Compressor fault early warning method based on electronic starter controller and electronic starter controller

Abstract

The invention discloses a compressor fault early warning method based on an electronic starter controller and the electronic starter controller, and relates to the technical field of electronic starter controllers, operating parameters and historical fault information of a compressor are collected, and the operating parameters comprise a power-on parameter, a starting parameter, a normal working parameter, a protection parameter and an environment industrial control parameter; correlation analysis is conducted on the operation parameters and historical fault information, and target operation parameters related to compressor faults are screened out; training a fault early warning model based on the target operation parameters; and obtaining current operation parameters of the compressor, inputting the current operation parameters into the trained fault early warning model, and automatically carrying out early warning on the compressor fault through the fault early warning model. The compressor fault early warning method solves the problem that a compressor fault early warning method in the prior art is low in accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic starting controller, and in particular relates to a compressor fault early warning method based on an electronic starting controller and an electronic starting controller. BACKGROUND

[0002] In recent years, as the core power equipment in the fields of refrigeration air conditioning, industrial manufacturing and new energy, the operation reliability and continuous operation ability of the compressor directly affect the production efficiency of the industry chain, the quality of the terminal product and the operation safety, and the industry continues to improve the timeliness, accuracy and intelligent level of the compressor fault early warning. In order to meet the basic operation monitoring requirements, the industry gradually introduces temperature sensors, pressure sensors, current sensors and other single or few types of sensing devices, and triggers an alarm by setting fixed thresholds (such as over-temperature, over-pressure and over-current), or estimates the fault probability based on simple historical fault frequency statistics.

[0003] However, most of the current compressor fault early warning schemes still have technical limitations: first, relying on data collected by single or few sensors, the monitoring dimension is limited, only the changes of local dominant parameters such as temperature and pressure can be captured, and it is difficult to cover the early features of hidden faults such as motor insulation aging and bearing wear. The one-sidedness of the data leads to delayed early warning and cannot identify the precursors of the fault; second, the alarm logic based on fixed thresholds lacks analysis of the dynamic change trend of the parameters, which is easy to produce false alarms due to instantaneous voltage fluctuations, sudden changes in environmental temperature and humidity and other interference factors, or to miss the alarm due to the slow drift of the fault parameters without reaching the threshold, and the reliability is insufficient; therefore, the current compressor fault early warning method has the problem of low accuracy. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a compressor fault early warning method based on an electronic starting controller and an electronic starting controller, which aims to solve the problem of low accuracy of the compressor fault early warning method in the prior art.

[0005] In one aspect, the present application provides a compressor fault early warning method based on an electronic starting controller, which comprises: Collecting the operating parameters and historical fault information of the compressor, the operating parameters including power-on parameters, start-up parameters, normal working parameters, protection parameters and environmental control parameters; Correlation analysis of the operating parameters and historical fault information to screen out target operating parameters related to the compressor fault; Training a fault early warning model based on the target operating parameters; Obtaining the current operating parameters of the compressor, inputting the current operating parameters into the trained fault early warning model, and automatically warning the compressor fault through the fault early warning model.

[0006] Further, the compressor fault early warning method based on the electronic starting controller, wherein the step of performing correlation analysis on the operation parameters and the historical fault information to screen out target operation parameters related to the compressor fault comprises: synchronously collecting the operation parameters of the compressor and the historical fault information, filtering the operation parameters, and aligning the operation parameters on a time axis according to operation stages; dividing the historical fault information into different types of faults according to fault causes, for each type of fault, using different correlation analysis algorithms to calculate the correlation degrees of corresponding operation parameters respectively; obtaining real-time operation state signals output by the compressor, assigning dynamic weights to operation parameters in different operation stages, and re-calculating the weighted correlation degrees of each operation parameter and corresponding fault types based on the dynamic weights to screen out high-correlation-degree parameters; collecting normal operation parameter samples of the compressor under different environmental working conditions, verifying the correlation stability of the high-correlation-degree parameters preliminarily screened out under different environmental conditions, and finally screening out operation parameters with correlation stability greater than a threshold value in the whole working condition range as target operation parameters strongly related to the compressor fault.

[0007] Further, the compressor fault early warning method based on the electronic starting controller, wherein the fault types include motor faults, gas path faults, and control faults, and the step of using different correlation analysis algorithms to calculate the correlation degrees of corresponding operation parameters comprises: for motor faults, using a Pearson correlation coefficient to calculate the linear correlation degree of linear characteristic parameters and the fault; for gas path faults, using a mutual information entropy algorithm to calculate the nonlinear correlation degree of nonlinear time sequence parameters and the fault; for control faults, using a Spearman rank correlation coefficient to calculate the sequential correlation degree of sequential change parameters and the fault.

[0008] Further, the compressor fault early warning method based on the electronic starting controller, wherein the step of using different correlation analysis algorithms to calculate the correlation degrees of corresponding operation parameters further comprises: for operation parameters corresponding to each fault type, screening a preset number of parameter combinations according to a fault type and parameter attribute matching principle; loading an improved time window Apriori algorithm, taking fault data of the same type of compressor in recent years and operation data of the current compressor in recent years as data sources, introducing a preset minute time window constraint before the fault, and mining the change trend of the parameter combination and the fault triggering association rule; Calculate the rule confidence and set the threshold value according to the fault type, and compare the rule trigger records of the current compressor in the near future with the actual fault maintenance records, eliminate invalid rules with an actual fault matching rate less than the threshold value, and retain valid rules; For the parameter combination corresponding to the valid rule, determine the parameter fault contribution degree according to the proportion of the number of independent early warnings of the parameter in the corresponding fault in the near future, and assign a weight, calculate the coupling correlation degree according to the single parameter correlation degree by using the weighted product method, if the coupling correlation degree is greater than the maximum value of each single parameter correlation degree, then the coupling correlation degree replaces the original single parameter correlation degree, if the coupling correlation degree is less than the maximum value of the single parameter correlation degree, then the original single parameter correlation degree is retained.

