Determination Method, Device and Electronic Device of Target Device

Abstract

The present specification discloses a method, apparatus, and electronic device for determining a target device. The method includes: obtaining characteristic data of multiple devices; obtaining, according to the characteristic data of the multiple devices, a predicted probability of a failure occurring in a predetermined time period starting from the characteristic data acquisition moment for each device; establishing a total cost objective function with a maintenance trigger threshold as a variable according to the predicted probabilities of failures occurring in each of the multiple devices, the maintenance costs in the case of failures of each device, and the maintenance costs of each device; solving the total cost objective function, and taking the maintenance trigger threshold when the value of the total cost objective function is the smallest as the target trigger threshold; and taking the devices with a predicted probability of failure greater than or equal to the target trigger threshold as target devices that need to be maintained. This solution can minimize the total maintenance cost of all devices, reduce the failure rate of the devices and the time when the devices cannot be used during failures, and reduce the workload of maintenance.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a method, apparatus, and electronic device for determining a target device. Background Art

[0002] Currently, the maintenance of many devices is performed regularly, for example, fixed maintenance is carried out four times a year. The total cost of repair costs and fixed maintenance costs after a device fails is relatively high.

[0003] When a device fails and needs to be repaired outside the fixed maintenance time, it is often necessary to contact a maintenance unit. The maintenance unit dispatches maintenance personnel, and the maintenance personnel arrange the maintenance time according to their workload and distance. Therefore, it often takes a long time from the occurrence of a failure to the elimination of the failure, especially in enterprises where failure repair requires hierarchical feedback. For example, the regular maintenance and temporary failure repair mechanisms of ATMs in banks are both hierarchical feedback methods, which require the participation of branches, higher-level branches, and maintenance providers. When an ATM fails, it is reported hierarchically from the branch to the maintenance provider, and the maintenance provider then dispatches maintenance personnel for repair, resulting in the situation that the faulty machine cannot be used for a long time.

[0004] In the current context of fierce competition among different enterprises and within the same enterprise, and high customer requirements, there are relatively high requirements for the availability of devices. The current device repair and maintenance mechanisms can no longer meet this requirement. Summary of the Invention

[0005] The purpose of the embodiments of this application is to provide a method, apparatus, and electronic device for determining a target device to solve the problems of high maintenance costs and long-time unusability of faulty devices in the existing device repair and maintenance mechanisms.

[0006] To solve the above technical problems, the first aspect of this specification provides a method for determining a target device, including: obtaining characteristic data of multiple devices; obtaining the predicted probability of each device failing within a predetermined time period starting from the moment when the characteristic data is collected according to the characteristic data of the multiple devices; establishing a total cost objective function with the maintenance trigger threshold as a variable according to the predicted probability of each device in the multiple devices failing, the repair cost in the case of each device failing, and the maintenance cost of each device; solving the total cost objective function, and taking the maintenance trigger threshold when the value of the total cost objective function is the smallest as the target trigger threshold; taking the devices whose predicted probability of failing is greater than or equal to the target trigger threshold as the target devices to be maintained.

[0007] In some embodiments, the total cost objective function with the maintenance trigger threshold as a variable includes one of the following: where K is the number of devices whose predicted probability is greater than or equal to the maintenance trigger threshold, pay kis the maintenance cost when the device fails; where K is the number of devices with a prediction probability greater than or equal to the maintenance trigger threshold, pay k is the maintenance cost when the device fails, and n is the maximum number of failures corresponding to non-zero prediction probabilities; where 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a prediction probability greater than or equal to the maintenance trigger threshold, pay k is the maintenance cost when the device fails; where 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a prediction probability greater than or equal to the maintenance trigger threshold, pay k is the maintenance cost when the device fails, and n is the maximum number of failures corresponding to non-zero prediction probabilities.

