Method and system for analyzing device status data applied to automation electrical

Abstract

The application provides a device state data analysis method and system applied to automatic electrical equipment, and the method comprises the following steps: obtaining historical operation data set of a target device; constructing a multi-level state analysis model based on the historical operation data set, wherein the multi-level state analysis model comprises at least one first analysis level and at least one second analysis level; extracting key operation parameters from a real-time operation data stream of the target device through an operation parameter screening strategy in the first analysis level, and generating a parameter screening result according to the key operation parameters; inputting the parameter screening result into a state evaluation strategy in the second analysis level, combining a dynamic threshold range corresponding to the key operation parameters, and generating a state evaluation index of the target device; and outputting a real-time operation state grade of the target device according to a comparison result of the state evaluation index and a preset state threshold. The application can realize efficient monitoring and accurate early warning of the health state of the device.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a device state data analysis method and system applied to automatic electrical equipment. BACKGROUND

[0002] The state data analysis of automatic electrical equipment is one of the cores in the field of industrial operation and maintenance, aiming to identify potential faults and optimize maintenance strategies through real-time monitoring of equipment operation parameters. In the prior art, the device state is usually directly evaluated based on a preset static threshold, such as selecting key parameters through fixed rules or manual experience, comparing the selected parameters with a constant threshold to determine the equipment state level, or using a general machine learning model to train all parameters without distinction to generate a state classification result. However, the static threshold cannot adapt to the dynamic parameter change characteristics of the equipment under different running stages and load conditions, resulting in normal fluctuations being misjudged as abnormal states or real abnormalities being missed. Secondly, the parameter screening and state evaluation links are highly coupled, and a hierarchical processing mechanism is not established, which makes the evaluation accuracy disturbed by redundant parameters and the model response speed affected by high-dimensional data processing pressure. Thirdly, the traditional method lacks consideration of the time sequence correlation and weight dynamic allocation among parameters, and cannot accurately capture the key feature changes in the equipment degradation process, resulting in a lack of foresight and accuracy in maintenance decisions, which easily leads to over-maintenance or maintenance lag under complex working conditions, seriously affecting the equipment operation efficiency and increasing the operation and maintenance cost. SUMMARY

[0003] Therefore, the present application provides a device state data analysis method and system applied to automatic electrical equipment. The technical solution of the present application is realized as follows:

[0004] In one aspect, the present application provides a device state data analysis method applied to automatic electrical equipment, the method comprising: obtaining a historical operation data set of a target device, the historical operation data set containing historical data sequences of multiple device operation parameters and corresponding device state labels; constructing a multi-level state analysis model based on the historical operation data set, the multi-level state analysis model containing at least one first analysis level and at least one second analysis level, wherein each first analysis level corresponds to an operation parameter screening strategy, each second analysis level corresponds to a state evaluation strategy, and the second analysis level is after the first analysis level; extracting key operation parameters from a real-time operation data stream of the target device through the operation parameter screening strategy in the first analysis level, and generating a parameter screening result according to the key operation parameters; inputting the parameter screening result into the state evaluation strategy in the second analysis level, combining the dynamic threshold range corresponding to the key operation parameters, and generating a state evaluation index of the target device; and outputting a real-time operation state level of the target device according to the comparison result of the state evaluation index and a preset state threshold.

[0005] In another aspect, the present application provides a data analysis system comprising a memory and a processor, the memory storing a computer program executable on the processor, and the processor implementing the steps of the above method when executing the program.

[0006] The application provides a device state data analysis method applied to automatic electrical equipment, which screens key operation parameters through a first analysis level in a multi-level state analysis model and generates parameter screening results, combines a state evaluation strategy based on a dynamic threshold range in a second analysis level to generate a state evaluation index of the target device, and finally outputs a real-time operation state grade by comparing a preset state threshold, which can effectively solve the problems of low calculation efficiency and high misjudgment rate caused by poor adaptability of static thresholds and parameter redundancy in traditional device state analysis; wherein, the first analysis level dynamically extracts key operation parameters and assigns weights through parameter correlation screening conditions and stability indexes, ensures that the parameter screening results input into the second analysis level focus on core data that significantly affect the device state, and reduces irrelevant parameter interference, and the second analysis level dynamically adjusts threshold boundaries in combination with historical operation data distribution intervals and real-time operation stages, so that the state evaluation index can accurately reflect the degradation trend of the device under different operation environments; through the sequential dependence and data cooperation mechanism of the first analysis level and the second analysis level, a closed-loop logic chain from parameter screening to state evaluation is realized, and the reliability and real-time performance of the analysis results are improved; the comprehensive deviation score generated based on the dynamic threshold range and the state abnormality grade can accurately locate the potential fault risk point of the device and quantify the abnormality degree, provide multi-dimensional data support for maintenance decision, and finally realize automatic output of the real-time operation state grade, significantly reduce the artificial inspection cost and shorten the fault response time, and avoid the problems of over-maintenance or insufficient maintenance caused by fixed thresholds, and realize efficient monitoring and accurate early warning of the device health state in complex industrial scenes. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 An implementation process schematic diagram of a device state data analysis method applied to automatic electrical equipment is provided for the embodiments of the application.

[0008] Figure 2 A hardware entity schematic diagram of a data analysis system is provided for the embodiments of the application. DETAILED DESCRIPTION

[0009] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The term "first / second / third" referred to is only to distinguish similar objects, and does not represent a specific order of the objects, and it can be understood that "first / second / third" can be interchanged in a specific order or sequence as allowed, so that the embodiments of the application described herein can be implemented in an order other than that illustrated or described herein.

[0010] The embodiment of the present application provides a device state data analysis method applied to automatic electrical equipment, which can be executed by a processor of a data analysis system.

[0011] Figure 1 An implementation flow diagram of the device state data analysis method applied to automatic electrical equipment provided by the embodiment of the present application is shown in Figure 1 The method comprises the following steps.

[0012] Step S100: Obtain a historical running data set of a target device, wherein the historical running data set comprises historical data sequences of multiple device running parameters and corresponding device state labels.

[0013] The device running parameter refers to various physical quantities or indexes that can reflect the working state of the device during the running process of the device, for example, voltage, current, temperature, rotating speed, etc. The historical data sequence is a data set recorded in time sequence in the past period of time. The device state label is to classify and identify the running state of the device at each time point, for example, normal, abnormal, fault, etc.

[0014] In the embodiment of the present application, obtaining the historical running data set of the target device is the basis for subsequent device state analysis. Various sensors installed on the device can be used to collect device running parameters in real time, and these data are stored in the database. With the passage of time, these data form the historical running data set. For example, for a motor, current sensor, temperature sensor and rotating speed sensor can be installed to collect current, temperature and rotating speed data during the running process, and the corresponding device state label is labeled for each time point, such as normal running, overload running, fault, etc.

[0015] Step S200: Construct a multi-level state analysis model based on the historical running data set, wherein the multi-level state analysis model comprises at least one first analysis level and at least one second analysis level, wherein each first analysis level corresponds to a running parameter screening strategy, each second analysis level corresponds to a state evaluation strategy, and the second analysis level is after the first analysis level.

[0016] The running parameter screening strategy refers to a method for screening key running parameters that have an impact on device state analysis from a large number of device running parameters. The state evaluation strategy is a method for evaluating and judging the current state of the device according to the screened key running parameters.

[0017] The first analysis level can remove parameters that have less impact on the device state by running a parameter screening strategy, reduce the complexity of data processing, and improve analysis efficiency. The second analysis level uses a state evaluation strategy to evaluate the device state in combination with key operating parameters to obtain more accurate state evaluation results.

[0018] For example, the first analysis level can use a correlation analysis method to screen key operating parameters, calculate the correlation between each operating parameter and the device state label, and select parameters with high correlation as key operating parameters. The second analysis level can use any feasible machine learning model, such as a support vector machine, a neural network, etc., to classify and evaluate the device state according to the key operating parameters.

[0019] Step S300: Extract key operating parameters from the real-time operating data stream of the target device through the operating parameter screening strategy in the first analysis level, and generate a parameter screening result according to the key operating parameters.

[0020] The real-time operating data stream refers to the device operating parameter data generated by the target device in the current running process. The key operating parameters are parameters that are important for analyzing the device state and are screened from the real-time operating data stream. The parameter screening result is related information containing the key operating parameters, such as real-time measurement values of the parameters, parameter weights, etc.

[0021] In this step, the real-time operating data stream is processed by the operating parameter screening strategy, which can quickly and accurately extract the key operating parameters. These key operating parameters can more effectively reflect the current running state of the device and provide an important basis for subsequent state evaluation.

[0022] For example, for a transformer in a power system, the real-time operating data stream contains multiple parameters such as voltage, current, oil temperature, winding temperature, etc. Through the operating parameter screening strategy of the first analysis level, it is found that the voltage and oil temperature have a higher correlation with the state of the transformer, so the voltage and oil temperature are extracted as key operating parameters, and the parameter screening result is generated according to their real-time measurement values and related calculations.

[0023] As an implementation, step S300 can specifically include the following steps:

[0024] Step S310: Extract the operating parameter sequence in the current time window from the real-time operating data stream, and the operating parameter sequence contains real-time measurement values of multiple device operating parameters.

[0025] The time window refers to a time period range selected from the real-time operating data stream. The operating parameter sequence is a sequence composed of real-time measurement values of multiple device operating parameters collected in the time window in chronological order.

[0026] In practical applications, since the real-time running data stream is continuously generated, in order to facilitate analysis and processing, it is necessary to intercept data in a suitable time window. For example, the time window can be set to 1 minute, and the real-time measurement values of the voltage, current, temperature and other device running parameters in the last 1 minute are intercepted from the real-time running data stream to form a running parameter sequence.

[0027] Step S320: Based on the parameter correlation screening condition in the first analysis level, the association degree coefficient of each running parameter in the plurality of device running parameters and the device state label in the historical running data set is calculated.

[0028] The parameter correlation screening condition is a rule or standard pre-set for evaluating the correlation between the running parameter and the device state label. The association degree coefficient is a numerical value for measuring the association degree between each running parameter and the device state label. The greater the association degree coefficient, the closer the association between the running parameter and the device state.

[0029] In this step, by calculating the association degree coefficient, the influence degree of each running parameter on the device state can be quantitatively evaluated. For example, Pearson correlation coefficient, mutual information and other methods can be used to calculate the association degree coefficient. For example, for an industrial robot, its running parameters include joint angle, motor current, load torque, etc. By calculating the association degree coefficient of these parameters and the device state label (such as normal, fault, etc.) in the historical running data set, it can be determined which parameters are more important for the state evaluation of the robot.

