A load-adaptive based intelligent step compressor control method and system

Abstract

The application provides a load-adaptive-based intelligent staged-compressor control method and system, and belongs to the technical field of compressor control. The method comprises the following steps: obtaining multi-dimensional data of a compressor unit through a distributed sensor network; pre-processing the multi-dimensional data and performing feature extraction to obtain a feature vector; inputting the feature vector into a load prediction model to obtain a predicted load state; adjusting a decision threshold of a staged decision tree based on the predicted load state to obtain a target staged decision tree; inputting the feature vector into the target staged decision tree to generate an execution scheme; obtaining energy efficiency feedback data of the execution scheme, and inputting the predicted load state and the energy efficiency feedback data into a reinforcement learning model to obtain a control strategy with the Pareto optimality as the target; and adjusting the control action of the compressor unit based on the control strategy. The application improves the operation energy efficiency of the compressor system and reduces the mechanical loss and energy consumption of the equipment.

Description

Technical Field

[0001] This application relates to the technical field of compressor control, and in particular to an intelligent hierarchical compressor control method and system based on load adaptation. Background Technology

[0002] Currently, in the field of compressor control technology, while variable frequency speed control can optimize the efficiency of a single machine, its hardware costs are high, and it cannot solve the problem of multi-machine coordination. Meanwhile, preset threshold hierarchical control, due to its rigid logic, cannot adapt to irregular load fluctuations, leading to frequent start-ups and shutdowns or inefficient operation. These problems not only increase energy consumption and maintenance costs but also shorten equipment lifespan, hindering the stability and economy of industrial production. There are still many shortcomings in the field of compressor control technology regarding the aforementioned related technologies.

[0003] Therefore, there is an urgent need for an intelligent hierarchical compressor control method and system based on load adaptation to ensure the efficient, safe and stable operation of the compressor under various working conditions. Summary of the Invention

[0004] To address the aforementioned technical problems, this application presents a method and system for intelligent hierarchical compressor control based on load adaptation.

[0005] A first aspect of this application provides a load-adaptive intelligent hierarchical compressor control method, comprising: Multidimensional data of the compressor unit is acquired through a distributed sensor network. The multidimensional data includes process parameters, equipment operating status, and energy consumption data. The multidimensional data is preprocessed and features are extracted to obtain feature vectors; The feature vector is input into the load prediction model to obtain the predicted load status; The decision threshold of the hierarchical decision tree is adjusted based on the predicted load state to obtain the target hierarchical decision tree; The multidimensional data is input into the target hierarchical decision tree to generate an execution plan; Obtain energy efficiency feedback data for executing the execution scheme, input the predicted load state and the energy efficiency feedback data into the reinforcement learning model, and obtain the control strategy with Pareto optimality as the objective; The control actions of the compressor unit are adjusted based on the control strategy.

[0006] A second aspect of this application provides a load-adaptive intelligent staged compressor control system, comprising: The data acquisition module is used to acquire multi-dimensional data of the compressor unit through a distributed sensor network. The multi-dimensional data includes process parameters, equipment operating status and energy consumption data. The data processing module is used to preprocess the multidimensional data and extract features to obtain feature vectors; The load prediction module is used to input the feature vector into the load prediction model to obtain the predicted load status. The scheme generation module is used to adjust the decision threshold of the hierarchical decision tree based on the predicted load status to obtain the target hierarchical decision tree; and to input the multidimensional data into the target hierarchical decision tree to generate an execution scheme. The control strategy module is used to acquire energy efficiency feedback data for executing the execution scheme, and input the predicted load state and the energy efficiency feedback data into the reinforcement learning model to obtain the control strategy with Pareto optimality as the objective. The strategy execution module is used to adjust the control actions of the compressor unit based on the control strategy.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described intelligent hierarchical compressor control method based on load adaptation.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described intelligent hierarchical compressor control method based on load adaptation.

[0009] The beneficial effects of the intelligent hierarchical compressor control method and system based on load adaptation provided in this application are as follows: This application acquires multi-dimensional data through a distributed sensor network, enabling a comprehensive understanding of the compressor unit's operating status. Based on multi-dimensional data and a load prediction model, it can accurately predict the load state, allowing the control to adapt to load changes in advance and reducing energy waste. Furthermore, by adjusting the threshold of the hierarchical decision tree based on the predicted load state, the generated execution scheme is more in line with actual needs, improving control flexibility and targeting. Simultaneously, based on energy efficiency feedback data and a reinforcement learning model, the control strategy obtained with Pareto optimality as the objective can maximize energy efficiency while ensuring compression performance, significantly reducing energy consumption costs. In addition, adjusting control actions can reduce frequent start-ups and overload operation of the unit, extend equipment lifespan, reduce maintenance costs, and ensure long-term stable and efficient operation of the compressor unit. Attached Figure Description

[0010] Figure 1 A schematic flowchart of a load-adaptive intelligent hierarchical compressor control method provided in an embodiment of this application; Figure 2 A structural block diagram of a load-adaptive intelligent hierarchical compressor control system provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a load-adaptive intelligent hierarchical compressor control method according to an embodiment of this application. The method includes: S101: Acquire multi-dimensional data of the compressor unit through a distributed sensor network. The multi-dimensional data includes process parameters, equipment operating status, and energy consumption data.

[0014] In this embodiment, the distributed sensor network adopts a redundant deployment method, setting up primary and backup sensors at key monitoring points to ensure the continuity and accuracy of data acquisition. The sampling frequency is dynamically adjusted according to the operating conditions of the compressor unit. The distributed sensor network includes: pressure sensors, temperature sensors, a three-phase power monitoring module, and a mechanical condition monitoring module; process parameters include: inlet pressure, exhaust pressure, inlet temperature, exhaust temperature, gas flow rate, and gas composition; equipment operating status includes: motor speed, vibration frequency, bearing temperature, valve opening, and cooling system flow rate; energy consumption data includes real-time power, cumulative power consumption, and energy consumption per unit time.

[0015] For example, a pressure sensor is used to accurately capture dynamic changes in the pressure field, with a range of 0-10 MPa and an accuracy of ±0.05%FS; a temperature sensor is used to meet the requirement of rapid temperature sensing over a wide temperature range, with a temperature sensing range of -50℃ to 600℃ and a response time of less than 0.3s; an electromagnetic flowmeter is used to achieve high-precision measurement of fluid flow across the entire range, with a range ratio of 1:100 and repeatability of ±0.1%. A three-phase power monitoring module is used to track power consumption and energy efficiency status in real time. A mechanical condition monitoring module is used to deploy vibration and acceleration sensors on key mechanical components such as compressor crankshafts and bearings to capture abnormal vibration characteristics of the equipment in real time, providing data support for early warning of mechanical failures.

