Recovery control system based on MVR system

Abstract

The invention discloses a recovery control system based on an MVR system, and relates to the field of recovery control of the MVR system, and the key points of the technical scheme are that a data collection module is used for collecting comprehensive state data of the MVR system, carrying out the fusion processing of the collected data, and obtaining a comprehensive data matrix; the segmentation module obtains an initial countermeasure set from the comprehensive data matrix through an anomaly recognition model, and performs clustering analysis on the initial countermeasure set to obtain k groups of sub-countermeasure sets; the scheme output module performs optimization on the k groups of sub-countermeasure sets by using an optimization algorithm to obtain a to-be-selected set, and performs optimization on the to-be-selected set again by using the optimization algorithm to obtain an optimal regulation and control scheme for recovery control of the MVP system; all the modules are connected in a wired and / or wireless mode, and stability and energy optimization in the recovery control process based on the MVR system are achieved.

Description

Technical Field

[0001] The present invention relates to the field of recovery control of MVR system, and more particularly, to a recovery control system based on MVR system. Background Art

[0002] MVR system, full name Mechanical Vapor Recompression system, is an advanced energy-saving evaporation technology, widely used in industrial wastewater treatment, food processing, chemical product concentration, seawater desalination and other fields. The core of MVR technology is to use the secondary steam generated in the system, increase its energy and temperature through mechanical compression, and then recycle it as a heat source, thereby reducing the demand for fresh steam and achieving energy saving.

[0003] The MVR system involves multiple technical links, including evaporation, compression, heat exchange, etc. These links require precise control to achieve the best results and need to be integrated with existing factory equipment and process flows. This may involve complex engineering and coordination work. The operation of each link requires professional and accurate instructions. Improper instructions may lead to reduced system efficiency or equipment damage, reduce the stability of the MVR system during operation, and cause compressor surge. In addition, although the MVR system is designed for energy saving, in actual operation it may lead to increased energy consumption due to various reasons. Therefore, the MVR system needs to adapt to different environmental conditions, such as temperature, humidity, etc., to maintain optimal performance. Therefore, the recovery control of the MVR system needs to be designed. Summary of the invention

[0004] In view of the problems existing in the prior art, the object of the present invention is to provide a recycling control system based on an MVR system to realize recycling control of the MVR system.

[0005] To achieve the above object, the present invention provides the following technical solution: a recycling control system based on an MVR system, the recycling control system based on the MVR system comprising:

[0006] Data acquisition module: used to collect the comprehensive status data of the MVR system, and perform fusion processing on the collected data to obtain a comprehensive data matrix;

[0007] Segmentation module: obtain the initial countermeasure set from the comprehensive data matrix through the anomaly recognition model, perform cluster analysis on the initial countermeasure set, and obtain k groups of sub-countermeasure sets;

[0008] Solution output module: Use optimization algorithm to optimize k groups of sub-strategy sets to obtain the candidate set, and then use optimization algorithm to optimize the candidate set again to obtain the optimal control solution of MVP system recycling control;

[0009] The modules are connected to each other via wired and / or wireless means.

[0010] Preferably, the data categories of the comprehensive status data include evaporator data, compressor data, heat exchanger data, condenser data and separator data, the evaporator data includes the liquid level of the evaporator, the compressor data includes the steam pressure, temperature and flow of the compressor, the heat exchanger data includes the temperature of the heat exchanger, the condenser data includes the flow, pressure and temperature of the condenser, and the separator data includes the liquid level of the separator.

[0011] Preferably, the method of fusing the collected data includes:

[0012] Assume that the comprehensive state data is provided by q sensors, and at time point t, obtain the measurement values ​​of w data categories to obtain the measurement vector Among them, z_t i represents the measurement value of the i-th sensor, and i=1,...,q;

[0013] The compressor characteristic coefficient and the condenser characteristic coefficient are obtained through the compressor data and the condenser data, the measured values ​​corresponding to the steam pressure, temperature and flow rate of the compressor in the measurement vector, and the measured values ​​of the flow rate, pressure and temperature of the condenser are deleted, the compressor characteristic coefficient and the condenser characteristic coefficient are filled into the measurement vector, and the labels are distributed to obtain the fused feature vector;

[0014] Taking time point t as the base point, the feature vectors at the previous p_a time points are continuously collected, arranged in chronological order, and combined into a time series matrix, which is recorded as the comprehensive data matrix at time point t.

[0015] Preferably, the method of obtaining the compressor characteristic coefficient and the condenser characteristic coefficient through the compressor data and the condenser data comprises:

[0016] Extract the measured values ​​corresponding to the steam pressure, temperature and flow of the compressor in the measurement vector, as well as the measured values ​​of the flow, pressure and temperature of the condenser, perform curve fitting to obtain the characteristic coefficients of the compressor and the condenser, and set the curve fitting formula as Among them, W_Y o It is used to represent the characteristic coefficient of the compressor or the characteristic coefficient of the condenser at the time point o, and o=t or 1, ..., e, ...r, te represents the time points before the time point o; Obtained by least squares method, x_g y represents the correlation coefficient between the yth measurement value and the fault, represents the measurement error variance of the yth sensor at time point t-1, z_t yrepresents the measurement value of the yth sensor, and Y represents the number of measurements.