[0009] Further, the above-mentioned compressor fault early warning method based on an electronic starting controller, wherein the improved Apriori algorithm is loaded, the near-fault data of the same type of compressor and the near-future operation data of the current compressor are used as the data source, a pre-set minute time window constraint before the fault is introduced, and the steps of mining the change trend of the parameter combination and the fault trigger association rule include: Sort the data source according to the time stamp, set the change quantitative threshold value of each dedicated parameter, and convert the parameter into a standardized representation of the parameter name, change direction and change amplitude; Call the fault type and time window mapping table, and find the corresponding time window of each fault type in the fault type and time window mapping table; Divide the transaction set according to the corresponding time window, each transaction set contains the standardized representation of all dedicated parameters in a single window, and based on the transaction set, increase the parameter change order constraint when generating the frequent item set to form the ordered item set; According to the minimum weighted support degree corresponding to the fault type, the ordered frequent item set that meets the threshold value is selected from the ordered item set, and the change trend of the parameter combination and the fault trigger association rule are extracted from the ordered frequent item set.

[0010] Further, the above-mentioned compressor fault early warning method based on an electronic starting controller, wherein the calculation formula of the minimum weighted support degree is: ; Wherein, N is the total number of transaction sets in the data source, i is the i th occurrence record of the ordered item set, n is the total number of occurrences of the ordered item set, λ is the exponential decay coefficient, is the time length from the collection time of the i th occurrence record to the current calculation time, is the effective identification of the i th occurrence record.

[0011] Further, the compressor fault early warning method based on the electronic starting controller, wherein the step of calling the fault type and time window mapping table and searching for the time sequence window corresponding to each fault type in the fault type and time window mapping table comprises: An initial fault type and time window mapping table is constructed, and each fault type in the initial mapping table corresponds to a basic time sequence window, which is determined based on the fault precursor appearance period statistical value of the historical fault data of the same model compressor; The health index of the current running stage of the compressor is monitored in real time, and the health index is calculated based on the cumulative running time of the compressor, the parameter fluctuation amplitude within the last three maintenance intervals, and the failure rate decay curve of the same batch model; The basic time sequence window is dynamically corrected according to the health index, and when the health index is lower than a set threshold, the time sequence window corresponding to the fault type is enlarged by a preset proportion, and when the health index is higher than the set threshold, the time sequence window corresponding to the fault type is reduced by a preset proportion.

[0012] Further, the compressor fault early warning method based on the electronic starting controller, wherein the step of dynamically correcting the basic time sequence window according to the health index further comprises: The time difference between the actual compressor fault early warning record and the real fault occurrence time is collected, a correction factor library is established, and the time window correction coefficients corresponding to different seasons and different load rate intervals are stored in the correction factor library; The matching correction coefficient is retrieved from the correction factor library in combination with the current environmental parameters and the load rate, the time sequence window corrected by the health index is calibrated again, and the final dynamic time sequence window is obtained.

[0013] Further, the compressor fault early warning method based on the electronic starting controller, wherein the step of training the fault early warning model based on the target running parameters comprises: A multi-modal fusion network is used to construct the fault early warning model, and the multi-modal fusion network comprises a parameter feature extraction layer, a time sequence association layer, and a fault classification layer; wherein the parameter feature extraction layer designs exclusive convolution kernels and loop units for different types of parameters in the target running parameters, and the feature weights of the parameter combinations with high coupling degrees are strengthened through an attention mechanism; The time sequence association layer introduces a dynamic time sequence window as a time dimension constraint, concatenates the parameter change trend features in different windows according to the time sequence, uses the association rules mined by the improved time window Apriori algorithm as priori knowledge, and models the topological association between the parameter combinations and the fault types through a graph neural network; The fault classification layer trains the model by using a weighted loss function, the weighted loss function gives different weights to the misjudgment cost of different fault types: a high penalty weight is set for fatal faults, and a low penalty weight is set for slight faults, and a deviation rate of a dynamic time window is introduced as a regularization term to constrain the adaptability of the model to window fluctuations.

[0014] Another object of the present application is to provide an electronic control device for implementing the above-mentioned compressor fault early warning method based on the electronic control device, the electronic control device comprising: The acquisition module is configured to collect the operating parameters and historical fault information of the compressor, wherein the operating parameters include power-on parameters, start-up parameters, normal working parameters, protection parameters, and environmental control parameters. The screening module is configured to perform correlation analysis on the operating parameters and the historical fault information, and screen out target operating parameters related to the compressor faults. The training module is configured to train the fault early warning model based on the target operating parameters. The early warning module is configured to acquire the current operating parameters of the compressor, input the current operating parameters into the trained fault early warning model, and automatically perform early warning on the compressor faults through the fault early warning model.

[0015] Another object of the present application is to provide a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the above-mentioned method.

[0016] Another object of the present application is to provide an electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the processor implements the steps of the above-mentioned method when executing the program.

[0017] The present application breaks through the problem of limited monitoring dimension of a single or a small number of sensors by collecting multi-dimensional operating parameters of the compressor such as power-on parameters, start-up parameters, normal working parameters, protection parameters, and environmental control parameters, can comprehensively capture early features including hidden faults such as motor insulation aging and bearing wear, solves the problem of early warning lag caused by one-sidedness of data, realizes early identification of fault precursors, and at the same time, in combination with the historical fault information, performs correlation analysis on the operating parameters to screen out target operating parameters, trains the fault early warning model based on the target operating parameters, replaces the traditional alarm logic based on fixed threshold, can accurately analyze the dynamic change trend of the parameters, effectively eliminates interference factors such as instantaneous voltage fluctuation and sudden change of environmental temperature and humidity, reduces false positives and false negatives, and finally analyzes the current operating parameters through the trained fault early warning model to realize automatic early warning, comprehensively improves the timeliness, accuracy and intelligent level of the compressor fault early warning, and solves the problem of low accuracy of the compressor fault early warning method in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 Figure 1 is a flow chart of the compressor fault early warning method based on the electronic starting controller in the first embodiment of the present application; Figure 2 Figure 4 is a structure block diagram of the compressor fault early warning device based on the electronic starting controller in the third embodiment of the present application.

