Hybrid control system and method for gas compressor

Abstract

The invention discloses a hybrid control system and method for a gas compressor, and relates to the technical field of mechanical engineering, and the method comprises the following steps: during the operation period of the gas compressor, collecting multi-dimensional operation parameter data generated by each controller in the compressor in the operation process in real time; and preprocessing the collected original multi-dimensional operation parameter data, and organizing the processed effective data according to a time sequence structure to form a data set for subsequent feature extraction and model analysis. According to the method, machine learning and multi-dimensional feature analysis are fused, and intelligent recognition and dynamic adjustment of lagging interval overlapping of the multiple controllers of the gas compressor are achieved. By collecting operation parameters and extracting key indexes, combining with a machine learning model to identify a control blind area, and based on a control priority and an error state, dynamically optimizing lagging area setting, a phenomenon that a controller has no response collectively is effectively avoided, and responsiveness, stability and operation efficiency of a system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical engineering, and particularly to a hybrid control system and method for a gas compressor. Background Art

[0002] The hybrid control of a gas compressor refers to the comprehensive application of multiple control strategies (such as pressure control, flow control, temperature control, start-stop control, etc.) in a control system. By coordinating the operation of each control loop, the optimal regulation of the compressor's operating state is achieved. Hybrid control can not only automatically adjust the operating parameters of the compressor according to the change of the system load to maintain the stable and efficient operation of the system, but also avoid energy consumption waste, reduce mechanical wear, and improve the overall economy and reliability of the system by means of master-slave control and load distribution when multiple compressors are operating in parallel. This control method is often combined with advanced sensor technology and automation platforms such as PLC and DCS to achieve refined management.

[0003] The prior art has the following deficiencies:

[0004] The prior art usually sets a hysteresis interval in the automatic control system to suppress frequent regulation caused by minute fluctuations and prevent the controller or actuator from overresponding. In the event of a sudden load fluctuation, to prevent misoperation caused by instantaneous interference, the system automatically enters the "Suppression Mode", and enhances system stability by temporarily expanding the hysteresis interval. However, when multiple controllers are operating in parallel and there are cross-coupling relationships among their controlled objects (such as pressure, flow, temperature, etc.), if the hysteresis intervals are simultaneously expanded without unified coordination among the controllers in the suppression mode, it may lead to the overlap of the hysteresis intervals, and the error ranges of some key variables are jointly ignored by multiple controllers, thus forming a "blind area" where no one responds. If the load disturbance persists at this time and the controller does not intervene in the regulation for a long time due to the expansion of the hysteresis interval, it may cause the system parameters to gradually deviate from the set values, ultimately leading to serious operation risks such as equipment instability, overpressure, and overload.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The objective of the present invention is to provide a hybrid control system and method for a gas compressor. By integrating machine learning and multi-dimensional feature analysis, it realizes the intelligent identification and dynamic adjustment of the overlapping lag intervals of multiple controllers of the gas compressor. By collecting operating parameters and extracting key indicators, combining with a machine learning model to identify control blind spots, and dynamically optimizing the lag zone settings based on control priorities and error states, it effectively avoids the phenomenon of collective non-response of controllers, significantly improves the responsiveness, stability, and operating efficiency of the system, so as to solve the problems in the above-mentioned background technology.

[0007] To achieve the above objective, the present invention provides the following technical solution: A hybrid control method for a gas compressor, including the following steps:

[0008] During the operation of the gas compressor, multi-dimensional operating parameter data generated by each controller inside the compressor during operation is collected in real time;

[0009] The collected original multi-dimensional operating parameter data is preprocessed, and the processed valid data is organized into a data set according to the time series structure for subsequent feature extraction and model analysis;

[0010] Through feature engineering and statistical analysis techniques, key dimension indicators reflecting the overlap of the lag intervals of each controller are extracted from the data set, and the extracted key dimension indicators are comprehensively analyzed to identify potential overlap relationships between the controller lag intervals;

[0011] The key dimension indicators after analysis and processing are used as input data and input into a pre-trained machine learning model. Through the machine learning model, it is identified and judged whether there is an overlap in the lag intervals between each controller;

[0012] When the machine learning model identifies that there is an overlap relationship in the lag intervals between each controller, according to the control importance levels of each control loop and the current error state, a priority evaluation is performed on all control loops to determine the target control loop that needs to be intervened first; after completing the determination of the target control loop, a real-time working condition evaluation algorithm is triggered, and according to the current load level, disturbance frequency, and the priority relationship between controllers, the lag interval ranges of each controller are dynamically adjusted.

[0013] Preferably, the specific steps for preprocessing the collected original multi-dimensional operating parameter data include:

[0014] First, the collected data is denoised, and a filtering algorithm is used to eliminate random noise caused by sensor fluctuations or instantaneous interference;

[0015] Secondly, outlier detection and removal are performed to identify and remove data mutation points caused by sensor failures, communication interruptions, and physical anomalies;

[0016] Subsequently, the data is aligned and synchronized to ensure that the parameters collected by different controllers are unified under the same time reference, and data completion and interpolation are completed to repair the data gaps caused by instantaneous packet loss and signal absence;

[0017] Then, all the data is normalized to unify the dimension and eliminate the interference of dimensional differences on model analysis;

[0018] The processed valid data is sorted according to the timestamps and organized into a data set with a time series structure according to the sampling period, providing a stable, continuous, and high-quality data basis for subsequent feature extraction and machine learning model analysis.

[0019] Preferably, through feature engineering and statistical analysis techniques, key dimensional indicators reflecting the overlap of the lag intervals of each controller are extracted from the data set. The extracted indicators include the proportion of the residence time of "no response" within the intersection range of the lag intervals of the variable and the abnormal residence time of the process variable within any unstable value segment. The proportion of the residence time of "no response" of the extracted variable within the intersection range of the lag intervals and the abnormal residence time of the process variable within any unstable value segment are analyzed under the detection window to generate a blind area residence reference value and an unexpected residence reference value respectively, and the potential overlap relationship between the controller lag intervals is identified through the blind area residence reference value and the unexpected residence reference value.