[0030] As an implementation manner, step S320 can specifically include the following steps:

[0031] Step S321: Extracting data distribution features of each device running parameter under different device state labels from the historical running data set, the data distribution features including parameter mean, variance and distribution shape index.

[0032] The parameter mean refers to the average value of all historical data of the device running parameter under a certain device state label. The variance is a statistical quantity for measuring the degree of data dispersion, reflecting the fluctuation of data relative to the mean. The distribution shape index is used to describe the shape of the data distribution, such as skewness, kurtosis, etc.

[0033] In this step, extracting data distribution features helps to understand the variation law of each device running parameter under different device states. For example, for the spindle speed parameter of a numerical control machine tool, under normal running state, its mean value may be stable around a set value, and the variance is small; while under fault state, the mean value may change significantly, and the variance may also increase. By analyzing these data distribution features, important basis can be provided for subsequent calculation of the association degree coefficient.

[0034] Step S322: Calculate the discriminant value of each device operating parameter between different device state labels according to the data distribution characteristics. The discriminant value is determined based on the combination of parameter mean difference and variance ratio.

[0035] The discriminant value is an indicator for measuring the discrimination ability of a device operating parameter between different device state labels. The parameter mean difference reflects the gap between the average values of the parameter under different device states, and the variance ratio considers the degree of dispersion of the data. By combining the parameter mean difference and the variance ratio, a more comprehensive discriminant value can be obtained.

[0036] In actual calculation, the discriminant value can be calculated using the following formula: discriminant value = (mean1 - mean2) / (variance1 + variance2), where mean1 and mean2 are the means of the parameter under two different device state labels, and variance1 and variance2 are the corresponding variances. For example, for the flow parameter of a water pump, by calculating its discriminant value under normal operation and fault operation, it can be determined whether the parameter can effectively distinguish between the two states.

[0037] Step S323: Obtain the parameter change trend curve of each device operating parameter in the historical operation data set, and extract the number of peak points and the position of trend inflection points in the parameter change trend curve.

[0038] The parameter change trend curve is a curve formed by connecting the historical data of the device operating parameter in chronological order, which reflects the change trend of the parameter over time. The peak point is the point of local maximum value in the curve, and the trend inflection point is the point where the slope of the curve changes.

[0039] Extracting the number of peak points and the position of trend inflection points helps to find abnormal changes in device operating parameters. For example, for the power parameter of a wind turbine, a sudden increase in the number of peak points or abnormal movement of the trend inflection point in the parameter change trend curve may indicate that the turbine has failed or has been affected by external environment.

[0040] Step S324: Generate the correlation coefficient combining the discriminant value, the number of peak points and the position of trend inflection points, wherein the correlation coefficient is positively correlated with the discriminant value and the number of peak points, and negatively correlated with the dispersion of the position of trend inflection points.

[0041] The generation of the correlation coefficient comprehensively considers multiple factors such as the discriminant value, the number of peak points, and the position of the trend inflection point. The greater the discriminant value, the stronger the distinguishing ability of the parameter between different equipment states, and the greater the influence of the equipment state, so the correlation coefficient is positively correlated with the discriminant value. The more the number of peak points, the more intense the change of the parameter, and the closer the association with the equipment state, so the correlation coefficient is positively correlated with the number of peak points. The smaller the dispersion of the position of the trend inflection point, the more stable the change trend of the parameter, and the higher the correlation with the equipment state, so the correlation coefficient is negatively correlated with the dispersion of the position of the trend inflection point. In actual calculation, a weighted summation method can be used to generate the correlation coefficient: correlation coefficient = a*discriminant value + b*number of peak points - c*dispersion of the position of the trend inflection point, where a, b, and c are pre-set weight coefficients.

[0042] Step S325: Normalizing the correlation coefficient and dynamically matching and verifying the normalized correlation coefficient with the time sequence change of the equipment state label in the historical operation data set.

[0043] The normalization processing is to convert the correlation coefficient to a pre-set interval, such as [0, 1], so as to facilitate comparison and analysis between different parameters. The dynamic matching verification is to compare the normalized correlation coefficient with the time sequence change of the equipment state label in the historical operation data set, and check whether the correlation coefficient can accurately reflect the change of the equipment state.

[0044] In this step, the normalization processing can use common normalization methods such as minimum-maximum normalization, Z-score normalization, etc. Through dynamic matching verification, the accuracy and reliability of the correlation coefficient can be verified. For example, if the normalized correlation coefficient can be adjusted in time when the equipment state changes, it indicates that the correlation coefficient can better reflect the change of the equipment state, and has high accuracy and reliability.

[0045] Step S330: From the multiple equipment operation parameters, a candidate operation parameter with a correlation coefficient greater than a pre-set correlation threshold is selected, and a parameter stability index is determined according to the fluctuation amplitude of the real-time measurement value of the candidate operation parameter.

[0046] The pre-set correlation threshold is a pre-set critical value for judging whether the correlation degree of an operation parameter with the equipment state is strong enough. The candidate operation parameter refers to an operation parameter with a correlation coefficient greater than the pre-set correlation threshold. The parameter stability index is an index for measuring the fluctuation of the real-time measurement value of the candidate operation parameter, and the smaller the fluctuation amplitude, the higher the parameter stability.

[0047] In this step, by screening the candidate running parameters, the amount of data for subsequent processing can be reduced, and the analysis efficiency can be improved. According to the fluctuation amplitude of the real-time measurement value of the candidate running parameter, the parameter stability index is determined, which helps to understand the stability of these parameters, and provides an important basis for subsequent parameter sorting and selection of key running parameters. For example, for a plurality of running parameters collected by sensors in an automatic production line, candidate running parameters with a correlation coefficient greater than a preset correlation threshold are screened out, and then the fluctuation amplitude of the real-time measurement value of these candidate running parameters is calculated to determine their parameter stability index.

[0048] Step S340: Sort the candidate running parameters according to the parameter stability index, and select a preset number of candidate running parameters in the front as the key running parameters.

[0049] The preset number is the number of key running parameters to be selected. By sorting the candidate running parameters according to the parameter stability index, parameters with higher stability can be preferentially selected as key running parameters.

[0050] In practical applications, parameters with higher stability can more accurately reflect the running state of the equipment and reduce errors caused by parameter fluctuations. For example, for a plurality of monitoring parameters in a power system, after sorting the candidate running parameters according to the parameter stability index, the top 5 candidate running parameters are selected as the key running parameters for subsequent device state analysis.

[0051] Step S350: Generate a parameter screening result containing parameter weight distribution based on the real-time measurement value of the key running parameters, and the parameter weight is dynamically adjusted according to the correlation coefficient and the parameter stability index.

[0052] The parameter weight refers to the proportion of each key running parameter in the parameter screening result, which is used to reflect the importance of the parameter to the device state analysis. Dynamic adjustment refers to real-time adjustment of the parameter weight according to the changes of the correlation coefficient and the parameter stability index.

[0053] In this step, by generating a parameter screening result containing parameter weight distribution, the influence of each key running parameter on the device state can be more accurately considered. The parameter weight is dynamically adjusted according to the correlation coefficient and the parameter stability index, which can make the parameter screening result more consistent with the actual running situation of the equipment. For example, for key running parameters in an industrial control system, the parameter weight is dynamically adjusted according to the correlation coefficient and the parameter stability index, and a parameter screening result containing different weights is generated for subsequent state evaluation.

[0054] As an implementation manner, step S350 can specifically include the following steps:

[0055] Step S351: Calculate the real-time offset according to the real-time measurement value of the key operating parameter and the historical mean value of the corresponding parameter in the historical operating data set.

[0056] The real-time offset is the difference between the real-time measurement value of the key operating parameter and the mean value of the parameter in the historical operating data set. It reflects the degree of deviation of the current parameter value from the historical average level.

[0057] In this step, calculating the real-time offset helps to discover abnormal changes in the key operating parameters. For example, for a temperature parameter of an air conditioning system, the historical mean value is 25°C, and the current real-time measurement value is 28°C, then the real-time offset is 3°C. By monitoring the real-time offset in real time, whether the temperature parameter has abnormal fluctuations can be found in time.

[0058] Step S352: Determine the real-time abnormal probability of each key operating parameter based on the real-time offset and the parameter stability index.

[0059] The real-time abnormal probability refers to the possibility of the occurrence of an abnormality in a key operating parameter at the current time. It is calculated based on the real-time offset and the parameter stability index.

[0060] In actual calculation, a probability model can be used to determine the real-time abnormal probability. For example, when the real-time offset is large and the parameter stability index is low, it indicates that the parameter has a high probability of abnormality, and the real-time abnormal probability will also increase accordingly. For a speed parameter of an elevator, if the real-time offset exceeds the preset range and the parameter stability is poor, the real-time abnormal probability of the speed parameter will increase.

[0061] Step S353: Generate a dynamic weight distribution ratio according to the real-time abnormal probability and the correlation coefficient, wherein the dynamic weight distribution ratio is positively related to the product of the real-time abnormal probability and the correlation coefficient.

[0062] The dynamic weight distribution ratio refers to the specific weight proportion of each key operating parameter in the parameter screening result. It is positively related to the product of the real-time abnormal probability and the correlation coefficient, that is, the larger the real-time abnormal probability and the correlation coefficient, the larger the dynamic weight distribution ratio.

[0063] In this step, by generating the dynamic weight distribution ratio, the weights of the key operating parameters can be more reasonably allocated, and the parameter screening result can more accurately reflect the operating state of the equipment. For example, for multiple key operating parameters in a sewage treatment system, the dynamic weight distribution ratio is generated according to their real-time abnormal probability and correlation coefficient, so that the parameters with higher abnormal probability and higher correlation degree with the equipment state have higher weights in the parameter screening result.

[0064] Step S354: Weighted fusion of real-time measurement values of key operating parameters with dynamic weight distribution ratios to generate parameter screening results containing weighted parameters.

[0065] Weighted fusion refers to multiplying real-time measurement values of key operating parameters by corresponding dynamic weight distribution ratios, and then adding these products to obtain a comprehensive result. Parameter screening results containing weighted parameters can more comprehensively reflect the influence of each key operating parameter on the device state.

[0066] In practical applications, weighted fusion can highlight the role of key operating parameters with larger weights, while reducing the influence of parameters with smaller weights. For example, for key operating parameters collected by multiple sensors in a smart home system, their real-time measurement values are weighted and fused with dynamic weight distribution ratios to generate parameter screening results containing weighted parameters, which are used for subsequent device state evaluation.