[0008] In some embodiments, to solve the total cost objective function and take the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold, the method includes: taking 0 to 1 as the target interval, selecting a value at a predetermined interval within the target interval as the maintenance trigger threshold, substituting it into the total cost objective function, and obtaining the total cost objective function value; selecting the maintenance trigger threshold corresponding to the minimum total cost objective function value as the reference value; repeatedly executing the following steps until the currently determined reference value is equal to the previously determined reference value in the previous cycle: taking an interval with a predetermined length centered on the current reference value as the current target interval, selecting a value at a predetermined interval within the current target interval as the maintenance trigger threshold, substituting it into the total cost objective function, and obtaining the total cost objective function value, selecting the maintenance trigger threshold corresponding to the minimum total cost objective function value as the current reference value; taking the currently determined reference value as the target trigger threshold.

[0009] In some embodiments, to obtain the prediction probability of each device failing within a predetermined period starting from the feature data acquisition moment according to the feature data of the multiple devices, the method includes: inputting the feature data of the multiple devices into a pre-established prediction model; taking the output of the prediction model as the prediction probability of each device failing within a predetermined period starting from the feature data acquisition moment.

[0010] In some embodiments, the prediction model is trained as follows: Obtain sample pairs of multiple devices, where the sample pairs include feature data and label values, and the label values are used to label whether a device fails within a predetermined time period starting from the moment when the feature data is collected; Based on the sample pairs of the multiple devices, train the prediction model.

[0011] In some embodiments, when the prediction model is a lightGBM model, the data set on the leaf node to be split in the lightGBM model is selected as follows: Calculate the sum of the mutual distances between all the data in the data set on each leaf node; Use the data set with the largest mutual distance between the data as the data set on the leaf node to be split.

[0012] In some embodiments, when the prediction model is a lightGBM model, the first data set on the leaf node is split into a second data set on the leaf node according to the following method: Calculate the sum of the distances between each data in the first data set and the rest of the data; Use the data with the largest sum of the distances to the rest of the data as the data in the second data set; Repeat the following steps until the difference is negative: Calculate the mean of the distances between each data point in the first data set and the rest of the data in the first data set as the first data, and calculate the mean of the distances from each data point in the first data set to all the data points in the second data set as the second data, and calculate the difference between the first data and the second data; Divide the data point with the largest difference into the second data set.

[0013] In some embodiments, before training the lightGBM model based on the sample pairs of the multiple devices, it further includes: Input the sample pairs into the lightGBM model; Determine the number of times each type of feature data is split as a leaf node; Screen out multiple types of feature data according to the number of splits; Correspondingly, when training the lightGBM model, use the sample pairs composed of the screened multiple types of feature data and label values to train the lightGBM model.

[0014] The second aspect of this specification provides a determining device for a target device, including: an acquisition module, including acquiring characteristic data of multiple devices; a prediction module, including obtaining a prediction probability of each device having a fault within a predetermined time period starting from the characteristic data acquisition moment according to the characteristic data of the multiple devices; a building module, configured to build a total cost objective function with a maintenance trigger threshold as a variable according to the prediction probability of each device in the multiple devices having a fault, the maintenance cost in the case of each device having a fault, and the maintenance cost of each device; a solving module, configured to solve the total cost objective function, and use the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold; a first determining module, configured to use the device whose prediction probability of having a fault is greater than or equal to the target trigger threshold as the target device to be maintained.

[0015] In some embodiments, the total cost objective function with the maintenance trigger threshold as a variable includes one of the following: where K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device having a fault; where K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device having a fault, and n is the maximum number of times of having a fault corresponding to the non-zero prediction probability; where 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device having a fault; where 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device having a fault, and n is the maximum number of times of having a fault corresponding to the non-zero prediction probability.

[0016] In some embodiments, the solving module includes: an obtaining sub-module, configured to use 0 to 1 as the target interval, select a value at a predetermined interval within the target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value; a first determination sub-module, configured to select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the reference value; the obtaining sub-module and the first determination sub-module are further configured to repeatedly execute the following steps until the currently determined reference value is equal to the currently determined reference value in the previous cycle: use an interval with a predetermined length centered on the current reference value as the current target interval, select a value at a predetermined interval within the current target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value, select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the current reference value; a second determination sub-module, configured to use the currently determined reference value as the target trigger threshold.