[0016] S102: Preprocess the multidimensional data and extract features to obtain feature vectors.

[0017] In this embodiment, the preprocessing method for multidimensional data includes data cleaning, outlier handling, data normalization, and denoising. Outlier handling can use the 3σ criterion to identify and remove outliers from the multidimensional data. For missing data in the multidimensional data, linear interpolation or LSTM neural network prediction can be used to fill in the missing data based on its time series characteristics. Denoising can be based on the Kalman filter algorithm to suppress dynamic noise. The Kalman filter algorithm can effectively filter out random interference in signals such as pressure and flow rate based on the time correlation of states and the statistical characteristics of observation noise, improving the data signal-to-noise ratio and providing a reliable data foundation for load assessment. This embodiment uses state-space equations to optimally estimate noisy signals such as pressure and flow rate, filtering out random interference to improve the data signal-to-noise ratio. The formula for the state-space equation is:

[0018] in, This represents the current state value. The Kalman gain is updated in real time, representing the weighting of observed and predicted values. These are real-time observations from the sensors; This is the predicted state value from the previous moment.

[0019] Secondly, multi-dimensional feature extraction is performed on the preprocessed data. The vibration signal is decomposed into 12 time-frequency features such as energy, entropy, and peak factor. The current signal is transformed by Park to obtain the DC flow feature in the synchronous rotating coordinate system. The pressure signal is combined with the ARIMA model to extract trend offset and periodic anomaly features. At the same time, the parameters such as pressure, flow, temperature, and current are normalized to obtain the load coefficient. Finally, the data are fused to form a feature vector that includes process parameters, equipment operating status, and energy consumption characteristics.

[0020] Among them, data normalization is achieved by weighted fusion of multiple physical quantities to output the load factor (L), and the formula is as follows:

[0021] Where P represents pressure, Q represents flow rate, T represents temperature, and I represents current, corresponding to system load intensity, energy transfer efficiency, thermal state, and motor load characteristics, respectively. The weighting method is pressure-dominant, flow rate-secondary, temperature-assisted, and current-corrected; for example, α1=0.4, α2=0.3, α3=0.2, α4=0.1, to reflect the differentiated impact of key process parameters on load conditions.

[0022] S103: Input the feature vector into the load prediction model to obtain the predicted load status; In this embodiment, the load prediction model is built on a long short-term memory network. It takes high-frequency sampled historical data and real-time feature vectors as input and trains it through a network structure with three hidden layers, an Adam optimizer, and a mean squared error loss function. Through multi-dimensional feature encoding and temporal dependency modeling, it can predict the load status in future periods. The load status includes the load rate and the load change trend, which includes: rising, falling, and stable.

[0023] In this embodiment, the prediction model also includes an update mechanism. Specifically, an online learning-driven model calibration system is designed, which continuously injects real-time data streams and performs incremental training to construct a closed loop of data acquisition, feature update, and model iteration. The network weights and prediction parameters of the prediction model are adjusted based on the latest production data to adapt to complex time-varying scenarios such as production process switching and equipment load fluctuations, ensuring that the prediction model maintains high fitting accuracy over the long term.

[0024] S104: Adjust the decision threshold of the hierarchical decision tree based on the predicted load status to obtain the target hierarchical decision tree; input the feature vector into the target hierarchical decision tree to generate an execution plan; In this embodiment, the hierarchical decision tree is initially constructed based on the ID3 algorithm, using energy efficiency, equipment operation stability, and response speed as decision indicators. Decision thresholds corresponding to light load, medium load, and heavy load are preset, and each level of decision threshold is determined based on the percentile of historical operating data. For example, the decision thresholds of the hierarchical decision tree can be set as: a first threshold, a second threshold, and a third threshold, corresponding to light load, medium load, and heavy load states, respectively. For example, the first threshold is set to 0.35, the second threshold to 0.65, and the third threshold to 0.85.

[0025] In this embodiment, the decision threshold is adjusted based on the predicted load change trend to obtain a target hierarchical decision tree. The specific adjustment method includes: when the load change trend increases, the decision threshold is increased by 0.05; when the load change trend decreases, the decision threshold is decreased by 0.05; and when the load change trend is stable, the decision threshold remains unchanged.

[0026] In this embodiment, the execution scheme includes: generating one variable frequency compressor when the load coefficient is less than the adjusted first threshold; generating one fixed-speed compressor plus one variable frequency compressor when it is between the first and second thresholds; generating two fixed-speed compressors plus one variable frequency compressor when it is between the second and third thresholds; and generating three fixed-speed compressors plus one variable frequency compressor when it is greater than the third threshold. S105: Obtain energy efficiency feedback data for the execution plan, input the predicted load state and energy efficiency feedback data into the reinforcement learning model, and obtain the control strategy with Pareto optimality as the goal of reinforcement learning; In this embodiment, the energy efficiency feedback data for executing the plan includes unit gas production energy consumption, compressor operating efficiency, equipment temperature rise rate, and fault warning index.

[0027] An agent based on a deep Q-network reinforcement learning model takes predicted load state and energy efficiency feedback data as state inputs and compressor speed, start / stop status, and valve opening as action outputs. A policy iteration is triggered every 5 minutes. A multi-objective optimization problem is decomposed into single-objective sub-problems using an objective decomposition algorithm. Candidate control strategies are obtained through interaction sampling between the single-objective sub-problems and the environment. Pareto non-dominated strategies are then obtained through non-dominated ranking. If the solution set exceeds a preset capacity, clustering based on feature similarity is used to select representative strategies, ultimately yielding a control strategy with Pareto optimality as the objective. The multi-objective optimization problems include minimizing energy consumption, pressure fluctuations, temperature deviations, and equipment wear.

[0028] S106: Adjust the control actions of the compressor unit based on the control strategy.

[0029] In this embodiment, the control action adjustment adopts a hierarchical execution method and uses an intelligent controller with a heterogeneous architecture of ARM Cortex-A53 multi-core processor and FPGA coprocessor to convert the control strategy obtained by reinforcement learning into specific control instructions. Precise adjustment is achieved through fuzzy PID and model predictive control algorithms. The fuzzy rule base in PID is shown in Table 1. Table 1 Fuzzy Rule Base

[0030] As shown in Table 1, the input variables load deviation (e) and deviation change rate (ec) are both divided into five linguistic values: NB, NM, ZE, PM, and PB. Among them, the load deviation uses a triangular / trapezoidal membership function to achieve discrete fuzzy mapping of continuous deviation values, and the deviation change rate uses a Gaussian membership function to finely characterize dynamic trends. The same five linguistic values ​​of the output variable ΔK correspond to single-point membership functions, and the defuzzification process is simplified by directly mapping the determined values.

