A key indicator trend early warning method, system, terminal and storage medium

Abstract

The application relates to a key indicator trend early warning method and system, a terminal and a storage medium, and belongs to the technical field of data processing. The method comprises the following steps: acquiring original data, a first term coefficient of a convolution kernel and the number of convolution kernels; constructing a convolution kernel model according to the first term coefficient of the convolution kernel and the number of the convolution kernels; scanning the original data by using the convolution kernel model to obtain a scanning interface, and marking local original data located in the scanning interface as a target vector; performing bucketing on the target vector to obtain bucketing characteristic values; calculating the bucketing characteristic values adjacent to each other by using a first-order difference function to obtain a difference vector; calculating a trend probability of the scanning interface according to the difference vector, qualitatively determining the trend of the scanning interface; and calculating a trend growth rate of the scanning interface according to the bucketing characteristic values, quantitatively determining the trend of the scanning interface. The application can achieve the purpose of intelligently operating and maintaining industrial equipment.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, system, terminal and storage medium for early warning of key indicator trends. Background Technology

[0002] Condition monitoring, as one of the means of early warning for rotating machinery faults, mainly involves monitoring the key characteristic indicators of the rotating machinery and establishing a monitoring method with threshold alarm prompts based on historical experience or national and ISO standards. However, this traditional indicator monitoring method has the following shortcomings:

[0003] 1. Simple threshold comparison or distributed control system (DCS) using fixed upper and lower limit alarms are easily affected by actual working conditions, human factors, and the performance of the data acquisition sensors themselves, which can lead to false alarms.

[0004] 2. Simple threshold comparison can only tell you the magnitude of the current indicator value, but it cannot reflect the current deterioration state of the unit equipment (such as a downward trend or an upward trend). Therefore, it cannot detect the future deterioration trend of the unit equipment in time, resulting in missing the best opportunity to stop the deterioration of the unit equipment in advance.

[0005] Since the monitoring and early warning of key indicators are directly related to the dispatch of work orders for on-site workshop personnel's inspections and maintenance, failure to promptly report the alarm level and deterioration status of the unit equipment to the workshop inspection personnel will seriously threaten the normal operation of the unit equipment and personal safety. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this application provides a key indicator trend early warning method, system, terminal and storage medium, which can realize adaptive detection and early warning of long-term and short-term trends of key indicators of industrial equipment, as well as accurately locate the time nodes of trend occurrence and end, which greatly facilitates equipment inspection and maintenance personnel to troubleshoot and solve problems in a targeted and efficient manner, and realizes intelligent operation and maintenance of industrial equipment.

[0007] The purpose of this application is to provide a method for early warning of trends in key indicators.

[0008] The aforementioned objective of this application is achieved through the following technical solution:

[0009] A key indicator trend early warning method includes:

[0010] Obtain the original data, the first term coefficient of the convolution kernel, and the number of convolution kernels;

[0011] Construct a convolution kernel model based on the first term coefficient of the convolution kernel and the number of convolution kernels;

[0012] The original data is scanned using the convolution kernel model to obtain a scanning interface, and the local original data located in the scanning interface is marked as a target vector;

[0013] The target vector is bucketed to obtain bucket eigenvalues;

[0014] The first-order difference function is used to calculate the difference vector of pairwise adjacent bucket eigenvalues;

[0015] The trend probability of the scanning interface is calculated based on the difference vector, and the trend of the scanning interface is qualitatively determined;

[0016] The trend growth rate of the scanning interface is calculated based on the bucket eigenvalues, and the trend of the scanning interface is quantitatively determined.

[0017] In a preferred example, the present application can be further configured as: the construction of the convolution kernel model according to the leading coefficient of the convolution kernel and the number of convolution kernels includes: [[ID=,18]]

[0018] Calculate the tolerance: where a is the leading coefficient of the convolution kernel, 0 < a < 1, and N is the number of convolution kernels;

[0019] According to the tolerance d, the leading coefficient a of the convolution kernel, and N convolution kernels, the general formula for the width of the scanning window when scanning the original data is obtained: C n = L(a + (n - l)d), where n <= N and L is the data length of the original data.

[0020] [[ID=,29]]Obtain the span factor f, and calculate the moving step size based on the span factor f and the general formula C for the width of the scanning window n Calculate the moving step size: B n = (1 - f)*C n , where 0 <= f < 1. <,

[0021] In a preferred example, the present application can be further configured as: the scanning of the original data using the convolution kernel model to obtain a scanning interface includes: with the width of the scanning window being C n and the distance between two adjacent scanning windows being B n , the original data is self-drivenly scanned from left to right to obtain a scanning interface.

[0022] In a preferred example, the present application can be further configured as: the bucketing of the target vector to obtain bucket eigenvalues includes: equally dividing the target vector into K parts of bucket eigenvalues.

[0023] In a preferred example, the present application can be further configured as: the calculation of the difference vector of pairwise adjacent bucket eigenvalues using the first-order difference function includes: dpcy ii∈[0,K-1]= pcy(ii+1)-pcy(ii), where pcy(ii+1) and pcy(ii) are the pairwise adjacent bucket feature values, and dpcy ii∈[0 , K-1] It is a difference vector, and K is the K bucket eigenvalues.