[0017] Preferably, the correlation coefficient x_g y It is used to evaluate the impact of data categories represented by each measurement value and equipment failure. The equipment state is set to Z_u, and Z_u=1 indicates failure, Z_u=0 indicates normal, and the fault association model is set to Among them, β0 is the intercept term, βy is the y-th measurement value z_t y The regression coefficient corresponding to the data category, and the regression coefficient βy is estimated by maximizing the likelihood function, x_g y 9Z_u=1|X_y) represents the regression coefficient corresponding to the data category of y measurement values ​​and the equipment failure as the correlation coefficient;

[0018] The correlation coefficient x_g at time point t y It is obtained by using the measured value at time point t-1 through the fault correlation model calculation.

[0019] Preferably, the step of obtaining an initial set of countermeasures from a comprehensive data matrix through an anomaly recognition model specifically includes:

[0020] Set a sample set, the sample set includes H_k group subsets, each subset includes a comprehensive data matrix, anomaly measurement values ​​and an initial countermeasure set corresponding to the anomaly measurement values, the comprehensive data matrix is ​​marked as the input of the anomaly recognition model, the anomaly recognition model matches the anomaly measurement value with the fixed countermeasure set, outputs the initial countermeasure set, the anomaly measurement value and the initial countermeasure set are marked as the output of the anomaly recognition model, the sample set is divided into a training set and a validation set, and they are input into the anomaly recognition model in sequence;

[0021] The fixed countermeasure set includes all control strategies when the abnormal measurement value is regulated to the normal state. Each control strategy includes the parameter settings used by the equipment when the MVP system is adjusted. The abnormal identification model uses the optimization algorithm to search for the optimal control strategy of group z_c in the fixed countermeasure set to form the initial countermeasure set.

[0022] Set up an anomaly recognition model based on RBF and LSTM. Optimize model parameters by minimizing the loss function, use the validation set for cross-validation, adjust model hyperparameters, use accuracy, recall, and F1-score as performance evaluation indicators, and stop when the expected performance evaluation indicators are reached. Then obtain a trained anomaly recognition model and export the trained anomaly recognition model into a deployable format.

[0023] Import the comprehensive data matrix at time point t into the trained anomaly recognition model to obtain the initial countermeasure set at time point t.

[0024] Preferably, the method of performing cluster analysis on the initial countermeasure set includes:

[0025] Step S1: Set the number of clusters k_zg and randomly select k_zg control strategies as the initial centroid;

[0026] Step S2: Calculate the distance between each control strategy and each centroid, and assign each control strategy to the cluster represented by the centroid closest to it;

[0027] Step S3: Calculate a new centroid for all control strategies in each cluster, where the centroid is the mean of all control strategies in the cluster;

[0028] Step S4: repeat steps S2 to S3 until the centroid does not change between two consecutive times;

[0029] Step S5: Output the control strategy in each cluster and obtain a set of k sub-strategies.

[0030] Preferably, the method for obtaining the number of clusters k_zg includes:

[0031] A cluster number selection data set K = {k_1, ..., k_n, ..., k_N} is preset, where k_n represents the number of clusters represented by the nth value of the cluster number selection data set, and n = 1, ..., N;

[0032] Randomly select a k_n as the pre-set number of clusters and calculate the average silhouette coefficient when the number of clusters is k_n Among them, z_c represents the number of control strategies in the initial countermeasure set;

[0033] Traverse and calculate the average silhouette coefficient of all clusters in the selected data set, select the cluster number with the largest silhouette coefficient as the optimal cluster number, and set the optimal cluster number as the clustering cluster number k_zg;

[0034] For any control strategy, calculate the average distance between it and all other control strategies in the same cluster Where |C_sj| represents the number of control strategies in cluster C_j, d(j,l) represents the distance between the jth control strategy and the lth control strategy; and calculates the average distance between it and the points in all other clusters to find the nearest cluster. Among them, cluster C_h is different from cluster C_j, d(j,z) is the distance between the j-th regulation strategy and the z-th regulation strategy in cluster C_h, and C_sh represents the number of regulation strategies in cluster C_h.

[0035] Preferably, the method of using an optimization algorithm to optimize the k groups of sub-strategy sets to obtain the set to be selected includes:

[0036] Each control strategy in any set of sub-strategies is regarded as a hunting dog individual. The hunting dogs cooperate with each other and constantly adjust their positions to get closer to the optimal prey.

[0037] Set the position update formula for each hound individual: in, represents the position of the vth hound at the oth iteration, wp v is the local optimal position of the vth hound, wp is the global optimal position, β1 and β2 are learning factors, and ε1 and ε2 are random factors between 0 and 1;

[0038] Setting fitness in, and are the stability and energy consumption index of the control strategy after the vth hound is updated, γ1 and γ2 are weight coefficients, if the fitness If it exceeds the current global optimal fitness, the global optimal position is updated, that is, When the change rate of two consecutive fitness values ​​is less than the set threshold in several consecutive iterations, the process stops.

[0039] Set the constraints to Where SF represents the number of stability determining parameters, θ wx represents the weight coefficient of the wxth parameter, Δh_l represents the absolute difference between the updated control strategy and the standard value, f_op and YU_D are pre-set thresholds, θ1 and θ2 represent weight coefficients, and Q RE represents the recovered heat, Q IN represents the input heat, Q FR Indicates the usage of fresh steam;

[0040] Output the current global optimal position wp, which is the corresponding optimal control strategy in the sub-strategy set.