[0020] Preferably, the specific steps for analyzing the proportion of the residence time of "no response" of the variable within the intersection range of the lag intervals under the detection window to generate a blind area residence reference value are as follows:

[0021] First, it is necessary to identify whether the error value of the key control variable during operation falls into the intersection area of the lag intervals of multiple controllers, and construct a residence mapping function for quantification. The defined formula is as follows:

[0022] M i = δ(E i ∈D)·ω i

[0023] , where M i is the blind area mark value of the i-th sampling point, E i is the target set value of the i-th sampling point, D is the intersection area of the lag intervals, δ(E i ∈D) is a logical decision function used to detect whether the error E i is in the intersection area D, and ω i is the error depth weight factor;

[0024] After obtaining the blind area state sequence, a response confidence weight function is introduced to evaluate the continuity of the blind area residence and the severity of the control gap, and a blind area residence reference value is generated. The generation formula is as follows:

[0025]

[0026] , where BZD is the blind zone residence reference value, and φ(S i ) is the response confidence weight function, which is used to assign a higher risk weight to the "continuous no-response" behavior for a longer time. S i is the residence length that measures the continuous blind zone state of the variable starting from the i-th sampling point, and κ i is the controller response missing factor.

[0027] Preferably, the specific steps for analyzing the abnormal residence time of the process variable within any unstable value segment under the detection window to generate the unexpected residence reference value are as follows:

[0028] Perform point-by-point analysis on the process variable. Based on the preset stable target range, determine whether the variable is in the unexpected interval, and define the residence marking function. The formula is as follows:

[0029]

[0030] , where V(q) is the process variable data collected at the q-th time point, [L, H] is the expected stable interval of the process variable, L is the upper limit of the expected stable interval, H is the lower limit of the expected stable interval, ε is the micro-variation threshold for identifying the "static residence" state, and S(q) is the output value of the residence marking function;

[0031] After obtaining the residence marking sequence, introduce a non-linear enhancement function to calculate the unexpected residence reference value. The calculation expression is as follows:

[0032]

[0033] , where UD is the unexpected residence reference value, m is the total number of sampling points within the detection window, D(q) is the residence history weighting factor for enhancing the influence degree of continuous residence. The formula is as follows:

[0034]

[0035] , where indicates how many points among the h sampling points before the current point q belong to the "unexpected residence" state, and S(q - j) is the unexpected residence mark at the q - j-th time point, representing the residence state of the past j-th sampling point.

[0036] Preferably, the blind area residence reference value and the unexpected stay reference value obtained through analysis and processing are used as input data and input into a pre-trained machine learning model. The machine learning model generates a lag interval overlap risk coefficient, and whether there is a lag interval overlap between each controller is intelligently identified and judged through the lag interval overlap risk coefficient.

[0037] Preferably, the lag interval overlap risk coefficient generated when the machine learning model identifies and judges whether there is a lag interval overlap between each controller is compared and analyzed with a preset lag interval overlap risk coefficient reference threshold to judge whether there is a lag interval overlap between each controller. The judgment steps are as follows:

[0038] If the lag interval overlap risk coefficient is greater than the preset lag interval overlap risk coefficient reference threshold, it is judged that there is a lag interval overlap between each controller; if the lag interval overlap risk coefficient is less than or equal to the preset lag interval overlap risk coefficient reference threshold, it is judged that there is no lag interval overlap between each controller.

[0039] Preferably, when the machine learning model identifies that there is a lag interval overlap relationship between each controller, a priority evaluation is performed on all control loops to determine the target control loop that needs to be intervened first; after the determination of the target control loop is completed, a real-time working condition evaluation algorithm is triggered, and the specific steps for dynamically adjusting the lag interval range of each controller are as follows:

[0040] When the machine learning model identifies that there is a lag interval overlap between controllers, first a comprehensive priority evaluation is performed on all involved control loops. The formula is as follows:

[0041]

[0042] , where PRI(a) is the priority evaluation index of the a-th controller, W imp (a) is the control importance level weight coefficient of the a-th controller, E(a) is the error state collected in real time by the a-th controller, DOR is the lag interval overlap risk coefficient, DOR ref is the lag interval overlap risk coefficient reference threshold;

[0043] After the priority evaluation of the target control loop is completed, the real-time working condition evaluation algorithm is immediately started to dynamically adjust the lag interval of each controller. The specific formula is:

[0044]

[0045] , where ADJ(a) is the adjusted lag interval range of the a-th controller, DB ini (a) is the initial lag interval range of the a-th controller, fscan (a) is the current real-time scanning and monitoring frequency of the a-th control loop, f ini is the initial flow scanning frequency, PRI avg is the average value of the real-time priority indices of all controllers.

[0046] Hybrid control system for a gas compressor, comprising a multi-dimensional data acquisition module, a data preprocessing and sequence construction module, a feature extraction and overlapping relationship identification module, a machine learning discrimination module, and a priority determination and hysteresis zone dynamic adjustment module:

[0047] Multi-dimensional data acquisition module, during the operation of the gas compressor, it acquires in real time multi-dimensional operation parameter data generated by each controller inside the compressor during the operation process;

[0048] Data preprocessing and sequence construction module, preprocesses the acquired original multi-dimensional operation parameter data, and organizes the processed valid data into a data set according to the time series structure for subsequent feature extraction and model analysis;

[0049] Feature extraction and overlapping relationship identification module, through feature engineering and statistical analysis techniques, extracts key dimension indicators reflecting the overlap of the hysteresis intervals of each controller from the data set, and comprehensively analyzes the extracted key dimension indicators to identify potential overlapping relationships between the hysteresis intervals of the controllers;

[0050] Machine learning discrimination module, takes the key dimension indicators after analysis and processing as input data, inputs them into a pre-trained machine learning model, and identifies and determines whether there is an overlap of hysteresis intervals between each controller through the machine learning model;

[0051] Priority determination and hysteresis zone dynamic adjustment module, when the machine learning model identifies an overlap relationship of hysteresis intervals between each controller, evaluates the priorities of all control loops according to the control importance levels and current error states of each control loop to determine the target control loop that needs to be intervened first; after completing the determination of the target control loop, triggers a real-time working condition evaluation algorithm, and dynamically adjusts the hysteresis interval ranges of each controller according to the current load level, disturbance frequency, and priority relationship between the controllers.

