A method and system for comprehensively evaluating health indicators of a genset device

Abstract

The application discloses a kind of generator set equipment health degree index comprehensive evaluation method and system.The technical scheme for using in the application is: with the historical data of all state variables of the same equipment as input, using the method of clustering, classify the data, and filter out the working state of different classes of equipment;Each class of working state obtained by clustering establishes a rule base to score or deduct each variable;Determine how many variable scores by judging which class of working state the data belongs to at a certain time and whether it exceeds the upper and lower limits of each rule;Output the importance of each variable and determine the weight of each variable in health degree fusion by combining the preset health degree weight coefficient;Through variable score, and combining the weight of each variable in health degree fusion, weighted summation is obtained to obtain the equipment health degree index score, and a hard rule for increasing or decreasing the equipment health degree index score is added.The application can accurately grasp the real-time running state of the equipment.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of power generation, in particular to a generator set equipment health index comprehensive evaluation method and system. BACKGROUND

[0002] The equipment predictive maintenance is a maintenance mode of equipment in which the position and cause of fault are judged and the service life of the equipment is predicted by using instruments to collect and process signals and analyze information generated by the equipment without disassembling the equipment basically. The state detection avoids the sudden failure of mechanical equipment, thereby avoiding forced shutdown and affecting production. The state detection analysis provides a reliable basis for predicting the maintenance period of mechanical equipment, that is, maintenance is performed only when necessary, and maintenance plans can be arranged in time, thereby overcoming unnecessary economic losses caused by regular maintenance and greatly shortening the maintenance period.

[0003] Therefore, by means of advanced technical means, the equipment state monitoring, predictability maintenance and fault diagnosis are implemented, early fault hidden dangers and causes of the equipment are found as much as possible, the downtime of the equipment is reduced, the service life of the equipment is prolonged, and the operation and maintenance cost of the equipment is reduced. SUMMARY

[0004] The application aims to provide a generator set equipment health index comprehensive evaluation method and system, which calculates the real-time health of the equipment, enables monitoring personnel to accurately grasp the real-time operation state of the equipment, finds early fault hidden dangers and causes of the equipment as much as possible, thereby reducing the downtime of the equipment, prolonging the service life of the equipment, and reducing the operation and maintenance cost of the equipment.

[0005] To this end, one technical solution adopted by the application is as follows: a generator set equipment health index comprehensive evaluation method, which comprises the following steps:

[0006] Step 1: determining all state variables related to the health index of the generator set to be evaluated;

[0007] Step 2: using the historical data of all state variables of the same equipment as input, using the clustering method to classify different data, and screening the working states of different classes of the equipment;

[0008] Step 3: establishing a rule base for each class of working state obtained by clustering to score or deduct each state variable, the rule base uses four rules to analyze the data change speed, data state abnormality, data state abnormality duration and data fluctuation amplitude of the state variable from four angles, and the historical data of each state variable is counted;

[0009] Step 4, determine the score of the state variable by judging which category the data belongs to at a certain moment and whether it exceeds the upper and lower limits of each rule, and then determine how much the score of the state variable is added or reduced;

[0010] Step 5, use the linear regression algorithm to output the importance of each state variable, combine the preset health degree weight coefficient, and determine the weight of each state variable in health degree fusion;

[0011] Step 6, add the state variable score obtained in step 4, and combine the weight of each state variable in health degree fusion to obtain the device health degree index score, and add the hard rule of increasing or decreasing the device health degree index score.

[0012] Further, in step 1, the state variable selection can reflect the real-time running state of the device.

[0013] Further, in step 2, all historical data of the device are clustered, and the state variables of the same device are selected, and the correlation between a certain state variable and other state variables in the same device is calculated.

[0014] Further, in step 3, each type of working state obtained by clustering is analyzed from four angles of data change speed, whether the related state variable is abnormal, duration of state variable abnormality, and data fluctuation amplitude. The rule base includes the following five categories:

[0015] 1)relevance rule

[0016] From the perspective of whether the related state variable in the same device is abnormal;

[0017] 2)recent rule

[0018] From the perspective of the current fluctuation amplitude of the data;

[0019] 3)duration rule

[0020] From the perspective of whether the data is out of limit and the time of exceeding the limit;

[0021] 4)vibrate rule

[0022] From the perspective of data change rate;

[0023] 5)section rule

[0024] From the perspective of whether the most basic numerical value exceeds the limit.