[0067] Step S355: Generating parameter screening codes according to weighted parameter values in parameter screening results, which are used to identify the priority order of different key operating parameters in the second analysis level.

[0068] Parameter screening codes are a way to encode key operating parameters, which determine the priority order of different key operating parameters in the second analysis level according to weighted parameter values in parameter screening results.

[0069] In this step, generating parameter screening codes helps more efficiently process key operating parameters in the second analysis level. For example, for multiple key operating parameters in an automated warehouse system, generating parameter screening codes according to their weighted parameter values allows these parameters to be processed in priority order in the second analysis level, improving the efficiency and accuracy of state evaluation.

[0070] Step S400: Inputting parameter screening results into state evaluation strategies in the second analysis level, combining dynamic threshold ranges corresponding to key operating parameters, to generate state evaluation indicators of the target device.

[0071] State evaluation strategies are methods for evaluating and judging the operating state of a device based on input parameter screening results. Dynamic threshold ranges refer to reasonable value ranges of key operating parameters under different device operating stages and conditions, which will be adaptively adjusted with changes in device running time and environment, etc. State evaluation indicators are a set of indicators for comprehensively reflecting the current operating state of the target device, such as abnormality level, abnormality duration, etc.

[0072] In this step, the parameter screening result is input into the state evaluation strategy and analyzed in combination with the dynamic threshold range, so that the running state of the target device can be more accurately evaluated. For example, for a wind turbine, the parameter screening result is input into the state evaluation strategy, and the dynamic threshold range of key operating parameters such as generator temperature and speed is combined to generate the state evaluation index of the wind turbine, such as whether there is an abnormality, the severity of the abnormality, etc.

[0073] As an implementation, step S400 can specifically include the following steps:

[0074] Step S410: Analyzing the real-time measurement value of the key operating parameter and the corresponding parameter weight distribution proportion from the parameter screening result.

[0075] In the parameter screening result, the real-time measurement value of the key operating parameter and the corresponding weight distribution proportion of each parameter are included. These information is analyzed to accurately use these data for subsequent state evaluation.

[0076] In this step, by analyzing the parameter screening result, the real-time measurement value of the key operating parameter and the parameter weight distribution proportion can be extracted. For example, for the parameter screening result of a chemical production device, the real-time measurement value of the key operating parameters such as temperature and pressure and their corresponding parameter weight distribution proportion are analyzed to provide basic data for subsequent state evaluation.

[0077] Step S420: According to the parameter distribution interval of different device state labels in the historical operation data set, the dynamic threshold range of the key operating parameter is determined, and the dynamic threshold range is adaptively adjusted with the device running time.

[0078] The parameter distribution interval refers to the value range of the key operating parameter under different device state labels. The dynamic threshold range is determined according to the parameter distribution interval and is adaptively adjusted with the change of the device running time to adapt to the dynamic change of the device running state.

[0079] In this step, by analyzing the parameter distribution interval of different device state labels in the historical operation data set, the reasonable value range of the key operating parameter under different states can be determined. With the increase of the device running time, the performance and running state of the device may change, so the dynamic threshold range also needs to be adjusted accordingly. For example, for the battery parameters of an electric vehicle, according to the parameter distribution interval of different charging states and driving states in the historical operation data set, the dynamic threshold range of the key operating parameters such as battery voltage and current is determined and adaptively adjusted with the increase of the battery usage time and the number of charging and discharging.

[0080] As an implementation, in step S420, determining the dynamic threshold range of the key operating parameter can specifically include the following steps:

[0081] Step S421: Extracting the historical maximum and minimum values of the key operating parameter in different equipment operating stages from the historical operating data set.

[0082] The historical maximum and minimum values refer to the maximum and minimum values of the key operating parameter in different equipment operating stages in the historical operating data set. They reflect the value range of the parameter in the historical operating process.

[0083] In this step, extracting the historical maximum and minimum values of the key operating parameter helps to understand the change range of the parameter in different operating stages. For example, for the temperature parameter of a steel smelting furnace, the historical maximum and minimum values in the melting stage, refining stage and other different operating stages are extracted from the historical operating data set, providing a reference for subsequent determination of the dynamic threshold range.

[0084] Step S422: Calculating the initial threshold range according to the historical maximum and minimum values, and determining the current operating stage based on the real-time operating time of the target equipment.

[0085] The initial threshold range is a preliminary threshold range calculated from the historical maximum and minimum values, which provides a basis for subsequent dynamic correction. The current operating stage is determined according to the real-time operating time of the target equipment and the operating law of the equipment. Different operating stages may correspond to different parameter value ranges.

[0086] In actual calculation, the initial threshold range can be taken as the interval between the historical maximum and minimum values. For example, for the water level parameter of an industrial boiler, the initial threshold range is calculated as [water level minimum value, water level maximum value] according to the historical maximum and minimum values. Then, according to the real-time operating time of the boiler, it is determined whether it is in the startup stage, normal operating stage or shutdown stage, etc.

[0087] Step S423: Obtaining the sliding average and sliding variance of the real-time measurement value of the key operating parameter in the current operating stage.

[0088] The sliding average refers to the average value of the real-time measurement value of the key operating parameter in a sliding window in the current operating stage. The sliding variance is a statistical quantity that measures the dispersion degree of these real-time measurement values in the sliding window.

[0089] In this step, calculating the moving average and the moving variance helps to understand the real-time changes of the key operating parameters within the current operating stage. For example, for the feed speed parameter of a numerical control machine tool, by calculating its moving average and moving variance within the current operating stage, the stability and fluctuation of the parameter can be judged.

[0090] Step S424: dynamically correct the initial threshold range according to the moving average and the moving variance, and generate a dynamic threshold range containing a correction offset.

[0091] The correction offset refers to the amount of adjustment needed for the upper and lower limits of the threshold range when dynamically correcting the initial threshold range. By dynamically correcting the initial threshold range according to the moving average and the moving variance, the dynamic threshold range can be more in line with the current operating state of the device.

[0092] In this step, the process of dynamic correction can adjust the upper and lower limits of the initial threshold range according to the changes of the moving average and the moving variance. For example, if the moving average changes significantly, it means that the average level of the parameter has changed, and the threshold range needs to be adjusted accordingly; if the moving variance increases, it means that the fluctuation of the parameter has intensified, and the threshold range also needs to be adjusted appropriately.

[0093] As an implementation, step S424 can specifically include the following steps:

[0094] Step S4241: calculate the difference between the moving average and the median of the initial threshold range as the mean offset.

[0095] The mean offset refers to the difference between the moving average and the median of the initial threshold range, which reflects the deviation of the current parameter average level from the initial threshold range.

[0096] In this step, calculating the mean offset helps to determine the direction and size of the adjustment needed for the initial threshold range. For example, for the pressure parameter of a refrigeration system, calculating the difference between its moving average and the median of the initial threshold range, if the difference is positive, it means that the average level of the current pressure is higher than the median of the initial threshold range, and the threshold range needs to be adjusted accordingly.

[0097] Step S4242: determine the variance correction coefficient according to the ratio of the moving variance to the historical variance.

[0098] The variance correction coefficient is a coefficient calculated according to the ratio of the moving variance to the historical variance, which is used to adjust the width of the threshold range to adapt to the changes in parameter fluctuation.

[0099] In this step, by comparing the size of the sliding variance and the historical variance, it can be judged whether the fluctuation of the parameter has changed. If the sliding variance is greater than the historical variance, it means that the fluctuation of the parameter is intensified, and the variance correction coefficient needs to be increased, so as to expand the threshold range; on the contrary, if the sliding variance is less than the historical variance, it means that the fluctuation of the parameter is reduced, and the variance correction coefficient needs to be reduced, so as to narrow the threshold range.

[0100] Step S4243: multiplying the mean offset and the variance correction coefficient to generate a dynamic correction amount.

[0101] The dynamic correction amount is a value calculated according to the mean offset and the variance correction coefficient, which is used to adjust the upper limit value and the lower limit value of the initial threshold range.

[0102] In this step, the mean offset and the variance correction coefficient are multiplied to consider the deviation of the parameter average level and the change of the fluctuation, and a reasonable dynamic correction amount is obtained. For example, for the oil temperature parameter of a power transformer, the dynamic correction amount is obtained by multiplying the calculated mean offset and the variance correction coefficient, which is used to adjust the dynamic threshold range of the oil temperature.

[0103] Step S4244: adding the dynamic correction amount to the upper limit value and the lower limit value of the initial threshold range respectively to generate the corrected dynamic threshold range.

[0104] In this step, the dynamic correction amount is added to the upper limit value and the lower limit value of the initial threshold range respectively, and the corrected dynamic threshold range is obtained. For example, for the speed parameter of an elevator, the initial threshold range is [speed lower limit, speed upper limit], and the dynamic correction amount is Δv, then the corrected dynamic threshold range is [speed lower limit + Δv, speed upper limit + Δv].

[0105] Step S4245: boundary constraint processing is performed on the corrected dynamic threshold range to ensure that the upper limit value of the dynamic threshold range does not exceed the preset percentage of the historical maximum value, and the lower limit value does not fall below the preset percentage of the historical minimum value.

[0106] The boundary constraint processing is to ensure that the dynamic threshold range does not exceed a reasonable range. The preset percentage is a proportion value set in advance, which is used to limit the upper limit value and the lower limit value of the dynamic threshold range.

[0107] In this step, through the boundary constraint processing, the rationality and reliability of the dynamic threshold range can be guaranteed. For example, for the temperature parameter of an automatic production line, the preset percentage of the historical maximum value is set to 110%, and the preset percentage of the historical minimum value is set to 90%, and the boundary constraint processing is performed on the corrected dynamic threshold range, so that the upper limit value does not exceed 110% of the historical maximum value, and the lower limit value does not fall below 90% of the historical minimum value.

[0108] Step S425: Matching verification is performed between the dynamic threshold range and the historical change rate of the key operating parameter, so that the change rate of the dynamic threshold range does not exceed a preset rate limit.

[0109] The historical change rate refers to the change speed of the key operating parameter in the historical operation process. The preset rate limit is a critical value preset to limit the change speed of the dynamic threshold range.

[0110] In this step, by matching verification between the dynamic threshold range and the historical change rate of the key operating parameter, it can be avoided that the change of the dynamic threshold range is too violent, and the stability and reliability of the dynamic threshold range are ensured. For example, for the pressure parameter of a chemical reaction kettle, the change rate of the dynamic threshold range is compared with the historical change rate of the pressure parameter, to ensure that the change rate of the dynamic threshold range does not exceed the preset rate limit.