[0041] Preferably, the step of again optimizing the selected set by an optimization algorithm to obtain the optimal control scheme for the recycling control of the MVP system includes:

[0042] Collect the optimal control strategies of all sub-strategy sets to form a candidate set. Use the optimization algorithm to optimize the control strategies in the candidate set again, and output the optimal control strategy in the candidate set, which is recorded as the optimal control strategy at time point t. The MVP system adjusts the parameters at the next time point based on the optimal control strategy at this time.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] In the present invention, compressor characteristic coefficients and condenser characteristic coefficients are used to represent compressor data and condenser data, which not only simplifies the data volume, but also reflects the comprehensive characteristics of the data, improves the data processing speed, enhances the data mining depth, and realizes the high-quality input of subsequent data. The recognition model often learns the changing trend of the measured values ​​at multiple time points in the comprehensive data matrix based on RBF and LSTM, analyzes and predicts the numerical values, and obtains the measured values ​​of abnormal or dangerous states, which is convenient for the subsequent analysis of the abnormal measured values, reduces the processing of the data volume, and improves the optimized use of the data. The optimization algorithm realized by the present invention realizes multi-layer optimization of the control strategy and outputs the optimal control strategy. Through the multi-layer control strategy segmentation, it is convenient for the centralized analysis and processing of data with the same characteristics. The optimization layer by layer not only reduces the data processing volume, but also improves the operation rate through parallel calculation. The adaptive control optimization mechanism continuously improves the overall efficiency of intelligent analysis, and multiple modules cooperate with each other to realize comprehensive data collection, processing and analysis, and provide real-time and accurate analysis of detection data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention proposes a structural schematic diagram of a recycling control system based on an MVR system;

[0046] Figure 2 It is a schematic diagram of the method applied to the recovery control system based on the MVR system in the present invention. DETAILED DESCRIPTION

[0047] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0048] Embodiment 1

[0049] Reference Figure 1 and Figure 2 Embodiment 1 further illustrates a recycling control system based on an MVR system proposed by the present invention.

[0050] MVR system, full name Mechanical Vapor Recompression system, is an advanced energy-saving evaporation technology, widely used in industrial wastewater treatment, food processing, chemical product concentration, seawater desalination and other fields. The core of MVR technology is to use the secondary steam generated in the system, increase its energy and temperature through mechanical compression, and then recycle it as a heat source, thereby reducing the demand for fresh steam and achieving energy saving.

[0051] The working principle of the MVR system includes:

[0052] Steam generation: The material is heated in the evaporator to generate steam;

[0053] Steam compression: The secondary steam generated is compressed by a compressor to increase its temperature and pressure;

[0054] Thermal energy utilization: The compressed steam returns to the evaporator as a heat source to continue heating the material;

[0055] Evaporation concentration: The material continues to evaporate under the action of high-temperature steam to achieve the purpose of concentration or crystallization;

[0056] Heat recovery: The condensed water and concentrated liquid generated by the system recover heat energy through the heat exchanger to preheat the feed;

[0057] Recycling: Steam is recycled throughout the process, reducing the need for fresh steam.

[0058] The MVR system recovers secondary steam with extremely high economic value, for example:

[0059] A. Energy saving and high efficiency: The MVR system reduces the demand for fresh steam or external heat source by recovering secondary steam, thus significantly reducing energy consumption.

[0060] B. Reduce operating costs: Reducing dependence on external energy directly reduces energy procurement costs, thereby reducing overall operating costs.

[0061] C. Improve thermal efficiency: After the recovered secondary steam is compressed by the compressor, its heat energy is reused, which improves the thermal efficiency of the entire system.

[0062] D. Reduce environmental impact: Reducing dependence on fossil fuels reduces greenhouse gas emissions and other pollutants, which has a positive effect on environmental protection.

[0063] E. Achieve clean production: MVR technology, as a clean energy technology, helps to achieve cleaner industrial production and complies with current environmental regulations and sustainable development goals.

[0064] F. Improve system economics: Efficient energy utilization makes MVR systems more economically attractive, especially in areas with fluctuating energy prices or high energy costs.

[0065] G. Reduce cooling water demand: Since the MVR system reduces the use of fresh steam, the demand for cooling water is correspondingly reduced, further reducing the energy consumption and cost of the system.

[0066] H. Optimize the process flow: Recycling and utilizing secondary steam can optimize the entire process flow and improve production efficiency and product quality.

[0067] I. Reduce system scale: Due to the high thermal efficiency of the MVR system, the required equipment scale may be smaller, reducing equipment investment and floor space.

[0068] J. Improve system reliability and stability: Closed-loop circulation reduces the impact of external factors on system stability and improves production reliability.

[0069] K. Strong adaptability: The MVR system can adapt to different material characteristics and processing requirements, and has good flexibility and adaptability.

[0070] L. Technological advancement: The use of MVR technology reflects the company's application of advanced technology and enhances the company's technological image and market competitiveness.

[0071] In short, the recycling of MVR system is based on economic, environmental and technical considerations, and it represents an efficient, environmentally friendly and sustainable industrial production method.