[0024] In a preferred embodiment, this application can be further configured such that: calculating the trend probability of the scanned interface based on the difference vector includes:

[0025] Calculate the probability of an upward trend Where count(dpcy) represents the total number of elements in the difference vector in the scanning interface, and count(dpcy(x>0)) represents the number of difference vector values ​​greater than 0 in the scanning interface;

[0026] Calculate the probability of a downward trend Here, count(dpcy) represents the total number of elements in the difference vector in the scanning interface, and count(dpcy(x<0)) represents the number of difference vector values ​​less than 0 in the scanning interface.

[0027] In a preferred embodiment, this application can be further configured such that: the calculation of the trend growth rate of the scanning interface based on the bucket feature values ​​includes:

[0028] The value V1 is obtained based on the binning feature values ​​of the first H bins on the scanning interface, and the value V2 is obtained based on the binning feature values ​​of the last H bins on the scanning interface. When:

[0029] When both values ​​V1 and V2 are positive, calculate the trend growth rate: When the value V1 is negative, the trend growth rate is calculated as follows: When the value V1 is positive and the value V2 is negative, calculate the trend growth rate: When both values ​​V1 and V2 are negative, calculate the trend growth rate:

[0030] In a preferred embodiment, this application can be further configured such that the trend form includes an upward trend, a downward trend, a simultaneous upward and downward trend, or no trend;

[0031] The quantitative trend analysis of the scanned interface includes:

[0032] Get the preset threshold th2, when:

[0033] When the upward trend probability p(T1) reaches a preset threshold th1 and the trend growth rate tr(T1) reaches a preset threshold th2, the local raw data in the scanning interface is marked as an upward trend.

[0034] When the probability of a downward trend p(T2) reaches a preset threshold th1 and the trend growth rate tr(T2) reaches a preset threshold th2, the local raw data in the scanning interface is identified as having a downward trend.

[0035] When the upward trend probability p(T1) reaches a preset threshold th1 and the trend growth rate tr(T1) reaches a preset threshold th2, and the downward trend probability p(T2) reaches a preset threshold th1 and the trend growth rate tr(T2) reaches a preset threshold th2, the local raw data in the scanning interface is marked as having both an upward and downward trend.

[0036] When the upward trend probability p(T1) does not reach the preset threshold th1 and / or the trend growth rate tr(T1) does not reach the preset threshold th2, and the downward trend probability p(T2) does not reach the preset threshold th1 and / or the trend growth rate tr(T2) does not reach the preset threshold th2, the local raw data in the scanning interface is marked as having no trend.

[0037] tr(T1) represents the trend growth rate of an upward trend, and tr(T2) represents the trend growth rate of a downward trend.

[0038] In a preferred embodiment, this application can be further configured such that, after quantifying the trend of the scanned interface, the method further includes:

[0039] From the scan interfaces generated by scanning within the same scan window, select the scan interface corresponding to the target vector with the largest or smallest trend growth rate as the optimal trend interface;

[0040] Compare the optimal trend interface selected under each scanning window, and select the trend form of the optimal trend interface with the largest or smallest trend growth rate as the overall trend form of the original data.

[0041] This application discloses a key indicator trend early warning method, which can construct a convolution kernel model by selecting the first term coefficient and the number of convolution kernels when the original data is obtained. Then, the convolution kernel model is used to scan the original data to obtain multiple scanning interfaces. Qualitative and quantitative trend calculations are then performed on each scanning interface. Finally, based on the quantitative trend calculation, the most representative optimal trend interface is selected from the scanning interfaces. This allows managers to intuitively see the trend form of industrial equipment and the start and end times of the trend form, thereby monitoring the potential risks of industrial equipment in real time and achieving the goal of intelligent operation and maintenance of industrial equipment.

[0042] The second objective of this application is to provide a key indicator trend early warning system.

[0043] The second objective of this application is achieved through the following technical solution:

[0044] A key indicator trend early warning system, used to execute the aforementioned key indicator trend early warning method, includes:

[0045] The data acquisition module is used to acquire raw data, the first term coefficient of the convolution kernel, and the number of convolution kernels;

[0046] The first processing module is used to construct a convolution kernel model based on the first term coefficient of the convolution kernel and the number of convolution kernels;

[0047] The second processing module is used to scan the original data using the convolution kernel model to obtain a scanning interface, and to mark the local original data located in the scanning interface as target vectors;

[0048] The third processing module is used to divide the target vector into buckets to obtain bucketed feature values;

[0049] The fourth processing module is used to calculate the difference vector by using a first-order difference function to calculate the difference vector of the pairwise adjacent bucket feature values;

[0050] The first generation module is used to calculate the trend probability of the scanning interface based on the difference vector, and to perform trend qualitative analysis on the scanning interface.

[0051] The second generation module is used to calculate the trend growth rate of the scanning interface based on the bucket feature values, and to quantify the trend of the scanning interface.