[0031] Secondly, based on the frequency adjustment value in the control strategy, the variable frequency compressor motor is smoothly speed-regulated using a vector control type variable frequency drive. During the speed regulation process, the frequency change rate is ensured not to exceed 5Hz / s to avoid current surges, and the speed regulation accuracy is ±0.01%. For fixed-speed compressors, the start and stop are controlled by solid-state relays with a response time ≤10ms.

[0032] The valve opening is adjusted using a PID control algorithm to keep the deviation between the actual and target opening within ±2%. Finally, the cooling system flow rate is adjusted by regulating the water pump speed to achieve precise flow control. Simultaneously, a dynamic programming algorithm is used to pre-plan the start-up and shutdown sequence and speed adjustment gradient to ensure pressure fluctuations are controlled within ±0.03 MPa and the system's dynamic response time is ≤2 seconds.

[0033] Throughout the adjustment process, the operating parameters of the compressor unit are monitored in real time. When a parameter exceeds the limit, the protection mechanism is immediately triggered, the control action is suspended and an alarm signal is issued. The operation will continue after the fault is cleared, thereby enabling the compressor unit to operate efficiently, safely and stably under various operating conditions, and improving the overall operating economy and reliability.

[0034] As can be seen from the above, this application acquires multi-dimensional data through a distributed sensor network, enabling a comprehensive understanding of the compressor unit's operating status. Based on multi-dimensional data and a load prediction model, it can accurately predict load conditions, allowing control to adapt to load changes in advance and reducing energy waste. Furthermore, by adjusting the threshold of the hierarchical decision tree based on predicted load conditions, the generated execution scheme is more closely aligned with actual needs, improving control flexibility and targeting. Simultaneously, based on energy efficiency feedback data and a reinforcement learning model, the control strategy, aiming for Pareto optimality, can maximize energy efficiency while ensuring compression performance, significantly reducing energy consumption costs. In addition, adjusting control actions in this application can reduce frequent start-ups and overload operation of the unit, extend equipment lifespan, reduce maintenance costs, and ensure the long-term stable and efficient operation of the compressor unit.

[0035] In one embodiment of this application, the intelligent hierarchical compressor control method based on load adaptation further includes: The target feature vector is obtained by extracting features from the equipment operating status, and the deviation of each feature in the target feature vector from its preset normal operating benchmark threshold is calculated. The deviations corresponding to each feature in the target feature vector are fused at the feature level to generate a multi-source fused deviation feature vector. Based on the multi-source fusion deviation feature vector and the preset basic probability allocation function and its weight, the uncertainty reasoning and evidence combination of multi-source evidence are carried out through the evidence theory framework to obtain the comprehensive failure probability value. The corresponding early warning level is determined based on the preset risk level range in which the comprehensive failure probability value falls; If the overall failure probability value exceeds the safety threshold, a control action degradation command is triggered, generating a safe execution plan. The degradation control command takes precedence over the execution plan output by the target hierarchical decision tree, providing safety protection control for the compressor unit.

[0036] In this embodiment, the feature vector of the equipment operating status is further extracted to obtain fault-related features. For example, motor speed fluctuation rate, vibration frequency peak offset, bearing temperature gradient, valve opening response delay time, and cooling system flow fluctuation coefficient.

[0037] The preset normal operation reference thresholds are determined by collecting stable operating data of the compressor unit under rated operating conditions. For example, the reference threshold for motor speed fluctuation rate is set to ±2%, the reference threshold for vibration frequency peak offset is ±5Hz, the reference threshold for bearing temperature gradient is ≤2℃ / min, the reference threshold for valve opening response delay time is ≤0.5s, and the reference threshold for cooling system flow fluctuation coefficient is ±3%.

[0038] The formula for calculating the deviation in this embodiment is: Deviation = (Actual feature value - benchmark threshold) / benchmark threshold. If the benchmark threshold is 0, then the absolute deviation is used for calculation.

[0039] In this embodiment, the deviations corresponding to each feature in the target feature vector are fused at the feature level. A weighted average method is used to assign fusion coefficients based on the weight of each feature's impact on equipment failure. For example, a feature importance judgment matrix is ​​constructed using the analytic hierarchy process (AHP), and equipment maintenance experts and fault diagnosis engineers are invited to score the degree of failure impact of each feature. Next, the deviation of each feature is multiplied by its corresponding weight to obtain a weighted deviation. Finally, all weighted deviations are summed to generate a multi-source fused deviation feature vector. The multi-source fused deviation feature vector maintains consistency with the target feature vector, with values ​​ranging from -1 to 1. Positive values ​​indicate feature values ​​higher than a baseline threshold, while negative values ​​indicate values ​​lower than a baseline threshold.

[0040] In this embodiment, the Basic Probability Assignment Function (BPA) setting method includes: dividing the multi-source fusion deviation feature vector into multiple intervals based on its value range, and setting a corresponding BPA function for each interval, i.e., the confidence level assignment of each fault hypothesis under that interval; wherein, the fault hypothesis includes no fault, minor fault, moderate fault, and severe fault. For example, when the fusion deviation value is <10%, the BPA is assigned as: no fault 0.9, minor fault 0.1, moderate fault 0, severe fault 0; when the fusion deviation value is 10%-30%, the BPA is assigned as: no fault 0.3, minor fault 0.5, moderate fault 0.2, severe fault 0; when the fusion deviation value is 30%-50%, the BPA is assigned as: no fault 0, minor fault 0.2, moderate fault 0.6, severe fault 0.2; when the fusion deviation value is >50%, the BPA is assigned as: no fault 0, minor fault 0, moderate fault 0.3, severe fault 0.7. The BPA function needs to be calibrated based on a large number of historical failure cases (such as collecting data from 1,000+ compressor failures and statistically analyzing the proportion of actual failure types corresponding to different fusion deviation values) to ensure that the trust level allocation conforms to the actual failure patterns.

[0041] The Dempster-Shafer combination rule in evidence theory is used to combine the weighted BPA evidence. First, the degree of contradiction represented by the conflict coefficient between each piece of evidence is calculated to be less than or equal to a preset threshold. Then, the confidence levels of the multi-source evidence are fused through the combination rule to obtain the final confidence levels of four types of hypotheses: no fault, minor fault, moderate fault, and severe fault. Finally, the confidence levels of these four types of hypotheses are summed to obtain the comprehensive failure probability value, which ranges from [0,1]. The larger the value, the higher the risk of failure.