[0025] Further, in step 4, the specific process of adding or reducing each state variable according to the rules is as follows:

[0026] 1)relevance rule

[0027] Select a historical data, calculate the mean and standard deviation of each state variable in the historical data, obtain the set interval of relevance rule: [mean-6*standard deviation, mean+2*standard deviation], when the first two state variables with the strongest correlation with the state variable simultaneously exceed the interval, then the state variable is reduced, the reduced or added score is determined according to the part of the state variable and its related state variable exceeding or not exceeding the corresponding interval;

[0028] 2)recent rule

[0029] By skewness processing on historical data and data to be judged, and calculating the mean and variance of the ratio between the processed data to be judged and the historical data, the set interval of recent rule is obtained: [mean-6*standard deviation, mean+1.5*standard deviation], whether the fluctuation value of the current time data compared with the data of the previous period is in this interval is judged, if not in the interval, the score is reduced by the plus-minus score rule, if in the interval, the score is increased by the plus-minus score rule;

[0030] 3)duration rule

[0031] Determine whether the data is abnormal, determine the upper limits G1, G2, G3, G4 and the lower limits L1, L2, L3, L4 of eight limits and divide four intervals through historical data, the upper and lower limits of the four intervals are L1-G1: [mean-4*standard deviation, mean+3*standard deviation], L2-G2: [mean-5*standard deviation, mean+4*standard deviation], L3-G3: [mean-7*standard deviation, mean+5*standard deviation], L4-G4: (minimum limit, maximum limit); the data exceeding different intervals has different deduction coefficients; when the data of a certain time exceeds the limit, the over-limit time needs to be recorded, only when the over-limit time is greater than the minimum duration of the limit, the score will be reduced;

[0032] 4)vibrate rule

[0033] The maximum and minimum values of the historical data are counted, the maximum value and the minimum value*0.2 are recorded as the amplitude range, and the mean and standard deviation of each state variable in the historical data are calculated, to obtain the set interval of vibrate rule: [mean-5*standard deviation, mean+3*standard deviation], when the data appears twice with a difference greater than the amplitude range, the interval of the data is judged, and different scores are reduced according to the degree of data deviation;

[0034] 5)section rule

[0035] A most basic interval is determined through historical data, and the mean and standard deviation of each state variable in the historical data are calculated to obtain the set interval of the section rule: [mean-5*standard deviation, mean+3*standard deviation], and if the data exceeds the interval limit, points are deducted without judging other conditions.

[0036] Further, the specific content of step 5 is as follows:

[0037] 1) The importance of the state variable is output by using a linear regression algorithm, and is arranged into an importance matrix;

[0038] 2) Each column of the importance matrix is summed up to obtain the importance of each state variable to the device;

[0039] 3) The high-correlation state variable set in the device is screened out by using the constructed device-state variable dictionary type data, and a representative state variable is selected from each state variable set and is given a preset high weight coefficient, and the remaining state variables are given a preset low weight coefficient;

[0040] 4) The importance of each state variable to the device is multiplied by the corresponding weight coefficient, and finally the final weight of the device health degree fusion is obtained.

[0041] Further, in step 6,

[0042] 1) The device health index score is obtained by weighting and summing the obtained state variable score and the health degree fusion weight;

[0043] 2) When a device health index score is reported to be abnormal, the device health index score is deducted;

[0044] 3) When evaluating a device health index, if the device health index is a level 1 index and the device health index score is less than 60 points, the device in which the device health index is located is classified as an abnormal working state level, and a device warning is given.

[0045] Another technical solution adopted by the application is:

[0046] Variable input unit: all state variables related to the health index of the generator set to be evaluated;

[0047] Clustering and dividing unit: clustering all historical data of the device, classifying different data, screening out different working states of the device, selecting state variables of the same device, and calculating the correlation between each state variable and other state variables;

[0048] Rule base unit: each class of working state obtained by clustering is analyzed from four angles of data change speed, whether the related state variable is abnormal, the duration of state variable abnormality and the amplitude of data fluctuation;

[0049] Limit value judging unit: the number of state variable points is determined by judging which class of working state the data at a moment belongs to and whether it exceeds the upper and lower limits of each rule;

[0050] Health degree fusion unit: the importance of each state variable is output by using a linear regression algorithm, and the weight of each state variable in health degree fusion is determined by combining a preset health degree weight;

[0051] Device level scoring unit: the device health degree index score is obtained by weighted summation of state variable points and the weight of each state variable in health degree fusion, and a hard rule of increasing or decreasing the device health degree index score is added.

[0052] The health degree obtained by the application is a comprehensive index reflecting the real-time state of the device, and the real-time running state of the device can be accurately grasped and the reasonable maintenance opportunity can be determined through the index. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a flowchart of the generator set device health degree index comprehensive evaluation method of embodiment 1 of the application;

[0054] Figure 2 is a cluster analysis diagram obtained by clustering the historical data of the outlet air temperature of the coal mill of embodiment 1 of the application;

[0055] Figure 3 is a correlation analysis diagram of all state variables related to the health degree of the coal mill of embodiment 1 of the application;

[0056] Figure 4 is a state variable score diagram given by recent rule judgment analysis of the current of the coal mill of embodiment 1 of the application;

[0057] Figure 5 is a device health degree index score diagram obtained by rule base analysis and judgment of all real-time state variables of a coal mill of embodiment 1 of the application, and by weight fusion of multiple state variables. DETAILED DESCRIPTION

[0058] The specific embodiments of the application will be further described below in combination with the drawings and examples. The following examples are used to illustrate the application, but not to limit the scope of the application.