[0052] The third objective of this application is to provide a terminal.

[0053] The aforementioned objective three of this application is achieved through the following technical solution:

[0054] A terminal includes a memory and a processor, wherein the memory stores computer program instructions for the aforementioned key indicator trend early warning method that can be loaded and executed by the processor.

[0055] The fourth objective of this application is to provide a computer medium capable of storing corresponding programs.

[0056] The fourth objective of this application is achieved through the following technical solution:

[0057] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed for any of the aforementioned key indicator trend early warning methods.

[0058] In summary, this application includes at least one of the following beneficial technical effects:

[0059] 1. This application can construct a convolution kernel model by selecting the first term coefficient and the number of convolution kernels when the original data is obtained. Then, the convolution kernel model is used to scan the original data to obtain multiple scanning interfaces. Qualitative and quantitative trend calculations are then performed on each scanning interface. Finally, the most representative optimal trend interface is selected from the scanning interfaces based on the quantitative trend calculation. This allows managers to intuitively see the trend form of industrial equipment and the start and end times of the trend form, thereby enabling real-time monitoring of potential risks of industrial equipment and achieving the goal of intelligent operation and maintenance of industrial equipment.

[0060] 2. In addition, when judging the trend of local raw data in each scanning interface, this application not only judges the local trend by calculating the characteristic values ​​of each pair of adjacent buckets, but also judges the trend of local raw data in each scanning interface by calculating the trend probability and trend growth rate of the scanning interface. Finally, the scanning interface with the maximum or minimum trend growth rate is selected as the optimal trend interface, thereby ensuring the accuracy of the overall trend of the final raw data. Attached Figure Description

[0061] Figure 1 This is a flowchart of a key indicator trend early warning method according to an embodiment of this application.

[0062] Figure 2 This is an example diagram of a convolutional kernel model scanning the original data in a key indicator trend early warning method according to an embodiment of this application.

[0063] Figure 3 This is an example diagram of the scanning interface generated by the convolutional kernel model scanning the original data in a key indicator trend early warning method according to an embodiment of this application.

[0064] Figure 4 This is an example diagram illustrating the determination of local raw data trend in a key indicator trend early warning method according to an embodiment of this application.

[0065] Figure 5 This is a conditional graph for judging the trend of local raw data in a key indicator trend early warning method according to an embodiment of this application.

[0066] Figure 6 This is a block diagram of a key indicator trend early warning system according to an embodiment of this application.

[0067] Explanation of reference numerals in the attached diagram: 1. Data acquisition module; 2. First processing module; 3. Second processing module; 4. Third processing module; 5. Fourth processing module; 6. First generation module; 7. Second generation module. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] Current technologies are unable to provide trend warnings for key indicators of industrial equipment, nor can they accurately identify and locate trends, meaning they cannot precisely provide the time points when trends begin and end, hindering equipment maintenance personnel from troubleshooting and quickly resolving issues. This application provides a key indicator trend warning method. This method can not only provide trend warnings and location for high-frequency vibration indicators, but also for other indicator values ​​(such as temperature, flow rate, pressure, speed, wind speed, current, and voltage). Therefore, this application can be applied to scenarios such as monitoring performance degradation in industrial equipment, monitoring changes in process conditions, and addressing vibration sensor detachment or loosening, exhibiting broad coverage and strong practicality.

[0070] like Figure 1 The main process of a key indicator trend early warning method is described below.

[0071] Step S1: Obtain the original data, the first term coefficient of the convolution kernel, and the number of convolution kernels.

[0072] First, the process of acquiring raw data is as follows: starting from the current sampling time, acquire the key indicator data of industrial equipment within a specified time period as raw data y. The specified time can be set as needed. For example, when the sensors used to collect key indicator data generated by industrial equipment collect data very frequently, i.e., a large amount of key indicators can be acquired in a short period, the specified time can be shorter, such as one day; conversely, when the collection frequency is very low, in order to acquire sufficient raw data y, the specified time can be longer, such as one week.

[0073] The aforementioned industrial equipment refers to rotating mechanical devices in environments such as factories, power plants, and water plants, while data acquisition sensors refer to sensors that have the function of collecting key indicator data (such as vibration values, temperature, flow rate, rotational speed, pressure, etc.), such as vibration sensors, temperature sensors, flow meters, accelerometers, etc.

[0074] After acquiring key indicator data from the data acquisition sensor, the key indicator data is input as raw data y, and the data length L of the raw data y is calculated. The length L of the raw data y refers to the length of the data transmitted or stored in the computer. Simply put, after the key indicator data is input into the computer, the computer converts the key indicator data into binary values, which are the raw data y, and each value in the raw data y is a length unit.