[0042] This embodiment, based on the compressor fault handling response time requirements and the degree of fault impact, divides the comprehensive fault probability value into four risk level intervals, corresponding to a four-level early warning mechanism: Level 1, Level 2, and Level 3. The fault risk level intervals can be divided into four levels: no warning when the comprehensive fault probability value is in the first interval; Level 1 warning when the comprehensive fault probability value is in the second interval; Level 2 warning when the comprehensive fault probability value is in the third interval; and Level 3 warning when the comprehensive fault probability value is in the fourth interval. Specifically, the first interval is set to 0-0.2; the second interval to 0.2-0.4; the third interval to 0.4-0.6; and the fourth interval to 0.6-1.0. Different warning levels correspond to different response measures. Level 1 warnings only display warning information on the monitoring system; Level 2 warnings activate enhanced equipment status monitoring, increasing the sampling frequency to 15Hz; and Level 3 warnings trigger audible and visual alarms and automatically push warning information to the mobile terminals of maintenance personnel, triggering a downgrade control action command.

[0043] The three-level early warning response in this embodiment includes: Level 1 early warning, i.e., yellow warning: when the characteristic parameter deviates by 10%-20%, encrypted monitoring is triggered, for example, the vibration sampling frequency is increased to 50kHz; Level 2 early warning, i.e. orange warning: when the deviation is 20%-50% or multiple signals are abnormal, the redundant compressor pre-operation program is automatically started, for example, the start-up preparation is completed within 10 seconds; Level 3 early warning, i.e. red warning: when the deviation is >50% or the safety threshold is reached, for example, the pressure is >9MPa, the shutdown is triggered and the fault point is locked within 0.1ms.

[0044] If the overall failure probability value exceeds the safety threshold, a control action degradation instruction is triggered to generate a safe execution plan. For example, if the safety threshold is set to 0.5, when the overall failure probability value > 0.5, the control action degradation is executed immediately.

[0045] This embodiment's safe execution scheme simultaneously records equipment status data, early warning information, and degradation control parameters at the time of a fault, generating a fault diagnosis report. In safe execution mode, the equipment recalculates the comprehensive fault probability value every 5 minutes. When the calculated value is ≤0.3 for three consecutive times, the degradation command can be gradually lifted, restoring the device to normal control mode.

[0046] In summary, this embodiment can promptly detect abnormal situations in equipment operation by calculating the deviation between the equipment's operating status characteristics and a preset benchmark threshold. Secondly, feature-level fusion and multi-source evidence reasoning on the deviation can integrate multiple factors to more accurately assess the equipment's failure probability, further improving the accuracy and reliability of fault diagnosis. Determining the warning level based on the comprehensive failure probability value and triggering a control action downgrade command when necessary allows for timely measures to be taken when equipment failure risks occur, ensuring safe equipment operation and reducing downtime and maintenance costs.

[0047] In one embodiment of this application, based on a multi-source fusion deviation feature vector and a preset basic probability allocation function and its weights, uncertainty reasoning and evidence combination of multi-source evidence are performed through an evidence theory framework to obtain a comprehensive failure probability value, including: The conflict degree between the basic probability assignment functions of any two pieces of evidence is calculated using the Jousselme distance formula, and a conflict degree matrix is ​​constructed. The conflict level between the evidence is determined based on the conflict degree matrix. When the conflict level is greater than the preset conflict threshold, the conflict resolution strategy is used to process the conflicting evidence and obtain a comprehensive probability allocation. The confidence level of each fault type is determined based on the comprehensive probability allocation, and the fault probability value corresponding to the highest confidence level is taken as the comprehensive fault probability value.

[0048] For example, vibration signals are assigned a probability of 0.7 to mechanical faults, current signals to electrical faults to 0.8, and pressure signals to fluid faults to 0.8. The conflict degree between any two pieces of evidence is calculated using the Jousselme distance formula, yielding a conflict degree of approximately 0.53 for vibration and current signals, approximately 0.61 for vibration and pressure signals, and approximately 0.59 for current and pressure signals. A conflict degree matrix is ​​then constructed based on this. Next, based on the risk classification of the "three-level early warning mechanism" and combined with the false alarm rate requirements for fault diagnosis in industrial scenarios, a conflict threshold is preset. If the conflict degrees of vibration and pressure signals, and current and pressure signals, as determined by the conflict degree matrix, are all within the acceptable range... If the threshold is exceeded, conflict resolution needs to be initiated. Next, the m-functions of each piece of evidence are weighted and corrected. After correction, a comprehensive probability allocation result is obtained through the DS evidence combination rule. Based on the confidence level of the comprehensive probability allocation result, for example, the probability allocation of the normal state is excluded to avoid interference. Specifically, after correction and combination, the probability allocations for mechanical failure, electrical failure, fluid failure, and normal state are 0.32, 0.29, 0.25, and 0.14, respectively. The confidence levels for mechanical failure are approximately 0.37, electrical failure approximately 0.34, and fluid failure approximately 0.29. The highest confidence level corresponds to mechanical failure, and its probability allocation value of 0.32 is the comprehensive failure probability value.

[0049] The process from calculating evidence conflict to determining the comprehensive fault probability is supported by the core logic of the "fault self-diagnosis module" to achieve high-precision diagnosis. Ultimately, relying on multi-source signal fusion diagnosis, the fault location accuracy rate reaches over 95%, the false alarm rate is reduced by 80%, and the early fault warning time is advanced to 2-4 hours before the fault occurs. For example, in the nascent stage of the fault, such as when the bearing temperature rises slightly or the vibration is abnormal, an alarm is issued, allowing sufficient maintenance time. A typical case is the fault warning of wind turbine gearbox. The alarm 4 hours in advance can effectively avoid sudden unit shutdown.

[0050] In this embodiment, the Jousselme distance formula is used to calculate the conflict degree and construct a conflict degree matrix, which can quantify the degree of conflict between evidence and provide a basis for conflict resolution. Secondly, by processing conflicting evidence through conflict resolution strategies, the accuracy of multi-source evidence fusion can be improved, resulting in a more reliable comprehensive failure probability value, thereby improving the accuracy of fault diagnosis. Combined with a redundant architecture to ensure continuous production, a full-link fault management system is formed.

[0051] In one embodiment of this application, the conflict level between pieces of evidence is determined based on a conflict degree matrix. When the conflict level is greater than a preset conflict threshold, a conflict resolution strategy is used to process the conflicting evidence to obtain a comprehensive probability allocation, including: The average Jousselme distance between each piece of evidence was calculated based on the conflict degree matrix. The reliability weight of each piece of evidence is obtained by normalizing the inverse of the average Jousselme distance between each piece of evidence. Based on the reliability weights of each piece of evidence, a weighted average is calculated on the basic probability allocation functions of all evidence to obtain the fused basic probability allocation function, and the fused basic probability allocation function is used as a single piece of evidence. Based on the evidence theory combination rules, evidence is synthesized from a single piece of evidence to obtain a comprehensive probability allocation.