[0017] In some embodiments, the prediction module includes: an input sub-module, configured to input the feature data of the multiple devices into a pre-established prediction model; a third determination sub-module, configured to use the output of the prediction model as the prediction probability of each device having a failure within a predetermined time period starting from the feature data acquisition moment.

[0018] In some embodiments, a training module is further included, configured to train the prediction model; the training module includes: an acquisition sub-module, configured to acquire a sample pair of multiple devices, the sample pair including feature data and a marking value, the marking value being used to mark whether a device has a failure within a predetermined time period starting from the feature data acquisition moment; a training sub-module, configured to train the prediction model based on the sample pairs of the multiple devices.

[0019] In some embodiments, when the prediction model is a lightGBM model, the training module further includes a selection sub-module, configured to select the data set on the leaf node to be split in the lightGBM model; the selection sub-module includes: a first calculation sub-module, configured to calculate the sum of the mutual distances between all the data in the data set on each leaf node; a fourth determination sub-module, configured to use the data set with the maximum mutual distance between the data as the data set on the leaf node to be split.

[0020] In some embodiments, when the prediction model is a lightGBM model, the training module further includes a splitting sub-module for splitting the first data set on the leaf node into a second data set on the leaf node according to the following method; the splitting sub-module includes: a second calculation sub-module for calculating the sum of the distances between each data in the first data set and the rest of the data; a fifth determination sub-module for taking the data with the largest sum of the distances from the rest of the data as the data in the second data set; the second calculation sub-module and the fifth determination sub-module are further configured to repeatedly execute the following steps until the difference is negative: calculating the mean of the distances between each data point in the first data set and the rest of the data in the first data set as the first data, and calculating the mean of the distances from each data point in the first data set to all data points in the second data set as the second data, and calculating the difference between the first data and the second data; dividing the data point with the largest difference into the second data set.

[0021] In some embodiments, when the prediction model is a lightGBM model, it further includes: an input module for inputting the sample pair into the lightGBM model; a second determination module for determining the number of times each type of feature data is split as a leaf node; a screening module for screening out multiple types of feature data according to the number of splits; correspondingly, when training the lightGBM model, the sample pairs composed of the screened multiple types of feature data and the labeled values are used to train the lightGBM model.

[0022] The third aspect of this specification provides an electronic device, including: a memory and a processor, the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor realizes the steps of the method according to the first aspect or any of its embodiments by executing the computer instructions.

[0023] The fourth aspect of this specification provides a computer storage medium, the computer storage medium stores computer program instructions, and the computer program instructions, when executed, realize the steps of the method according to the first aspect or any of its embodiments.

[0024] The method, device and electronic device for determining a target device provided by the embodiments of this specification predict the probability of each device having a failure within a predetermined time period starting from the feature data acquisition moment according to the feature data of multiple devices, and use the device with a prediction probability greater than or equal to the target maintenance trigger threshold as the target device. Since the target maintenance trigger threshold corresponds to the situation where the total cost objective function has the minimum value, performing maintenance on the target device obtained according to the target maintenance trigger threshold in advance can minimize the total maintenance cost of all devices; it can also reduce the failure rate of the devices and the time when the devices cannot be used during a failure; during each maintenance, only the target device needs to be maintained, and there is no need to check each device one by one, reducing the workload of maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0026] Figure 1 Shows a flowchart of a method for determining a target device according to an embodiment of this specification;

[0027] Figure 2 Shows a flowchart of a method for determining a target maintenance trigger threshold in an embodiment;

[0028] Figure 3 Shows a flowchart of a method for obtaining a prediction probability;

[0029] Figure 4 Shows a flowchart of a method for training a prediction model;

[0030] Figure 5 Shows a flowchart of a method for selecting a data set on a leaf node to be segmented;

[0031] Figure 6 Shows a flowchart of a method for segmenting a first data set on a leaf node to obtain a second data set on the leaf node;

[0032] Figure 7 Shows a flowchart of a method for screening feature data types;

[0033] Figure 8 Shows a schematic block diagram of a device for determining a target device according to an embodiment of this specification;

[0034] Figure 9 Shows a schematic block diagram of an electronic device according to an embodiment of the present invention. Detailed implementation manners

[0035] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0036] This specification proposes a new equipment maintenance mechanism, that is, before the equipment is about to fail, it is maintained in advance, so as to reduce the failure rate and maintenance cost of the equipment, that is, changing the regular maintenance to a maintenance with flexible time.