[0052] In the above technical solution, the technical effects and advantages provided by the present invention:

[0053] The present invention realizes the efficient identification and precise intervention of the overlapping problem of multiple controller lag intervals during the operation of a gas compressor by introducing a controller lag interval identification and dynamic adjustment mechanism based on machine learning and multi-dimensional feature analysis. By collecting the multi-controller operation parameters in real time, constructing a time series data set and extracting key dimension indicators, it can effectively identify the "control blind area" generated due to the intersection of lag intervals under specific working conditions; combined with the high-precision judgment ability of the machine learning model, it further improves the system's ability to predict potential control conflicts; after identifying the overlapping relationship, according to the control importance level and the current error state of the system, the lag interval is dynamically optimized and adjusted, effectively avoiding the phenomenon that multiple controllers are collectively silent when the key variables deviate, thus significantly improving the responsiveness, stability and overall operation efficiency of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0055] Figure 1 It is a method flow chart of the hybrid control method for the gas compressor of the present invention.

[0056] Figure 2 It is a module schematic diagram of the hybrid control system for the gas compressor of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Now, the exemplary embodiments will be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided so that the present disclosure will be more complete and comprehensive, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0058] The present invention provides a hybrid control method for a gas compressor as Figure 1 shown, including the following steps:

[0059] During the operation of the gas compressor, multi-dimensional operation parameter data generated by each controller inside the compressor during operation is collected in real time;

[0060] Multi-dimensional operating parameter data includes but is not limited to multi-dimensional information such as pressure, flow, temperature, controller output signal, and control target value. By deploying sensors or data acquisition modules at the sampling nodes of the control system, comprehensive monitoring of the current working conditions of each controller can be ensured, and data updates can be completed within a very short time interval. The specific role of this step is to provide a high-precision and timely data foundation for subsequent analysis and processing, and to provide a variety of input samples for feature extraction and judgment of machine learning models, thereby ensuring the effectiveness and accuracy of monitoring results.

[0061] Preprocess the collected original multi-dimensional operating parameter data, and organize the processed valid data into a data set according to the time series structure for subsequent feature extraction and model analysis;

[0062] The preprocessing process includes conventional operations such as denoising, outlier filtering, data alignment and normalization, and more complex strategies such as interpolation correction and data smoothing can be used according to actual needs. The key significance of this step is to ensure that the data used in subsequent steps has good consistency and accuracy, and to exclude data anomalies caused by external interference as much as possible, thereby reducing the misleading of feature extraction and machine learning model judgment, and improving the reliability of overall recognition.

[0063] The data set not only contains the single parameters of a single controller, but also includes the parallel observation data of multiple controllers, as well as the possible interaction information between the controllers. By mapping the data to a suitable multidimensional data structure (such as a multidimensional matrix, tensor, or hierarchical database), data of different dimensions can be uniformly managed and efficiently retrieved. The purpose of this step is to build a data foundation for further performing feature extraction, pattern recognition, or machine learning algorithms, avoiding processing difficulties caused by scattered data or inconsistent structures in subsequent analysis.

[0064] The specific steps of preprocessing the collected original multi-dimensional operating parameter data include: first, denoising the collected data, using filtering algorithms (such as sliding average, median filtering, etc.) to eliminate random noise caused by sensor fluctuations or instantaneous interference; second, performing outlier detection and elimination, identifying and removing data mutation points caused by sensor failures, communication interruptions, and physical anomalies; then, aligning and synchronizing the data to ensure that the parameters collected by different controllers are unified under the same time base, and completing data completion and interpolation to repair data gaps caused by instantaneous packet loss and signal loss; then normalizing all data, unifying the dimensions, and eliminating the interference of dimensional differences on model analysis. After completing the above steps, the processed valid data is sorted according to the timestamp and organized into a data set with a time series structure according to the sampling period, providing a stable, continuous, and high-quality data foundation for subsequent feature extraction and machine learning model analysis.

[0065] Through feature engineering and statistical analysis techniques, key dimensional indicators reflecting the overlap of the lag intervals of each controller are extracted from the data set, and the extracted key dimensional indicators are comprehensively analyzed to identify potential overlap relationships between the controller lag intervals;

[0066] Through feature engineering and statistical analysis techniques, key dimensional indicators reflecting the overlap of the lag intervals of each controller are extracted from the data set. The extracted indicators include the proportion of the residence time of "no response" within the intersection range of the lag intervals of the variable and the abnormal residence time of the process variable within any unstable value segment. The proportion of the residence time of "no response" within the intersection range of the lag intervals of the extracted variable and the abnormal residence time of the process variable within any unstable value segment are analyzed under the detection window to generate a blind area residence reference value and an unexpected residence reference value respectively, and potential overlap relationships between the controller lag intervals are identified through the blind area residence reference value and the unexpected residence reference value.

[0067] When the proportion of the residence time of "no response" within the intersection range of the lag intervals of the variable increases significantly, it usually indicates clearly that there is a phenomenon of overlap of multiple controller lag intervals in the system. The reason is that the lag interval is the non-response error range set by each controller to suppress minor fluctuations. If multiple controllers control the same or mutually coupled variables and their lag intervals intersect, a "control blind area" may be formed. When the process variable error stays in this intersection area for a long time and no controller generates an effective adjustment action, that is, it shows a state of "no response", it means that these controllers all regard this error as a tolerable range and thus give up the response. This "collective silence" is the direct consequence of the overlap of the lag intervals, and as the intersection area expands and the load fluctuation intensifies, the duration of this state will increase significantly. Therefore, the "increase in the proportion of the residence time in the blind area" essentially reflects the coincidence failure of multiple controllers in error identification and response range, and is one of the typical dynamic characteristics for identifying the problem of lag interval overlap.