[0019] The following detailed description will further describe the present application in combination with the above-mentioned drawings. DETAILED DESCRIPTION

[0020] In order to facilitate the understanding of the present application, the present application will be described more fully below in connection with the related drawings. The drawings show several embodiments of the present application. However, the present application can be realized in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0021] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly on the other element or there can be a middle element. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or there can be a middle element. The terms "vertical", "horizontal", "left", "right", and similar expressions used herein are for illustrative purposes only.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terminology used in the description of the present application herein only for the purpose of describing specific embodiments and is not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0023] Embodiment One Please refer to Figure 1 Figure 1 is a flow chart of the compressor fault early warning method based on the electronic starting controller in the first embodiment of the present application, which comprises steps S10-S13.

[0024] Step S10, collect the running parameters and historical fault information of the compressor, the running parameters including power-on parameters, start-up parameters, normal working parameters, protection parameters and environmental control parameters.

[0025] The running parameters of the compressor and historical fault information are collected, and the running parameters specifically cover power-on parameters such as power-on voltage, current peak, start-up delay, start-up parameters such as start-up current, start-up power, start-up time length, and key data of the device from the start-up initialization data to the transition stage of operation, normal working parameters such as running current, exhaust pressure, body temperature, and running frequency in the rated working condition, stable working state data, protection parameters such as over-current protection threshold trigger record, over-temperature protection action signal, and under-voltage protection response data related to the self-protection mechanism of the device, and environmental working parameters such as environmental temperature and humidity, and external working condition data. Through multi-dimensional and full-stage parameter collection, the state characteristics of the whole life cycle of the compressor are fully covered, and the problem of insufficient monitoring dimension of traditional single sensor is completely solved. In combination with the historical fault information, a complete fault analysis data basis is constructed, and effective samples are provided for subsequent correlation analysis.

[0026] In step S11, the correlation between the running parameters and the historical fault information is analyzed, and target running parameters related to the compressor fault are screened out.

[0027] In the method, the internal correlation between various running parameters and different fault types is mined through statistical or machine learning algorithms, irrelevant or weakly correlated redundant parameters are eliminated, core parameters with high indication significance for fault warning are focused on, the calculation complexity of subsequent model training is reduced, and the warning efficiency is improved.

[0028] Specifically, the running parameters and the historical fault information of the compressor are synchronously collected, the synchronous recording of the running parameters and the fault information in the time dimension is realized, and it is ensured that each fault information can correspond to a complete running parameter sequence before and after the fault occurs; the running parameters are subjected to noise filtering, and algorithms such as Kalman filtering, median filtering, or wavelet transform are used to remove abnormal fluctuation data generated in the process of sensor data collection due to electromagnetic interference, mechanical vibration, environmental noise, etc., and the true change trend of the parameters is restored; and the running parameters are time axis aligned according to the running stage, the running stage switching signals of the compressor such as the start-up completion signal and the stable running signal are taken as time nodes, the collected running parameters are accurately divided into corresponding time intervals such as the power-on stage, the start-up stage, and the normal working stage, and the strict matching of the running parameters and the fault information in the time dimension is ensured, thereby providing a high-quality and consistent data basis for subsequent type-specific fault correlation analysis.

[0029] The historical fault information is divided into different types of faults according to fault causes, and combined with the structure and working principle of the compressor, the faults are divided into main types such as motor type, gas path type and control type, which can be further subdivided into specific fault subclasses such as motor type faults, which can be divided into winding overheating, bearing wear and motor locked-rotor, etc. Through fault classification, targeted analysis is realized to avoid interference of parameter characteristics of different cause faults with each other; for each type of fault, different correlation analysis algorithms are used to calculate the correlation degree of the corresponding operating parameters, and the algorithm is selected according to the parameter characteristic difference of different fault types to ensure the accuracy of the correlation degree calculation and provide a scientific basis for subsequent parameter screening.

[0030] The running state signals output by the compressor in real time, such as running stage identification signals, load change signals and equipment health state feedback signals, are obtained, and dynamic weights are assigned to the operating parameters in different running stages. According to the fault occurrence probability and parameter importance of the compressor in different running stages, different weights are given to the start-up parameters in the start-up stage and the working parameters in the normal working stage through weight distribution algorithms such as analytic hierarchy process and entropy weight method, for example, the weights of parameters such as start-up current and start-up power are emphasized in the start-up stage, and the weights of parameters such as running temperature and discharge pressure are emphasized in the normal working stage; The weighted correlation degree of each operating parameter and the corresponding fault type is recalculated based on the dynamic weights, the contribution of key stage parameters to the fault correlation degree is strengthened through weighted calculation, and high correlation degree parameters are obtained through setting weight threshold to ensure that the selected parameters are more suitable for the fault warning needs of the actual running state of the equipment. The normal operating parameter samples of the compressor under different environmental working parameters are collected, the normal operating parameter range and variation law of the compressor under various working conditions are recorded by changing the environmental temperature, humidity and load rate, and a full-working-condition normal parameter sample library is constructed; The high correlation degree parameters preliminarily selected are verified for their correlation degree stability under different environmental conditions, the fluctuation amplitude of the correlation degree of the same parameter under different working conditions and the corresponding fault type is calculated to judge the stability of the parameter correlation degree; finally, the operating parameters with correlation degree stability greater than the threshold in the full-working-condition range are selected as the target operating parameters with strong correlation with the compressor faults, to ensure that the selected target parameters can stably and reliably indicate the fault risk in the complex and changeable actual working environment, and avoid the failure of the parameters under specific working conditions to cause warning deviation.

[0031] For example, for motor class faults, the Pearson correlation coefficient is used to calculate the linear correlation degree of the linear feature parameters and the faults. Many feature parameters of motor class faults, such as the motor operating current and the winding temperature, and the power supply voltage and the motor power, often show obvious linear change relationship. The Pearson correlation coefficient can accurately quantify this linear correlation by calculating the ratio of the covariance and the standard deviation between two variables. The value range is between -1 and 1. The closer the absolute value is to 1, the stronger the linear correlation. This algorithm can effectively identify the feature parameters in the motor system that have a linear relationship with the faults, and provide accurate basis for motor class fault warning.