[0052] In this embodiment, the multi-source fusion deviation feature vector corresponds to n types of evidence, where n includes: monitoring data related to mechanical faults, electrical faults, control system faults, cooling system faults, and normal state.

[0053] Based on the characteristic deviation range, a basic probability allocation function is preset. For example, when the deviation of the main peak of vibration frequency is ≥0.7, the basic probability allocation function of mechanical fault is set to 0.9, and the basic probability allocation function of normal state is set to 0.1; when the deviation of motor speed fluctuation rate is ≤-0.6, the basic probability allocation function of electrical fault is set to 0.85, and the basic probability allocation function of normal state is set to 0.15.

[0054] Secondly, when using the Jousselme distance formula to calculate the distance between any two basic probability assignment functions of evidence in the conflict matrix, the basic probability assignment function vector is first substituted into the formula to obtain the conflict degree value. For example, the conflict degree calculation result of evidence 1 and evidence 2 is 0.85, indicating that the two are significantly conflicted. Among them, evidence 1 is associated with mechanical faults and evidence 2 is associated with electrical faults.

[0055] A conflict degree matrix constructed based on conflict values ​​is used to present the conflict relationships between various pieces of evidence, thereby determining the conflict level among them. When the conflict level exceeds a preset conflict threshold, a conflict resolution strategy is employed to process the conflicting evidence, resulting in a comprehensive probability allocation. For example, the preset conflict threshold of 0.6 is set based on historical failure case statistics. When the average conflict value is greater than 0.6, it indicates that more than 60% of the evidence pairs have significant conflicts, requiring resolution to be initiated. In the weighted correction method for conflict resolution, the credibility of evidence is calculated by first determining the average conflict degree of each piece of evidence with all other evidence. For example, the average conflict degree of evidence 3 is 0.7, and its credibility is determined by the formula: The calculation shows that Evidence 3 is a correlation of control system failures.

[0056] The average probability assignment function is the arithmetic mean of the basic probability assignment functions of n pieces of evidence. For example, the average probability assignment function for the mechanical failure dimension is the average of the basic probability assignment functions for mechanical failure of each piece of evidence. The corrected basic probability assignment function needs to be re-normalized to ensure that the sum of the probabilities of all pieces of evidence is 1.

[0057] This embodiment uses Dempster's combination rule. If an empty set probability occurs after combination, it is proportionally distributed to each non-empty proposition. The final comprehensive probability distribution must satisfy the condition that the sum of all evidence probabilities is 1. In this embodiment, the confidence level of each fault type is determined based on the allocation of comprehensive probabilities, and the fault probability value corresponding to the highest confidence level is used as the comprehensive fault probability value. The allocation of comprehensive fault probability values ​​includes the confidence levels of n pieces of evidence; in the comprehensive probability allocation, the confidence level of the normal state is the highest, and there is a special case where multiple fault types have the same highest confidence level.

[0058] In summary, this embodiment calculates the reliability weights of evidence based on the conflict matrix and performs a weighted average operation on the basic probability allocation function. This allows for the reasonable utilization of information from each piece of evidence and reduces the impact of conflicting evidence on the results. Furthermore, by using the fused function as a single piece of evidence for evidence synthesis, the effectiveness of evidence fusion is ensured, while simultaneously improving the efficiency and accuracy of fault diagnosis.

[0059] In one embodiment of this application, the preset basic probability allocation function generation method includes: Calculate the weight factors corresponding to each feature in the multi-source fusion deviation feature vector; The deviations of each feature are weighted and fused based on the weighting factors to generate a basic probability allocation function.

[0060] In this embodiment, the calculation of the weighting factor needs to consider multiple factors. This embodiment determines the weighting factor through the entropy method and the analytic hierarchy process (AHP). Specifically, the AHP is used to construct a judgment matrix, compare the importance of each feature pairwise, and then perform consistency checks and calculations to obtain the weighting factor of each feature. The entropy method determines the weight based on the dispersion of the feature data. The greater the dispersion, the more information the feature contains, and the larger the weighting factor.

[0061] This embodiment multiplies the deviation of each feature by its corresponding weighting factor to obtain a weighted deviation. Based on these deviations, the data is integrated, and a basic probability allocation function is derived through mathematical calculations. The core of this process is to weigh the deviations of different features using weighting factors, enabling the generated basic probability allocation function to more accurately characterize the degree to which each feature's deviation supports the occurrence of device failure. For example, for features with larger weighting factors, their deviations have a greater impact on the basic probability allocation function after weighted fusion. When the deviation of such a feature is large, the basic probability allocation function will assign a higher probability value to the device failure.

[0062] In this embodiment, the analytic hierarchy process (AHP) or entropy weighting method is used to calculate feature weighting factors. Based on the varying degrees of influence of each feature on the equipment's operating status, weights are rationally allocated to make the basic probability allocation function more representative of the actual situation. Secondly, the deviation is weighted and fused based on the weighting factors to generate the basic probability allocation function, providing a more accurate foundation for subsequent multi-source evidence fusion and fault probability calculation, thus helping to improve the accuracy of fault diagnosis.

[0063] In one embodiment of this application, the predicted load status includes load rate and load change trend. The decision threshold of the hierarchical decision tree is adjusted based on the predicted load status to obtain the target hierarchical decision tree, including: The corresponding baseline adjustment template in the template library is determined based on the load rate; the template library includes multiple baseline adjustment templates. The threshold boundaries in the baseline adjustment template are corrected based on the load change trend and the compressor's energy efficiency value to obtain the target adjustment template; The decision threshold of the hierarchical decision tree is adjusted based on the target adjustment template to obtain the target hierarchical decision tree.

[0064] In this embodiment, the load rate is an important indicator reflecting the current load level of the compressor unit, expressed as the ratio of the actual load to the rated load. A template library is constructed based on the operating characteristics and historical adjustment experience of the compressor unit under different load rate ranges. This template library contains multiple benchmark adjustment templates. Each benchmark adjustment template corresponds to a specific load rate range, including the decision threshold settings for each decision node in the hierarchical decision tree within that load rate range. For example, when the load rate is between 30% and 50%, it corresponds to template A, whose decision threshold settings favor energy-saving operation; when the load rate is between 70% and 90%, it corresponds to template B, whose decision threshold settings focus more on the operational stability of the equipment.