[0072] However, the difficulties or problems that the MVR system may encounter during the recycling process mainly include:

[0073] The MVR system involves multiple technical links, including evaporation, compression, heat exchange, etc. These links require precise control to achieve the best results and need to be integrated with existing factory equipment and process flows. This may involve complex engineering and coordination work. The operation of each link requires professional and accurate instructions. Improper instructions may lead to reduced system efficiency or equipment damage, reduce the stability of the MVR system during operation, and cause compressor surge. In addition, although the MVR system is designed for energy saving, in actual operation it may lead to increased energy consumption due to various reasons. Therefore, the MVR system needs to adapt to different environmental conditions, such as temperature, humidity, etc., to maintain optimal performance. Therefore, the recovery control of the MVR system needs to be designed.

[0074] The MVR system needs to be integrated with existing plant equipment and process flows, which may involve complex engineering and coordination work. Comprehensive consideration of system design, equipment selection, operator training, maintenance strategy and environmental adaptability ensures that the MVR system can effectively recover and reuse secondary steam to achieve energy saving and environmental protection goals.

[0075] The core components of the MVR system are centrifugal or Roots compressors, which have high volume flow rates in the compression ratio range of 1:1.2 to 1:2. The MVR evaporator can be of different types such as falling film evaporator, forced circulation evaporator or crystallizer to meet different process requirements. The condensed water and concentrated liquid are preheated through the waste heat recovery device to improve energy efficiency.

[0076] When designing the system, it is necessary to consider maximizing heat recovery and reducing energy loss. MVR systems are usually equipped with automated control systems to achieve real-time monitoring and adjustment of key parameters such as temperature, pressure, and flow to ensure stable operation of the system. Designing a complete MVR (mechanical vapor recompression) system recovery control solution requires comprehensive consideration of the system's stability, efficiency, energy consumption, and integration with other process flows. The following is a detailed design plan:

[0077] In the present invention, the control target is set as:

[0078] 1. Stabilize system operation: Ensure that the system can operate stably under different working conditions, maintain stable operation of the compressor and evaporator, and prevent compressor surge.

[0079] 2. Improve energy efficiency: minimize energy consumption and optimize the energy utilization efficiency of the system, including maximizing the recovery of heat from secondary steam and reducing the use of fresh steam.

[0080] 3. Adapt to process changes: realize automatic control and real-time adjustment of the system, reduce human intervention, flexibly respond to changes in raw material concentration, temperature, etc., and maintain optimal system performance.

[0081] 4. Safe operation: monitor and prevent safety hazards such as system overvoltage and overtemperature.

[0082] Data acquisition module: used to collect the comprehensive status data of the MVR system, and perform fusion processing on the collected data to obtain a comprehensive data matrix;

[0083] The data categories of the comprehensive status data include evaporator data, compressor data, heat exchanger data, condenser data, and separator data.

[0084] The evaporator data includes the liquid level of the evaporator. The liquid level in the evaporator is monitored by a liquid level sensor, and the liquid level is kept stable by adjusting the flow of the feed pump.

[0085] Compressor data includes the steam pressure, temperature and flow of the compressor. Pressure sensors, temperature sensors and flow sensors are used to monitor the import and export parameters of the compressor in real time.

[0086] The surge protection of the compressor can be achieved by installing an anti-surge controller to adjust the compressor's operating point in real time. For example, a bypass valve can be added to reduce the compressor load and prevent surge.

[0087] The heat exchanger data includes the temperature of the heat exchanger. According to the process requirements, the flow valve on the steam side of the heat exchanger is adjusted to control the steam flow in the heat exchanger, thereby regulating the temperature of the liquid to be evaporated, so that the outlet temperature of the liquid to be evaporated is maintained within the set range, ensuring the stability of the evaporation process.

[0088] The condenser data includes the flow rate, pressure and temperature of the condenser. The main control target is to condense the remaining steam into liquid and recover heat. By adjusting the flow rate of cooling water, the temperature and pressure controllers are used to adjust the cooling water flow rate and the pressure in the condenser respectively to keep the pressure and temperature in the condenser at the target value.

[0089] The separator data includes the liquid level of the separator. In the separator, the separation of steam and liquid is linked to the drain valve through the liquid level controller (PID) to maintain a stable liquid level. The liquid level changes are monitored by the liquid level sensor to adjust the opening of the drain valve.

[0090] The above-mentioned temperature, liquid level, pressure and flow are all detected by temperature sensors, liquid level sensors, pressure sensors and flow sensors.

[0091] The methods for fusing the collected data include:

[0092] Assume that the comprehensive state data is provided by q sensors. At time point t, obtain the measurement values ​​of w data categories and distribute labels to each sensor. The labels can be numerically encoded through one-hot. The labels are used to mark the measurement values ​​and retrieve the sensors corresponding to the measurement values, so as to correspond the measurement values ​​to the measured objects and obtain the measurement vector

[0093] The compressor characteristic coefficient and the condenser characteristic coefficient are obtained through the compressor data and the condenser data, the measured values ​​corresponding to the steam pressure, temperature and flow rate of the compressor in the measurement vector, and the measured values ​​of the flow rate, pressure and temperature of the condenser are deleted, the compressor characteristic coefficient and the condenser characteristic coefficient are filled into the measurement vector, and the labels are distributed to obtain the fused feature vector;

[0094] Taking time point t as the base point, the feature vectors at the previous p_a time points are continuously collected, arranged in chronological order, and combined into a time series matrix, which is recorded as the comprehensive data matrix at time point t.