[0059] Example 1

[0060] As attached Figure 1 As shown, the present invention provides a comprehensive evaluation method for the health index of a generator set, comprising the following steps:

[0061] Step 1: Determine all state variables related to the health indicator of the equipment to be evaluated. Example 1: Taking the coal mill equipment in a thermal power generator as an example, the corresponding state variables include current, oil pressure, inlet and outlet differential pressure, outlet pressure, inlet air and powder temperature, outlet air and powder temperature, motor bearing temperature, coil temperature, etc.

[0062] Step 2: Taking the historical data of all state variables of the same device as input, clustering method is used to classify different data and filter out different types of working states of the device.

[0063] Select historical data on device state variables and cluster all data related to the device. Example 1 uses Kmeans 2*2 for two-layer clustering: the first layer of clustering is 2 clusters, and the second layer of clustering is to cluster each cluster of the first layer with 2 clusters. Based on the clustering results, select the state variables of the same device, calculate the correlation between each state variable and other state variables, and finally generate a correlation dictionary for the same device.

[0064] Attachment Figure 2 The cluster analysis diagram of the historical data of the outlet air temperature of a coal mill is given, and a total of 6 types of working states are obtained.

[0065] Attachment Figure 3 The correlation analysis diagram of the current state variable of the coal mill and all state variables related to health is given. Figure 3 The horizontal coordinates from left to right are: a is the coal feed rate feedback measurement point of coal feeder B, b is the automatic controlled value (simulation) of coal feeder B, c is the differential pressure measurement point of the upper and lower parts of the grinding bowl of coal mill B, d is the current measurement point of coal feeder B, e is the F2 oil fire detection flame intensity measurement point, f is the temperature 1 measurement point of the motor coil of coal mill B, g is the differential pressure measurement point of the upper and lower parts of the grinding bowl of coal mill A, h is the current measurement point of coal mill A, i is the coal feed rate feedback measurement point of coal feeder A, j is the air powder temperature measurement point of the outlet pipe 1 of coal mill B, k is the temperature measurement point of the inlet of coal mill E The primary air pressure 1 measuring point, l is the air-powder temperature measuring point at the outlet pipe 5 of coal mill B, m is the automatically adjusted quantity (simulation) of coal feeder A, n is the air-powder temperature measuring point at the outlet pipe 3 of coal mill B, o is the air-powder temperature measuring point at the outlet pipe 2 of coal mill B, p is the air-powder temperature measuring point at the outlet pipe 4 of coal mill B, q is the outlet pressure measuring point of coal mill B, r is the outlet temperature 1 measuring point of coal mill B, s is the primary air pressure 1 measuring point at the inlet of coal mill A, t is the air-powder temperature measuring point at the outlet pipe 5 of coal mill F, and u is the wind speed measuring point of the powder pipe 1 of coal mill B.

[0066] Step 3, a rule base is established for each class of working state obtained by clustering to score or deduct each state variable, the rule base uses four rules to analyze the data change speed, data state whether abnormal, data state abnormal duration and data fluctuation amplitude from four angles, and the history data of each state variable is counted, the rules mainly include the following five types:

[0067] 1) relevance rule

[0068] from the perspective of whether the related state variables in the same device are abnormal;

[0069] 2) recent rule

[0070] from the perspective of the fluctuation amplitude of the data at the current time;

[0071] 3) duration rule

[0072] from the perspective of whether the data is out of limit and the out of limit time;

[0073] 4) vibrate rule

[0074] from the perspective of data change rate;

[0075] 5) section rule

[0076] from the perspective of whether the most basic numerical value exceeds the limit.

[0077] Step 4, the number of points of the state variable is determined by judging which class of working state the data at a certain time belongs to and whether it exceeds the upper and lower limits of each rule, and the state variable score is obtained. The specific process of scoring or deducting each state variable according to the rules is as follows:

[0078] 1) relevance rule

[0079] select a period of historical data, calculate the mean and standard deviation of each state variable in the historical data, and obtain the set interval of relevance rule: [mean-6*standard deviation, mean+2*standard deviation], when the two state variables with the strongest correlation with this state variable simultaneously exceed the interval, then score or deduct this state variable, the score of deduction or addition is determined according to the part of the state variable and its related state variable exceeding or not exceeding the corresponding interval.

[0080] The specific steps are as follows:

[0081] ① calculate the mean and standard deviation of each state variable in the historical data (one month), and set the interval: [mean-6*standard deviation, mean+2*standard deviation];

[0082] 2) When the state variable itself is out of range:

[0083] According to the correlation dictionary, find the top two state variables with the strongest correlation with the state variable. When the two state variables with the strongest correlation with the state variable are also out of the set interval, reduce the score;

[0084] 3) Real-time judgment of whether it meets the rules.