[0065] This embodiment illustrates the load change trend, showing the direction and rate of load rate change over a period of time, such as an upward, downward, or stable trend. Based on the load change trend and the compressor's energy efficiency value, the threshold boundaries in the baseline adjustment template are corrected to obtain the target adjustment template. The compressor's energy efficiency value reflects the energy utilization efficiency of the equipment under the current operating state; a higher energy efficiency value indicates more efficient energy utilization. Based on the baseline adjustment template, threshold boundaries are corrected according to the load change trend and energy efficiency value. For example, if the load change trend is upward and the current energy efficiency value is low, it indicates that the equipment needs to raise some decision thresholds in advance to cope with the upcoming high load; if the load change trend is downward and the energy efficiency value is high, some decision thresholds can be appropriately lowered to further optimize energy-saving effects while ensuring stable operation. Through such dynamic correction, the target adjustment template is obtained. The target adjustment template includes the corrected threshold boundaries for each decision node. These thresholds are applied to the hierarchical decision tree to replace or adjust the original decision thresholds, enabling the hierarchical decision tree to more accurately adapt to real-time load changes and energy efficiency status, thereby generating a better execution plan.

[0066] In this embodiment, the baseline adjustment template is determined based on the load rate, which provides a basic framework for the adjustment of the hierarchical decision tree, thereby improving the efficiency and rationality of the adjustment.

[0067] In one embodiment of this application, the threshold boundary in the baseline adjustment template is corrected based on the load change trend and the compressor's energy efficiency value to obtain the target adjustment template, including: Based on preset mapping rules, the load change trend is mapped to a threshold offset; Determine the energy efficiency adjustment coefficient based on the energy efficiency value; The threshold in the baseline adjustment template is corrected based on the threshold offset and energy efficiency adjustment coefficient to obtain the target adjustment template.

[0068] In one embodiment, the load change trend is mapped to a threshold offset based on a preset mapping rule, including: If the load change trend is upward, then adjust the compressor start threshold downward and the shutdown threshold upward. If the load change trend is downward, then adjust the compressor start threshold upward and the shutdown threshold downward. The greater the magnitude of the load change trend, the greater the corresponding threshold correction magnitude.

[0069] In this embodiment, the preset mapping rule machine determines the offset direction and magnitude of the threshold based on the load change trend. For example, if the load change trend is upward, it means that the load will increase further in the future. At this time, it is necessary to adjust the compressor start threshold downward so that the compressor can start earlier to cope with the upcoming higher load; at the same time, the shutdown threshold is adjusted upward. If the load change trend is downward, it means that the load will decrease in the future. Therefore, the start threshold is adjusted upward to reduce unnecessary compressor starts to save energy; the shutdown threshold is adjusted downward so that the compressor stops in time when the load is low, avoiding ineffective operation.

[0070] In this embodiment, the greater the magnitude of the load change trend, the greater the corresponding threshold correction magnitude. For example, when the load rate rises rapidly at a rate of 10% per hour, the threshold is corrected downwards by 8%; while when it rises slowly at a rate of 3% per hour, the threshold only needs to be corrected downwards by 3%. This embodiment uses mapping rules to ensure that the threshold offset accurately reflects the trend and intensity of load changes, providing direction and basic values ​​for threshold correction.

[0071] In this embodiment, the determination of the energy efficiency adjustment coefficient is related to the energy efficiency value. An energy efficiency benchmark value can be set. When the actual energy efficiency value is higher than the benchmark value, the energy efficiency adjustment coefficient is greater than 1, indicating that the threshold correction restriction can be relaxed to a certain extent to maintain the current high-efficiency operation. When the actual energy efficiency value is lower than the benchmark value, the energy efficiency adjustment coefficient is less than 1, meaning that the threshold correction needs to be strengthened, and the energy efficiency of the equipment needs to be improved through stricter threshold control. For example, if the energy efficiency benchmark value is 0.8, and the actual energy efficiency value is 0.9, the energy efficiency adjustment coefficient can be set to 1.1; similarly, when the actual energy efficiency value is 0.7, the energy efficiency adjustment coefficient can be set to 0.9, and so on.

[0072] This embodiment corrects the thresholds in the baseline adjustment template based on threshold offset and energy efficiency adjustment coefficient to obtain the target adjustment template. Specifically, each threshold in the baseline adjustment template is multiplied by its corresponding threshold offset, and then multiplied by the energy efficiency adjustment coefficient to obtain the corrected threshold. In this embodiment, the threshold settings of the baseline adjustment template take into account load change trends and energy efficiency status, making the obtained target adjustment template more accurate.

[0073] In this embodiment, the load change trend is mapped to a threshold offset, facilitating accurate operation when adjusting the decision tree threshold. Determining the energy efficiency adjustment coefficient based on the energy efficiency value takes into account the compressor's energy efficiency, ensuring that the adjusted decision tree not only adapts to load changes but also optimizes energy efficiency, thus contributing to energy-saving operation of the compressor.

[0074] In one embodiment of this application, a control strategy is obtained with Pareto optimality as the objective, including: Based on at least one preset performance optimization objective, a multi-objective optimization problem for controlling the compressor unit is constructed. The multi-objective optimization problem is mapped to the reward function of the multi-objective reinforcement learning model, where each optimization objective corresponds to a reward component; By using Pareto optimality to perform online optimization of the reinforcement learning model, we can obtain the Pareto optimal policy solution set. Based on the optimization objective priority of the current equipment operating status, the optimal objective control strategy is determined from the Pareto optimal strategy solution set.

[0075] In this embodiment, at least one performance optimization objective is preset to construct a multi-objective optimization problem for controlling the compressor unit. The preset performance optimization objectives mainly include minimum energy consumption, minimum equipment loss, and fastest response speed.

[0076] The mathematical model of the multi-objective optimization problem in this embodiment is as follows: under the premise of satisfying the safety constraints of compressor unit operation, minimize energy consumption, minimize equipment damage, and maximize response speed. Specifically, the multi-objective optimization problem is mapped to the reward function of a multi-objective reinforcement learning model, where each optimization objective corresponds to a reward component. The reward function is designed as a weighted combination R of the reward components, and the formula is: , in, The weights of each reward component, energy consumption reward component. Equipment loss reward Response speed bonus .

[0077] In this embodiment, the formula for the energy consumption reward component is:

[0078] Where E is the actual energy consumption in the current cycle. Maximum energy consumption under this operating condition The value ranges from [0,1]. The lower the energy consumption, the better. The larger, The value range is [0,1].