[0095] The method of obtaining the compressor characteristic coefficient and the condenser characteristic coefficient through the compressor data and the condenser data includes:

[0096] Extract the measured values ​​corresponding to the steam pressure, temperature and flow of the compressor in the measurement vector, as well as the measured values ​​of the flow, pressure and temperature of the condenser, perform curve fitting to obtain the characteristic coefficients of the compressor and the condenser, and set the curve fitting formula as Among them, W_Y o It is used to represent the characteristic coefficient of the compressor or the characteristic coefficient of the condenser at the time point o, and o=t or 1, ..., e, ...r, te represents the time points before the time point o; Obtained by least squares method, x_g y represents the correlation coefficient between the yth measurement value and the fault, represents the measurement error variance of the yth sensor at time point t-1, z_t y represents the measurement value of the yth sensor, and Y represents the number of measurements.

[0097] The compressor characteristic coefficient and the condenser characteristic coefficient are used to represent the compressor data and the condenser data, which not only simplifies the data volume, but also reflects the comprehensive characteristics of the data, improves the data processing speed, enhances the data mining depth, and realizes the high-quality input of subsequent data.

[0098] Correlation coefficient x_g y It is used to evaluate the data categories represented by each measurement value and the degree of influence of the equipment failure. In the present invention, the data categories represented by each measurement value are respectively the steam pressure, temperature and flow of the compressor or the flow, pressure and temperature of the condenser. The equipment failure is the compressor or condenser failure. The equipment state is set to Z_u, and Z_u=1 indicates failure, Z_u=0 indicates normal, and the fault association model is set to Among them, β0 is the intercept term, βy is the y-th measurement value z_t y The regression coefficient corresponding to the data category, and the regression coefficient βy is estimated by maximizing the likelihood function, x_g y (Z_u=1|X_y) represents the regression coefficient corresponding to the data category of y measurement values ​​and the equipment failure as the correlation coefficient;

[0099] The correlation coefficient x_g at time point t y It is obtained by using the measured value at time point t-1 through the fault correlation model calculation.

[0100] Segmentation module: obtain the initial countermeasure set from the comprehensive data matrix through the anomaly recognition model, perform cluster analysis on the initial countermeasure set, and obtain k groups of sub-countermeasure sets;

[0101] The initial set of countermeasures is obtained from the comprehensive data matrix through the anomaly identification model, including:

[0102] Set a sample set, the sample set includes H_k group subsets, each subset includes a comprehensive data matrix, anomaly measurement values ​​and an initial countermeasure set corresponding to the anomaly measurement values, the comprehensive data matrix is ​​marked as the input of the anomaly recognition model, the anomaly recognition model matches the anomaly measurement value with the fixed countermeasure set, outputs the initial countermeasure set, the anomaly measurement value and the initial countermeasure set are marked as the output of the anomaly recognition model, the sample set is divided into a training set and a validation set, and they are input into the anomaly recognition model in sequence;

[0103] The fixed countermeasure set includes all the control strategies when the abnormal measurement value is regulated to the normal state. Each control strategy includes the parameter settings used by the equipment when the MVP system is adjusted. The abnormal recognition model uses an optimization algorithm (such as simulated annealing, genetic algorithm) to search for the optimal control strategy of group z_c in the fixed countermeasure set to form an initial countermeasure set; the fixed countermeasure set can also be a set of control strategies used at all historical time points, or it can be an integration of control strategies independently set by staff based on experience or experimental data analysis, or it can be a set of strategies automatically generated by some learning machines based on known data.

[0104] Set up an anomaly recognition model based on RBF and LSTM. Optimize model parameters by minimizing the loss function, use the validation set for cross-validation, adjust model hyperparameters, use accuracy, recall, and F1-score as performance evaluation indicators, and stop when the expected performance evaluation indicators are reached. Then obtain a trained anomaly recognition model and export the trained anomaly recognition model into a deployable format.

[0105] Import the comprehensive data matrix at time point t into the trained anomaly recognition model to obtain the initial countermeasure set at time point t.

[0106] The anomaly recognition model is based on RBF and LSTM learning to analyze and predict the changing trends of the measured values ​​at multiple time points in the comprehensive data matrix, and obtain the measured values ​​with abnormal or dangerous states, which facilitates the subsequent analysis of abnormal measured values ​​and reduces the amount of data processing.

[0107] Methods for clustering analysis of the initial set of countermeasures include:

[0108] Step S1: Set the number of clusters k_zg and randomly select k_zg control strategies as the initial centroid;

[0109] Step S2: Calculate the distance between each control strategy and each centroid, and assign each control strategy to the cluster represented by the centroid closest to it;

[0110] Step S3: Calculate a new centroid for all control strategies in each cluster, where the centroid is the mean of all control strategies in the cluster;

[0111] Step S4: repeat steps S2 to S3 until the centroid does not change between two consecutive times;

[0112] Step S5: Output the control strategy in each cluster and obtain a set of k sub-strategies.

[0113] In the present invention, when calculating between control strategies, the calculation is performed through the numerical values ​​of the same category in the control strategies.