[0037] For this reason, this specification also provides a method for determining a target device, which is used to determine which devices need to be maintained in advance within a period of time in the future. As Figure 1 shown, the method for determining the target device includes the following steps.

[0038] S110: Obtain the characteristic data of multiple devices.

[0039] The devices described in the embodiments of this specification may be self-service teller machines (such as ATM machines in banks), transportation means (such as taxis, utility vehicles, trains, subways, airplanes, etc.), household appliances (such as televisions, refrigerators, washing machines, air conditioners, etc.), electronic devices (such as computers, mobile phones, earphones, etc.). Of course, they may also be other devices.

[0040] In some embodiments, the "device" in this specification may also be each module or component included in a large device, and the specific scope covered by the "device" can be changed according to the actual situation.

[0041] The specific meaning of the "device" can be determined according to the actual situation. Or components or modules in the device. This specification will not list them one by one.

[0042] The key used components or modules of various devices are different, and the types of failures are also different. Some types of failures are predictable, while some types of failures are not predictable. The method for determining the target device provided in the embodiments of this specification only focuses on the predictable failure types of the device. For example, a bank ATM has line failures, cassette failures, receipt failures, card reader failures, etc. Some line failures are considered not predictable, and this method can focus on the failures of the cash dispensing and cash receiving modules.

[0043] The characteristic data of a device can include multiple types.

[0044] In some embodiments, the feature data may include duration information such as the duration from the data collection time to the production date of the device, the duration from the data collection time to the initial use date of the device, etc.

[0045] In some embodiments, the feature data may include the usage frequency information of the device. This information may be the average value, maximum value, and sum of the usage frequencies per time unit within a predetermined duration before the data collection time. For example, the average value, maximum value of the usage frequency per month within 6 months before the data collection time, and the sum of the usage frequencies within 6 months.

[0046] In some embodiments, the feature data may further include the brand information of the device.

[0047] In some embodiments, the usage frequency further includes the environmental information of the device usage. For example, it may be the average value, maximum value, etc. per time unit within a predetermined duration before the data collection time, such as the average temperature, maximum temperature, average humidity, maximum humidity, etc. per day within 6 months before the collection time for the environment where an outdoor unattended sensor device is located.

[0048] The difference between the collection times of these feature data should be within a predetermined time difference range. For example, the difference between the collection times of any two feature data is within 1 hour or 10 minutes. Correspondingly, the average value of the collection times of each feature data or any one of the collection times of the feature data can be used as the collection time of these feature data.

[0049] S120: Obtain the predicted probability of each device having a failure within a predetermined time period starting from the feature data collection time according to the feature data of multiple devices.

[0050] In some embodiments, the "multiple devices" may be of the same type. For example, the devices are all ATMs of a bank.

[0051] In some embodiments, the "multiple devices" may also be of different types. For example, the devices may include ATMs of a bank and self-service payment machines, where the self-service payment machines can be used to pay water bills, electricity bills, gas bills, etc. with a bank card.

[0052] The "predetermined time period" may be one week, that is, predict the predicted probability of each device having a failure within one week starting from the feature data collection time. Correspondingly, the prediction method of the target device shown can be executed once a week. Figure 1 The prediction method of the target device shown.

[0053] In some embodiments, the predicted probability obtained by the model may be a probability value, and this probability value indicates whether a failure occurs.

[0054] In some embodiments, the predicted probabilities obtained by the model may also be two or more non-zero probability values. Each probability value can correspond to a number of times, representing the probability of the device having a corresponding number of faults. The number of non-zero probability values corresponding to different devices may be different. For example, for device A, the non-zero probability values are: the probability of having 1 fault is 0.6, and the probability of having 2 faults is 0.1; for device B, the non-zero probability value is: the probability of having 1 fault is 0.4.