[0068] The specific steps for analyzing the proportion of the residence time of "no response" within the intersection range of the lag intervals of the variable under the detection window to generate a blind area residence reference value are as follows:

[0069] First, it is necessary to identify whether the error value of the key control variable falls into the intersection area of the lag intervals of multiple controllers during operation, and construct a residence mapping function for quantification. The defined formula is as follows:

[0070] M i =δ(E i ∈D)·ω i

[0071] , where M i is the blind area mark value of the i-th sampling point, Ei is the target set value of the i-th sampling point, D is the intersection region of the lag intervals, and δ(E i ∈ D) is a logical decision function used to detect the error E i whether it is in the intersection region D, if the output is 1, it means that the current point "is in the blind area"; if the output is 0, it means that the current point does not form a blind area, ω i is the error depth weight factor, indicating the "depth" of the current error from the center of the lag intersection. The smaller the deviation (the deeper the error into the intersection center), the greater the weight; the larger the deviation (close to the edge), the smaller the weight;

[0072] The above steps use the logical decision function to mark each sampling point as whether it is in the "lag overlap blind area", thereby constructing the "blind area state sequence" of the variable during the entire running time.

[0073] After obtaining the blind area state sequence, a response confidence weight function is introduced to evaluate the continuity of the stay in the blind area and the severity of the control gap, generating a reference value for the blind area stay. The generation formula is as follows:

[0074]

[0075] , where BZD is the reference value for the blind area stay, φ(S i ) is the response confidence weight function, which assigns a higher risk weight to the "continuous no-response" behavior for a longer time, and S i is a measure of the stay length of the variable continuously in the blind area state starting from the i-th sampling point (i.e., the number of consecutive M = 1), and κ i is the controller response missing factor, which measures how many controllers in the system should respond but do not respond in the blind area state, that is, the "degree of response missing".

[0076] The above steps use the response confidence weight function to weight the continuous stay behavior of the variable in the intersection of the lag intervals, highlighting the control risk brought by the "long-time no-response". This method can avoid missing the continuity risk in simple statistics and achieve accurate quantification and amplified recognition of the potential blind area problem caused by the overlap of lag intervals.

[0077] The larger the blind area residence reference value generated after analyzing the proportion of the residence time of the variable in the "unresponsive" state within the intersection range of the lag intervals under the detection window, the larger the control blind area in the system, and the variable remains within the error range jointly ignored by the controllers for a long time, significantly reflecting the overlapping of the lag intervals of multiple controllers. This indicates that there is a lack of coordination among the controllers when setting the lag intervals, resulting in some errors falling into the area where all controllers consider "no response required". On the contrary, if the blind area residence reference value remains at a low level for a long time, it means that most errors within the monitoring window can be effectively recognized and adjusted by at least one controller, and the control system does not experience a long-term "collective silence". It can be considered that the lag intervals of each controller are set reasonably and there is no significant overlap.

[0078] When the abnormal residence time of the process variable within any unstable value segment is significantly prolonged, it can indicate the overlap of the lag intervals of multiple controllers. The reason is that under normal control logic, once the process variable deviates from the set value, at least one controller should be quickly triggered to adjust and restore the system to the target range. However, when multiple controllers are respectively provided with lag intervals and there is an intersection among these lag intervals, the system may have a "misjudged as normal" blind area - that is, the error value falls within the overlapping range of the lag intervals of all controllers, resulting in no response from all controllers, thus forming a control vacuum area without intervention. At this time, even if the process variable has deviated from the stable interval, the system will not generate effective control actions, causing the variable to stay in the non-target value segment for a long time, forming an abnormal residence phenomenon. Therefore, the prolongation of the abnormal residence time is not only a direct manifestation of the failure of the controller response, but also a strong correlation performance of the "collective neglect" of errors caused by the overlap of the lag intervals, and can be used as an important dynamic feature for identifying the problem of multi-controller lag overlap.

[0079] The specific steps for generating the unexpected residence reference value by analyzing the abnormal residence time of the process variable within any unstable value segment under the detection window are as follows:

[0080] Perform point-by-point analysis on the process variable. Based on the preset stable target range, judge whether the variable is in the unexpected interval, and define the residence marking function. The formula is as follows:

[0081]

[0082] , where V(q) is the process variable data collected at the qth time point, [L, H] is the expected stable interval of the process variable, L is the upper limit of the expected stable interval, H is the lower limit of the expected stable interval, ε is the micro-variation threshold used to identify the "static residence" state and prevent slight fluctuations of the variable from being misjudged as adjustment behavior, and S(q) is the output value of the residence marking function. S(q)=1 indicates that the qth sampling point meets the "abnormal residence" criterion; S(q)=0 indicates that this point does not constitute an unexpected residence point;

[0083] The function of the above function is as follows: Only when the variable deviates from the set range and the change amplitude is very small, it is considered to be in the "abnormal stay" state, excluding the normal transition section during the adjustment process. By traversing the window to generate the binary sequence S, all "unexpected stay points" can be efficiently identified.

[0084] After obtaining the stay mark sequence, a non-linear enhancement function is introduced to calculate the unexpected stay reference value. The non-linear enhancement function not only considers the number of stay points, but also weights the continuity, identifies the control vacuum characteristics of "long-term unregulated", and the calculation formula is as follows:

[0085]

[0086] , where UD is the unexpected stay reference value, m is the total number of sampling points in the detection window, and D(q) is the stay history weighting factor, which is used to enhance the influence degree of continuous stay. The formula is as follows:

[0087]

[0088] , where represents how many points among the h sampling points before the current point q belong to the "unexpected stay" state. S(q - j) is the unexpected stay mark at the q - jth time point, indicating the stay state of the jth sampling point in the past.