[0085] Score reduction rules:

[0086] Deviation ratio: value of state variable out of set interval / width of set interval;

[0087] If the state variable itself and the state variable with strong correlation deviate at the same time, calculate the set deviation ratio of the state variable itself and the state variable with strong correlation respectively, denoted as:

[0088] State variable itself deviation ratio: P0,

[0089] Strongly correlated state variable 1: P1,

[0090] Strongly correlated state variable 2: P2,

[0091] Score reduction = 100 (points) * (P0 + (P1 + P2) / 2) * W (coefficient temporarily set to 0.1), finally get the score to be reduced.

[0092] Score increase rules:

[0093] Deviation ratio: difference M between the value and the middle line of the upper and lower interval in the rule base, then calculate the percentage A of M in half the interval size;

[0094] If no state variable with strong correlation deviates, get the deviation ratio A0, A1, A2 of each state variable, A = max(A0, A1, A2), score increase = lost score * (1-A) * W (W is a coefficient temporarily set to 0.1).

[0095] 2) recent rule

[0096] By skewness processing on historical data and data to be judged, and calculating the mean and variance of the ratio between the processed data to be judged and the historical data, the set interval of the recent rule can be obtained: [mean-6*standard deviation, mean+1.5*standard deviation], judge whether the fluctuation value of the current time data compared with the previous period data is within this interval, if not within the interval, reduce the score through the score increase and decrease rules, if within the interval, increase the score through the score increase and decrease rules.

[0097] The specific steps are:

[0098] Firstly, the historical data is classified into three levels according to skewness, and the data is normalized after the skewness is calculated. Different skewness means different deviation of data distribution from normal distribution, so different transformation is needed.

[0099] The definition is as follows:

[0100]

[0101] Where X is each data point, μ is the mean of the data set, σ is the standard deviation of the data set, and k2 and k3 represent the second and third central moments, respectively.

[0102] Symmetry: -0.5 to 0.5,

[0103] Moderate positive / negative skewness: -0.5 to -1.0 and 0.5 to 1.0,

[0104] High positive / negative skewness: <-1.0 and >1.0.

[0105] Secondly, the data after classification is processed, and different data processing rules are used for different levels.

[0106] -0.5 < skewness < 0.5: no processing,

[0107] -1 < skewness < -0.5, 0.5 < skewness < 1: square root of data,

[0108] skewness <-1, 1 < skewness: take the logarithm of the data.

[0109] Thirdly, the setting interval is calculated

[0110] Sliding window (window length is 10 minutes), calculate the ratio of the data at the next time point to the average of the data in the window, and calculate the average and variance of the ratio.

[0111] Setting interval: [mean - 6*standard deviation, mean + 1.5*standard deviation].

[0112] Fourthly, it is judged in real time whether it meets the rules

[0113] Real-time calculation of the ratio of the current value to the average of the previous data (10 minutes of data at this time), and judge whether the ratio is within the setting interval.

[0114] Penalty rule:

[0115] If the value is not within the setting interval, calculate the proportion of the value deviating from the setting interval, and the deviation percentage refers to the value exceeding the setting interval / the width of the setting interval. The score to be deducted is 100 (full score)*deviation percentage*W (W is the deviation coefficient, generally set to 0.1), and finally the score to be deducted is obtained.

[0116] Bonus rule:

[0117] If the value is in the rule base interval, calculate the difference M between the value and the middle of the upper and lower interval of the rule base, then calculate the percentage A of M in the size of half interval, add the score of the lost score*(1-A)*W, and finally get the score to be added.

[0118] 3) duration rule

[0119] Determine whether the data is abnormal, determine the upper limit G1, G2, G3, G4 and the lower limit L1, L2, L3, L4 of eight limits and divide four intervals, the upper and lower limits of the four intervals are L1-G1:[mean-4*standard deviation, mean+3*standard deviation], L2-G2:[mean-5*standard deviation, mean+4*standard deviation], L3-G3:[mean-7*standard deviation, mean+5*standard deviation], L4-G4:(minimum limit, maximum limit); data exceeding different intervals have different deduction coefficients; when the data at a certain time exceeds the limit, the over-limit time needs to be recorded, only when the over-limit time is greater than the minimum duration of this limit, will the score be deducted.

[0120] The specific steps are:

[0121] ① Set interval

[0122] According to the condition that the device switching value or current is less than 1, remove the downtime, and then calculate the mean and variance.

[0123] Set three limits:

[0124] L1:[mean-4*standard deviation, mean+3*standard deviation],

[0125] L2:[mean-5*standard deviation, mean+4*standard deviation],

[0126] L3:[mean-7*standard deviation, mean+5*standard deviation],

[0127] L4:(inf, mean-7*standard deviation) and (mean+5*standard deviation, inf).

[0128] Over-limit duration statistics:

[0129] Record and count the duration of each state variable in the historical data that exceeds the limit (count the points that exceed L1, L2, L3, L4 respectively), calculate the minimum value, maximum value, median, mode, and mean.