[0032] For air path class faults, the mutual information entropy algorithm is used to calculate the nonlinear correlation degree of nonlinear time series parameters and faults. Air path system faults such as pressure changes caused by leaks and flow fluctuations caused by damaged air valves often show complex nonlinear time series characteristics. Traditional linear algorithms cannot accurately capture these characteristics. The mutual information entropy algorithm measures the information sharing degree between two random variables without assuming the linear relationship between the variables. It can effectively quantify the dependence relationship between nonlinear parameters and faults. Even if the parameter changes have no obvious linear rules, it can accurately calculate the correlation degree between the parameters and the faults, breaking through the application limitations of linear algorithms.

[0033] For control class faults, the Spearman rank correlation coefficient is used to calculate the sequential correlation degree of sequential change parameters and faults. Control class faults often show characteristics of parameter changes in a specific order. For example, control signal delay can cause a sequence of abnormalities such as "command issuance - actuator response - parameter change". The Spearman rank correlation coefficient calculates the correlation between the rankings of variables by sorting and converting the data. It does not depend on the distribution characteristics of the data and can effectively capture the correlation between the order of parameter changes and the occurrence of faults. It is especially suitable for analyzing the indicating effect of sequential change parameters on control class faults.

[0034] Step S12, training a fault warning model based on the target operating parameters.

[0035] The fault warning model is trained based on the target operating parameters using deep learning, machine learning, or other algorithms. The filtered target operating parameters are used as input features, and the historical fault information is used as labels to build and train a model that can learn the parameter change and fault occurrence rules. The model has the ability to identify fault precursors from real-time parameters.

[0036] Step S13, obtaining the current operating parameters of the compressor, and inputting the current operating parameters into the trained fault warning model to automatically warn the compressor faults through the fault warning model.

[0037] The current operation parameters of the compressor are acquired, various operation parameters of the compressor under the current working condition are collected in real time through the electronic starting controller, the real-time and accuracy of the data are ensured, the current operation parameters are input into the trained fault early warning model, the model automatically judges whether there is a fault risk and a fault type through feature extraction and rule matching of the real-time parameters, and then the compressor fault is automatically warned through the fault early warning model, the warning information can be output through sound and light alarm, system pop-up window, remote push and the like, the timely discovery and early intervention of the fault are realized, and the stable operation of the equipment is ensured.

[0038] In summary, the compressor fault early warning method based on the electronic starting controller in the above embodiments of the present application breaks through the problem of limited monitoring dimension of a single or a few sensors, can comprehensively capture early features including hidden faults such as motor insulation aging and bearing wear, solves the problem of early warning lag caused by one-sided data, realizes early identification of fault precursors, and at the same time, the target operation parameters are screened out by correlation analysis of the operation parameters combined with historical fault information, and then the fault early warning model is trained based on the target operation parameters, which replaces the traditional alarm logic based on fixed threshold, can accurately analyze the dynamic change trend of the parameters, effectively eliminates interference factors such as instantaneous voltage fluctuation and sudden change of environmental temperature and humidity, reduces false positives and false negatives, and finally realizes automatic warning by analyzing the current operation parameters through the trained fault early warning model, comprehensively improves the timeliness, accuracy and intelligent level of the compressor fault early warning, and solves the problem of low accuracy of the compressor fault early warning method in the prior art.

[0039] Embodiment Two The present embodiment also proposes a compressor fault early warning method based on an electronic starting controller, which is different from the compressor fault early warning method based on an electronic starting controller in Embodiment One in that: The step of calculating the correlation degree of the corresponding operation parameters by using different correlation analysis algorithms further comprises: For the operation parameters corresponding to each fault type, a preset number of parameter combinations are screened according to the fault type and parameter attribute matching principle; An improved time window Apriori algorithm is loaded, the fault data of the same type of compressor in recent years and the operation data of the current compressor in recent years are taken as data sources, a preset minute time window constraint before the fault is introduced, and the change trend of the parameter combination and the fault triggering association rule are mined; The rule confidence is calculated and a threshold value is set according to the fault type, and the rule triggering record of the current compressor in the recent month is compared with the actual fault maintenance record, invalid rules with an actual fault matching rate less than the threshold value are removed, and effective rules are retained; For the parameter combination corresponding to the effective rule, the parameter fault contribution degree is determined according to the proportion of the number of independent early warnings of the parameter in the corresponding fault in recent years, and a weight is allocated, the coupling correlation degree is calculated according to the single parameter correlation degree by using the weighted product method, if the coupling correlation degree is greater than the maximum value of each single parameter correlation degree, the coupling correlation degree is used to replace the original single parameter correlation degree, if the coupling correlation degree is less than the maximum value of the single parameter correlation degree, the original single parameter correlation degree is retained.

[0040] Among them, for each fault type corresponding operating parameter, according to the fault type and parameter attribute matching principle, a preset number of parameter combinations are screened, the attribute related parameters are combined according to the causes and characteristics of different fault types, for example, the motor class fault can combine the motor operating current, winding temperature, bearing vibration frequency and other parameters directly related to the motor working state, the gas path class fault can combine the exhaust pressure, inlet flow, gas valve opening and other gas path system core parameters, the control class fault can combine the control signal transmission delay, actuator response time, sensor sampling frequency and other control loop parameters, and the number of parameter combinations corresponding to each fault type is preset, such as 3-5 groups, to avoid too many combinations leading to calculation redundancy, and to ensure the pertinence and effectiveness of the combination.

[0041] An improved time window Apriori algorithm is loaded, which is an optimized version of the traditional Apriori algorithm by introducing a time window constraint, and the near-year fault data of the same type of compressor and the near-year operation data of the current compressor are used as the data source. The data of the same type of compressor ensures the generality of the fault rule, and the current compressor data ensures the individual adaptability of the rule, effectively expands the sample size and improves the pertinence of the rule; introduce a preset minute time window constraint before the fault, according to the precursor evolution period of different fault types, set a time window of a certain time length such as 30 minutes, 1 hour, 2 hours before the fault occurs, focus on the parameter change interval before the fault occurs, and avoid irrelevant time period data interference; the change trend of the parameter combination and the fault triggering association rule are mined through the algorithm, that is, the regularity knowledge of "when a certain group of parameters changes in a certain trend such as continuous rise of current and synchronous rise of temperature, a certain type of fault will be triggered within a preset time window".