[0079] The formula for equipment loss reward is:

[0080] in, This represents the incremental equipment damage for the current period. Maximum allowable damage increment The value range is [0,1]. The smaller the damage, the better. The larger.

[0081] The formula for response speed bonus components is:

[0082] in, For the current response time, To the maximum allowable response time, The larger the value, the faster the response speed.

[0083] In this embodiment, the reinforcement learning model employs a multi-objective deep deterministic policy gradient algorithm. During the interaction with the environment, the agent explores different control policies, collecting corresponding state, action, reward, and next-state data. During optimization, for each policy, its performance metrics across the three optimization objectives are calculated. If no other policy outperforms this policy across all objectives, then this policy is considered Pareto optimal. Through iterative learning, the policy library is gradually updated, ultimately forming a Pareto optimal policy solution set. This solution set includes multiple non-dominant policies, each with its own emphasis on different objectives. For example, some policies have lower energy consumption but slower response times, while others have faster response times but slightly higher equipment wear. Based on the optimization objective priorities according to the current equipment operating status, the optimal objective control strategy is determined from the Pareto optimal strategy solution set. The optimization objective priorities are dynamically adjusted according to the equipment operating status. For example, when the equipment is operating under high load, the priority of minimizing equipment loss is set to the highest, followed by the lowest energy consumption, and the fastest response speed is set to the lowest. Specifically, the weight of equipment loss can be set to 0.4, the weight of lowest energy consumption to 0.35, and the weight of response speed to 0.25.

[0084] In this embodiment, based on the determined priority weights, the comprehensive performance score of each strategy in the Pareto optimal strategy solution set is calculated, and the strategy with the highest score is selected as the optimal target control strategy. For example, under high load conditions, a certain strategy scores 0.9 for equipment loss target, 0.7 for energy consumption target, and 0.6 for response speed target. Its comprehensive score is 0.9×0.4+0.7×0.35+0.6×0.25=0.36+0.245+0.15=0.755. If 0.755 is the highest in the solution set, it is selected as the optimal control strategy. In this embodiment, a multi-objective optimization problem is constructed and solved through reinforcement learning. This approach comprehensively considers multiple performance optimization objectives, such as energy consumption and stability, resulting in a more comprehensive and superior control strategy.

[0085] In one embodiment of this application, the optimal target control strategy is determined from the Pareto optimal strategy solution set based on the optimization target priority of the current device operating state, including: If the size of the Pareto optimal policy solution set is greater than the preset capacity limit, then the policies are clustered based on the feature similarity between them. In each cluster, at least one representative policy is selected to form a new Pareto optimal policy solution set. Based on the optimization objective priority of the current equipment operating status, the strategies in the new Pareto optimal strategy solution set are sorted or selected to obtain the optimal objective control strategy.

[0086] In this embodiment, when the size of the Pareto optimal strategy solution set exceeds a preset capacity limit, clustering is first performed based on the feature similarity between strategies. Feature similarity can be quantified and evaluated using the Euclidean distance between the control parameter distribution and performance index vectors. The control parameters include: compressor start-stop combination, speed regulation range, and valve group opening interval. Similar strategies are grouped into one cluster, and 1-2 representative strategies are selected from each cluster to form a new Pareto optimal strategy solution set. Secondly, the optimization objective priority is determined based on the current equipment operating state; the strategies in the new solution set are then sorted or selected according to the optimization objective priority.

[0087] In one embodiment of this application, a multi-objective optimization problem for compressor unit control is constructed, including: Set decision variables for compressor unit control, including compressor start / stop combination, guide vane opening setpoint, and target pressure setpoint; Based on the preset performance optimization goals, target optimization functions corresponding to the performance optimization goals are established. The target optimization functions include energy consumption target function, stability target function and equipment life loss target function. A mathematical model for a multi-objective optimization problem is constructed based on decision variables, objective optimization functions, and preset equipment operation constraints. The preset equipment operation constraints include pressure constraints, temperature constraints, and current constraints.

[0088] In one embodiment, a hybrid drive architecture of "fixed speed + variable frequency" is constructed to achieve adaptive adjustment under multiple operating conditions. The fixed-speed compressor unit uses solid-state relays to construct a contactless switching control loop, supporting millisecond-level response speeds and ≥10... 8 With a reliable switching life, it is suitable for high-frequency start-stop scenarios, ensuring high reliability and long-term stability of equipment operation; the variable frequency compressor unit is configured with a vector control type variable frequency drive, covering a power range of 75-5000kW, with a high-precision speed regulation capability of ±0.01%, and realizes linear and smooth adjustment of motor speed through field-oriented control technology, meeting the precise power matching requirements under complex loads.

[0089] The decision variables are set based on two dimensions: equipment operating status and process parameters. These include: determining the compressor start-stop combination, based on a "fixed speed + variable frequency" dual-mode drive architecture, selecting the number of fixed-speed compressors and variable frequency compressors to operate according to dynamic load trends, such as "1 fixed speed + 1 variable frequency" or "2 fixed speed + 1 variable frequency"; setting the guide vane opening, dynamically adjusting it in conjunction with real-time flow data collected by electromagnetic flowmeters to match process flow requirements; and determining the target pressure, dynamically setting it within safe boundaries based on pipeline pressure monitored by pressure sensors to reduce the impact of pressure fluctuations on system stability.

[0090] Corresponding to the load-adaptive intelligent hierarchical compressor control method in the above embodiment, Figure 2 This is a structural block diagram of a load-adaptive intelligent staged compressor control system according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The AI-based production data monitoring system 20 includes: a data acquisition module 21, a data processing module 22, a load prediction module 23, a scheme generation module 24, a control strategy module 25, and a strategy execution module 26.

[0091] The data acquisition module 21 is used to acquire multi-dimensional data of the compressor unit through a distributed sensor network. The multi-dimensional data includes process parameters, equipment operating status and energy consumption data. Data processing module 22 is used to preprocess the multidimensional data and extract features to obtain feature vectors; Load prediction module 23 is used to input the feature vector into the load prediction model to obtain the predicted load status; The scheme generation module 24 is used to adjust the decision threshold of the hierarchical decision tree based on the predicted load status to obtain the target hierarchical decision tree; and to input the feature vector into the target hierarchical decision tree to generate an execution scheme. Control strategy module 25 is used to acquire energy efficiency feedback data for executing the execution scheme, and input the predicted load state and the energy efficiency feedback data into the reinforcement learning model to obtain the control strategy with Pareto optimality as the objective. The strategy execution module 26 is used to adjust the control actions of the compressor unit based on the control strategy.

[0092] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the data acquisition module 21, data processing module 22, load prediction module 23, scheme generation module 24, control strategy module 25, and strategy execution module 26 are shown.