[0075] After obtaining the original data y, scanning windows with different widths are generated based on the preset kernel leading coefficient 'a' and the number of kernels N. Specifically, the kernel leading coefficient 'a' is a percentage value representing the smallest unit size in the scanning window, such as 1 / 100, 2 / 100, ..., 100 / 100. When the administrator obtains the original data y, they can select an appropriate kernel leading coefficient 'a' based on the length of the original data y. The principle for selecting the leading coefficient 'a' of the convolution kernel is that the longer the length of the original data 'y', the larger the leading coefficient 'a' should be. For example, if there are leading coefficients of 1 / 100 and 5 / 100, and the lengths of the received original data 'y' are 100 and 10000 respectively, to reduce computer energy consumption, a leading coefficient of 1 / 100 is chosen when the original data 'y' is 100, and a leading coefficient of 5 / 100 is chosen when the original data 'y' is 1000. This ensures that the local data material extracted from the original data 'y' is sufficiently rich each time, while also reducing computer energy consumption. Of course, if the computer's processing power is sufficient, the length of the original data 'y' can be ignored for more refined analysis of the local data, and the smallest unit scale can be selected each time.

[0076] The principle for determining the number of convolutional kernels N is that the larger N is, the more convolutional kernels of different scales are generated (more scanning windows of different widths are generated), the richer the material of local data extracted from the original data y, and the greater the computing power consumption. Therefore, when choosing the number of convolutional kernels N, it is also necessary to match a suitable number of convolutional kernels N with the processing power of the computer. The correspondence between the computer's processing power and the number of convolutional kernels N can be obtained through a finite number of trials.

[0077] Step S2: Construct a convolution kernel model based on the first term coefficient and the number of convolution kernels.

[0078] After obtaining the first term coefficient 'a' and the number of convolution kernels 'N', the first step is to construct a convolution kernel model, which generates scanning windows with different widths. Then, the convolution kernel model is used to automatically scan the original data 'y'. Simply put, the convolution kernel model has the function of automatically scanning the original data 'y'.

[0079] Specifically, the process of constructing the convolution kernel model is as follows: First, calculate the tolerance. Among them, a is the leading coefficient of the convolution kernel, where 0 < a < 1, and N is the number of convolution kernels. Since the leading coefficient a of the convolution kernel and the number N of convolution kernels have been described in step S1, they will not be elaborated here. After obtaining the tolerance d, calculate the general formula for the width of the scanning window when scanning the original data y each time: C n = L(a + (n - 1)d), where C n refers to the width of the scanning window when scanning the original data y for the nth time being C n . In addition, L is the data length of the original data y, a is the leading coefficient of the convolution kernel, and d is the tolerance. Finally, calculate the moving step B of each scanning window n = (1 - f) * C n . The moving step B n refers to when scanning the original data y for the nth time, with the width of the scanning window being C n and the distance between two adjacent scanning windows being B n , self-driven scanning of the original data y from left to right. As Figure 2 shown, the width of the scanning window in the figure is 2 data lengths. If the moving step B n = 3, the scanning window will move from scanning position A to scanning position B.

[0080] Step S3: Use the convolution kernel model to scan the original data to obtain a scanning interface, and mark the local original data located in the scanning interface as the target vector. As Figure 4 shown, each target vector will execute steps S4 - S7. U corresponds to a scanning interface, and the local original data in this scanning interface is the target vector. V represents the bucket feature value corresponding to step S4, W represents the difference vector corresponding to step S5, X represents the estimated trend probability corresponding to step S6, and Y represents the calculated trend growth rate corresponding to step S7.

[0081] Step S4: Bucket the target vector to obtain the bucket feature value.

[0082] Step S5: Use the first-order difference function to calculate the difference vector of two adjacent bucket feature values.

[0083] Step S6: Calculate the trend probability based on the difference vector and qualitatively determine the trend of the scanning interface. <000022><0>Step S7: Calculate the trend growth rate based on the bucket feature value and quantitatively determine the trend of the scanning interface.

[0085] Through steps S3 to S7, the target vector cy in the scanning interface can be specifically analyzed, and the judgment logic for determining the overall trend form of the original data based on the target vector cy is as follows:

[0086] First, the convolution kernel model calculates the width C of the scanning window. n and the movement step size B n The self-driving system scans the original data y from left to right to obtain the number of scanned interfaces: Among them, S n Take the integer part; if it is not divisible, like Figure 3 As shown, assuming the first scanning window C1 scans to obtain the 1st, 2nd, ..., S1th scanning interfaces, the scanning interface corresponding to the target vector cy with the largest trend growth rate (upward trend) or the smallest trend growth rate (downward trend) in S1 is selected as the optimal trend interface and defined as F1; similarly, when scanning with the second scanning window C2, the scanning interface corresponding to the target vector cy with the largest trend growth rate (upward trend) or the smallest trend growth rate (downward trend) in S2 is selected as the optimal trend interface and defined as F2; ​​when scanning with the nth scanning window, S... n A scanning interface, in S n The optimal trend interface is selected as the scanning interface corresponding to the target vector cy with the largest trend growth rate (upward trend) or the smallest trend growth rate (downward trend), and is defined as F. n Thus, we obtain the set F = {F1, F2, ..., F...} n Finally, the trend growth rate, trend type, and trend position coordinates of the optimal trend interface with the largest trend growth rate (upward trend) or the smallest trend growth rate (downward trend) in set F are taken as the final trend analysis result of the original data y. The trend position coordinates (A1, B1) return the upper left corner of the optimal trend interface, and the trend position coordinates (A2, B2) return the lower right corner of the optimal trend interface. A1 and A2 represent the start and end times of the optimal trend interface, respectively, and B1 and B2 represent the maximum and minimum values ​​of the original data y in the optimal trend interface, respectively. Therefore, this application can accurately locate the start and end times of a trend, enabling managers to quickly trace and investigate the person in charge of the shift or specific action at the time the problem occurred, accelerating problem-solving efficiency and achieving the goal of intelligent operation and maintenance of industrial equipment.