[0089] The larger the finally obtained unexpected stay reference value UD, the longer the process variable stays in the state of "should be adjusted but not adjusted", indicating that there is a serious problem of lack of control response in the system, which is highly likely to be caused by the overlap of the lag intervals of multiple controllers.

[0090] The larger the unexpected stay reference value generated by analyzing the abnormal stay time of the process variable in any unstable value segment under the detection window, the more it indicates that the system responds untimely or not at all when facing deviations. In a system with multiple controllers running in parallel, if there is an overlap in the lag intervals, when the error value of the process variable falls into the common non-responsive area of these controllers, the system cannot trigger any adjustment behavior, resulting in the variable staying in the unstable section for a long time. Therefore, the continuous increase of the unexpected stay reference value often means that the controller fails to intervene in time, strongly indicating the problem of overlap of the lag intervals. On the contrary, if this reference value remains low, it means that even if the variable deviates, at least one controller can adjust in time, and the process variable quickly returns to the set value, indicating that the lag intervals of each controller are reasonably distributed and there is no overlap, and the system is in a good control response state.

[0091] Use the key dimension indicators that have been analyzed and processed as input data and input them into a machine learning model that has been pre-trained. Identify and determine whether there is an overlap in the lag intervals between each controller through the machine learning model.

[0092] Use the blind spot residence reference value and the unanticipated stay reference value that have been analyzed and processed as input data and input them into a machine learning model that has been pre-trained. Generate a lag interval overlap risk coefficient through the machine learning model, and intelligently identify and determine whether there is an overlap in the lag intervals between each controller through the lag interval overlap risk coefficient.

[0093] The "pre-trained machine learning model" refers to an intelligent recognition algorithm model that has completed modeling and parameter optimization through an offline training process based on a large amount of historical operation data, working condition samples, and labeled controller lag interval overlap status data before actual application. In the training process of this model, supervised or unsupervised learning methods are used to learn and fit the input feature variables (such as blind spot residence reference value, unanticipated stay reference value, controller response time, error trend, etc.), so as to master the internal relationship and behavior pattern between variables. For example, through learning a large number of samples, the model can identify situations that are very likely to represent the overlap of controller lag intervals under a certain combination of features, and establish a mathematical mapping relationship accordingly. After the training is completed, the model is "frozen", that is, its parameters are no longer adjusted, and it can be directly deployed in the actual control system to quickly infer and judge whether the current state belongs to the high-risk area of lag interval overlap. Such models can adopt structures such as decision trees, random forests, support vector machines (SVMs), or neural networks, and are selected and optimized according to system complexity and data characteristics.

[0094] In addition, the pre-trained machine learning model has good generalization ability and real-time advantages. During the online monitoring process, the model does not need to be re-trained or perform large-scale calculations. It only needs to receive the current working condition data (such as blind spot residence reference value and unanticipated stay reference value), and can quickly generate a lag interval overlap risk coefficient, which is used as an important basis for intelligent judgment. This risk coefficient is a quantitative indicator, usually expressed as a value between 0 and 1, and is used to measure the likelihood of lag zone overlap between multiple controllers in the current control system. The closer the value is to 1, the higher the lag overlap risk, and the more the system needs to adjust the controller lag interval or issue a warning. Compared with the traditional judgment method based on fixed thresholds or hard-coded rules, the pre-trained machine learning model has stronger adaptability, robustness, and self-learning ability. It can extract hidden behavior rules from complex, multi-dimensional, and non-linear operation data, and achieve accurate, efficient, and real-time identification of lag overlap phenomena, providing solid data support and decision-making basis for the intelligent and safe operation of complex equipment such as compressors.

[0095] The machine learning model is not limited herein, and any machine learning model that can comprehensively analyze the blind zone residence reference value BZD and the unexpected stay reference value UD to generate the lag interval overlap risk coefficient DOR can be used. To implement the technical solution of the present invention, the present invention provides a specific implementation manner;

[0096] The formula for generating the lag interval overlap risk coefficient is as follows: DOR = k 1 ·BZD + k 2 ·UD, where k 1 and k 2 are respectively the preset proportionality coefficients of the blind zone residence reference value BZD and the unexpected stay reference value UD, and k 1 and k 2 are both greater than 0.

[0097] The preset proportionality coefficient refers to the weight parameter set artificially in the calculation process of generating the lag interval overlap risk coefficient DOR, which is used to adjust the proportion of the blind zone residence reference value BZD and the unexpected stay reference value UD in the overall risk calculation. Specifically, the coefficients k 1 and k 2 respectively represent the importance or influence of the blind zone residence reference value BZD and the unexpected stay reference value UD in the final risk result DOR, and are preset values based on the actual application scenario, historical data analysis experience or system security sensitivity. These two coefficients are usually set by means of model tuning, expert experience, sensitivity analysis, etc., and have adjustability and scalability. Their functions are as follows: if the system has a low tolerance for "no response in the blind zone", a larger k 1 value can be set, so that DOR is more sensitive to BZD; on the contrary, if the system pays more attention to the long-term abnormal drift of variables, the weight of k 2 can be increased. In short, the preset proportionality coefficient is a regulating factor used to quantify the contribution degree of different indicators to the comprehensive risk assessment. The larger its value, the more significant the impact on the final risk result.

[0098] From the lag interval overlap risk coefficient, it can be seen that the larger the blind zone residence reference value generated by analyzing the proportion of the "no response" residence time of the variable within the intersection range of the lag interval under the detection window, and the larger the unexpected stay reference value generated by analyzing the abnormal stay time of the process variable within any unstable value segment under the detection window, the larger the lag interval overlap risk coefficient generated by the machine learning model that has been pre-trained to identify and judge whether there is a lag interval overlap between each controller, indicating that the probability of there being a lag interval overlap between each controller is greater. On the contrary, it indicates that the probability of there being a lag interval overlap between each controller is smaller.