[0093] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0094] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0095] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0096] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the intelligent hierarchical compressor control method based on load adaptation provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0097] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0098] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0099] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0100] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0102] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0103] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A load-adaptive intelligent hierarchical compressor control method, characterized in that, include: Multidimensional data of the compressor unit is acquired through a distributed sensor network. The multidimensional data includes process parameters, equipment operating status, and energy consumption data. The multidimensional data is preprocessed and features are extracted to obtain feature vectors; The feature vector is input into the load prediction model to obtain the predicted load status; The decision threshold of the hierarchical decision tree is adjusted based on the predicted load status to obtain the target hierarchical decision tree; wherein, the predicted load status includes load rate and load change trend; The step of adjusting the decision threshold of the hierarchical decision tree based on the predicted load state to obtain the target hierarchical decision tree includes: Based on the load rate, a corresponding baseline adjustment template is determined in the template library; the template library includes multiple baseline adjustment templates; based on the load change trend and the compressor's energy efficiency value, the threshold boundaries in the baseline adjustment template are corrected to obtain the target adjustment template; based on the target adjustment template, the decision threshold of the hierarchical decision tree is adjusted to obtain the target hierarchical decision tree; The step of correcting the threshold boundary in the baseline adjustment template based on the load change trend and the compressor's energy efficiency value to obtain the target adjustment template includes: The load change trend is mapped to a threshold offset based on a preset mapping rule; an energy efficiency adjustment coefficient is determined based on the energy efficiency value; and the threshold in the benchmark adjustment template is corrected based on the threshold offset and the energy efficiency adjustment coefficient to obtain the target adjustment template. The feature vector is input into the target hierarchical decision tree to generate an execution plan; The energy efficiency feedback data for executing the execution scheme is obtained, and the predicted load state and the energy efficiency feedback data are input into the reinforcement learning model to obtain the control strategy with Pareto optimality as the objective. The control actions of the compressor unit are adjusted based on the control strategy.

2. The intelligent hierarchical compressor control method based on load adaptation according to claim 1, characterized in that, Also includes: The operating status of the device is subjected to feature extraction to obtain a target feature vector, and the deviation of each feature in the target feature vector from its preset normal operating benchmark threshold is calculated. The deviations corresponding to each feature in the target feature vector are fused at the feature level to generate a multi-source fused deviation feature vector. Based on the multi-source fusion deviation feature vector and the preset basic probability allocation function and its weights, uncertainty reasoning and evidence combination of multi-source evidence are carried out through the evidence theory framework to obtain a comprehensive failure probability value. The corresponding early warning level is determined based on the preset risk level range in which the comprehensive failure probability value falls; If the overall failure probability value is greater than the safety threshold, a control action degradation instruction is triggered to generate a safe execution plan.

3. The intelligent hierarchical compressor control method based on load adaptation according to claim 2, characterized in that, Based on the multi-source fusion deviation feature vector and the preset basic probability allocation function and its weights, uncertainty reasoning and evidence combination of multi-source evidence are performed through the evidence theory framework to obtain a comprehensive failure probability value, including: The conflict degree between the basic probability assignment functions of any two pieces of evidence is calculated using the Jousselme distance formula, and a conflict degree matrix is ​​constructed. The conflict level between the evidence is determined based on the conflict degree matrix. When the conflict level is greater than the preset conflict threshold, a conflict resolution strategy is adopted to process the conflicting evidence and obtain a comprehensive probability allocation. The confidence level of each fault type is determined based on the comprehensive probability allocation, and the fault probability value corresponding to the highest confidence level is taken as the comprehensive fault probability value.

4. The intelligent hierarchical compressor control method based on load adaptation according to claim 3, characterized in that, The step of determining the conflict level among evidence based on the conflict degree matrix, and when the conflict level is greater than a preset conflict threshold, employing a conflict resolution strategy to process the conflicting evidence to obtain a comprehensive probability allocation, includes: The average Jousselme distance between each piece of evidence is calculated based on the conflict degree matrix. The reliability weight of each piece of evidence is obtained by normalizing the inverse of the average Jousselme distance between the evidences. Based on the reliability weights of each piece of evidence, a weighted average is calculated on the basic probability allocation functions of all evidence to obtain a fused basic probability allocation function, and the fused basic probability allocation function is used as a single piece of evidence. Based on the evidence theory combination rules, the single piece of evidence is synthesized to obtain a comprehensive probability allocation.

5. The intelligent hierarchical compressor control method based on load adaptation according to claim 2, characterized in that, The preset basic probability allocation function generation method includes: Calculate the weight factor corresponding to each feature in the multi-source fusion deviation feature vector; The deviations of each feature are weighted and fused based on the weighting factors to generate a basic probability allocation function.

6. A load-adaptive intelligent staged compressor control system, characterized in that, The data acquisition module is used to acquire multi-dimensional data of the compressor unit through a distributed sensor network. The multi-dimensional data includes process parameters, equipment operating status and energy consumption data. The data processing module is used to preprocess the multidimensional data and extract features to obtain feature vectors; The load prediction module is used to input the feature vector into the load prediction model to obtain the predicted load status. The scheme generation module is used to adjust the decision threshold of the hierarchical decision tree based on the predicted load status to obtain a target hierarchical decision tree; input the feature vector into the target hierarchical decision tree to generate an execution scheme; wherein, the predicted load status includes load rate and load change trend; The step of adjusting the decision threshold of the hierarchical decision tree based on the predicted load state to obtain the target hierarchical decision tree includes: Based on the load rate, a corresponding baseline adjustment template is determined in the template library; the template library includes multiple baseline adjustment templates; based on the load change trend and the compressor's energy efficiency value, the threshold boundaries in the baseline adjustment template are corrected to obtain the target adjustment template; based on the target adjustment template, the decision threshold of the hierarchical decision tree is adjusted to obtain the target hierarchical decision tree; The step of correcting the threshold boundary in the baseline adjustment template based on the load change trend and the compressor's energy efficiency value to obtain the target adjustment template includes: The load change trend is mapped to a threshold offset based on a preset mapping rule; an energy efficiency adjustment coefficient is determined based on the energy efficiency value; and the threshold in the benchmark adjustment template is corrected based on the threshold offset and the energy efficiency adjustment coefficient to obtain the target adjustment template. The control strategy module is used to acquire energy efficiency feedback data for executing the execution scheme, and input the predicted load state and the energy efficiency feedback data into the reinforcement learning model to obtain the control strategy with Pareto optimality as the objective. The strategy execution module is used to adjust the control actions of the compressor unit based on the control strategy.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images