[0087] The following details the process of determining the overall trend of the original data y. Specifically:

[0088] First, step S4 involves downsampling the target vector cy using a bucketing approach to reduce noise. This involves dividing the target vector cy into K equal parts of bucketed feature values ​​pcy. If equal division is not possible, the excess target vector cy is collected in a single bucket. For example... Figure 4As shown, if column U represents a scanning interface containing values ​​generated between January 2nd and January 11th, 2023, these values ​​are collectively referred to as the target vector. This target vector needs to be divided into 5 equal parts, each representing a bucket. The bucket characteristic value is then calculated for each bucket. The bucket characteristic value refers to the average or median value of the elements within the same bucket. In this example, the bucket characteristic value is the average of the elements within the same bucket, such as... Figure 4 In column V, the bucket feature value is 1.35 for elements 1.2 and 1.5 in the first bucket, 2.20 for elements 1.9 and 2.5 in the second bucket, 1.45 for elements 2.4 and 0.5 in the third bucket, 3.4 for elements 3.2 and 3.6 in the fourth bucket, and 4.20 for elements 4.1 and 4.3 in the fifth bucket. This results in a bucket vector [1.35, 2.20, 1.45, 3.40, 4.20] composed of the five bucket feature values ​​in column V. It should be noted that the buckets in this example are virtual boundaries, not physical objects.

[0089] Then, in step S5, the first-order difference function is used to calculate the difference vector dpcy from the pairwise adjacent bucket feature values. The first-order difference function is: dpcy ii∈[0,K-1] = pcy(ii+1)-pcy(ii), where pcy(ii+1) and pcy(ii) are the pairwise adjacent bucket feature values, and dpcy ii∈[0,K-1] It is a difference vector. For example... Figure 4 As shown, column V represents the binning feature values, and column W represents the differences between pairwise adjacent binning feature values, which is also a difference vector. That is, subtracting the first binning feature value (1.35) from the second binning feature value (2.20) in column V yields the first element of the difference vector W, which is 0.85. Similarly, the second element of the difference vector W is -0.75 (1.45 - 2.20), the third element is 1.95 (3.40 - 1.45), and the fourth element is 0.80 (4.20 - 3.40). The difference vector is dpcy. ii∈[0,K-1] The magnitude of dpcy can reflect the local trend between two adjacent bucket eigenvalues ​​pcy, such as dpcy. ii∈[0,K-1] =0 indicates that the characteristic values ​​of the two adjacent buckets are equal, which means there is no trend; if dpcy ii∈[0,K-1] A value >0 indicates that the characteristic value of the bucket in the later bucket is greater than the characteristic value of the bucket in the earlier bucket, suggesting a local upward trend; if dpcy ii∈[0,K-1]A value less than 0 indicates that the characteristic value of the latter bucket is less than that of the former bucket, suggesting a local downward trend. Therefore, the correlation of local trends among multiple adjacent buckets can be used to determine the trend changes in local raw data within the scanning interface. Local raw data refers to the raw data located within the scanning interface.

[0090] Next, the trend probability of the difference vector is calculated in step S6 to qualitatively characterize the trend of the target vector interface. The trend in this application includes four types: upward trend, downward trend, simultaneous upward and downward trends, and no trend (periodic fluctuations are also a form of no trend). These four trend forms cover all the possible forms that the original data y should have.

[0091] In this example, since the local trends obtained in step S5 are easily affected by factors such as transient operating conditions, oscillating fluctuations, abrupt changes, and outliers, it is necessary to statistically analyze the various local trends contained in the scanned interface to estimate the trend probability of the entire scanned interface, thereby enhancing the generalization performance and robustness of the algorithm. The process for initially determining which type of trend the local raw data in the scanned interface belongs to is as follows: First, calculate the upward trend probability. Here, `count(dpcy)` represents the total number of elements in the difference vector of the scanned interface, and `count(dpcy(x>0))` represents the number of difference vector values ​​greater than 0 in the scanned interface. Additionally, the probability of a downward trend is also calculated. Where count(dpcy) represents the total number of elements in the difference vector in the scanning interface, and count(dpcy(x<0)) represents the number of difference vector values ​​less than 0 in the scanning interface. For example... Figure 4 As shown, in column W, the total number of elements in the difference vector in the scanning interface is 4, and the number of difference vector values ​​greater than 0 is 3. Therefore, p(T1) = 3 / 4 = 0.75, and p(T2) = 1 / 4 = 0.25.