[0130] ② Real-time judgment whether it meets the rules

[0131] Real-time determine whether the data is out of bounds. If the limit occurs, determine that the limit is exceeded, and the duration reaches the minimum duration of the limit in historical data, then reduce the score of the state variable.

[0132] The deduction rule is:

[0133] When the value exceeds the L2 or L3 interval, calculate the percentage of the excess value in the width of the grade interval (the grade interval - the last grade interval), denoted as A, and the deduction = 100 (points) * A * W (the coefficient is temporarily set to 0.1).

[0134] When the value exceeds the L4 interval, calculate the percentage of the excess value in the width of the third grade interval (the upper limit of the third interval - the lower limit of the third interval), denoted as A, and the deduction = 100 (points) * A * W (the coefficient is temporarily set to 0.1).

[0135] The deduction rule is:

[0136] When the value is in the L1 interval, calculate the difference M between the value and the middle of the upper and lower interval in the rule library, and then calculate the percentage A of M in half the interval size, and the deduction = the loss score * (1-A) * W (W is the coefficient, temporarily set to 0.1).

[0137] 4) vibrate rule

[0138] The maximum and minimum values of the historical data are counted, and the maximum value and the minimum value * 0.2 are recorded as the amplitude range, and the mean and standard deviation of each state variable in the historical data are calculated, and the setting interval of the vibrate rule is obtained: [mean-5*standard deviation, mean+3*standard deviation], when the data appears twice, the difference is greater than the amplitude range, the data is judged by interval, and different deductions are made according to the degree of data deviation.

[0139] The specific steps are:

[0140] ① Set the interval

[0141] The maximum and minimum values of the historical data are counted, and the maximum value and the minimum value * 0.2 are recorded as b, as the amplitude range.

[0142] Set the interval: [mean-5*standard deviation, mean+2.5*standard deviation]

[0143] ② Real-time determine whether the rule is met

[0144] The fixed window is 20 seconds, and the difference between the current time and the time 1 second ago is calculated to determine whether the difference is greater than b. If the difference is greater than 1 in the window, then reduce the score.

[0145] The deduction rule is:

[0146] If the value is not in the set interval, calculate the proportion of the value deviating from the set interval.

[0147] Deviation proportion: value of state variable exceeding the set interval / width of the set interval.

[0148] Penalty = 100 (points) * deviation percentage * W (coefficient temporarily set to 0.1), finally the number of points to be reduced is obtained.

[0149] Bonus rules:

[0150] If the value is in the rule base interval, calculate the difference M between the value and the middle line of the upper and lower intervals of the rule base, and then calculate the percentage A of M in half the interval size. Bonus = loss points * (1-A) * W (W is a coefficient temporarily set to 0.1).

[0151] 5) section rules

[0152] A most basic interval is determined through historical data, and the mean and standard deviation of each state variable in the historical data are calculated to obtain the set interval of the section rules: [mean-5*standard deviation, mean+3*standard deviation], if the data exceeds the interval limit, a penalty is given without the need to judge other conditions.

[0153] ① Set interval

[0154] Calculate the mean and variance of historical data

[0155] Set interval: [mean-5*standard deviation, mean+3*standard deviation]

[0156] ② Real-time judgment of whether it meets the rules

[0157] Penalty rules:

[0158] If the value is not in the set interval, calculate the proportion of the value deviating from the set interval.

[0159] Deviation proportion: value of state variable exceeding the set interval / width of the set interval.

[0160] Penalty = 100 (points) * deviation percentage * W (coefficient temporarily set to 0.01), finally the number of points to be reduced is obtained.

[0161] Bonus rules:

[0162] If the value is in the rule base interval, calculate the difference M between the value and the middle line of the upper and lower intervals of the rule base, and then calculate the percentage A of M in half the interval size. Bonus = loss points * (1-A) * W (W is a coefficient temporarily set to 0.1).

[0163] Appendix Figure 4The state variable score graph given by the rule base judgment analysis of the outlet air temperature of the coal mill is given, and the upper limit of the score is 100 points and the lower limit is 0 points.

[0164] Step 5, using the linear regression algorithm, the importance of each state variable is output, and the weight coefficient of the preset health degree is combined to determine the weight of each state variable in the health degree fusion.

[0165] 1) Using the linear regression algorithm, the importance of the state variable is output, and the importance matrix is arranged;

[0166] 2) Sum each column of the importance matrix to obtain the importance of each state variable to the device;

[0167] 3) Using the constructed “device-state variable” dictionary type data, the state variable set with high correlation in the device is selected, and the representative state variable is selected from each state variable set and given a preset high weight coefficient, and the remaining state variables are given a preset low weight coefficient;

[0168] 4) Multiply the importance of each state variable to the device by the corresponding weight coefficient, and finally obtain the maximum weight in the health degree fusion of the device.

