressor operation characteristics and the update requirements of the model; S244. Conduct quality assessment and preprocessing on the newly collected data, conduct online verification and evaluation of the model through the new data, and evaluate the accuracy and performance of the model by comparing the model prediction results with the actual simulation results; S245. According to the results of the online verification, adjust the parameters of the deep reinforcement learning model in real time, and based on the adaptive learning rate adjustment mechanism, dynamically adjust the learning rate according to the learning progress and error change of the model; S246. Update the model parameters and policies obtained from online learning to the compressor control system in actual operation, and through the policy optimization algorithm, combine historical data and real-time data to iteratively optimize the control policy.

5. The compressor control method according to claim 4, characterized in that, The S243 includes: Evaluate the timeliness of the data by comparing the timestamp, change rate of the data, and its relevance to other parameters, and judge whether the data still represents the current operating state of the compressor; Design a dynamically adjustable time window size based on the change speed of the compressor operation characteristics and the update requirements of the model; Set the sliding strategy of the time window based on a fixed time interval or the change of data volume, and based on the data automatic update mechanism, when new data enters the time window, automatically replace the oldest data; Through the model feedback loop, re-enter the updated data into the deep reinforcement learning model for a new round of training and verification; Identify and filter out abnormal data in the simulation environment through an anomaly detection algorithm; formulate a data filtering strategy according to the results of the anomaly detection; Quantify the adaptability of the model to new data according to the adaptability evaluation index, and dynamically adjust the optimization strategy of the model according to the results of the adaptability evaluation.

6. The compressor control method according to claim 1, wherein The S3 includes: S31. Select an adaptive neural network model, design the input layer and output layer of the model. The input layer is used to receive the preprocessed compressor operation data, and the output layer is used to predict the operation state and performance parameters of the compressor in the future for a period of time; S32. Through the backpropagation algorithm and the Adam optimizer, conduct offline training on the model so that the model can predict the operation state and performance parameters of the compressor in the future for a period of time; S33. Dynamically adjust the structure and parameters of the adaptive neural network according to the current working conditions and the compressor historical data, and prevent the model from overfitting through regularization; S34. Based on the preset model update strategy, when new data arrives, update the model parameters according to the importance of the data; use the prediction results of the adaptive neural network as one of the inputs of the deep reinforcement learning model to assist it in making decisions.

7. The compressor control method according to claim 1, wherein The S41 includes: Extract the control strategy optimized based on immediate reward and long-term return under the current working conditions from the deep reinforcement learning module; Conduct an adaptability assessment on the extracted strategy, and analyze the stability and robustness of the extracted measurements under different working conditions; Use an adaptive neural network to predict the future operating state of the compressor, and verify the accuracy of the prediction model by comparing the prediction results with the actual operating data; Through the fusion algorithm architecture, take the outputs of the deep reinforcement learning decision module and the adaptive neural network prediction module as inputs, and generate a comprehensive control strategy through non-linear transformation, etc.; Use machine learning algorithms to optimize the weights in the fusion algorithm, and conduct a preliminary verification of the fusion strategy in the simulation environment; Conduct an adaptability analysis of the fusion strategy according to the operating characteristics of the compressor under different working conditions; according to the results of the working condition adaptability analysis, adjust and optimize the parameters of the fusion algorithm and the control strategy.

8. The compressor control method according to claim 1, characterized in that The S5 described above includes: S51. According to the comprehensive control strategy, intelligently adjust the operating parameters of the compressor, and record the key data during the adjustment process; S52. Based on the closed-loop feedback mechanism, continuously monitor the operating state and performance parameters of the compressor. When abnormalities or deviations from expectations are detected, promptly adjust the control strategy; S53. Through the exception handling process, when an abnormality is detected, immediately initiate emergency protection measures, and record the key events and abnormal states during the operation of the compressor; S54. Through the fault warning system, when a potential fault is predicted, notify the operator in advance to take corresponding measures.

9. A system for implementing the compressor control method as described in claim 1, characterized in that, The system described above includes: Data acquisition module: Through a multi-parameter sensor network, real-time collect multi-dimensional data during the operation of the compressor, preprocess the collected multi-dimensional data, and construct a data set; Optimization learning module: Through a reinforcement learning model based on a deep neural network, learn and optimize the compressor control strategy based on historical data and the current environmental state; through simulation and online learning, continuously iterate and update; Real-time prediction module: Based on an adaptive neural network model, real-time predict the operating state and performance parameters of the compressor in the future for a period of time, and dynamically adjust the network structure and parameters according to the current working conditions and the compressor historical data; Strategy generation module: Integrate the outputs of the deep reinforcement learning decision module and the adaptive neural network prediction module to generate a comprehensive control strategy; Closed-loop control module: According to the comprehensive control strategy, intelligently adjust the operating parameters of the compressor; at the same time, through the closed-loop feedback mechanism, continuously monitor the operating state of the compressor and promptly adjust the control strategy.

Images