[0042] The confidence of the rule is calculated and a threshold is set according to the fault type. The confidence represents the probability of the fault occurrence when the parameter combination meets the rule condition. The confidence is calculated by counting the frequency of the actual fault occurrence after the rule condition appears in the historical data. Different confidence thresholds are set according to the severity and early warning needs of different fault types. For example, the confidence threshold of a fatal fault is set to 90%, and the confidence threshold of a slight fault is set to 70%. At the same time, the rule triggering records and actual fault maintenance records of the compressor in the current month are compared. The proportion of actual faults occurring after each rule triggering is counted. Invalid rules with a matching rate less than the threshold, such as rules with a matching rate less than 50%, are removed. Valid rules are retained to ensure that the final association rules have high reliability and actual early warning value.

[0043] For the parameter combination corresponding to the valid rules, the parameter fault contribution degree is determined according to the proportion of the independent early warning times of the parameter in the corresponding fault in the recent years, and the weight is allocated. For example, in a certain parameter combination, parameter A accounts for 60% of the independent early warning in the corresponding fault, and parameter B accounts for 40%. Then, 0.6 and 0.4 weights are respectively assigned to parameter A and parameter B. The weighted product method is used to calculate the coupling association degree according to the single parameter association degree. The synergistic association effect of the parameter combination is quantified by this method. If the coupling association degree is greater than the maximum value of each single parameter association degree, it means that the synergistic effect of the parameter combination is stronger than that of a single parameter in indicating the fault. At this time, the coupling association degree replaces the original single parameter association degree, and the sensitivity of fault identification is improved. If the coupling association degree is less than the maximum value of the single parameter association degree, it means that the early warning effect of a single parameter is better. The original single parameter association degree is retained to ensure that the final association index has the optimal early warning performance.

[0044] Further, the data source is sorted by timestamp to arrange the historical data of the same type of compressor and the current compressor in the order of data collection time, ensuring the time sequence continuity of the data and providing a basis for subsequent time window division and parameter change trend analysis; the change quantization threshold of each dedicated parameter is set, and the parameter change determination threshold is set according to the physical characteristics and normal fluctuation range of different parameters, such as current change amplitude exceeding 10%, temperature change exceeding 5°C, pressure fluctuation exceeding 0.2 MPa, etc., to clearly define the determination criteria of parameter "rise", "fall" and "stability"; the parameter is converted into a standardized representation of parameter name, change direction and change amplitude, such as "exhaust pressure rises from 0.8 MPa to 1.0 MPa" standardized to "exhaust pressure - rise - 25%", and "motor current falls from 5A to 4.5A" standardized to "motor current - fall - 10%", so that the algorithm can quickly identify and process parameter change characteristics and improve algorithm processing efficiency. The fault type and time window mapping table is called, which is a pre-constructed association table containing different fault types and corresponding basic time windows, stored in the storage module of the electronic starter, and the corresponding time window of each fault type is obtained by searching in the fault type and time window mapping table, such as 30-minute time window for motor stall fault and 1-hour time window for air path leakage fault. The corresponding window is quickly matched by looking up the table to ensure accurate adaptation of time sequence constraints and fault types.

[0045] The transaction set is divided according to the corresponding time window, with the fault occurrence time as the end point, and the corresponding time window is traced back for a certain period of time. All parameter standardized representations in this time period are taken as a transaction set, and each transaction set completely contains the change characteristics of all dedicated parameters in a single window, such as the 30-minute time window of motor stall fault, in which the standardized representations of motor current, temperature, vibration frequency and other parameters in this time period are integrated into a transaction set. Based on the transaction set, the parameter change order constraint is added when generating the frequent item set to form an ordered item set. The frequent item set refers to the parameter combination that appears in the transaction set with a frequency exceeding the set support threshold. The traditional Apriori algorithm does not consider the parameter change order, and this step ensures that the generated ordered item set can reflect the order of parameter change, such as the ordered combination of "current rise → temperature rise", which is more consistent with the actual evolution process of parameters before the fault occurs.

[0046] filtering out the ordered frequent item sets satisfying the threshold from the ordered item set according to the minimum weighted support corresponding to the fault type, the minimum weighted support being a dynamic support threshold set in combination with the timeliness and effectiveness of data and being calculated through the formula of claim 6, different fault types corresponding to different minimum weighted support thresholds;extracting the parameter combination change trend and the fault triggering association rule from the ordered frequent item sets satisfying the threshold, for example, extracting the association rule of "motor current-rising-20%→winding temperature-rising-15%→triggering motor overheating fault within 30 minutes" from the ordered frequent item sets, and completing the mining of the association rule.

[0047] For example, the calculation formula of the minimum weighted support is as follows: ; Among them, N is the total number of transaction sets in the data source, i is the i th occurrence record of the ordered item set, n is the total number of occurrences of the ordered item set, λ is an exponential decay coefficient, is the time length of the i th occurrence record from the current calculation time, is the effective identification of the i th occurrence record.

[0048] In addition, in some optional embodiments of the present application, the dynamic adaptation of the time sequence window is realized through initial mapping table construction and health index correction, and the problem that the fixed time window cannot adapt to the aging state change of the compressor is solved.

[0049] Constructing the initial fault type and time window mapping table, first collecting a large amount of historical fault data of compressors of the same model, for each fault type, counting the period of appearance of the precursor before the fault occurs, i.e. the time interval from the appearance of the first abnormal parameter to the occurrence of the fault, calculating the statistical value of the period such as the average value, the 90% quantile value, etc., taking the statistical value as the basic time sequence window corresponding to each fault type, for example, the average period of appearance of the precursor of motor faults is 1 hour, then 1 hour is set as the basic time sequence window of motor faults, and the corresponding relationship between all fault types and corresponding basic time sequence windows is arranged as the initial fault type and time window mapping table and stored in the system as the basis for subsequent window adjustment.