[0114] The number of clusters k_zg can be obtained by:

[0115] A cluster number selection data set K = {k_1, ..., k_n, ..., k_N} is preset, where k_n represents the number of clusters represented by the nth value of the cluster number selection data set, and n = 1, ..., N;

[0116] Randomly select a k_n as the pre-set number of clusters and calculate the average silhouette coefficient when the number of clusters is k_n Among them, z_c represents the number of control strategies in the initial countermeasure set;

[0117] Traverse and calculate the average silhouette coefficient of all clusters in the selected data set, select the cluster number with the largest silhouette coefficient as the optimal cluster number, and set the optimal cluster number as the clustering cluster number k_zg;

[0118] For any control strategy, calculate the average distance between it and all other control strategies in the same cluster Where |C_sj| represents the number of control strategies in cluster C_j, and d(j,l) represents the distance between the jth control strategy and the lth control strategy (e.g., Euclidean distance). The average distance between it and the points in all other clusters is calculated to find the nearest cluster (i.e., the cluster with the smallest average distance). Among them, cluster C_h is different from cluster C_j, d(j,z) is the distance between the j-th regulation strategy and the z-th regulation strategy in cluster C_h, and C_sh represents the number of regulation strategies in cluster C_h.

[0119] By dividing the initial countermeasure set into multiple sub-countermeasure sets, similar control strategies are clustered, reducing the complexity of the data, simplifying subsequent analysis and processing tasks, reducing local difficulties caused by excessive data dispersion during the optimization process, reducing the computing power used in data retrieval, optimizing resource allocation and strategy formulation, and facilitating accurate optimization among different similar strategy groups, thereby achieving high-quality data processing.

[0120] Solution output module: Use optimization algorithm to optimize k groups of sub-strategy sets to obtain the candidate set, and then use optimization algorithm to optimize the candidate set again to obtain the optimal control solution of MVP system recycling control;

[0121] Methods of using optimization algorithms to optimize k groups of sub-strategy sets to obtain candidate sets include:

[0122] Each control strategy in any set of sub-strategies is regarded as a hunting dog. The hunting dogs cooperate with each other and constantly adjust their positions (i.e., the parameter settings in the control strategy) to get closer to the optimal prey (optimal control strategy).

[0123] Set the position update formula for each hound individual: in, represents the position of the vth hound at the oth iteration, wp v is the local optimal position of the vth hound, wp is the global optimal position, β1 and β2 are learning factors, and ε1 and ε2 are random factors between 0 and 1;

[0124] Setting fitness in, and are the stability and energy consumption index of the control strategy after the vth hound is updated, γ1 and γ2 are weight coefficients, if the fitness If it exceeds the current global optimal fitness, the global optimal position is updated, that is, When the change rate of two consecutive fitness values ​​is less than the set threshold in several consecutive iterations, the process stops.

[0125] Set the constraints to Where SF represents the number of stability determining parameters, θ wx represents the weight coefficient of the wxth parameter, Δh_l represents the absolute difference between the updated control strategy and the standard value, f_op and YU_D are pre-set thresholds, θ1 and θ2 represent weight coefficients, and Q RE represents the recovered heat, Q IN represents the input heat, Q FR Indicates the usage of fresh steam;

[0126] Output the current global optimal position wp, which is the corresponding optimal control strategy in the sub-strategy set.

[0127] The optimal control scheme for the MVP system recycling control is obtained again through the optimization algorithm for the selected set, including:

[0128] Collect the optimal control strategies of all sub-strategy sets to form a candidate set. Use the optimization algorithm to optimize the control strategies in the candidate set again, and output the optimal control strategy in the candidate set, which is recorded as the optimal control strategy at time point t. The MVP system adjusts the parameters at the next time point based on the optimal control strategy at this time.

[0129] In this embodiment, the present invention uses compressor characteristic coefficients and condenser characteristic coefficients to represent compressor data and condenser data, which not only simplifies the data volume, but also reflects the comprehensive characteristics of the data, improves the data processing speed, enhances the data mining depth, and realizes the high-quality input of subsequent data. The recognition model often learns the changing trend of the measured values ​​at multiple time points in the comprehensive data matrix based on RBF and LSTM, analyzes and predicts the numerical values, and obtains the measured values ​​of abnormal or dangerous states, which is convenient for the subsequent analysis of abnormal measured values, reduces the processing of data volume, and improves the optimal use of data. The optimization algorithm realized by the present invention realizes multi-layer optimization of the control strategy and outputs the optimal control strategy. Through the multi-layer control strategy segmentation, it is convenient for the centralized analysis and processing of data with the same characteristics. The optimization layer by layer not only reduces the data processing volume, but also improves the operation rate through parallel calculation; the adaptive control optimization mechanism continuously improves the overall efficiency of intelligent analysis, and multiple modules cooperate with each other to realize comprehensive data collection, processing and analysis, and provide real-time and accurate analysis of detection data.

[0130] Embodiment 2

[0131] Reference Figure 1 and Figure 2 , Example 2 further illustrates a recycling control system based on an MVR system proposed by the present invention.