[0055] S130: Establish a total cost objective function with the maintenance trigger threshold as a variable according to the predicted probabilities of each device in multiple devices having faults, the maintenance costs in the case of each device having a fault, and the maintenance costs of each device.

[0056] The maintenance trigger threshold, that is, when the predicted probability is greater than or equal to the maintenance trigger threshold, it means that the device corresponding to the predicted probability should be maintained.

[0057] In some embodiments, the total maintenance cost in the case of the device having a fault can be included in the total cost objective function. Whether the device will have a fault within a predetermined time period starting from the feature data acquisition moment is determined by whether the predicted probability of the device is greater than or equal to the maintenance trigger threshold.

[0058] For example, the total cost objective function with the maintenance trigger threshold as a variable can be where K is the number of devices whose predicted probabilities are greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device having a fault.

[0059] Another example, the total cost objective function with the maintenance trigger threshold as a variable can be where K is the number of devices whose predicted probabilities are greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device having a fault, and n is the maximum number of times of having a fault corresponding to the non-zero predicted probability.

[0060] In some embodiments, multiple devices can also be regarded as a whole. The predicted probabilities corresponding to the devices having faults are summed to obtain P, and the value of P represents the overall performance of the multiple devices. Specifically, the smaller the value of P, the lower the degree of maintenance required, the better the overall performance of the multiple devices, and the less the maintenance cost; on the contrary, the larger the value of P, the higher the degree of maintenance required, the worse the overall performance of the multiple devices, and the more the maintenance cost. In this case, P representing performance and maintenance cost can be converted into cost and included in the total cost objective function. Whether the device will have a fault within a predetermined time period starting from the feature data acquisition moment is determined by whether the predicted probability of the device is greater than or equal to the maintenance trigger threshold.

[0061] For example, the total cost objective function with the maintenance trigger threshold as a variable can be

[0062] where 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a predicted probability greater than or equal to the maintenance trigger threshold, and pay k is the repair cost when a device fails.

[0063] Again, for example, the total cost objective function with the maintenance trigger threshold as a variable can be

[0064] where 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a predicted probability greater than or equal to the maintenance trigger threshold, pay k is the repair cost when a device fails, and n is the maximum number of failures corresponding to non-zero predicted probabilities.

[0065] S140: Solve the total cost objective function, and take the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold.

[0066] In the above total cost objective function, since the maintenance trigger threshold is a variable, the value of the maintenance trigger threshold can be adjusted within the range of 0 to 1 to obtain different values of the total cost objective function, and then the target maintenance trigger threshold corresponding to the minimum value of the total cost objective function can be determined.

[0067] S150: Take the devices with a predicted probability of failure greater than or equal to the target trigger threshold as the target devices to be maintained.

[0068] The target devices are the devices that need to be maintained starting from the feature data collection moment. Preferably, the maintenance work is completed within a predetermined time period starting from the feature data collection moment.

[0069] For the above method for determining target devices, the predicted probability of failure of each device within a predetermined time period starting from the feature data collection moment is predicted based on the feature data of multiple devices, and the devices with a predicted probability greater than or equal to the target maintenance trigger threshold are taken as target devices. Since the target maintenance trigger threshold corresponds to the situation where the total cost objective function takes the minimum value, maintaining the target devices in advance according to the target maintenance trigger threshold can minimize the total maintenance cost of all devices; it can also reduce the failure rate of the devices and the time when the devices cannot be used during failure; each time maintenance is performed, only the target devices need to be maintained, without the need to check each device one by one, reducing the workload of maintenance.

[0070] In some embodiments, as Figure 2 shown, step S140 may include the following steps.

[0071] S141: Using 0 to 1 as the target interval, select a value at a predetermined interval within the target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value.

[0072] S142: Select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the reference value.

[0073] S143: Use an interval with a predetermined length centered on the current reference value as the current target interval. Select a value at a predetermined interval within the current target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value. Select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the current reference value.

[0074] Each time step S143 is executed, its predetermined interval is different. Generally, the predetermined interval is less than half of the current target interval.