[0050] The health index of the current running stage of the compressor is monitored in real time. The health index is a quantitative index that comprehensively reflects the current aging state and running stability of the compressor. The health index is calculated by fusing multiple dimensions of running data. Specifically, the cumulative running time of the compressor, i.e., the total running time of the equipment from being put into use to the present, is integrated. The longer the cumulative running time, the lower the health index. The parameter fluctuation range within the last three maintenance intervals, i.e., the fluctuation standard deviation or variation coefficient of key running parameters such as current, temperature, and pressure between the last three equipment maintenance periods, is integrated. The larger the fluctuation range, the more unstable the running, and the lower the health index. The failure rate decay curve of the same batch of models, i.e., the failure rate change curve of the same batch of models as the current compressor with the change of the use time, is integrated. The health of the current compressor is evaluated by correcting the health index through the curve to avoid misjudgment caused by individual differences. The above three dimensions of index are quantified and integrated into the health index by using a weighted summation algorithm. The index range can be set to 0-100. The higher the value, the better the health state of the equipment.

[0051] The basic time sequence window is dynamically corrected according to the health index. A threshold of the health index, such as 60, is preset. When the health index is lower than the preset threshold, it indicates that the compressor is seriously aged, and the evolution period of the failure precursor may be prolonged. At this time, the time sequence window of the corresponding failure type is expanded by a preset ratio, such as 1.2-1.5 times, to ensure that the early appearing failure signs can be captured, and the early warning lag is avoided. When the health index is higher than the preset threshold, it indicates that the compressor is in good condition, and the evolution period of the failure precursor is short. The time sequence window of the corresponding failure type is reduced by a preset ratio, such as 0.7-0.9 times, to reduce the inclusion of irrelevant data and improve the accuracy of rule mining. Through this dynamic correction, the time sequence window can adapt to the failure evolution law of the compressor under different health states.

[0052] Further, on the basis of the health index correction, a secondary calibration step of environmental and load factors is added to further improve the accuracy of the timing window and solve the problem of different fault evolution periods under different working conditions. The actual fault warning record and the time difference between the real fault occurrence time of the compressor are collected. The fault warning record of the compressor in the past period, such as 1-2 years, including the warning trigger time, the warning fault type, and the corresponding real fault occurrence time, is collected. The time difference between each warning trigger and the real fault occurrence is calculated, which reflects the deviation of the timing window and the actual fault evolution period. A correction factor library is established. According to the calculated time difference, combined with the corresponding environmental parameters such as season, environmental temperature, humidity, etc. and the load rate interval such as light load 0-30%, medium load 30%-70%, heavy load 70%-100%, the change rule of the time difference under different environmental and load conditions is analyzed, for example, the warning time difference of motor class faults in high temperature environment in summer is 20% shorter than that in normal temperature environment, and the warning time difference of air path class faults in heavy load working condition is 30% longer than that in heavy load working condition. Based on these rules, the corresponding time window correction coefficient is set for different seasons such as spring, summer, autumn, winter, and different load rate intervals. The correction coefficient is stored in association with the corresponding environment and load condition to build a correction factor library. The correction coefficient is usually represented in decimal form, such as 1.2, 0.8, etc., which is used to adjust the size of the timing window. Combined with the current environmental parameters and the load rate, the environmental temperature, humidity and other environmental parameters of the current compressor are collected in real time by the electronic control device, and the load rate is calculated based on the current load demand signal. In the correction factor library, the corresponding season or environmental interval is determined according to the real-time environmental parameters, and the corresponding load rate interval is determined according to the real-time load rate, and then the correction coefficient that accurately matches the environment and load interval is retrieved. The timing window corrected by the health index is calibrated again. The health corrected timing window value is multiplied by the retrieved correction coefficient to obtain the final dynamic timing window, for example, the health corrected window is 1 hour, and the correction coefficient corresponding to the current environment and load is 1.2. The final dynamic timing window is 1.2 hours. Through secondary calibration, the timing window is adapted to the health status of the compressor, the environmental conditions and the load demand, and is maximized to approach the actual fault evolution period, thereby improving the accuracy of the association rule mining.

[0053] In addition, in some optional embodiments of the present application, a multi-modal fusion network is used to construct the model, which improves the fault identification accuracy and adaptability of the model through hierarchical design and targeted optimization. The multi-modal fusion network is used to construct the fault warning model. The multi-modal fusion network can process different types and dimensions of parameter features at the same time, and has stronger feature extraction and pattern recognition ability than traditional single models. The network specifically includes a parameter feature extraction layer, a timing association layer, and a fault classification layer, which work together to complete the complete inference process from parameter input to fault warning.

[0054] Among them, the parameter feature extraction layer designs exclusive convolution kernels and loop units for different types of parameters in the target operating parameters, including linear feature parameters, nonlinear time series parameters, sequential change parameters, etc. For linear feature parameters such as motor operating current and power supply voltage, ordinary convolution kernels such as 1D convolution kernels are designed to extract local linear features. For nonlinear time series parameters such as air path pressure fluctuations and temperature change curves, loop units such as LSTM and GRU units are designed to capture time series nonlinear features. For sequential change parameters such as control signal transmission timing, a hybrid structure of combined convolution kernels and loop units is designed to extract sequential features. At the same time, the attention mechanism is used to strengthen the feature weights of high-coupling correlation parameter combinations. The attention mechanism calculates the feature importance scores of each parameter combination and gives higher weights to high-coupling correlation parameter combinations, so that the model focuses on parameter combination features that are more indicative of fault warning during feature extraction, improving the relevance and effectiveness of feature extraction.

[0055] The time sequence correlation layer introduces the final dynamic time window as a time dimension constraint, concatenates the parameter change trend features in different dynamic time windows in time sequence order to form time sequence data containing multiple time period features, ensuring that the model can capture the parameter evolution law at different stages before the fault occurs. Combined with the association rules mined by the improved time window Apriori algorithm as prior knowledge, the association rules are converted into model recognizable constraints such as association weights between features, change order constraints, etc., and are integrated into the model training process to improve the reasoning efficiency and accuracy of the model. Through graph neural network modeling of the topological association between parameter combinations and fault types, parameter combinations are taken as graph nodes, and the association between parameter combinations and fault types is taken as graph edges. The graph neural network learns the topological structure features between nodes to accurately capture the complex association between parameter combinations and different fault types, further improving the recognition accuracy of the model for fault types.