[0092] It should be noted that each scanned interface needs to undergo the above-mentioned process of calculating the probability of the trend. Then, a preset threshold th1 is obtained, where th1 represents the probability threshold for the overall trend. The relationship between the preset threshold th1 and the probability of an upward trend p(T1) and the probability of a downward trend p(T2) is then determined. When:

[0093] When the aforementioned upward trend probability p(T1) reaches the preset threshold th1, the local raw data in the scanning interface is identified as having an upward trend, indicating that the preliminary trend detection conditions have been met, and only then can step S7 be performed to calculate the trend growth rate.

[0094] When the aforementioned downward trend probability p(T2) reaches the preset threshold th1, the local raw data in the scanning interface is identified as having a downward trend, indicating that the preliminary trend detection conditions have been met, and only then can step S7 be performed to calculate the trend growth rate.

[0095] Therefore, by comparing the relationship between the preset threshold th1 and the probability of an upward trend p(T1) and the probability of a downward trend p(T2), we can preliminarily determine the trend of local raw data in a scanning interface.

[0096] In this application, the trend quantitative calculation in step S7 can only be performed after p(T1) or p(T2) reaches a preset threshold th1. Specifically: First, the bucket feature values ​​of the first H buckets in the scanning interface are extracted to obtain the value V1. Here, the bucket feature value refers to the average or median value. At the same time, the bucket feature values ​​of the last H buckets in the scanning interface are extracted as the value V2. Similarly, the value V2 here is the average or median value. It should be noted that if the bucket feature values ​​of the first H buckets are selected as the average, then the bucket feature values ​​of the last H buckets are also selected as the average, that is, it is necessary to ensure that the values ​​V1 and V2 contain the same content. Then, the trend growth rate of the target vector cy is calculated. When:

[0097] When both values ​​V1 and V2 are positive, calculate the trend growth rate: When value V1 is negative and value V2 is positive, calculate the trend growth rate: When value V1 is positive and value V2 is negative, calculate the trend growth rate: When both values ​​V1 and V2 are negative, calculate the trend growth rate:

[0098] For ease of representation, tr(T1) represents the growth rate of an upward trend, and tr(T2) represents the growth rate of a downward trend.

[0099] like Figure 4 As shown, in column Y, assuming H=1, which means the average value of the elements in the first bucket is selected, then the value V1=1.35. The average value of the elements in the last bucket, 4.20, is used as the value V2, thus obtaining tr(T1)=(4.20-1.35) / 1.35*100=211.11%.

[0100] Finally, combining the relationship between steps S6 and S7 and the preset thresholds th1 and th2, the trend warning logic is as follows: Figure 5 As shown. In this application, when:

[0101] When the upward trend probability p(T1) reaches a preset threshold th1 and its corresponding trend growth rate tr(T1) reaches a preset threshold th2, and the downward trend probability p(T2) reaches a preset threshold th1 and its corresponding trend growth rate tr(T2) reaches a preset threshold th2, the local raw data in the scanning interface is marked as having both an upward and downward trend.

[0102] When the upward trend probability p(T1) reaches the preset threshold th1 and its corresponding trend growth rate tr(T1) reaches the preset threshold th2, the local raw data in the scanning interface is marked as an upward trend.

[0103] When the downward trend probability p(T2) reaches a preset threshold th1 and its corresponding trend growth rate tr(T2) reaches a preset threshold th2, the local raw data in the scanning interface is marked as a downward trend.

[0104] When the upward trend probability p(T1) does not reach the preset threshold th1 and / or its corresponding trend growth rate tr(T1) does not reach the preset threshold th2, and the downward trend probability p(T2) does not reach the preset threshold th1 and / or its corresponding trend growth rate tr(T2) does not reach the preset threshold th2, the local raw data in the scanning interface is marked as having no trend.

[0105] Therefore, by comparing the above-mentioned probability p(T1) and the probability p(T2) of trend decline with the preset threshold th1, and comparing the trend growth rate tr(T1) and the trend growth rate tr(T2) with the preset threshold th2, the trend form of the local original data in each scanning interface can be determined.

[0106] In summary, only when the qualitative trend reaches a preset threshold th1 and the quantitative trend reaches a preset threshold th2 can it be finally determined that the local raw data in the scanning interface exhibits an upward trend, a downward trend, or both upward and downward trends; otherwise, it is treated as having no trend. Furthermore, after determining the trend form of the local raw data in the scanning interface and calculating the trend growth rate of the local raw data in each scanning interface, the scanning interface corresponding to the target vector cy with the largest or smallest trend growth rate is selected from the scanning interfaces generated from the same scanning window as the optimal trend interface. For example, the first scanning window C1 scans to obtain S1 scanning interfaces; the scanning interface corresponding to the target vector cy with the largest trend growth rate is selected as the optimal trend interface from S1. Then, the optimal trend interfaces selected under each scanning window are compared, i.e., the numerical values ​​in the set F are compared, and finally, the optimal trend interface with the largest value in set F is selected as the overall trend of the raw data y. Therefore, as... Figure 4As shown in column Z, the trend analysis results of this scanning interface show that the trend type is upward, the trend growth rate is 211.11%, and the trend position coordinates are (A1,B1) = (2023 / 1 / 2,4.3) and (A2,B2) = (2023 / 1 / 11,0.5).