[0099] Compare and analyze the lag interval overlap risk coefficient generated when a machine learning model completed through pre-training identifies and determines whether there is an overlap in the lag intervals between each controller with a pre-set reference threshold for the lag interval overlap risk coefficient to determine whether there is an overlap in the lag intervals between each controller. The judgment steps are as follows:

[0100] If the lag interval overlap risk coefficient is greater than the pre-set reference threshold for the lag interval overlap risk coefficient, it is determined that there is an overlap in the lag intervals between each controller; if the lag interval overlap risk coefficient is less than or equal to the pre-set reference threshold for the lag interval overlap risk coefficient, it is determined that there is no overlap in the lag intervals between each controller.

[0101] When the machine learning model identifies an overlap relationship in the lag intervals between each controller, evaluate the priority of all control loops according to the control importance level of each control loop and the current error state to determine the target control loop that needs to be intervened first; after completing the determination of the target control loop, trigger the real-time working condition evaluation algorithm, and dynamically adjust the lag interval range of each controller according to the current load level, disturbance frequency, and the priority relationship between the controllers;

[0102] The above steps play a key role in response decision-making and dynamic optimization in the entire lag interval overlap identification and adaptive regulation process. Its core purpose is to quickly and accurately determine which control loop should be intervened first after identifying an overlap relationship in the lag intervals between the controllers, and through the dynamic adjustment mechanism, eliminate the blind area conflict between the controllers and restore the regulation ability and stability of the system. Specifically, by evaluating the control importance level of each control loop, the priority can be divided based on the influence degree of the control variable on the system safety, operation efficiency, or product quality. For example, pressure control is usually higher than temperature control, and the main control loop is superior to the auxiliary loop, etc.; combined with the current error state, it can be identified which control loop's deviation has approached the limit tolerance value and needs to be immediately responded to prevent failures or performance degradation.

[0103] On this basis, the system triggers the real-time working condition evaluation algorithm, and further dynamically fine-tune the lag intervals of each controller according to the current system operating state (such as load level, disturbance intensity, process fluctuation frequency, etc.) and the cooperation relationship between the controllers, ensuring that the controller with a higher priority can intervene in a more sensitive error range in a timely manner, while the controller with a lower priority appropriately expands the lag interval to avoid control interference, thereby realizing the intelligent division of labor and dynamic coordination between multiple controllers. This mechanism significantly improves the adaptive ability and robustness of the control system in a complex and multi-disturbance environment, avoiding control vacuum, response delay, or frequent misregulation caused by the intersection of lag intervals, and is the core step to realize the closed-loop control logic of "identification - decision - response - optimization".

[0104] When the machine learning model identifies an overlapping relationship of hysteresis intervals among controllers, a priority evaluation is performed on all control loops to determine the target control loop that needs to be intervened first; after determining the target control loop, the real-time working condition evaluation algorithm is triggered, and the specific steps for dynamically adjusting the hysteresis interval range of each controller are as follows:

[0105] When the machine learning model identifies an overlap of hysteresis intervals among controllers, first, a comprehensive priority evaluation is performed on all involved control loops, and the formula is as follows:

[0106]

[0107] , where PRI(a) is the priority evaluation index of the a-th controller. The higher this value, the more priority is required for adjustment and intervention. W imp (a) is the weight coefficient of the control importance level of the a-th controller, which is usually obtained from expert experience or set during the initial stage of system design. The higher the value, the higher the importance of this loop to the overall working condition stability. E(a) is the error state collected in real time by the a-th controller, representing the amplitude of the current error of the control loop (the difference between the actual value and the target value). The larger the absolute value, the more serious the deviation of the error from the target. DOR is the hysteresis interval overlap risk coefficient, and DOR ref is the reference threshold of the hysteresis interval overlap risk coefficient;

[0108] The above steps combine the control importance and the current error state, and at the same time, with the help of the standardized hysteresis interval overlap risk coefficient, comprehensively and accurately judge the intervention order of all control loops to determine a target control loop that most urgently needs to be adjusted and intervened first, avoiding disordered intervention order or insufficient intervention intensity, and ensuring the maximization of the system working condition recovery efficiency.

[0109] After completing the priority evaluation of the target control loop, immediately start the real-time working condition evaluation algorithm to dynamically adjust the hysteresis interval of each controller. The specific formula is:

[0110]

[0111] , where ADJ(a) is the adjusted hysteresis interval range of the a-th controller, and DB ini (a) is the initial hysteresis interval range of the a-th controller, and f scan (a) is the current real-time scanning and monitoring frequency of the a-th control loop, which is used to represent the speed of real-time data acquisition and processing of the control loop. The larger f scan (a), the stronger the sensitivity of the loop to the change of system error. f ini is the initial flow rate scanning frequency, representing the reference scanning frequency of monitoring data and control response during the initial design stage, and PRI avgIt is the average value of the real-time priority indices of all controllers and is used to reflect the relative priority level of the current controller.

[0112] In the formula, if PRI(a) is greater than the average PRI avg , then the whole within the parentheses is greater than 1, making ADJ(a) larger than the original DB ini (α), enhancing the sensitivity of the controller to errors; if PRI(a) is less than the average value, then the whole is less than 1, slightly increasing the hysteresis interval and reducing the control sensitivity.

[0113] The function of the above steps is to dynamically adapt to the real-time operating state of the system. By adjusting the hysteresis interval range of each controller in real time, it gives priority to solving the control blind area caused by the overlap of the hysteresis intervals in the high-risk area, ensuring that the control system can stably and quickly return to the normal operating condition, reflecting the control idea of real-time optimization and intelligent adjustment.

[0114] The present invention realizes the efficient identification and precise intervention of the problem of overlapping hysteresis intervals of multiple controllers during the operation of the gas compressor by introducing a mechanism for identifying and dynamically adjusting the hysteresis intervals of the controller based on machine learning and multi-dimensional feature analysis. By collecting the operating parameters of multiple controllers in real time, constructing a time series data set and extracting key dimension indicators, it can effectively identify the "control blind area" generated due to the intersection of hysteresis intervals under specific operating conditions; combined with the high-precision judgment ability of the machine learning model, it further improves the system's ability to predict potential control conflicts; after identifying the overlapping relationship, according to the control importance level and the current error state of the system, the hysteresis interval is dynamically optimized and adjusted, effectively avoiding the phenomenon of multiple controllers being collectively silent when the key variables deviate, thus significantly improving the responsiveness, stability and overall operating efficiency of the control system.