[0056] The fault classification layer trains the model by using a weighted loss function, the weighted loss function gives different weights to the misjudgment cost of different fault types, solves the problem that the traditional loss function treats all fault types equally, sets a high penalty weight such as 10 for fatal faults such as motor short circuit, gas explosion and the like, which may cause equipment damage and safety accidents, ensures that the identification accuracy of the model for such faults is improved in priority, and reduces the risk of misjudgment; for slight faults such as slight delay of control signals, small amplitude fluctuation of parameters and the like, which do not affect the normal operation of the equipment, a low penalty weight such as 1 is set, which reduces the model training burden while ensuring the identification effect; and a deviation rate of a dynamic time window is introduced as a regularization term, the deviation rate of the dynamic time window is the difference between the actual fault evolution period and the dynamic time window, the deviation rate is integrated into the loss function, the model is constrained to adapt to the fluctuation of the dynamic time window during the training process, the adaptability of the model to parameter characteristics under different window conditions is improved, the performance of the model is prevented from being reduced due to the change of the window, and finally a fault early warning model with high identification accuracy and strong adaptability is trained.

[0057] In summary, the compressor fault early warning method based on the electronic starting controller in the above embodiment of the present application breaks through the problem of limited monitoring dimension of a single or a small number of sensors, can comprehensively capture early features including hidden faults such as motor insulation aging and bearing wear, solves the problem of early warning lag caused by one-sidedness of data, realizes early identification of fault precursors, and at the same time, the target operating parameters are screened out by combining the historical fault information with the correlation analysis of the operating parameters, and then the fault early warning model is trained based on the target operating parameters, which replaces the traditional alarm logic based on fixed threshold, can accurately analyze the dynamic change trend of the parameters, effectively eliminates interference factors such as instantaneous voltage fluctuation and sudden change of environmental temperature and humidity, reduces the false alarm and missed alarm phenomena, and finally realizes automatic early warning by analyzing the current operating parameters through the trained fault early warning model, comprehensively improves the timeliness, accuracy and intelligent level of the compressor fault early warning. The problem of low accuracy of the compressor fault early warning method in the prior art is solved.

[0058] Embodiment three Please refer to Figure 2 , which shows an electronic starting controller proposed in the third embodiment of the present application, the electronic starting controller comprises: The acquisition module 100 is configured to acquire the operating parameters and the historical fault information of the compressor, the operating parameters including power-on parameters, start-up parameters, normal working parameters, protection parameters and environmental control parameters; The screening module 200 is configured to perform correlation analysis on the operating parameters and the historical fault information, and screen out target operating parameters related to the compressor fault; The training module 300 is configured to train the fault early warning model based on the target operating parameter. The early warning module 400 is configured to acquire the current operating parameter of the compressor, input the current operating parameter into the trained fault early warning model, and automatically perform early warning on the compressor fault through the fault early warning model.

[0059] The functions or operation steps realized when the above modules are executed are generally the same as those of the above method embodiments, and thus will not be described herein.

[0060] Embodiment four In another aspect, the present application also provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, realizes the steps of the method according to any one of the above embodiments one to two.

[0061] Embodiment five In another aspect, the present application also provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor realizes the steps of the method according to any one of the above embodiments one to two when executing the program.

[0062] The technical features of each of the above embodiments can be combined in any manner. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0063] Those skilled in the art can understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a list of executable instructions for realizing the logic function, which can be specifically embodied in any computer readable storage medium for use by or in conjunction with an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from an instruction execution system, device or apparatus. For the present specification, the "computer readable storage medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in conjunction with an instruction execution system, device or apparatus, or in conjunction with these instruction execution systems, devices or apparatus.

[0064] More specific examples (a non-exhaustive list) of the computer-readable storage medium include the following: an electrical connection having one or more wires (electrical device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable storage medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then reproducible in a computer.

[0065] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques can be used to implement the hardware used in the described embodiments: a discrete logic circuit(s) having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), and / or the like.

[0066] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific feature, structure, material or characteristic being described is included in at least one embodiment or example of the application. The illustrative description of the above terms in the specification does not necessarily refer to the same embodiment or example. Moreover, the description of the specific feature, structure, material or characteristic can be combined in any one or more embodiments or examples in a suitable manner.

[0067] The above-described embodiments are merely some embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as limiting the scope of the patent of the present application. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the scope of protection of the patent of the present application should be subject to the appended claims.

Claims

1. A method for early warning of compressor failure based on electronic starting controller, characterized in that, The method comprises: Collecting operation parameters and historical fault information of the compressor, the operation parameters including power-on parameters, start-up parameters, normal working parameters, protection parameters and environmental industrial control parameters; Performing correlation analysis on the operation parameters and the historical fault information to screen out target operation parameters related to compressor faults; Training a fault early warning model based on the target operation parameters; Obtaining current operation parameters of the compressor, inputting the current operation parameters into the trained fault early warning model, and automatically warning the compressor faults through the fault early warning model.

2. The electronic governor based compressor failure warning method of claim 1, wherein, The step of performing correlation analysis on the operation parameters and the historical fault information to screen out target operation parameters related to compressor faults comprises: Synchronously collecting the operation parameters and the historical fault information of the compressor, filtering noise of the operation parameters, and aligning the operation parameters on a time axis according to operation stages; Dividing the historical fault information into different types of faults according to fault causes, and calculating the correlation degrees of corresponding operation parameters by using different correlation analysis algorithms for each type of fault; Obtaining real-time output operation state signals of the compressor, assigning dynamic weights to operation parameters in different operation stages, and screening high-correlation-degree parameters after recalculating the weighted correlation degrees of each operation parameter and corresponding fault types based on the dynamic weights. Collecting normal operation parameter samples of the compressor under different environmental industrial control parameters, verifying the correlation stability of the high-correlation-degree parameters preliminarily screened out under different environmental conditions, and finally screening out operation parameters with correlation stability greater than a threshold value in the full range of working conditions as target operation parameters strongly related to compressor faults.