[0107] In summary, the implementation principle of the key indicator trend early warning method in this application embodiment is as follows: when the original data y is obtained, a convolution kernel model is constructed by selecting the first term coefficient and the number of convolution kernels. Then, the convolution kernel model is used to scan the original data y to obtain multiple scanning interfaces. Qualitative trend calculation and quantitative trend calculation are then performed on each scanning interface. Finally, based on the quantitative trend calculation, the most representative optimal trend interface is selected from the scanning interfaces. This optimal trend interface not only includes coordinates but also the start and end times of the trend, so that managers can intuitively see the trend of industrial equipment and the start and end times of the trend, thereby monitoring the potential risks of industrial equipment in real time and achieving the purpose of intelligent operation and maintenance of industrial equipment.

[0108] This application provides a key indicator trend early warning system, such as Figure 6 As shown, a key indicator trend early warning system includes a data acquisition module 1, a first processing module 2, a second processing module 3, a third processing module 4, a fourth processing module 5, a first generation module 6, and a second generation module 7. The data acquisition module 1 is connected to the first processing module 2 and the second processing module 3, respectively. The first processing module 2, the second processing module 3, the third processing module 4, the fourth processing module 5, the first generation module 6, and the second generation module 7 are connected sequentially.

[0109] Specifically, data acquisition module 1 acquires the raw data, the first term coefficient of the convolution kernel, and the number of convolution kernels. First processing module 2 constructs a convolution kernel model based on the first term coefficient and the number of convolution kernels. Second processing module 3 scans the raw data using the convolution kernel model to obtain a scanning interface, and marks local raw data within the scanning interface as target vectors. Third processing module 4 bins the target vectors to obtain bin feature values. Fourth processing module 5 uses a first-order difference function to calculate the difference vector from the pairwise adjacent bin feature values. First generation module 6 calculates the trend probability of the scanning interface based on the difference vector and qualitatively characterizes the trend of the scanning interface. Second generation module calculates the trend growth rate of the scanning interface based on the bin feature values ​​and quantitatively characterizes the trend of the scanning interface. In short, data acquisition module 1, first processing module 2, second processing module 3, third processing module 4, fourth processing module 5, first generation module 6, and second generation module 7 work together to achieve real-time monitoring of potential risks in industrial equipment and to achieve trend-based localization.

[0110] To better execute the above-described method, this application also provides a terminal, which includes a memory and a processor.

[0111] The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the aforementioned key indicator trend early warning method; the data storage area may store data involved in the aforementioned key indicator trend early warning method.

[0112] A processor may include one or more processing cores. The processor executes instructions, programs, code sets, or instruction sets stored in memory, and calls data stored in memory to perform various functions and process data as described in this application. The processor may be at least one of a specific application-specific integrated circuit, a digital signal processor, a digital signal processing device, a programmable logic device, a field-programmable gate array, a central processing unit, a controller, a microcontroller, and a microprocessor. It is understood that, for different devices, the electronic devices used to implement the above-described processor functions may also be other types, and the embodiments of this application do not specifically limit the specific implementation.

[0113] This application also provides a computer-readable storage medium, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code. This computer-readable storage medium stores a computer program that can be loaded by a processor and executed using the aforementioned key indicator trend early warning method.

[0114] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for early warning of key indicator trends, characterized in that, include: Obtain the original data, the first term coefficient of the convolution kernel, and the number of convolution kernels; Construct a convolution kernel model based on the first term coefficient of the convolution kernel and the number of convolution kernels; The original data is scanned using the convolutional kernel model to obtain a scanning interface, and the local original data located in the scanning interface is marked as the target vector; The target vector is divided into buckets to obtain bucketed feature values; The difference vector is obtained by calculating the pairwise adjacent bucket feature values ​​using a first-order difference function; The trend probability of the scanned interface is calculated based on the difference vector, and the trend of the scanned interface is qualitatively characterized. The trend growth rate of the scanning interface is calculated based on the bucket characteristic value, and the trend of the scanning interface is quantitatively analyzed. The method further includes: From the scan interfaces generated by scanning within the same scan window, select the scan interface corresponding to the target vector with the largest or smallest trend growth rate as the optimal trend interface; Compare the optimal trend interface selected under each scanning window, and select the trend form of the optimal trend interface with the largest or smallest trend growth rate as the overall trend form of the original data.

2. The key indicator trend early warning method according to claim 1, characterized in that, The step of constructing a convolution kernel model based on the first term coefficient of the convolution kernel and the number of convolution kernels includes: Calculation tolerance: , where a is the leading coefficient of the convolution kernel, 0 < a < 1, and N is the number of convolution kernels; The general formula for the width of the scanning window when scanning the original data is calculated based on the tolerance d, the first term coefficient a of the convolution kernel, and N convolution kernels: Where n <= N, and L is the length of the original data; Obtain the span factor f, based on the span factor f and the general formula C for the width of the scanning window. n The calculated step size is: , where 0≤f<1.