[0115] The present invention provides a hybrid control system for a gas compressor as Figure 2 shown, including a multi-dimensional data acquisition module, a data preprocessing and sequence construction module, a feature extraction and overlap relationship identification module, a machine learning discrimination module, and a priority determination and hysteresis zone dynamic adjustment module:

[0116] The multi-dimensional data acquisition module, during the operation of the gas compressor, collects in real time the multi-dimensional operating parameter data generated by each controller inside the compressor during the operation process;

[0117] The data preprocessing and sequence construction module preprocesses the collected original multi-dimensional operating parameter data and organizes the processed valid data into a data set according to the time series structure for subsequent feature extraction and model analysis;

[0118] The feature extraction and overlap relationship recognition module extracts key dimensional indicators reflecting the overlap of the lag intervals of each controller from the data set through feature engineering and statistical analysis techniques, and comprehensively analyzes the extracted key dimensional indicators to identify potential overlap relationships between the lag intervals of the controllers;

[0119] The machine learning discrimination module takes the key dimensional indicators after analysis and processing as input data and inputs them into a pre-trained machine learning model to identify and determine whether there is an overlap in the lag intervals between each controller through the machine learning model;

[0120] The priority determination and lag zone dynamic adjustment module, when the machine learning model identifies an overlap relationship in the lag intervals between each controller, evaluates the priorities of all control loops according to the control importance levels of each control loop and the current error state to determine the target control loop that needs to be intervened first; after completing the determination of the target control loop, it triggers a real-time working condition evaluation algorithm to dynamically adjust the lag interval ranges of each controller according to the current load level, disturbance frequency, and the priority relationship between the controllers.

[0121] The hybrid control method of the gas compressor provided by the embodiment of the present invention is implemented through the above-mentioned hybrid control system of the gas compressor. The specific methods and processes of the hybrid control system of the gas compressor are detailed in the embodiments of the above-mentioned hybrid control method of the gas compressor, and will not be elaborated here.

[0122] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0123] Only some exemplary embodiments of the present invention have been described by way of illustration above. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

[0124] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0125] It should be understood that in various embodiments of the present application, the magnitudes of the serial numbers of the above processes do not imply the order of execution, and the order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0126] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present application.

[0127] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0128] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] In addition, in each embodiment of the present application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0130] As described above, this is only the specific implementation manner of the present application. However, the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

[0131] Only some exemplary embodiments of the present invention have been described above by way of illustration. Undoubtedly, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

Claims

1. A mixing control method for a gas compressor, characterized in that: The following steps are involved: During the operation of the gas compressor, multi-dimensional operating parameter data generated by each controller inside the compressor during operation is collected in real time; Preprocess the collected original multi-dimensional operating parameter data, and organize the processed valid data into a data set according to the time series structure for subsequent feature extraction and model analysis; Through feature engineering and statistical analysis techniques, key dimension indicators reflecting the overlap of controller hysteresis intervals are extracted from the data set, and the extracted key dimension indicators are comprehensively analyzed to identify the potential overlapping relationship between controller hysteresis intervals; The analyzed key dimension indicators are used as input data and input into the pre-trained machine learning model. The machine learning model is used to identify and determine whether there is overlap in the hysteresis intervals between the controllers. When the machine learning model identifies that there is an overlapping relationship between the hysteresis intervals of each controller, it performs a priority evaluation on all control loops according to the control importance level of each control loop and the current error state to determine the target control loop that requires priority intervention; after the target control loop is determined, the real-time operating condition assessment algorithm is triggered to dynamically adjust the hysteresis interval range of each controller according to the current load level, disturbance frequency and the priority relationship between controllers.

2. The mixing control method of a gas compressor according to claim 1, characterized in that: The specific steps of preprocessing the collected original multi-dimensional operating parameter data include: First, the collected data is denoised using a filtering algorithm to eliminate random noise caused by sensor fluctuations or instantaneous interference; Secondly, perform outlier detection and removal to identify and remove data mutation points caused by sensor failures, communication interruptions, and physical anomalies; Subsequently, the data is aligned and synchronized to ensure that the parameters collected by different controllers are unified under the same time base, and data completion and interpolation are completed to repair data gaps caused by instantaneous packet loss and signal loss; Then normalize all the data to unify the dimensions and eliminate the interference of dimension differences on model analysis; The processed valid data is sorted by timestamp and organized into a data set with a time series structure according to the sampling period, providing a stable, continuous and high-quality data foundation for subsequent feature extraction and machine learning model analysis.

3. The mixing control method of a gas compressor according to claim 1, characterized in that: Through feature engineering and statistical analysis techniques, key dimensional indicators reflecting the overlap of hysteresis intervals of various controllers are extracted from the data set. The extracted indicators include the proportion of "no response" residence time of variables in the intersection range of hysteresis intervals and the abnormal residence time of process variables in any unstable value segment. The proportion of "no response" residence time of the extracted variables in the intersection range of hysteresis intervals and the abnormal residence time of process variables in any unstable value segment are analyzed under the detection window to generate blind spot retention reference values ​​and unexpected retention reference values ​​respectively. The potential overlapping relationship between controller hysteresis intervals is identified through the blind spot retention reference values ​​and the unexpected retention reference values.