[0111] Step S430: The deviation degree of the real-time measurement value from the upper limit value and the lower limit value of the dynamic threshold range is calculated, and a comprehensive deviation score is generated in combination with the parameter weight distribution proportion.

[0112] The deviation degree refers to the absolute value of the difference between the real-time measurement value of the key operating parameter and the upper limit value or the lower limit value of the dynamic threshold range. The comprehensive deviation score is a score value calculated according to the deviation degree and the parameter weight distribution proportion, which is used to reflect the abnormality degree of the key operating parameter.

[0113] In this step, by calculating the comprehensive deviation score, the abnormality of the key operating parameter can be more comprehensively evaluated. For example, for the voltage parameter of a generator, the deviation degree of the real-time measurement value from the upper limit value and the lower limit value of the dynamic threshold range is calculated, and then the comprehensive deviation score is generated in combination with the weight distribution proportion of the parameter. If the comprehensive deviation score is high, it means that the abnormality degree of the voltage parameter is large.

[0114] Step S440: According to the comparison result of the comprehensive deviation score and the preset deviation threshold, the state abnormality level corresponding to the key operating parameter is determined.

[0115] The preset deviation threshold is a critical value preset to judge whether the abnormality degree of the key operating parameter reaches the set level. The state abnormality level is divided into different levels according to the comparison result of the comprehensive deviation score and the preset deviation threshold, such as mild abnormality, moderate abnormality, severe abnormality, etc.

[0116] In this step, by comparing the comprehensive deviation score with the preset deviation threshold, the state abnormality level corresponding to the key operating parameter can be accurately determined. For example, for a flow parameter of a water pump, if the comprehensive deviation score is greater than the preset deviation threshold and the deviation degree is large, it is determined that the state abnormality level corresponding to the flow parameter is severe abnormality.

[0117] Step S450: integrating the state abnormality levels of all key operating parameters to generate an overall state evaluation index of the target device, the overall state evaluation index including an abnormality level distribution and an abnormality duration.

[0118] The abnormality level distribution refers to the distribution of the state abnormality levels corresponding to the key operating parameters, which reflects the abnormality degree of the target device in different aspects. The abnormality duration refers to the duration of the target device in the abnormal state.

[0119] In this step, by integrating the state abnormality levels of all key operating parameters, the overall state evaluation index of the target device can be obtained. For example, for a production line of an automated factory, by integrating the state abnormality levels of various key operating parameters (such as temperature, pressure, speed, etc.), the overall state evaluation index of the production line is generated, including the abnormality level distribution (such as which parameters are in mild abnormality, which parameters are in moderate abnormality, etc.) and the abnormality duration, so as to comprehensively understand the running state of the production line.

[0120] Step S500: outputting the real-time running state level of the target device according to the comparison result of the state evaluation index and the preset state threshold.

[0121] The preset state threshold is a set of critical values preset for dividing the state evaluation index into different levels. The real-time running state level is determined according to the comparison result of the state evaluation index and the preset state threshold, which reflects the current running state of the target device, such as normal, warning, fault, etc.

[0122] In this step, by comparing the state evaluation index with the preset state threshold, the real-time running state level of the target device can be accurately output. For example, for a certain wind turbine of a wind farm, by comparing its state evaluation index (such as abnormality level distribution, abnormality duration, etc.) with the preset state threshold, if the state evaluation index is within the normal range, the real-time running state level is output as normal; if it exceeds the warning threshold but does not reach the fault threshold, the real-time running state level is output as warning; if it exceeds the fault threshold, the real-time running state level is output as fault.

[0123] As an implementation manner, the method provided by the embodiment of the present application can further include:

[0124] Step S210: Add a third analysis level in the multi-level state analysis model, the third analysis level is after the second analysis level.

[0125] The third analysis level is a newly added analysis level in the multi-level state analysis model, which is used for further analysis and prediction of the state of the equipment. Adding the third analysis level after the second analysis level can use the state evaluation index output by the second analysis level for more in-depth analysis.

[0126] In this step, adding the third analysis level can enhance the function of the multi-level state analysis model, so that it can provide more comprehensive equipment state information. For example, for a large data center server, adding a third analysis level based on the original first and second analysis levels can predict the future state of the server and discover potential hidden faults in advance.

[0127] Step S220: Predict the future state evolution path of the target equipment according to the historical trend of the state evaluation index through the state prediction strategy in the third analysis level.

[0128] The state prediction strategy is a method for predicting the future state of the target equipment according to the historical trend of the state evaluation index. The future state evolution path refers to the state change process that the target equipment may experience in the future.

[0129] In this step, the state prediction strategy can use historical data to predict the future state of the target equipment, providing a reference for the maintenance and management of the equipment. For example, for a bridge structural health monitoring system, through the state prediction strategy in the third analysis level, the structural safety state evolution path of the bridge in the future can be predicted according to the historical trend of the bridge state evaluation index, and appropriate maintenance measures can be taken in advance.

[0130] As an implementation, step S220 can specifically include the following steps:

[0131] Step S221: Extract time series data from the state evaluation index, the time series data containing the state evaluation index value of the target equipment at different time points and the real-time measurement value of the corresponding key operating parameters.

[0132] Time series data is a set of data arranged in chronological order, which reflects the changes of state evaluation index values and real-time measurement values of key operating parameters of the target equipment over time.

[0133] In this step, time series data is extracted for subsequent state prediction using these data. For example, for an aero-engine state evaluation index, the state evaluation index values at different time points in the past period and the corresponding real-time measurement values of key operating parameters (such as temperature, pressure, etc.) are extracted to form time series data.

[0134] Step S222: Multi-scale decomposition of time series data by state prediction strategy in the third analysis level to generate decomposed time series data containing long-term trend component, periodic fluctuation component and short-term noise component.

[0135] Multi-scale decomposition is a method of decomposing time series data into different scale components. The long-term trend component reflects the long-term change trend of the time series data, the periodic fluctuation component reflects the periodic change characteristics of the data, and the short-term noise component is the random fluctuation part of the data.

[0136] In this step, different features in the time series data can be separated by multi-scale decomposition, which facilitates subsequent analysis and processing of different components. For example, for the load prediction problem of a power system, multi-scale decomposition is performed on the time series data of the load to obtain long-term trend component, periodic fluctuation component and short-term noise component. Analysis and prediction of these components can improve the accuracy of prediction.

[0137] Step S223: Identify the overall degradation direction of the target device based on the long-term trend component, and extract the operating state change period of the target device in combination with the periodic fluctuation component.

[0138] The overall degradation direction refers to the direction in which the performance of the target device gradually decreases during long-term operation. The operating state change period refers to the time interval in which the operating state of the target device changes periodically according to a regular pattern.

[0139] In this step, the overall degradation of the target device can be understood by analyzing the long-term trend component, and the change period of the device operating state can be extracted in combination with the periodic fluctuation component. For example, for vibration monitoring data of a mechanical device, it is found that the vibration amplitude of the device gradually increases by analyzing the long-term trend component, indicating that the device has an overall degradation trend. The change period of the device vibration can be extracted in combination with the periodic fluctuation component to understand the periodic change rule of the device operating state.

[0140] Step S224: Generate a prediction sub-model set according to the overall degradation direction and the operating state change period, and the prediction sub-model set contains multiple prediction sub-models matched with different degradation stages and periodic fluctuation characteristics.

[0141] The set of prediction sub-models is a group of sub-models for predicting the future state of the target device, each of which is designed for different degradation stages and periodic fluctuation characteristics.

[0142] In this step, the set of prediction sub-models is generated according to the overall degradation direction and the running state change period, which can improve the accuracy and relevance of state prediction. For example, for the battery life prediction problem of an electric vehicle, according to the overall degradation direction and the charge-discharge period fluctuation characteristics of the battery, a plurality of prediction sub-models containing different degradation stages and periodic fluctuation characteristics are generated, such as early degradation stage sub-model, mid-term degradation stage sub-model, etc.

[0143] Step S225: input the decomposed time series data into each prediction sub-model in the set of prediction sub-models, generate future state prediction results under different confidence levels, and cross-validate the future state prediction results.

[0144] Cross-validation is a method for evaluating the accuracy and reliability of prediction models, which divides the data set into multiple subsets and uses different subsets for training and validation in turn.

[0145] In this step, the decomposed time series data is input into each prediction sub-model to generate future state prediction results under different confidence levels, and cross-validation is performed, which can filter out the most accurate and reliable prediction results. For example, for the quality prediction problem of an industrial production process, the decomposed time series data is input into each prediction sub-model in the set of prediction sub-models to generate product quality prediction results under different confidence levels, and then the accuracy of these results is evaluated through cross-validation to select the optimal prediction result.

[0146] Step S226: According to the cross-validation results, the target prediction sub-model with the smallest deviation from the latest state evaluation index value in the real-time running data stream of the target device is selected.

[0147] The target prediction sub-model is the sub-model selected from the set of prediction sub-models that is most suitable for the current state prediction of the target device, and it has the smallest deviation from the latest state evaluation index value in the real-time running data stream of the target device.

[0148] In this step, by selecting the target prediction sub-model, the accuracy of state prediction can be improved. For example, for the performance prediction problem of a communication base station, according to the cross-validation results, the target prediction sub-model with the smallest deviation from the latest state evaluation index value in the real-time running data stream of the base station is selected for subsequent base station performance prediction.

[0149] Step S227: The target prediction sub-model is used to correct the trend of the short-term noise component in the time series data, and the future state evolution path containing the corrected noise influence is generated.

[0150] Trend correction refers to processing the short-term noise component to remove random fluctuations therein and make it more consistent with the overall trend of change. The future state evolution path containing the corrected noise effect is the path of the future state change of the target device obtained after considering the correction effect of the short-term noise component.

[0151] In this step, by trend correction of the short-term noise component, the future state evolution path can be made smoother and more accurate. For example, for a weather forecasting problem, by trend correction of the short-term noise component in the meteorological data through the target prediction sub-model, a future weather change path containing the corrected noise effect is generated, improving the accuracy of weather forecasting.

[0152] Step S228: Correlate and match the future state evolution path with the dynamic threshold range of the key operating parameters in the real-time operating data stream, and identify the abnormal prediction interval in the future state evolution path that exceeds the dynamic threshold range.

[0153] Correlation and matching refers to comparing the future state evolution path with the dynamic threshold range of the key operating parameters to find the part of the future state evolution path that exceeds the dynamic threshold range. The abnormal prediction interval refers to the time period in the future state evolution path that exceeds the dynamic threshold range.