[0132] The recycling control method based on the MVR system comprises the following steps:

[0133] Step S1: Collect the comprehensive status data of the MVR system at time point t, and perform fusion processing on the collected data to obtain a comprehensive data matrix;

[0134] Step S2: construct an anomaly recognition model and train it, import the comprehensive data matrix at time point t into the trained anomaly recognition model, and obtain an initial countermeasure set;

[0135] Step S3: Set the number of clusters k_zg, perform cluster analysis on the initial countermeasure set, and obtain k groups of sub-countermeasure sets;

[0136] Step S4: Use the optimization algorithm to optimize the k groups of sub-strategy sets respectively, collect the optimized control strategies to form a candidate set, optimize the candidate set again through the optimization algorithm to obtain the optimal control plan, and apply the output optimal control plan to the recovery control of the MVP system.

[0137] The recycling control method based on the MVR system is applied to a recycling control system based on the MVR system and includes:

[0138] Data acquisition module: used to collect the comprehensive status data of the MVR system, and perform fusion processing on the collected data to obtain a comprehensive data matrix;

[0139] Segmentation module: obtain the initial countermeasure set from the comprehensive data matrix through the anomaly recognition model, perform cluster analysis on the initial countermeasure set, and obtain k groups of sub-countermeasure sets;

[0140] Solution output module: Use optimization algorithm to optimize k groups of sub-strategy sets to obtain the candidate set, and then use optimization algorithm to optimize the candidate set again to obtain the optimal control solution of MVP system recycling control;

[0141] The modules are connected to each other via wired and / or wireless means.

[0142] In addition, according to the implementation of the present application, the process described in the figure of a recycling control system based on an MVR system can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application. Of course, the architecture shown in the figure of a recycling control system based on an MVR system is only exemplary. When implementing different devices, adaptive selection or adjustment can be performed according to actual needs.

[0143] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0144] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A recycling control system based on an MVR system, characterized in that: The recycling control system based on the MVR system includes: Data acquisition module: used to collect the comprehensive status data of the MVR system, and perform fusion processing on the collected data to obtain a comprehensive data matrix; Segmentation module: obtain the initial countermeasure set from the comprehensive data matrix through the anomaly recognition model, perform cluster analysis on the initial countermeasure set, and obtain k groups of sub-countermeasure sets; Solution output module: Use optimization algorithm to optimize k groups of sub-strategy sets to obtain the candidate set, and then use optimization algorithm to optimize the candidate set again to obtain the optimal control solution of MVP system recycling control; The modules are connected to each other via wired and / or wireless means.

2. A recycling control system based on an MVR system according to claim 1, characterized in that: The data categories of the comprehensive status data include evaporator data, compressor data, heat exchanger data, condenser data and separator data. The evaporator data includes the liquid level of the evaporator, the compressor data includes the steam pressure, temperature and flow of the compressor, the heat exchanger data includes the temperature of the heat exchanger, the condenser data includes the flow, pressure and temperature of the condenser, and the separator data includes the liquid level of the separator.

3. A recycling control system based on the MVR system according to claim 2, characterized in that: The method of fusing the collected data includes: Assume that the comprehensive state data is provided by q sensors, and at time point t, obtain the measurement values ​​of w data categories to obtain the measurement vector Among them, z_t i represents the measurement value of the i-th sensor, and i=1,...,q; The compressor characteristic coefficient and the condenser characteristic coefficient are obtained through the compressor data and the condenser data, the measured values ​​corresponding to the steam pressure, temperature and flow rate of the compressor in the measurement vector, and the measured values ​​of the flow rate, pressure and temperature of the condenser are deleted, the compressor characteristic coefficient and the condenser characteristic coefficient are filled into the measurement vector, and the labels are distributed to obtain the fused feature vector; Taking time point t as the base point, the feature vectors at the previous p_a time points are continuously collected, arranged in chronological order, and combined into a time series matrix, which is recorded as the comprehensive data matrix at time point t.

4. A recycling control system based on the MVR system according to claim 3, characterized in that: The method for obtaining the compressor characteristic coefficient and the condenser characteristic coefficient through the compressor data and the condenser data includes: Extract the measured values ​​corresponding to the steam pressure, temperature and flow of the compressor in the measurement vector, as well as the measured values ​​of the flow, pressure and temperature of the condenser, perform curve fitting to obtain the characteristic coefficients of the compressor and the condenser, and set the curve fitting formula as Among them, W_Y o It is used to represent the characteristic coefficient of the compressor or the characteristic coefficient of the condenser at the time point o, and o=t or 1, ..., e, ...r, te represents the time points before the time point o; Obtained by least squares method, x_g y represents the correlation coefficient between the yth measurement value and the fault, represents the measurement error variance of the yth sensor at time point t-1, z_t y represents the measurement value of the yth sensor, and Y represents the number of measurements.

5. A recycling control system based on the MVR system according to claim 4, characterized in that: The correlation coefficient x_g y It is used to evaluate the impact of data categories represented by each measurement value and equipment failure. The equipment state is set to Z_u, and Z_u=1 indicates failure, Z_u=0 indicates normal, and the fault association model is set to Among them, β0 is the intercept term, βy is the y-th measurement value z_t y The regression coefficient corresponding to the data category, and the regression coefficient βy is estimated by maximizing the likelihood function, x_g y (Z_u=1|X_y) represents the regression coefficient corresponding to the data category of y measurement values ​​and the equipment failure as the correlation coefficient; The correlation coefficient x_g at time point t y It is obtained by using the measured value at time point t-1 through the fault correlation model calculation.