[0169] Step 6, the state variable score obtained in step 4 is combined with the weight of each state variable in the health degree fusion, and the device health degree index score is obtained by weighted summation, and the hard rule of increasing and decreasing the device health degree index score is added.

[0170] 1) The state variable score is obtained, combined with the weight in the health degree fusion, and the device health degree index score is obtained by weighted summation.

[0171] 2) When a device health degree index score is reported to be abnormal, the device health degree index score is reduced.

[0172] 3) When evaluating a device health degree index, if the device health degree index is a level 1 index and the device health degree index score is less than 60 points, the device score of the device health degree index is classified as an abnormal working state level, and a device warning is given. After the completion of this step, after a certain sampling period, return to step 4 and repeat the execution.

[0173] Appendix Figure 5 The device health degree index score (referred to as device score) graph obtained by analyzing and judging all real-time state variables of a coal mill through a rule base and by weight fusion of multiple state variables is given, and the upper limit of the score is 100 points and the lower limit is 0 points.

[0174] Example 2

[0175] The embodiment provides a generator set equipment health degree index comprehensive evaluation system, which comprises:

[0176] A variable input unit: all state variables related to the health degree index of the generator set equipment to be evaluated;

[0177] A clustering division unit: all historical data of the equipment are clustered, different data are classified, the working states of different classes of the equipment are screened out, the state variables of the same equipment are selected, and the correlation of each state variable with other state variables is calculated;

[0178] A rule base unit: each class of working state obtained through clustering is analyzed from four angles of data change speed, whether the related state variables are abnormal, the duration of state variable abnormality and the amplitude of data fluctuation;

[0179] A limit value judgment unit: the number of plus and minus scores of the state variables is determined by judging which class of working state the data at a moment belongs to and whether the upper and lower limits of each rule are exceeded, and the state variable score is obtained;

[0180] A health degree fusion unit: the importance of each state variable is output by using a linear regression algorithm, and the weight of each state variable in health degree fusion is determined by combining a preset health degree weight;

[0181] A device-level scoring unit: the device health degree index score is obtained by weighted summation of the state variable score and the weight of each state variable in health degree fusion, and a hard rule of increasing or decreasing the device health degree index score is added.

[0182] In the variable input unit, the state variables should be all related state variables capable of reflecting the real-time running state of the equipment.

[0183] In the clustering division unit, all historical data of the equipment are clustered, the state variables of the same equipment are selected, and the correlation of each state variable with other state variables is calculated.

[0184] In the rule base unit, each class of working state obtained through clustering is analyzed from four angles of data change speed, whether the related state variables are abnormal, the duration of state variable abnormality and the amplitude of data fluctuation, and the rule base comprises the following five types:

[0185] 1) relevance rule

[0186] From the angle of whether the related state variables in the same equipment are abnormal;

[0187] 2) recent rule

[0188] From the angle of the fluctuation amplitude of the data at the current moment;

[0189] 3) duration rule

[0190] From the perspective of whether the data is out of limit and the time of being out of limit;

[0191] 4) vibrate rule

[0192] From the perspective of the rate of data change;

[0193] 5) section rule

[0194] From the perspective of whether the most basic numerical value is out of limit.

[0195] In the limit value judgment unit, each state variable is scored or scored according to the rules:

[0196] 1) relevance rule

[0197] Select a section of historical data, calculate the mean and standard deviation of each state variable in the historical data, and obtain the set interval of the relevance rule: [mean-6*standard deviation, mean+2*standard deviation], when the two state variables with the strongest correlation with this state variable simultaneously exceed the interval, then score this state variable, the score is reduced or added according to the part of the state variable and its related state variable exceeding or not exceeding the corresponding interval;

[0198] 2) recent rule

[0199] By skewness processing on historical data and data to be judged, and calculating the mean and variance of the ratio between the processed data to be judged and the historical data, the set interval of the recent rule is obtained: [mean-6*standard deviation, mean+1.5*standard deviation], whether the fluctuation value of the current time data compared with the previous period data is in this interval, if not in the interval, then reduce the score through the score adding and reducing rule, if in the interval, then add the score through the score adding and reducing rule;

[0200] 3) duration rule

[0201] Determine whether the data is abnormal, determine the upper limits G1, G2, G3, G4 and the lower limits L1, L2, L3, L4 of eight limits by historical data and divide four intervals, the upper and lower limits of the four intervals are L1-G1: [mean-4*standard deviation, mean+3*standard deviation], L2-G2: [mean-5*standard deviation, mean+4*standard deviation], L3-G3: [mean-7*standard deviation, mean+5*standard deviation], L4-G4: (minimum limit, maximum limit); data exceeding different intervals has different deduction coefficients; when the data at a certain time exceeds the limit, the over-limit time needs to be recorded, only when the over-limit time is greater than the minimum duration of the limit, the deduction will be made;

[0202] 4) vibrate rule

[0203] The maximum and minimum values of the historical data are counted, the maximum value and the minimum value*0.2 are recorded as the amplitude range, and the mean and standard deviation of each state variable in the historical data are calculated to obtain the setting interval of the vibrate rule: [mean-5*standard deviation, mean+3*standard deviation], when the data appears twice with a difference greater than the amplitude range, the interval of the data is judged, and different deductions are made according to the degree of data deviation;

[0204] 5) section rule

[0205] A most basic interval is determined by historical data, and the mean and standard deviation of each state variable in the historical data are calculated to obtain the setting interval of the section rule: [mean-5*standard deviation, mean+3*standard deviation], if the data exceeds the interval limit, deduction is made without judging other conditions.