4. The mixing control method of a gas compressor according to claim 3, characterized in that: The specific steps for analyzing the proportion of "no response" residence time of the variable within the intersection range of the hysteresis interval under the detection window to generate the blind zone retention reference value are as follows: First, it is necessary to identify whether the error value of the key control variable falls into the intersection area of ​​multiple controller hysteresis intervals during operation, and construct a dwell mapping function for quantification. The definition formula is as follows: M i =δ(E i ∈D)·ω i , Where M i is the blind area mark value of the i-th sampling point, E i is the target setting value of the i-th sampling point, D is the intersection area of ​​the hysteresis interval, δ(E i ∈D) is a logical decision function used to detect the error E i Whether it is in the intersection area D, ω i is the error depth weight factor; After obtaining the blind spot state sequence, the response confidence weight function is introduced to evaluate the continuity of the blind spot stay and the severity of the control gap, and generate the blind spot stay reference value. The generation formula is as follows: ， Where BZD is the blind zone retention reference value, φ(S i ) is the response confidence weight function, which is used to assign higher risk weights to "continuous no response" behaviors for a longer period of time. i It is a measure of the length of time that the variable is continuously in the blind zone from the i-th sampling point, κ i is the controller response missing factor.

5. The mixing control method of a gas compressor according to claim 3, characterized in that: The specific steps for analyzing the abnormal residence time of the process variable in any unstable value segment under the detection window to generate the unexpected residence reference value are as follows: Analyze the process variables point by point, determine whether the variables are in an unexpected range based on the preset stable target range, and define the stay mark function. The formula is as follows: ， Where V(w) is the process variable data collected at the qth time point, [L, H] is the expected stable interval of the process variable, L is the upper limit of the expected stable interval, H is the lower limit of the expected stable interval, ε is the slight change threshold used to identify the "stationary stay" state, and S(q) is the output value of the stay mark function; After obtaining the stay mark sequence, a nonlinear enhancement function is introduced to calculate the unexpected stay reference value. The calculation expression is as follows: Where UD is the reference value of unexpected stay, m is the total number of sampling points in the detection window, and D(q) is the stay history weighting factor, which is used to enhance the influence of continuous stay. The formula is as follows: In the formula, It indicates how many points belong to the "unexpected stay" state among the h sampling points before the current point q, and S(qj) is the unexpected stay mark at the qjth time point, indicating the stay state of the jth sampling point in the past.

6. The mixing control method of a gas compressor according to claim 3, characterized in that: The analyzed blind spot retention reference value and unexpected retention reference value are used as input data and input into the pre-trained machine learning model. The hysteresis interval overlap risk coefficient is generated by the machine learning model. The hysteresis interval overlap risk coefficient is used to intelligently identify and judge whether there is hysteresis interval overlap between controllers.

7. The mixing control method of a gas compressor according to claim 6, characterized in that: The hysteresis interval overlap risk coefficient generated when the pre-trained machine learning model is used to identify and judge whether there is hysteresis interval overlap between the controllers is compared with the pre-set hysteresis interval overlap risk coefficient reference threshold to judge whether there is hysteresis interval overlap between the controllers. The judgment steps are as follows: If the hysteresis interval overlap risk coefficient is greater than the preset hysteresis interval overlap risk coefficient reference threshold, it is judged that there is hysteresis interval overlap between the controllers; if the hysteresis interval overlap risk coefficient is less than or equal to the preset hysteresis interval overlap risk coefficient reference threshold, it is judged that there is no hysteresis interval overlap between the controllers.

8. The mixing control method of a gas compressor according to claim 7, characterized in that: When the machine learning model identifies that there is an overlapping relationship between the hysteresis intervals of various controllers, priority evaluation is performed on all control loops to determine the target control loop that needs priority intervention; after the target control loop is determined, the real-time operating condition evaluation algorithm is triggered to dynamically adjust the hysteresis interval range of each controller. The specific steps are as follows: When the machine learning model identifies that there is overlap in the hysteresis intervals between controllers, it first performs a comprehensive priority evaluation of all involved control loops, as shown in the following formula: ， Where PRI(a) is the priority evaluation index of the a-th controller, W imp (a) is the control importance level weight coefficient of the a-th controller, E(a) is the error state collected in real time by the a-th controller, DOR is the hysteresis interval overlap risk coefficient, and DOR ref is the reference threshold of the overlap risk coefficient of the lag interval; After completing the priority evaluation of the target control loop, the real-time operating condition evaluation algorithm is immediately started to dynamically adjust the hysteresis interval of each controller. The specific formula is: ， Where ADJ(a) is the hysteresis range of the ath controller after adjustment, DB ini (a) is the initial hysteresis range of the a-th controller, f scan (a) is the current real-time scanning monitoring frequency of the a-th control loop, f ini is the initial flow scan frequency, PRI avg It is the average of the real-time priority indices of all controllers.

9. A hybrid control system for a gas compressor, used to implement the hybrid control method for a gas compressor as described in any one of claims 1 to 8, characterized in that: It includes multi-dimensional data acquisition module, data preprocessing and sequence construction module, feature extraction and overlapping relationship recognition module, machine learning discrimination module, and priority determination and hysteresis zone dynamic adjustment module: The multi-dimensional data acquisition module collects multi-dimensional operating parameter data generated by each controller inside the compressor during operation in real time during the operation of the gas compressor; The data preprocessing and sequence construction module preprocesses the collected original multi-dimensional operating parameter data and organizes the processed valid data into a data set according to the time series structure for subsequent feature extraction and model analysis; The feature extraction and overlapping relationship identification module extracts key dimension indicators reflecting the overlap of the hysteresis intervals of each controller from the data set through feature engineering and statistical analysis technology, and conducts a comprehensive analysis of the extracted key dimension indicators to identify the potential overlapping relationship between the hysteresis intervals of the controllers; The machine learning discrimination module uses the analyzed key dimension indicators as input data into the pre-trained machine learning model, and uses the machine learning model to identify and determine whether there is overlap in the hysteresis intervals between the controllers; Priority determination and hysteresis zone dynamic adjustment module. When the machine learning model identifies that there is an overlapping relationship between the hysteresis intervals of each controller, it performs a priority evaluation on all control loops according to the control importance level of each control loop and the current error state to determine the target control loop that requires priority intervention; after completing the target control loop determination, the real-time operating condition evaluation algorithm is triggered to dynamically adjust the hysteresis interval range of each controller according to the current load level, disturbance frequency and the priority relationship between controllers.

Images