3. The electronic governor based compressor fault warning method of claim 2, wherein, The fault types include motor faults, gas path faults and control faults, and the step of calculating the correlation degrees of corresponding operation parameters by using different correlation analysis algorithms comprises: For motor faults, calculating the linear correlation degree of linear feature parameters and faults by using a Pearson correlation coefficient; For gas path faults, calculating the nonlinear correlation degree of nonlinear time series parameters and faults by using a mutual information entropy algorithm; For control faults, calculating the sequential correlation degree of sequential change parameters and faults by using a Spearman rank correlation coefficient.

4. The electronic governor based compressor fault warning method of claim 3, wherein, After the step of calculating the correlation degrees of corresponding operation parameters by using different correlation analysis algorithms, the method further comprises: For operation parameters corresponding to each fault type, screening a preset number of parameter combinations according to a fault type and parameter attribute matching principle; Loading an improved time window Apriori algorithm, taking fault data of the same type of compressor in recent years and operation data of the current compressor in recent years as data sources, introducing a preset minute time window constraint before a fault, and mining the change trend of the parameter combinations and the fault triggering association rules; Calculating rule confidence and setting a threshold value according to the fault type, comparing rule triggering records of the current compressor in the recent month with actual fault maintenance records, eliminating invalid rules with an actual fault matching rate less than the threshold value, and retaining valid rules. For the parameter combination corresponding to the effective rule, the parameter fault contribution degree is determined according to the proportion of the number of independent early warnings of the parameter in the corresponding fault in recent years, and a weight is allocated, the coupling correlation is calculated according to the single parameter correlation by using the weighted product method, if the coupling correlation is greater than the maximum value of each single parameter correlation, the coupling correlation is used to replace the original single parameter correlation, if the coupling correlation is less than the maximum value of the single parameter correlation, the original single parameter correlation is retained.

5. The electronic governor based compressor fault warning method of claim 4, wherein, The improved time window Apriori algorithm is loaded, the fault data of the compressor of the same type in recent years and the current compressor operation data in recent years are used as the data source, a preset minute time window constraint before the fault is introduced, and the steps of mining the change trend of the parameter combination and the fault triggering association rule include: The data source is sorted according to the time stamp, the change quantitative threshold of each dedicated parameter is set, and the parameter is converted into the standardized expression of the parameter name, change direction and change amplitude; The fault type and time window mapping table is called, and the time window corresponding to each fault type is obtained in the fault type and time window mapping table; According to the corresponding time window, the transaction set is divided, each transaction set contains the standardized expression of all dedicated parameters in a single window, and based on the transaction set, the parameter change order constraint is increased when generating the frequent item set to form the ordered item set; According to the minimum weighted support degree corresponding to the fault type, the ordered frequent item set that meets the threshold is filtered out from the ordered item set, and the change trend of the parameter combination and the fault triggering association rule are extracted from the ordered frequent item set.

6. The electronic governor based compressor fault warning method of claim 5, wherein, The calculation formula of the minimum weighted support degree is: ； wherein, N is the total number of transaction sets in the data source, i is the first occurrence record of the ordered item set, i is the total number of occurrences of the ordered item set, n is the exponential decay coefficient, The step of calling the fault type and time window mapping table and obtaining the time window corresponding to each fault type in the fault type and time window mapping table includes: is the time length from the collection time of the th occurrence record to the current calculation time, i is the time length from the collection time of the th occurrence record to the current calculation time, i is the valid identification of the th occurrence record.

7. The electronic governor based compressor fault warning method of claim 5, wherein, An initial fault type and time window mapping table is constructed, each fault type in the initial mapping table corresponds to a basic time window, and the basic time window is determined based on the statistical value of the pre-fault symptom appearance period of the historical fault data of the compressor of the same type; The health index of the current operation stage of the compressor is monitored in real time, and the health index is calculated based on the cumulative operation time of the compressor, the parameter fluctuation amplitude within the latest three maintenance intervals and the failure rate decay curve of the same batch of models. According to the health index, the basic time window is dynamically corrected, when the health index is lower than the set threshold, the time window of the corresponding fault type is enlarged by a preset proportion, and when the health index is higher than the set threshold, the time window of the corresponding fault type is reduced by a preset proportion. The step of dynamically correcting the basic time window according to the health index further includes:

8. The electronic governor based compressor fault warning method of claim 7, wherein, The time difference between the actual fault warning record of the compressor and the real fault occurrence time is collected, a correction factor library is established, and the time window correction coefficients corresponding to different seasons and different load rate intervals are stored in the correction factor library; Combined with the current environmental parameters and the load rate, the matching correction coefficient is called from the correction factor library, the time window corrected by the health index is calibrated again, and the final dynamic time window is obtained. The step of training the fault warning model based on the target operation parameter includes:

9. The electronic governor based compressor fault warning method of claim 8, wherein, ​ A multi-modal fusion network is adopted to construct a fault early warning model, and the multi-modal fusion network comprises a parameter feature extraction layer, a time sequence association layer and a fault classification layer. The parameter feature extraction layer is designed with dedicated convolution kernels and loop units for different types of parameters in the target operating parameters, and the feature weights of high coupling degree parameter combinations are strengthened through an attention mechanism. The time sequence association layer introduces a dynamic time window as a time dimension constraint, concatenates the parameter trend features in different windows according to the time sequence, combines the association rules mined by the improved time window Apriori algorithm as priori knowledge, and models the topological association between the parameter combinations and the fault types through a graph neural network. The fault classification layer adopts a weighted loss function to train the model, the weighted loss function gives different weights to the misjudgment cost of different fault types, sets a high penalty weight for fatal faults and a low penalty weight for slight faults, and introduces a deviation rate of a dynamic time window as a regularization term to constrain the adaptability of the model to window fluctuations.

10. An electronic actuator, characterized in that The electronic starting controller-based compressor fault early warning method of any one of claims 1 to 9 comprises: an acquisition module configured to collect operating parameters and historical fault information of the compressor, wherein the operating parameters include power-on parameters, start-up parameters, normal working parameters, protection parameters and environmental control parameters; a screening module configured to perform correlation analysis on the operating parameters and the historical fault information, and screen out target operating parameters related to the compressor fault; a training module configured to train a fault early warning model based on the target operating parameters; and an early warning module configured to acquire current operating parameters of the compressor, input the current operating parameters into the trained fault early warning model, and automatically perform early warning on the compressor fault through the fault early warning model.

Images