[0206] The health degree fusion unit, the fusion process is as follows:

[0207] 1) Use linear regression algorithm to output the importance of state variables, and arrange into an importance matrix;

[0208] 2) Sum each column of the importance matrix to obtain the importance of each state variable to the device;

[0209] 3) Use the constructed "device-state variable" dictionary type data to select a set of state variables with high correlation in the device, and select a representative state variable from each state variable set and assign a preset high weight coefficient, and assign a preset low weight coefficient to the remaining state variables;

[0210] 4) Multiply the importance of each state variable to the device by the corresponding weight coefficient, and finally obtain the maximum weight of the device health degree fusion.

[0211] The device level scoring unit combines the weight of the health degree fusion to obtain a device health degree index, and adds a hard rule of increasing or decreasing the device health degree index

[0212] 1) Obtain a device health degree index score by combining the weight of the health degree fusion and the obtained state variable score;

[0213] 2) When an abnormality occurs in a device health degree index, the device health degree index is penalized;

[0214] 3) When a device health degree index is evaluated, if the device health degree index is a first-level index and the device health degree index score is lower than 60, the device in which the device health degree index is located is classified as an abnormal working state level, and a device warning is given.

[0215] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A comprehensive evaluation method for the health index of a generator set, characterized in that: include: Step 1: determine all state variables related to the health index of the generator set to be evaluated; Step 2: Using the historical data of all state variables of the same device as input, a clustering method is used to classify different data and filter out different types of working states of the device; Step 3: For each type of working state obtained through clustering, a rule base is established to add or subtract points for each state variable. The rule base uses four rules to analyze the state variable's data change speed, whether the data state is abnormal, the duration of the data state abnormality, and the amplitude of the data fluctuation. The historical data of each state variable is statistically analyzed. Step 4: By judging which working state the data at a certain moment belongs to and whether it exceeds the upper and lower limits of each rule, the state variable score is determined by adding or subtracting the score to obtain the state variable score; Step 5: Use the linear regression algorithm to output the importance of each state variable and combine it with the preset health weight coefficient to determine the weight of each state variable in health fusion; Step 6: The state variable scores obtained in step 4 are combined with the weighted sum of the weights of each state variable when the health fusion is performed to obtain the device health index score, and a hard rule for increasing or decreasing the device health index score is added; In step 4, the specific process of adding or subtracting points for each state variable according to the rules is as follows: 1) Relevance rules Select a period of historical data, calculate the mean and standard deviation of each state variable in the historical data, and obtain the setting interval of the relevance rule: [mean - 6 * standard deviation, mean + 2 * standard deviation]. When the first two state variables with the strongest correlation with the state variable exceed the interval at the same time, the state variable is deducted. The amount of deduction or addition is determined by the part of the state variable and its related state variables that exceeds or does not exceed the corresponding interval; 2) Recent rules By performing skewness processing on historical data and the data to be judged, and calculating the mean and variance of the ratio between the processed data to be judged and the historical data, we obtain the setting interval of the recent rule: [mean - 6 * standard deviation, mean + 1.5 * standard deviation]. We then determine whether the fluctuation value of the current data compared to the data of the previous period is within this interval. If it is not within this interval, points are deducted using the addition and subtraction rule. If it is within this interval, points are added using the addition and subtraction rule. 3)Duration rules To determine whether the data is abnormal, eight limits (high limits G1, G2, G3, G4) and low limits L1, L2, L3, L4) are determined based on historical data and divided into four intervals. The upper and lower limits of the four intervals are L1-G1: [mean - 4*standard deviation, mean + 3 standard deviations], L2-G2: [mean - 5*standard deviation, mean + 4 standard deviations], L3-G3: [mean - 7*standard deviation, mean + 5 standard deviations], and L4-G4: [lowest limit, highest limit]. Different deduction coefficients apply when data exceeds different intervals. When data at a certain moment exceeds a limit, the time it exceeds the limit needs to be recorded. Points will be deducted only when the time it exceeds the limit is greater than the minimum duration of this limit. 4) Vibrate Rules The maximum and minimum values ​​of the historical data are counted, and the maximum and minimum values*0.2 are recorded as the amplitude range. The mean and standard deviation of each state variable in the historical data are calculated to obtain the setting interval of the vibrate rule: [mean-5*standard deviation, mean+3*standard deviation]. When the difference between the two data before and after is greater than the amplitude range, the data is judged as an interval, and different points are deducted according to the degree of data deviation. 5) Section rules Determine a basic interval through historical data, and calculate the mean and standard deviation of each state variable in the historical data to obtain the set interval of the section rule: [mean - 5*standard deviation, mean + 3*standard deviation]. If the data exceeds the interval limit, the score is deducted without judging other conditions.