6. A recycling control system based on the MVR system according to claim 5, characterized in that: The initial countermeasure set is obtained from the comprehensive data matrix through the anomaly recognition model, specifically including: Set a sample set, the sample set includes H_k group subsets, each subset includes a comprehensive data matrix, anomaly measurement values ​​and an initial countermeasure set corresponding to the anomaly measurement values, the comprehensive data matrix is ​​marked as the input of the anomaly recognition model, the anomaly recognition model matches the anomaly measurement value with the fixed countermeasure set, outputs the initial countermeasure set, the anomaly measurement value and the initial countermeasure set are marked as the output of the anomaly recognition model, the sample set is divided into a training set and a validation set, and they are input into the anomaly recognition model in sequence; The fixed countermeasure set includes all control strategies when the abnormal measurement value is regulated to the normal state. Each control strategy includes the parameter settings used by the equipment when the MVP system is adjusted. The abnormal identification model uses the optimization algorithm to search for the optimal control strategy of group z_c in the fixed countermeasure set to form the initial countermeasure set. Set up an anomaly recognition model based on RBF and LSTM. Optimize model parameters by minimizing the loss function, use the validation set for cross-validation, adjust model hyperparameters, use accuracy, recall, and F1-score as performance evaluation indicators, and stop when the expected performance evaluation indicators are reached. Then obtain a trained anomaly recognition model and export the trained anomaly recognition model into a deployable format. Import the comprehensive data matrix at time point t into the trained anomaly recognition model to obtain the initial countermeasure set at time point t.

7. A recycling control system based on the MVR system according to claim 6, characterized in that: The method for clustering the initial countermeasure set comprises: Step S1: Set the number of clusters k_zg and randomly select k_zg control strategies as the initial centroid; Step S2: Calculate the distance between each control strategy and each centroid, and assign each control strategy to the cluster represented by the centroid closest to it; Step S3: Calculate a new centroid for all control strategies in each cluster, where the centroid is the mean of all control strategies in the cluster; Step S4: repeat steps S2 to S3 until the centroid does not change between two consecutive times; Step S5: Output the control strategy in each cluster and obtain a set of k sub-strategies.

8. A recycling control system based on the MVR system according to claim 7, characterized in that: The method for obtaining the number of clusters k_zg includes: A cluster number selection data set K = {k_1, ..., k_n, ..., k_N} is preset, where k_n represents the number of clusters represented by the nth value of the cluster number selection data set, and n = 1, ..., N; Randomly select a k_n as the pre-set number of clusters and calculate the average silhouette coefficient when the number of clusters is k_n Among them, z_c represents the number of control strategies in the initial countermeasure set; Traverse and calculate the average silhouette coefficient of all clusters in the selected data set, select the cluster number with the largest silhouette coefficient as the optimal cluster number, and set the optimal cluster number as the clustering cluster number k_zg; For any control strategy, calculate the average distance between it and all other control strategies in the same cluster Where |C_sj| represents the number of control strategies in cluster C_j, d9j,l) represents the distance between the jth control strategy and the lth control strategy; and calculates the average distance between it and the points in all other clusters to find the nearest cluster Among them, cluster C_h is different from cluster C_j, d9j,z) is the distance between the j-th regulation strategy and the z-th regulation strategy in cluster C_h, and C_sh represents the number of regulation strategies in cluster C_h.

9. A recycling control system based on the MVR system according to claim 8, characterized in that: The method of using the optimization algorithm to respectively optimize the k groups of sub-strategy sets to obtain the to-be-selected sets includes: Each control strategy in any set of sub-strategies is regarded as a hunting dog individual. The hunting dogs cooperate with each other and constantly adjust their positions to get closer to the optimal prey. Set the position update formula for each hound individual: in, represents the position of the vth hound at the oth iteration, wp v is the local optimal position of the vth hound, wp is the global optimal position, β1 and β2 are learning factors, and ε1 and ε2 are random factors between 0 and 1; Setting fitness in, and are the stability and energy consumption index of the control strategy after the vth hound is updated, γ1 and γ2 are weight coefficients, if the fitness If it exceeds the current global optimal fitness, the global optimal position is updated, that is, When the change rate of two consecutive fitness values ​​is less than the set threshold in several consecutive iterations, the process stops. Set the constraints to Where SF represents the number of stability determining parameters, θ wx represents the weight coefficient of the wxth parameter, Δh_l represents the absolute difference between the updated control strategy and the standard value, f_op and YU_D are pre-set thresholds, represents the weight coefficient, Q RE represents the recovered heat, Q IN represents the input heat, Q FR Indicates the usage of fresh steam; Output the current global optimal position wp, which is the corresponding optimal control strategy in the sub-strategy set.

10. A recycling control system based on the MVR system according to claim 9, characterized in that: The method of obtaining the optimal control scheme of the MVP system recycling control by optimizing the selected set again through the optimization algorithm includes: Collect the optimal control strategies of all sub-strategy sets to form a candidate set. Use the optimization algorithm to optimize the control strategies in the candidate set again, and output the optimal control strategy in the candidate set, which is recorded as the optimal control strategy at time point t. The MVP system adjusts the parameters at the next time point based on the optimal control strategy at this time.

Images