[0075] S144: Determine whether the current reference value is equal to the current reference value determined in the previous cycle. If they are equal, execute step S145; otherwise, continue to execute step S143.

[0076] When initially making the judgment in step S144, the reference value determined in step S142 can be used as the current reference value determined in the previous cycle.

[0077] S145: Use the determined current reference value as the target trigger threshold.

[0078] In some embodiments, as Figure 3 shown, step S120 can obtain the prediction probability through the following steps.

[0079] S310: Input the feature data of multiple devices into a pre-established prediction model.

[0080] S320: Use the output of the prediction model as the prediction probability that each device will fail within a predetermined time period starting from the feature data acquisition moment.

[0081] In some embodiments, as Figure 4 shown, the prediction model in step S310 can be trained according to the following method.

[0082] S410: Obtain sample pairs of multiple devices. The sample pair includes feature data and a marked value, and the marked value is used to mark whether the device fails within a predetermined time period starting from the feature data acquisition moment.

[0083] For the description of the feature data, please refer to S110.

[0084] In some embodiments, the prediction model can also predict the number of times a device fails and the corresponding prediction probability within a predetermined time period starting from the feature data collection moment. Correspondingly, when training the prediction model, the sample pair can also include the number of times a device fails within a predetermined time period starting from the feature data collection moment.

[0085] S420: Train the prediction model based on the sample pairs of multiple devices.

[0086] The prediction model can be any machine learning model, such as an RNN model, a decision tree model, etc. The decision tree model can be, for example, a lightGBM model.

[0087] In the case of using the lightGBM model, the histogram algorithm can be used to select the data set on the leaf node to be split and split the data set on the leaf node.

[0088] The embodiments of this specification provide a clustering method to select the data set on the leaf node to be split, as Figure 5 shown, and this method includes the following steps.

[0089] S510: Calculate the sum of the mutual distances between all the data in the data set on each leaf node.

[0090] That is, calculate the distance between each pair of data in the set and sum up all the distances.

[0091] For example, there are two data sets U1: (1, 5, 7, 4) and U2: (1, 8, 4, 20) on two leaf nodes.

[0092] Calculate the sum of distances d1 corresponding to U1: The sum of the distances between 1 and the other data is 4 + 6 + 3 = 13, the sum of the distances between 5 and the other data is 4 + 2 + 1 = 7, the sum of the distances between 7 and the other data is 6 + 2 + 3 = 11, the sum of the distances between 4 and the other data is 3 + 1 + 3 = 7, and the sum of distances d1 corresponding to U1 = 13 + 7 + 11 + 7 = 38.

[0093] Calculate the sum of distances d2 corresponding to U2: The sum of the distances between 1 and the other data is 7 + 3 + 19 = 29, the sum of the distances between 8 and the other data is 7 + 4 + 12 = 23, the sum of the distances between 4 and the other data is 3 + 4 + 16 = 23, the sum of the distances between 20 and the other data is 19 + 12 + 16 = 47, and the sum of distances d2 corresponding to U2 = 29 + 23 + 23 + 47 = 122.

[0094] S520: Use the data set with the largest mutual distance between data as the data set on the leaf node to be split.

[0095] Continuing with the above example, since d2 > d1, U2 is selected as the leaf node to be split.

[0096] An embodiment of this specification provides a method based on clustering to split the first data set on the leaf node to obtain the second data set on the leaf node, as Figure 6 shown. This method includes the following steps.

[0097] S610: Calculate the sum of the distances between each data in the first data set and the rest of the data.

[0098] S620: Use the data with the largest sum of distances from the rest of the data as the data in the second data set.

[0099] S630: Calculate the mean of the distances between each data point in the first data set and the rest of the data in the first data set as the first data, and calculate the mean of the distances from each data point in the first data set to all data points in the second data set as the second data, and calculate the difference between the first data and the second data.

[0100] S640: Divide the data point with the largest difference into the second data set.

[0101] S650: Determine whether the difference is negative. If so, end the split; if not, continue to execute step S630.