2. A comprehensive evaluation method for health indicators of generator set equipment according to claim 1, characterized in that: In step 1, the state variables are selected to be variables that can reflect the real-time operating state of the device.

3. A comprehensive evaluation method for health indicators of generator set equipment according to claim 1, characterized in that: In step 2, all historical data of the device are clustered, the state variables of the same device are selected, and the correlation between a state variable and other state variables in the same device is calculated.

4. A comprehensive evaluation method for health indicators of generator set equipment according to claim 1, characterized in that: In step 3, each type of working state obtained by clustering is analyzed from four perspectives: data change speed, whether the relevant state variables are abnormal, the duration of the state variable abnormality, and the amplitude of data fluctuation. The rule base includes the following five categories: 1) Relevance rules From the perspective of whether the relevant state variables in the same device are abnormal; 2) Recent rules From the perspective of the current fluctuation range of the data; 3)Duration rules From the perspective of whether the data exceeds the limit and the time of exceeding the limit; 4) Vibrate Rules From the perspective of data change rate; 5) Section rules Start from the perspective of whether the most basic value exceeds the limit.

5. A comprehensive evaluation method for health indicators of generator set equipment according to claim 1, characterized in that: The specific contents of step 5 are as follows: 1) Using the linear regression algorithm, the importance of the state variables is output and organized into an importance matrix; 2) Sum each column of the importance matrix to obtain the importance of each state variable to the device; 3) Using the constructed "device-state variable" dictionary type data, filter out each state variable set with high correlation in the device, and select representative state variables from each state variable set and assign them a preset high weight coefficient, and assign preset low weight coefficients to the remaining state variables; 4) Multiply the importance of each state variable to the device by the corresponding weight coefficient to obtain the final weight of the device health fusion.

6. A comprehensive evaluation method for health indicators of generator set equipment according to claim 1, characterized in that: In step 6, 1) The device health index score is obtained by combining the obtained state variable scores with the weighted sum of the health fusion weights; 2) When a device health index score is reported as abnormal, the device health index score will be deducted; 3) When evaluating a device health indicator, if the device health indicator is a level 1 indicator and the device health indicator score is lower than 60 points, the device where the device health indicator is located is classified as an abnormal working status level and an equipment warning is issued.

7. A comprehensive evaluation system for health indicators of generator set equipment, used to implement the comprehensive evaluation method for health indicators of generator set equipment according to any one of claims 1 to 6, characterized in that: include: Variable input unit: all state variables related to the health indicators of the generator set to be evaluated; Clustering unit: clusters all historical data of the equipment, classifies different data into categories, filters out different types of working status of the equipment, selects the state variables of the same equipment, and calculates the correlation between each state variable and other state variables; Rule base unit: Analyzes each type of working status obtained by clustering from four perspectives: data change speed, whether the relevant state variables are abnormal, the duration of state variable abnormality, and the amplitude of data fluctuation; Limit judgment unit: By judging which working state the data at a certain moment belongs to and whether it exceeds the upper and lower limits of each rule, the state variable score is determined by adding or subtracting the score. Health fusion unit: uses linear regression algorithm to output the importance of each state variable and combines it with the preset health weight to determine the weight of each state variable in health fusion; Device-level scoring unit: The device health index score is obtained by combining the state variable scores and the weighted sum of the weights of each state variable when the health is fused, and hard rules for increasing or decreasing the device health index score are added.

8. A comprehensive evaluation system for health indicators of power generation equipment according to claim 7, characterized in that: The fusion process of the health fusion unit is as follows: 1) Using the linear regression algorithm, the importance of the state variables is output and organized into an importance matrix; 2) Sum each column of the importance matrix to obtain the importance of each state variable to the device; 3) Using the constructed "device-state variable" dictionary type data, filter out each state variable set with high correlation in the device, and select representative state variables from each state variable set and assign them a preset high weight coefficient, and assign preset low weight coefficients to the remaining state variables; 4) Multiply the importance of each state variable to the device by the corresponding weight coefficient to obtain the final weight of the device health fusion.

9. A comprehensive evaluation system for health indicators of power generation equipment according to claim 7, characterized in that: In the device-level scoring unit, 1) The device health index score is obtained by combining the obtained state variable scores with the weighted sum of the health fusion weights; 2) When a device health indicator reports an abnormality, the device health indicator will be deducted; 3) When evaluating a device health indicator, if the device health indicator is a level 1 indicator and the device health indicator score is lower than 60 points, the device where the device health indicator is located is classified as an abnormal working status level and an equipment warning is issued.

Images