[0154] In this step, by correlation and matching, the time period when the target device may appear abnormal can be found in advance, providing a basis for preventive maintenance. For example, for a chemical production device, by correlating and matching its future state evolution path with the dynamic threshold range of temperature, pressure and other key operating parameters, the abnormal prediction interval in the future state evolution path that exceeds the dynamic threshold range of temperature or pressure is identified, and measures are taken in advance to avoid accidents.

[0155] Step S229: Based on the start time point, end time point and corresponding key operating parameter type of the abnormal prediction interval, generate a preventive maintenance suggestion containing maintenance trigger conditions and maintenance parameter range.

[0156] Maintenance trigger conditions refer to the conditions under which device maintenance is required, which are usually related to the start time point of the abnormal prediction interval and the abnormal condition of the key operating parameters. The maintenance parameter range refers to the reasonable value range of the parameters that need to be adjusted when the device is maintained.

[0157] In this step, preventive maintenance recommendations are generated based on the start time point, end time point of the abnormal prediction interval and the corresponding key operating parameter type, which can plan the maintenance work of the equipment in advance and improve the reliability and service life of the equipment. For example, for a power transformer, based on the start time point of the abnormal prediction interval and the corresponding temperature parameter abnormality, a preventive maintenance recommendation is generated, which includes maintenance trigger conditions (such as maintenance when the temperature exceeds a certain threshold) and maintenance parameter range (such as the temperature range for replacing cooling oil).

[0158] Step S230: Extracting key turning points in the future state evolution path and generating preventive maintenance recommendations in combination with real-time measurement values of key operating parameters.

[0159] The key turning point refers to a point in the future state evolution path where the state changes sharply, which usually indicates an important change in the state of the equipment. In combination with the real-time measurement values of the key operating parameters, the preventive maintenance recommendations can be more in line with the actual operation of the equipment.

[0160] In this step, key turning points are extracted and preventive maintenance recommendations are generated in combination with real-time measurement values, which can more accurately arrange the maintenance work of the equipment and avoid unnecessary maintenance and failure. For example, for an aircraft engine, key turning points in its future state evolution path are extracted, and in combination with real-time measurement values of temperature, pressure and other key operating parameters of the engine, targeted preventive maintenance recommendations are generated, such as engine overhaul and debugging before the key turning point.

[0161] As an implementation, step S230 can specifically include the following steps:

[0162] Step S231: Identifying an interval segment in the future state evolution path where the trend change rate of the predicted state curve exceeds a preset rate threshold, and extracting the start time point and end time point of the interval segment as initial turning points.

[0163] The predicted state curve is a graphical representation of the future state evolution path, which reflects the change of the future state of the target equipment over time. The trend change rate refers to the slope of the predicted state curve at a certain point, which indicates the speed of state change. The preset rate threshold is a critical value set in advance to determine whether the trend change rate is too large. The initial turning point refers to the start time point and end time point of the interval segment where the trend change rate of the predicted state curve exceeds the preset rate threshold.

[0164] In this step, by identifying the interval segment of the predicted state curve whose trend change rate exceeds the preset rate threshold, the part of the future state evolution path where the state changes more dramatically can be found, and the starting time point and the ending time point thereof are extracted as initial turning points. For example, for a running state prediction curve of an elevator, the interval segment of the curve whose trend change rate exceeds the preset rate threshold is identified, and the starting time point and the ending time point of the interval segment are extracted as initial turning points, which may indicate important changes in the running state of the elevator.

[0165] Step S232: Based on the fluctuation amplitude and the fluctuation direction of the predicted state curve between the initial turning points, the trend correlation degree between adjacent initial turning points is calculated, and the initial turning points whose trend correlation degree is less than the preset correlation threshold are selected as candidate turning points.

[0166] The trend correlation degree is an index for measuring the similarity of the predicted state curve between adjacent initial turning points, which is calculated according to the fluctuation amplitude and the fluctuation direction. The preset correlation threshold is a critical value preset for judging whether the trend correlation degree is small enough. The candidate turning points refer to the initial turning points whose trend correlation degree is less than the preset correlation threshold, and the state changes between these turning points may be more independent and important.

[0167] In this step, by calculating the trend correlation degree between adjacent initial turning points, the initial turning points whose trend correlation degree is less than the preset correlation threshold are selected as candidate turning points, which can further narrow the range of key turning points. For example, for a motion state prediction curve of an industrial robot, the trend correlation degree between adjacent initial turning points is calculated, and the initial turning points whose trend correlation degree is less than the preset correlation threshold are selected as candidate turning points, which may correspond to important changes in the motion state of the robot.

[0168] Step S233: According to the distribution density of the candidate turning points in the future state evolution path and the time interval between adjacent candidate turning points, the second derivative change feature of the predicted state curve is fused to determine the candidate turning points that meet the density condition and the time interval condition as key turning points.

[0169] The distribution density refers to the distribution density of the candidate turning points in the future state evolution path. The time interval refers to the time distance between adjacent candidate turning points. The second derivative change feature reflects the curvature change of the predicted state curve. The density condition and the time interval condition are preset standards for screening out the real key turning points.

[0170] In this step, by comprehensively considering the distribution density of candidate turning points, time interval and the second derivative variation characteristics of the predicted state curve, the candidate turning points meeting the conditions are determined as key turning points, which can more accurately find the most important state change point in the future state evolution path. For example, for a power prediction curve of a wind turbine generator, according to the distribution density of candidate turning points and the time interval of adjacent candidate turning points, combined with the second derivative variation characteristics of the curve, the candidate turning points meeting the density condition and time interval condition are determined as key turning points, which may correspond to the fault occurrence point or performance change point of the wind turbine generator.

[0171] Step S234: Obtain the real-time measurement value of the key operating parameter in the target time window corresponding to the key turning point, and extract the abnormal number and abnormal duration of the real-time measurement value exceeding the dynamic threshold range in the target time window.

[0172] The target time window refers to a time period centered on the key turning point, which is used to obtain the real-time measurement value of the key operating parameter. The abnormal number refers to the number of times that the real-time measurement value of the key operating parameter exceeds the dynamic threshold range in the target time window. The abnormal duration refers to the total duration of the key operating parameter in the abnormal state.

[0173] In this step, the real-time measurement value of the key operating parameter in the target time window corresponding to the key turning point is obtained, and the abnormal number and abnormal duration are extracted, which can understand the abnormal situation of the key operating parameter near the key turning point. For example, for the temperature parameter of a chemical reaction kettle, the temperature real-time measurement value in the target time window corresponding to the key turning point is obtained, and the abnormal number and abnormal duration of the temperature exceeding the dynamic threshold range in the time window are extracted. These information can provide basis for judging the running state of the reaction kettle and taking corresponding maintenance measures.

[0174] Step S235: Determine the state abnormality level of the key operating parameter at the key turning point according to the abnormal number and abnormal duration, and generate a maintenance type identifier combined with the trend change rate of the key turning point.

[0175] The state abnormality level is divided into different levels according to the abnormal number and abnormal duration, such as mild abnormality, moderate abnormality, severe abnormality, etc. The maintenance type identifier is a code used to identify the type of maintenance needed, which is generated according to the state abnormality level of the key operating parameter and the trend change rate of the key turning point.

[0176] In this step, the state abnormality level is determined according to the number of abnormalities and the duration of abnormalities, and the maintenance type identifier is generated in combination with the trend change rate, which can provide clear guidance for subsequent maintenance work. For example, for the oil temperature parameter of a power transformer, the state abnormality level is determined to be severe abnormality according to the number of abnormalities and the duration of abnormalities of the oil temperature near the key turning point, and the maintenance type identifier is generated in combination with the trend change rate of the key turning point, such as "emergency repair", indicating that the transformer needs to be repaired immediately.

[0177] Step S236: Match the maintenance action sequence in the preset maintenance strategy library based on the maintenance type identifier, and prioritize the maintenance action sequence according to the state abnormality level.

[0178] The preset maintenance strategy library is a database containing maintenance action sequences corresponding to various maintenance types. The maintenance action sequence refers to a series of maintenance actions needed to complete a certain type of maintenance. The priority ranking is to rank the maintenance action sequences according to the state abnormality level, so that the maintenance action sequence with higher state abnormality level has higher priority.

[0179] In this step, the maintenance action sequence in the preset maintenance strategy library is matched based on the maintenance type identifier, and the priority ranking is performed according to the state abnormality level, which can ensure that appropriate maintenance measures can be taken in time when the equipment appears abnormal. For example, for a device failure of an automated production line, the corresponding maintenance action sequence is matched from the maintenance strategy library according to the maintenance type identifier, such as replacing parts, adjusting parameters, etc., and then the priority ranking of these maintenance action sequences is performed according to the state abnormality level, and the maintenance action sequence with high state abnormality level is executed first.

[0180] Step S237: Generate a preventive maintenance suggestion containing maintenance type, maintenance time window and maintenance action execution order according to the time distribution characteristics of the key turning point and the priority ranking of the maintenance action sequence.

[0181] The time distribution characteristics refer to the distribution of the key turning points on the time axis, such as interval time, frequency of occurrence, etc. The maintenance time window refers to the appropriate time period for maintenance work. The maintenance action execution order refers to the execution order of each maintenance action in the maintenance action sequence.

[0182] In this step, according to the time distribution characteristics of the key turning points and the priority ranking of the maintenance action sequence, the preventive maintenance suggestion is generated, which can reasonably arrange the maintenance work of the equipment and improve the maintenance efficiency and the reliability of the equipment. For example, for a server equipment of a large data center, according to the time distribution characteristics of the key turning points and the priority ranking of the maintenance action sequence, a preventive maintenance suggestion containing the maintenance type (such as regular inspection, fault repair, etc.), the maintenance time window (such as the idle time of the weekend) and the maintenance action execution order (first check the hardware, then update the software, etc.) is generated.

[0183] As an implementation manner, the method provided by the embodiment of the application can further include the following steps after step S500:

[0184] Step S600: setting a feedback optimization mechanism in the multi-level state analysis model, the feedback optimization mechanism iteratively updating the operation parameter screening strategy and the state evaluation strategy according to the output result of the real-time running state level.

[0185] The feedback optimization mechanism is a mechanism for optimizing the multi-level state analysis model, which adjusts the operation parameter screening strategy and the state evaluation strategy according to the output result of the real-time running state level to improve the accuracy and adaptability of the model.

[0186] In this step, setting the feedback optimization mechanism can enable the multi-level state analysis model to continuously learn and improve to adapt to the changes of the equipment running state. For example, for a device state analysis model of a smart grid, through the feedback optimization mechanism, the operation parameter screening strategy is adjusted according to the output result of the real-time running state level to select more appropriate key operation parameters, and the state evaluation strategy is updated to improve the accuracy of state evaluation.