[0102] Continuing with the above example, split the first data set U2: (1, 8, 4, 20) according to the following method:

[0103] Calculated as follows: The sum of the distances between 1 and the rest of the data is 7 + 3 + 19 = 29, the sum of the distances between 8 and the rest of the data is 7 + 4 + 12 = 23, the sum of the distances between 4 and the rest of the data is 3 + 4 + 16 = 23, and the sum of the distances between 20 and the rest of the data is 19 + 12 + 16 = 47; 47 is the largest in the sum of distances, so the data 20 corresponding to 47 is used as the newly split data set, that is, the data in the second data set U3. After this split, the first data set U2: (1, 8, 4), and the second data set U3 (20).

[0104] Calculate in the first data set U2: The distance between 1 and the rest of the data is 7 + 3 = 10; the distance between 8 and the rest of the data is 7 + 4 = 11, and the distance between 4 and the rest of the data is 3 + 4 = 7; the average value of the distances is (10 + 11 + 7) ÷ 3 = 9.333. Calculate in the first data set U2: The distance between 1 and 20 in the second data set U3 is 19, the distance between 8 and 20 in the second data set U3 is 12, and the distance between 4 and 20 in the second data set U3 is 16; the average value of the distances is (19 + 12 + 16) ÷ 3 = 15.666.

[0105] Since 9.333 - 15.666 < 0, the segmentation ends. After the final segmentation, the first data set U2 is obtained: (1, 8, 4), and the second data set U3 is (20).

[0106] In some embodiments, before training the lightGBM model based on sample pairs of multiple devices, it is also necessary to screen the feature data to screen out the types of feature data that have a greater impact on the value of the prediction probability and discard the types of feature data that have a smaller impact on the value of the prediction probability, so as to reduce the computational amount of the model. The method of manual experience screening can be adopted, or the embodiments of this specification provide a method for screening the types of feature data, such as Figure 7 As shown, this screening method includes the following steps:

[0107] S710: Input the sample pair into the lightGBM model.

[0108] The lightGBM model in this step uses the initial parameter values.

[0109] After screening out the feature data, the process of training the model is the process of adjusting the parameter values in the model.

[0110] S720: Determine the number of times each type of feature data is split as a leaf node.

[0111] S730: Screen out multiple types of feature data according to the number of splits.

[0112] For the above method of screening feature data, taking the number of times each type of feature data is split as a leaf node as the importance of this type of feature data, the importance can be arranged from large to small, and the top predetermined number (for example, the top 10) of feature data can be screened out; or the importance can be arranged from large to small, and the ratio of the number of feature data screened from the front row to the total number of all feature data is a predetermined ratio, for example, screening out the feature data with the top 50% importance.

[0113] Corresponding to the above steps S710 to S730, when training the lightGBM model, a sample pair composed of the screened feature data of multiple types and the marked values is used to train the lightGBM model.

[0114] The embodiment of the present specification provides a determining device for a target device, which can be used to implement Figure 1 the described method for determining a target device. As Figure 8 shown, the device includes an acquisition module 10, a prediction module 20, a construction module 30, a solution module 40, and a first determination module 50.

[0115] The acquisition module 10 includes acquiring feature data of multiple devices.

[0116] The prediction module 20 includes obtaining, according to the feature data of multiple devices, the prediction probability that each device will fail within a predetermined time period starting from the feature data acquisition moment.

[0117] The construction module 30 is used to establish a total cost objective function with the maintenance trigger threshold as a variable according to the prediction probability of each device in multiple devices failing, the maintenance cost in the case of each device failing, and the maintenance cost of each device.

[0118] The solution module 40 is used to solve the total cost objective function, and take the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold.

[0119] The first determination module 50 is used to use the device whose prediction probability of failure is greater than or equal to the target trigger threshold as the target device that needs to be maintained.

[0120] In some embodiments, the total cost objective function with the maintenance trigger threshold as a variable includes one of the following: Among them, K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device failing; Among them, K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device failing, and n is the maximum number of failures corresponding to the non-zero prediction probability; Among them, 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices whose prediction probability is greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in the case of the device failing; Among them, 0 < p k < 1 and p k≥λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a predicted probability greater than or equal to the maintenance trigger threshold, and pay k is the repair cost in the event of a device failure, and n is the maximum number of failures corresponding to a non-zero predicted probability.