3. The key indicator trend early warning method according to claim 2, characterized in that, The step of scanning the original data using the convolutional kernel model to obtain the scanning interface includes: using a scanning window with a width of C... n And the distance between two adjacent scanning windows is B n The scanning interface is obtained by self-driving scanning the raw data from left to right.

4. The key indicator trend early warning method according to claim 1, characterized in that, The step of dividing the target vector into buckets to obtain bucketed feature values ​​includes: dividing the target vector into K equal parts of bucketed feature values.

5. The key indicator trend early warning method according to claim 1, characterized in that, The method of using a first-order difference function to calculate the difference vector of pairwise adjacent bucket feature values ​​includes: Where pcy(ii+1) and pcy(ii) are the pairwise adjacent bucket feature values, and dpcy ii∈[0,K-1] It is a difference vector, and K is the K bucket eigenvalues.

6. The key indicator trend early warning method according to claim 1, characterized in that, The calculation of the trend probability of the scanned interface based on the difference vector includes: Calculate the probability of an upward trend Here, count(dpcy) represents the total number of elements in the difference vector in the scanning interface, and count(dpcy(x>0)) represents the number of difference vector values ​​greater than 0 in the scanning interface; Calculate the probability of a downward trend Here, count(dpcy) represents the total number of elements in the difference vector in the scanning interface, and count(dpcy(x<0)) represents the number of difference vector values ​​less than 0 in the scanning interface.

7. The key indicator trend early warning method according to claim 6, characterized in that, The step of qualitatively analyzing the trend of the scanned interface includes: Get the preset threshold th1, when: When the upward trend probability p(T1) reaches the preset threshold th1, the local raw data in the scanning interface is characterized as an upward trend. When the downward trend probability p(T2) reaches a preset threshold th1, the local raw data in the scanning interface is characterized as a downward trend.

8. The key indicator trend early warning method according to claim 1, characterized in that, The calculation of the trend growth rate of the scanning interface based on the bucket feature values ​​includes: The value V1 is obtained based on the binning feature values ​​of the first H bins on the scanning interface, and the value V2 is obtained based on the binning feature values ​​of the last H bins on the scanning interface. When: When both values ​​V1 and V2 are positive, calculate the trend growth rate: ; When the value V1 is negative, the trend growth rate is calculated as follows: ; When the value V1 is positive and the value V2 is negative, calculate the trend growth rate: ; When both values ​​V1 and V2 are negative, calculate the trend growth rate: .

9. The key indicator trend early warning method according to any one of claims 7 or 8, characterized in that, The trend forms include an upward trend, a downward trend, a simultaneous upward and downward trend, and no trend. The quantitative trend analysis of the scanned interface includes: Get the preset threshold th2, when: When the upward trend probability p(T1) reaches a preset threshold th1 and the trend growth rate tr(T1) reaches a preset threshold th2, the local raw data in the scanning interface is marked as an upward trend. When the probability of a downward trend p(T2) reaches a preset threshold th1 and the trend growth rate tr(T2) reaches a preset threshold th2, the local raw data in the scanning interface is identified as having a downward trend. When the upward trend probability p(T1) reaches a preset threshold th1 and the trend growth rate tr(T1) reaches a preset threshold th2, and the downward trend probability p(T2) reaches a preset threshold th1 and the trend growth rate tr(T2) reaches a preset threshold th2, the local raw data in the scanning interface is marked as having both an upward and downward trend. When the upward trend probability p(T1) does not reach the preset threshold th1 and / or the trend growth rate tr(T1) does not reach the preset threshold th2, and the downward trend probability p(T2) does not reach the preset threshold th1 and / or the trend growth rate tr(T2) does not reach the preset threshold th2, the local raw data in the scanning interface is marked as having no trend. tr(T1) represents the trend growth rate of an upward trend, and tr(T2) represents the trend growth rate of a downward trend.

10. A key indicator trend early warning system, used to execute the method described in any one of claims 1-9, characterized in that, include: The data acquisition module (1) is used to acquire the original data, the first term coefficient of the convolution kernel, and the number of convolution kernels; The first processing module (2) is used to construct a convolution kernel model based on the first term coefficient of the convolution kernel and the number of convolution kernels; The second processing module (3) is used to scan the original data using the convolution kernel model to obtain a scanning interface, and to mark the local original data located in the scanning interface as a target vector; The third processing module (4) is used to divide the target vector into buckets to obtain bucket feature values; The fourth processing module (5) is used to calculate the difference vector by using the first-order difference function to calculate the difference vector of the pairwise adjacent bucket feature values; The first generation module (6) is used to calculate the trend probability of the scanning interface based on the difference vector and to perform trend qualitative analysis on the scanning interface. The second generation module (7) is used to calculate the trend growth rate of the scanning interface based on the bucket feature value and to quantify the trend of the scanning interface.

11. A terminal, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions that can be loaded by the processor and executed according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1-9.

Images