[0187] Step S700: obtaining the matching error between the real-time running state level and the actual equipment maintenance record, and calculating the model optimization weight according to the matching error.

[0188] The matching error refers to the difference between the real-time running state level and the actual equipment maintenance record. The model optimization weight is a weight value calculated according to the matching error, which is used to adjust the updating degree of the operation parameter screening strategy and the state evaluation strategy.

[0189] In this step, the matching error is obtained and the model optimization weight is calculated, which can quantify the inaccuracy of the model and provide a basis for the optimization of the model. For example, for a state analysis model of an industrial equipment, by comparing the real-time running state level with the actual equipment maintenance record, the matching error is calculated, and then the model optimization weight is calculated according to the matching error. When the matching error is large, the model optimization weight is increased to strengthen the updating of the operation parameter screening strategy and the state evaluation strategy.

[0190] Step S800: Adjust the parameter correlation screening condition in the first analysis level and the dynamic threshold range generation rule in the second analysis level based on the model optimization weight.

[0191] The parameter correlation screening condition is a condition used to screen key operating parameters in the first analysis level, and the dynamic threshold range generation rule is a rule used to determine the dynamic threshold range of the key operating parameters in the second analysis level. Adjusting these conditions and rules based on the model optimization weight can make the multi-level state analysis model more accurately screen key operating parameters and determine dynamic threshold ranges.

[0192] In this step, adjusting the parameter correlation screening condition and the dynamic threshold range generation rule based on the model optimization weight can gradually improve the performance of the model. For example, for a state analysis model of an elevator equipment, adjusting the parameter correlation screening condition in the first analysis level based on the model optimization weight selects more relevant operating parameters as key operating parameters; adjusting the dynamic threshold range generation rule in the second analysis level makes the dynamic threshold range more consistent with the actual operation of the elevator.

[0193] Step S900: Apply the adjusted parameter correlation screening condition and dynamic threshold range generation rule to the newly received real-time operation data stream to generate an updated multi-level state analysis model.

[0194] In this step, the adjusted parameter correlation screening condition and dynamic threshold range generation rule are applied to the newly received real-time operation data stream to generate an updated multi-level state analysis model through processing and analysis of new data. For example, for a state analysis model of a sewage treatment plant, the adjusted parameter correlation screening condition and dynamic threshold range generation rule are applied to the newly received real-time operation data stream to re-screen and determine the threshold of sewage flow, water quality and other parameters, and generate an updated multi-level state analysis model.

[0195] Step S1000: Backtracking verification of the historical operation data set is performed through the updated multi-level state analysis model to ensure that the prediction accuracy of the updated multi-level state analysis model is not less than a preset accuracy threshold.

[0196] Backtracking verification refers to using the updated multi-level state analysis model to analyze and predict the historical operation data set, and comparing the prediction results with the actual equipment state label to verify the accuracy of the model. The preset accuracy threshold is a critical value set in advance to measure whether the prediction accuracy of the model meets the requirements.

[0197] In this step, the performance of the updated multi-level state analysis model can be ensured to be improved through backtracking verification. For example, for a state analysis model of an aerospace equipment, the updated model is used to perform backtracking verification on a historical operation data set, and it is checked whether the prediction accuracy of the model is not lower than a preset accuracy threshold. If the requirement is not met, the parameters and rules of the model are continuously adjusted until the requirement is met.

[0198] As an implementation manner, the method provided by the embodiment of the present application can further include:

[0199] Step S1100: dynamically adjusting the boundary value range of the preset state threshold based on the historical distribution characteristics of the real-time operation state level.

[0200] The historical distribution characteristics of the real-time operation state level refer to the frequency and distribution interval of different real-time operation state levels in the past period of time. Dynamically adjusting the boundary value range of the preset state threshold refers to adjusting the upper limit value and the lower limit value of the preset state threshold according to the historical distribution characteristics, so as to make it more consistent with the actual operation of the equipment.

[0201] In this step, dynamically adjusting the boundary value range of the preset state threshold based on the historical distribution characteristics of the real-time operation state level can improve the accuracy of state evaluation. For example, for a device state analysis of a solar photovoltaic power generation system, according to the historical distribution characteristics of the real-time operation state level, it is found that the frequency of a certain state level is too high or too low, which indicates that the boundary value range of the preset state threshold may be unreasonable. At this time, the boundary value range of the preset state threshold is dynamically adjusted, so that the state evaluation more accurately reflects the actual operation state of the equipment.

[0202] Step S1200: obtaining a plurality of real-time operation state levels output by the target equipment in a preset time period, and counting the state evaluation index value distribution interval corresponding to each level.

[0203] The preset time period is a time range preset for obtaining the real-time operation state level of the target equipment. The state evaluation index value distribution interval refers to the value range of the state evaluation index under each real-time operation state level.

[0204] In this step, the plurality of real-time operation state levels of the target equipment in the preset time period are obtained, and the state evaluation index value distribution interval corresponding to each level is counted, so that the distribution law of the state evaluation index under different state levels can be understood. For example, for a device state analysis of a smart building, a plurality of real-time operation state levels output by the building equipment in a month are obtained, and the distribution interval of the temperature, humidity and other state evaluation index values corresponding to each level is counted, thereby providing data support for subsequent adjustment of the preset state threshold.

[0205] Step S1300: Match the state evaluation index value distribution interval with the parameter abnormal event occurrence frequency under the same device state label in the historical operation data set to determine the sensitive threshold interval that needs to be adjusted in the preset state threshold.

[0206] The parameter abnormal event occurrence frequency refers to the number of parameter abnormalities under the same device state label in the historical operation data set. The sensitive threshold interval refers to the interval in the preset state threshold that has a higher correlation with the parameter abnormal event occurrence frequency. Adjustment of these intervals may have a greater impact on the accuracy of state evaluation.

[0207] In this step, the state evaluation index value distribution interval is matched with the parameter abnormal event occurrence frequency to determine the sensitive threshold interval, which can be used to adjust the preset state threshold. For example, for the state analysis of an industrial robot, the state evaluation index value distribution interval is matched with the parameter abnormal event occurrence frequency under the same device state label in the historical operation data set, and it is found that the parameter abnormal event occurrence frequency is higher in a certain interval of a state evaluation index. This interval is the sensitive threshold interval, and the preset state threshold in this interval needs to be adjusted first.

[0208] Step S1400: Generate threshold offset correction coefficients for different real-time running state levels according to the fluctuation characteristics of the state evaluation index value in the sensitive threshold interval.

[0209] The fluctuation characteristics refer to the change amplitude and frequency of the state evaluation index value in the sensitive threshold interval. The threshold offset correction coefficient is a coefficient used to adjust the boundary value of the preset state threshold. It is generated according to the fluctuation characteristics of the state evaluation index value in the sensitive threshold interval, and different real-time running state levels may correspond to different threshold offset correction coefficients.

[0210] In this step, the threshold offset correction coefficient is generated according to the fluctuation characteristics of the state evaluation index value in the sensitive threshold interval, which can more accurately adjust the preset state threshold. For example, for the device state analysis of a power system, the threshold offset correction coefficients for different real-time running state levels are generated according to the fluctuation characteristics of the voltage, current, and other state evaluation index values in the sensitive threshold interval. When the state evaluation index value fluctuates greatly, the threshold offset correction coefficient is increased to expand the boundary value range of the preset state threshold.

[0211] Step S1500: Perform superposition operation on the threshold offset correction coefficient and the current boundary value of the preset state threshold to generate a dynamically adjusted preset state threshold.

[0212] In this step, the threshold offset correction coefficient is superimposed with the current boundary value of the preset state threshold to obtain a dynamically adjusted preset state threshold. For example, for the device state analysis of a communication base station, the current boundary value of the preset state threshold is [threshold lower limit, threshold upper limit], the threshold offset correction coefficient is Δ, and the dynamically adjusted preset state threshold is [threshold lower limit + Δ, threshold upper limit + Δ].

[0213] Step S1600: Reclassify the subsequent input real-time running state level according to the dynamically adjusted preset state threshold, and output the updated real-time running state level and the corresponding threshold adjustment log.

[0214] Reclassification refers to redividing and determining the subsequent input real-time running state level according to the dynamically adjusted preset state threshold. The threshold adjustment log is a log file recording the adjustment process and results of the preset state threshold, which contains information such as the time of adjustment, the parameters of adjustment, and the amplitude of adjustment.

[0215] In this step, reclassifying the real-time running state level according to the dynamically adjusted preset state threshold can make the state evaluation more accurately reflect the actual running state of the device. At the same time, outputting the threshold adjustment log can facilitate subsequent review and analysis of the threshold adjustment process. For example, for the device state analysis of an automated production line, reclassify the subsequent input real-time running state level according to the dynamically adjusted preset state threshold, output the updated real-time running state level and the corresponding threshold adjustment log, so as to timely discover the changes of the device state and evaluate the effect of threshold adjustment.

[0216] Step S1700: Associate and verify the threshold adjustment log with the running and maintenance records of the target device, and filter out abnormal threshold intervals in the dynamically adjusted preset state threshold that do not match the actual device failure events.

[0217] Association and verification refers to comparing the threshold adjustment log with the running and maintenance records of the target device to check whether the dynamically adjusted preset state threshold matches the actual device failure events. Abnormal threshold intervals refer to intervals in the dynamically adjusted preset state threshold that do not match the actual device failure events, which may lead to inaccurate state evaluation.

[0218] In this step, abnormal threshold intervals are filtered out through association and verification, which can further optimize the preset state threshold. For example, for the device state analysis of a wind power plant, associate and verify the threshold adjustment log with the running and maintenance records of the wind turbine, and find that the dynamically adjusted preset state threshold of a certain state evaluation indicator does not match the actual device failure events in a certain interval. This interval is the abnormal threshold interval, and the threshold in this interval needs to be adjusted again.

[0219] Step S1800: Based on the adjustment time node of the abnormal threshold interval and the corresponding state evaluation index value, the calculation logic of the reverse correction threshold offset correction coefficient is generated, and the preset state threshold is generated after secondary correction.

[0220] The calculation logic of the reverse correction threshold offset correction coefficient refers to adjusting and improving the method of generating the threshold offset correction coefficient according to the adjustment time node of the abnormal threshold interval and the corresponding state evaluation index value. The preset state threshold after secondary correction is a more accurate preset state threshold obtained after the calculation logic of the reverse correction threshold offset correction coefficient.

[0221] In this step, based on the adjustment time node of the abnormal threshold interval and the corresponding state evaluation index value, the calculation logic of the reverse correction threshold offset correction coefficient is generated, and the preset state threshold after secondary correction is generated, which can improve the accuracy and reliability of the preset state threshold. For example, for the state analysis of a chemical production equipment, according to the adjustment time node of the abnormal threshold interval and the corresponding temperature, pressure and other state evaluation index values, the calculation logic of the reverse correction threshold offset correction coefficient is generated, and the preset state threshold after secondary correction is generated, so that the state evaluation is more in line with the actual operation of the equipment.

[0222] Step S1900: The real-time running state level is reclassified again by using the preset state threshold after secondary correction, and the final calibrated real-time running state level and the calibration parameter set are output.

[0223] The secondary reclassification refers to redividing and determining the real-time running state level according to the preset state threshold after secondary correction. The calibration parameter set is a set containing the preset state threshold after secondary correction and other related calibration parameters, which records the final calibration result.

[0224] In this step, the final calibrated real-time running state level and the calibration parameter set are output by secondary reclassification, which can ensure that the accuracy of state evaluation reaches a high level. For example, for a large data center server equipment, the real-time running state level is reclassified again by using the preset state threshold after secondary correction, and the final calibrated real-time running state level such as normal, warning, failure, etc. is output, and the preset state threshold after secondary correction and related calibration parameters (such as adjustment coefficient, boundary value, etc.) are sorted into the calibration parameter set. This not only provides accurate equipment state information for the operation and maintenance personnel of the data center, but also provides a reliable basis for subsequent equipment management and maintenance decision-making.

[0225] As an implementation manner, the method provided by the embodiment of the present application can further include:

[0226] Step S2000: generating a device maintenance trigger instruction sequence according to the final calibrated real-time running state level.

[0227] The device maintenance trigger instruction sequence is a set of instructions for triggering device maintenance operations. Different final calibrated real-time running state levels correspond to different maintenance needs, so the corresponding maintenance trigger instructions can be generated according to these levels. For example, when the real-time running state level is "normal", the generated instruction may be regular inspection; when the level is "warning", the instruction may be to perform preliminary inspection and parameter adjustment of the device; when the level is "failure", the instruction may be to immediately shut down for repair, etc. In actual application, for a mechanical device on an automated production line, according to its final calibrated real-time running state level, the system will generate an instruction sequence containing specific maintenance operations and execution sequence, such as first performing power-off operation of the device, then performing fault diagnosis, etc.

[0228] Step S2100: extracting the real-time running state level duration corresponding to each instruction in the device maintenance trigger instruction sequence and the time interval between adjacent instructions.

[0229] The real-time running state level duration refers to the duration of the device in a certain real-time running state level, and the time interval between adjacent instructions refers to the time difference between the execution of two adjacent instructions in the device maintenance trigger instruction sequence. Extracting these information helps to reasonably arrange the time and sequence of maintenance work. For example, in a power system, for the maintenance trigger instruction sequence of a generator, extracting the real-time running state level duration corresponding to each instruction can determine the running stability of the generator in different states; extracting the time interval between adjacent instructions can optimize the maintenance plan to avoid too concentrated or too long interval of maintenance work.

[0230] For the extraction of the real-time running state level duration, the historical data of the device state monitoring system can be analyzed. The device state monitoring system will record the real-time running state level of the device at different time points, and through the arrangement and statistics of these records, the duration of each state level can be obtained. For example, for the elevator equipment of a smart building, its state monitoring system will record the running state of the elevator (normal, failure, etc.) in real time, and through the analysis of these records, the duration of the elevator in the failure state can be determined to evaluate the severity and impact range of the failure.

[0231] Step S2200: matching the maintenance action execution sequence in the preset maintenance strategy library based on the duration and time interval to generate a preliminary maintenance strategy scheme.

[0232] The preset maintenance strategy library is a database containing various device maintenance strategies and action execution sequences, which are classified and stored according to the type, state and maintenance requirements of the device. Based on the extracted real-time running state level duration and the time interval between adjacent instructions, the maintenance action execution sequence matching the same is searched in the preset maintenance strategy library, so as to generate a preliminary maintenance strategy scheme. For example, for a wind turbine generator, according to the real-time running state level duration and the time interval between adjacent instructions, the corresponding maintenance strategy is found in the preset maintenance strategy library, such as when the device is in a pre-warning state and the duration is long, the maintenance action sequence of first performing device inspection and then performing component debugging is executed, and a preliminary maintenance strategy scheme is generated.

[0233] Step S2300: Perform conflict detection on the execution time window of each maintenance action in the preliminary maintenance strategy scheme and the production plan schedule of the target device, and screen out the to-be-adjusted maintenance actions with time conflicts.

[0234] The execution time window refers to the time range in which each maintenance action can be executed, and the production plan schedule of the target device specifies the production tasks and operation arrangements of the device at different time periods. Conflict detection can avoid conflicts between maintenance work and production tasks, and ensure the normal operation of the device and the smooth execution of the production plan. For example, for the production line equipment of an automobile manufacturing factory, the execution time window of a certain maintenance action in the preliminary maintenance strategy scheme overlaps with the production peak period of the production line, and through conflict detection, this time conflict can be found, and the maintenance action is marked as a to-be-adjusted maintenance action.

[0235] Step S2400: Reassign the execution priority of the maintenance action according to the conflict type of the to-be-adjusted maintenance action and the emergency degree of the real-time running state level.

[0236] The conflict type can be divided into time conflict, resource conflict, etc., and the emergency degree of the real-time running state level reflects the severity of the device problem. According to these factors, the execution priority of the maintenance action is re-assigned, which can ensure that the maintenance work can be carried out efficiently and orderly. For example, for a substation device of a power system, a certain to-be-adjusted maintenance action needs to be rearranged due to time conflict, and the real-time running state level of the device is "failure", with a high degree of emergency, so the execution priority of the maintenance action should be correspondingly improved and arranged for execution in priority.

[0237] Step S2500: Reorganize the maintenance action execution sequence in the preliminary maintenance strategy scheme based on the re-assigned priority, and generate an optimized dynamic maintenance strategy.

[0238] According to the re-distributed maintenance action execution priority, the order of maintenance actions in the preliminary maintenance strategy scheme is adjusted and reorganized, so that the maintenance work can be more reasonable and efficient. For example, for a device of an automated warehouse system, after re-distributing the execution priority of the maintenance actions, the maintenance actions with high urgency are arranged in advance, and the related maintenance actions are reasonably combined to generate an optimized dynamic maintenance strategy.

[0239] Step S2600: Split the optimized dynamic maintenance strategy into multiple independently executable maintenance sub-tasks, and assign corresponding state evaluation index monitoring conditions to each maintenance sub-task.

[0240] Splitting the optimized dynamic maintenance strategy into multiple independently executable maintenance sub-tasks facilitates the operation and management of maintenance personnel. Assigning corresponding state evaluation index monitoring conditions to each maintenance sub-task can monitor the execution of the maintenance sub-tasks and the state changes of the equipment in real time. For example, for the maintenance of a ship power system, the optimized dynamic maintenance strategy is split into engine maintenance, oil line inspection, electrical system maintenance, and other maintenance sub-tasks, and corresponding state evaluation index monitoring conditions such as engine speed, oil temperature, and voltage are assigned to each sub-task.

[0241] Step S2700: Real-time monitor whether the state evaluation index monitoring condition meets the preset maintenance task activation threshold value, and trigger the execution instruction of the corresponding maintenance sub-task when it meets.

[0242] The preset maintenance task activation threshold value is a critical value set in advance to determine whether a maintenance sub-task needs to be executed. Real-time monitor the state evaluation index monitoring condition, and when the index value reaches or exceeds the preset maintenance task activation threshold value, trigger the execution instruction of the corresponding maintenance sub-task to ensure that the maintenance work can be carried out in a timely manner. For example, for the maintenance of an air conditioning system, real-time monitor the refrigeration efficiency of the air conditioner, and when the refrigeration efficiency is lower than the preset maintenance task activation threshold value, trigger the execution instruction of the maintenance sub-task of cleaning and debugging the air conditioner.

[0243] Step S2800: Continuously collect the key operating parameter change data of the target equipment during the execution of the maintenance sub-task, and generate a maintenance effect evaluation index.

[0244] During the execution of the maintenance sub-tasks, the key operating parameter change data of the target equipment, such as temperature, pressure, and rotating speed, are continuously collected. Through the analysis and processing of these data, the maintenance effect evaluation indexes are generated. These indexes can reflect the execution effect of the maintenance sub-tasks and the state improvement of the equipment. For example, for a machine tool maintenance sub-task, the machining accuracy and vibration frequency of the machine tool are continuously collected during the maintenance process, and the machining accuracy improvement rate and vibration frequency reduction rate are generated as maintenance effect evaluation indexes.

[0245] Step S2900: According to the difference value between the maintenance effect evaluation index and the expected maintenance target, the trigger condition and execution sequence of the unexecuted maintenance sub-task are adjusted.

[0246] The difference value between the maintenance effect evaluation index and the expected maintenance target reflects the gap between the execution effect of the maintenance sub-task and the expectation. According to this difference value, the trigger condition and execution sequence of the unexecuted maintenance sub-task are adjusted, so that the maintenance work can better meet the actual needs of the equipment. For example, for the maintenance of an industrial robot, after completing part of the maintenance sub-tasks, it is found that there is a difference between the maintenance effect evaluation index and the expected maintenance target, such as the robot's motion accuracy improvement amplitude does not reach the expectation. At this time, the trigger condition of the unexecuted maintenance sub-task can be adjusted, such as lowering the execution threshold of certain maintenance sub-tasks, or adjusting the execution sequence to preferentially execute the maintenance sub-tasks that are more helpful to improving the motion accuracy.

[0247] Step S3000: When all maintenance sub-tasks are executed, update the real-time running state level based on the latest state evaluation index value, and feedback to the dynamic adjustment process of the preset state threshold value.

[0248] When all maintenance sub-tasks are executed, the latest state evaluation index values, such as the performance parameters and running stability of the equipment, are collected. According to these index values, the real-time running state level of the equipment is re-evaluated. The updated real-time running state level is fed back to the dynamic adjustment process of the preset state threshold value, further optimizing the preset state threshold value and improving the accuracy of state evaluation. For example, for a wind turbine, after completing all maintenance sub-tasks, the power output and vibration condition of the unit are collected as state evaluation index values, and the real-time running state level is re-evaluated, such as updating from "warning" state to "normal" state. This update information is fed back to the dynamic adjustment module of the preset state threshold value, and the related threshold values are adjusted.

[0249] By continuously performing the above steps, i.e., updating the real-time running state level according to the latest state evaluation index value, and feeding back to the dynamic adjustment process of the preset state threshold, the multi-level state analysis model can be continuously optimized and improved, better adapting to the change of the running state of the equipment, improving the accuracy and reliability of the equipment state analysis, and providing more effective support for the maintenance and management of the equipment. For example, in a large manufacturing enterprise equipment management system, through continuous feedback and adjustment, the accuracy of equipment failure warning is continuously improved, the maintenance cost of the equipment is continuously reduced, and the production efficiency is improved.

[0250] Figure 2 A hardware entity schematic diagram of a data analysis system provided by an embodiment of the present application is shown in FIG. 1, which includes a processor 1001 and a memory 1002. The memory 1002 stores a computer program executable on the processor 1001, and the processor 1001 implements the steps in the method of any of the above embodiments when executing the program. Figure 2

[0251] The memory 1002 stores a computer program executable on the processor, and the memory 1002 is configured to store instructions and applications executable by the processor 1001, and can also cache data (e.g., image data, audio data, voice communication data, and video communication data) to be processed or having been processed by the processor 1001 and each module in the data analysis system 1000, which can be realized by FLASH or RAM.

[0252] The processor 1001 implements the steps of the device state data analysis method applied to the automated electrical equipment of any of the above embodiments when executing the program. The processor 1001 generally controls the overall operation of the data analysis system 1000.

[0253] The above is only an embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.​

Claims

1. A device state data analysis method applied to automation electricals, characterized by, The method comprises: obtaining a historical running data set of a target device, the historical running data set containing historical data sequences of multiple device running parameters and corresponding device state labels; constructing a multi-level state analysis model based on the historical running data set, the multi-level state analysis model containing at least one first analysis level and at least one second analysis level, wherein each first analysis level corresponds to a running parameter screening strategy, each second analysis level corresponds to a state evaluation strategy, and the second analysis level is after the first analysis level; extracting key running parameters from a real-time running data stream of the target device through the running parameter screening strategy in the first analysis level, and generating a parameter screening result according to the key running parameters; inputting the parameter screening result into the state evaluation strategy in the second analysis level, combining the dynamic threshold range corresponding to the key running parameters, and generating a state evaluation index of the target device; outputting a real-time running state level of the target device according to the comparison result of the state evaluation index and a preset state threshold; adding a third analysis level in the multi-level state analysis model, the third analysis level being after the second analysis level; predicting a future state evolution path of the target device according to the historical change trend of the state evaluation index through the state prediction strategy in the third analysis level; extracting key turning points in the future state evolution path, and generating a preventive maintenance suggestion in combination with real-time measurement values of the key running parameters; adjusting the response priorities of the parameter screening strategy and the state evaluation strategy in the multi-level state analysis model according to the matching result of the preventive maintenance suggestion and the real-time running state level; wherein the prediction of the future state evolution path of the target device according to the historical change trend of the state evaluation index through the state prediction strategy in the third analysis level comprises: extracting time series data from the state evaluation index, the time series data containing state evaluation index values of the target device at different time points and real-time measurement values of corresponding key running parameters; performing multi-scale decomposition on the time series data through the state prediction strategy in the third analysis level to generate decomposed time series data containing long-term trend components, periodic fluctuation components and short-term noise components; identifying the overall degradation direction of the target device based on the long-term trend components, and extracting the running state change period of the target device in combination with the periodic fluctuation components; generating a prediction sub-model set according to the overall degradation direction and the running state change period, the prediction sub-model set containing multiple prediction sub-models matched with different degradation stages and periodic fluctuation characteristics; inputting the decomposed time series data into each prediction sub-model in the prediction sub-model set to generate future state prediction results under different confidence levels, and cross- verifying the future state prediction results; screening out a target prediction sub-model with the smallest deviation from the latest state evaluation index value in the real-time running data stream of the target device according to the cross-verification result; The short-term noise component in the time series data is trend-corrected by the target prediction sub-model to generate a future state evolution path containing the corrected noise influence; The future state evolution path is associated and matched with the dynamic threshold range of the key operation parameter in the real-time operation data stream to identify an abnormal prediction interval in the future state evolution path that exceeds the dynamic threshold range; Based on the starting time point, the ending time point and the corresponding key operation parameter type of the abnormal prediction interval, a preventive maintenance suggestion containing a maintenance trigger condition and a maintenance parameter range is generated.

2. The method of claim 1, wherein, The key operation parameter is extracted from the real-time operation data stream of the target device by the operation parameter screening strategy in the first analysis level, and a parameter screening result is generated according to the key operation parameter, including: An operation parameter sequence within a current time window is intercepted from the real-time operation data stream, and the operation parameter sequence contains real-time measurement values of multiple device operation parameters; Based on the parameter correlation screening condition in the first analysis level, an association degree coefficient of each operation parameter in the multiple device operation parameters and a device state label in the historical operation data set is calculated; A candidate operation parameter with an association degree coefficient greater than a preset association threshold is screened from the multiple device operation parameters, and a parameter stability indicator is determined according to a real-time measurement value fluctuation amplitude of the candidate operation parameter; The candidate operation parameters are sorted according to the parameter stability indicator, and a preset number of candidate operation parameters in the front of the sorting are selected as the key operation parameters; A parameter screening result containing parameter weight allocation is generated based on the real-time measurement values of the key operation parameters, and the parameter weight is dynamically adjusted according to the association degree coefficient and the parameter stability indicator.

3. The method of claim 2, wherein, The association degree coefficient of each operation parameter in the multiple device operation parameters and the device state label in the historical operation data set is calculated based on the parameter correlation screening condition in the first analysis level, including: Data distribution features of each device operation parameter under different device state labels are extracted from the historical operation data set, and the data distribution features include parameter mean, variance and distribution form indicators; A discrimination value of each device operation parameter between different device state labels is calculated according to the data distribution features, and the discrimination value is determined based on the combination of parameter mean difference and variance ratio; A parameter change trend curve of each device operation parameter in the historical operation data set is obtained, and the number of peak points and the trend inflection point position in the parameter change trend curve are extracted; The association degree coefficient is generated in combination with the discrimination value, the number of peak points and the trend inflection point position, wherein the association degree coefficient is positively correlated with the discrimination value and the number of peak points, and is negatively correlated with the dispersion of the trend inflection point position; The association degree coefficient is normalized, and the normalized association degree coefficient is dynamically matched and verified with the time sequence change of the device state label in the historical operation data set.

4. The method of claim 3, wherein, The parameter screening result containing parameter weight allocation is generated based on the real-time measurement values of the key operation parameters, including: calculating a real-time offset according to the real-time measurement value of the key operation parameter and the historical mean value of the corresponding parameter in the historical operation data set; determining a real-time abnormal probability of each key operation parameter based on the real-time offset and the parameter stability index; generating a dynamic weight distribution ratio according to the real-time abnormal probability and the correlation coefficient, wherein the dynamic weight distribution ratio is positively correlated with the product of the real-time abnormal probability and the correlation coefficient; weighting and fusing the real-time measurement value of the key operation parameter and the dynamic weight distribution ratio to generate a parameter screening result containing a weighted parameter; generating a parameter screening code according to the weighted parameter value in the parameter screening result, wherein the parameter screening code is used to identify the priority order of different key operation parameters in the second analysis level.

5. The method of claim 1, wherein, inputting the parameter screening result into a state evaluation strategy in the second analysis level, combining the dynamic threshold range of the corresponding key operation parameter, and generating a state evaluation index of the target device, including: parsing the real-time measurement value of the key operation parameter and the corresponding parameter weight distribution ratio from the parameter screening result; determining a dynamic threshold range of the key operation parameter according to the parameter distribution interval under different device state labels in the historical operation data set, wherein the dynamic threshold range is self-adaptively adjusted with the device running time; calculating the deviation degree of the real-time measurement value from the upper limit value and the lower limit value of the dynamic threshold range, and generating a comprehensive deviation score in combination with the parameter weight distribution ratio; determining the state abnormality level corresponding to the key operation parameter according to the comparison result of the comprehensive deviation score and the preset deviation threshold; integrating the state abnormality levels of all key operation parameters to generate an overall state evaluation index of the target device, wherein the overall state evaluation index contains an abnormality level distribution and an abnormality duration.

6. The method of claim 5, wherein, The determination of the dynamic threshold range of the key operation parameter includes: extracting the historical maximum value and the historical minimum value of the key operation parameter in different device running stages from the historical operation data set; calculating an initial threshold range according to the historical maximum value and the historical minimum value, and determining a current running stage based on the real-time running time of the target device; obtaining a sliding average value and a sliding variance value of the real-time measurement value of the key operation parameter in the current running stage; dynamically correcting the initial threshold range according to the sliding average value and the sliding variance value to generate a dynamic threshold range containing a correction offset; matching and verifying the dynamic threshold range with the historical change rate of the key operation parameter, so that the change rate of the dynamic threshold range does not exceed a preset rate limit; wherein the dynamic correction of the initial threshold range according to the sliding average value and the sliding variance value to generate a dynamic threshold range containing a correction offset includes: calculating the difference between the sliding average value and the median value of the initial threshold range as a mean offset; determining a variance correction coefficient according to the ratio of the sliding variance value to the historical variance; multiplying the mean offset and the variance correction coefficient to generate a dynamic correction amount; add the dynamic correction quantity to the upper limit value and the lower limit value of the initial threshold range respectively to generate a corrected dynamic threshold range; perform boundary constraint processing on the corrected dynamic threshold range, so that the upper limit value of the dynamic threshold range does not exceed a preset percentage of the historical maximum value, and the lower limit value does not fall below a preset percentage of the historical minimum value.

7. The method of claim 1, wherein, The method further comprises: setting a feedback optimization mechanism in the multi-level state analysis model, the feedback optimization mechanism iteratively updates the operation parameter screening strategy and the state evaluation strategy according to the output result of the real-time operation state level; obtaining a matching error between the real-time operation state level and the actual equipment maintenance record, and calculating a model optimization weight according to the matching error; adjusting the parameter correlation screening condition in the first analysis level and the dynamic threshold range generation rule in the second analysis level based on the model optimization weight; applying the adjusted parameter correlation screening condition and dynamic threshold range generation rule to the newly received real-time operation data stream to generate an updated multi-level state analysis model; performing backtracking verification on the historical operation data set through the updated multi-level state analysis model, so that the prediction accuracy of the updated multi-level state analysis model is not less than a preset accuracy threshold.

8. A data analysis system comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, The processor implements the steps in the method of any one of claims 1-7 when executing the program. The processor implements the steps in the method of any one of claims 1-7 when executing the program.

Images