[0121] In some embodiments, the solution module 40 includes an obtaining sub-module 41, a first determination sub-module 42, and a second determination sub-module 43.

[0122] The obtaining sub-module 41 is configured to use 0 to 1 as the target interval, select a value at a predetermined interval within the target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value.

[0123] The first determination sub-module 42 is configured to select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the reference value.

[0124] The obtaining sub-module 41 and the first determination sub-module 42 are further configured to repeatedly execute the following steps until the currently determined reference value is equal to the currently determined reference value in the previous cycle: use the interval with a predetermined length centered on the current reference value as the current target interval, select a value at a predetermined interval within the current target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value, and select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the current reference value.

[0125] The second determination sub-module 43 is configured to use the currently determined reference value as the target trigger threshold.

[0126] In some embodiments, the prediction module 20 includes an input sub-module 21 and a third determination sub-module 22.

[0127] The input sub-module 21 is configured to input the feature data of multiple devices into a pre-established prediction model. The third determination sub-module 22 is configured to use the output of the prediction model as the predicted probability of a device failing within a predetermined time period starting from the feature data acquisition moment.

[0128] In some embodiments, the device further includes a training module 60 for training the prediction model. The training module 60 may include an acquisition sub-module 61 and a training sub-module 62.

[0129] The acquisition sub-module 61 is configured to acquire sample pairs of multiple devices. The sample pair includes feature data and a label value, and the label value is used to label whether a device fails within a predetermined time period starting from the feature data acquisition moment. The training sub-module 62 is configured to train the prediction model based on the sample pairs of multiple devices.

[0130] In some embodiments, when the prediction model is a LightGBM model, the training module 60 further includes a selection sub-module 63 for selecting the data set on the leaf node to be split in the LightGBM model.

[0131] The selection sub-module 63 may include a first calculation sub-module 631 and a fourth determination sub-module 632.

[0132] The first calculation sub-module 631 is used to calculate the sum of the mutual distances between all the data in the data set on each leaf node. The fourth determination sub-module 632 is used to use the data set with the largest mutual distance between the data as the data set on the leaf node to be split.

[0133] In some embodiments, when the prediction model is a LightGBM model, the training module 60 further includes a splitting sub-module 64 for splitting the first data set on the leaf node according to the following method to obtain a second data set on the leaf node.

[0134] The splitting sub-module 64 may include a second calculation sub-module 641 and a fifth determination sub-module 642.

[0135] The second calculation sub-module 641 is used to calculate the sum of the distances between each data in the first data set and the rest of the data.

[0136] The fifth determination sub-module 642 is used to use the data with the largest sum of distances from the rest of the data as the data in the second data set.

[0137] The second calculation sub-module 641 and the fifth determination sub-module 642 are further used to repeatedly execute the following steps until the difference is negative: calculate the mean of the distances between each data point in the first data set and the rest of the data in the first data set as the first data, and calculate the mean of the distances from each data point in the first data set to all the data points in the second data set as the second data, and calculate the difference between the first data and the second data; divide the data point with the largest difference into the second data set.

[0138] In some embodiments, when the prediction model is a LightGBM model, the device further includes an input module 70, a second determination module 80, and a screening module 90.

[0139] The input module 70 is used to input the sample pair into the LightGBM model.

[0140] The second determination module 80 is used to determine the number of times each type of feature data is split as a leaf node.

[0141] The screening module 90 is used to screen out multiple types of feature data according to the number of splits.

[0142] Accordingly, when training the lightGBM model, multiple types of feature data and sample pairs composed of labeled values are used to train the lightGBM model.

[0143] For specific details of the above device, reference can be made to Figures 1 to 7 the corresponding relevant descriptions and effects in the embodiments of

[0144] An embodiment of the present invention also provides an electronic device, as Figure 9 shown. The electronic device may include a processor 91 and a memory 92. The processor 91 and the memory 92 may be connected by a bus or other means. Figure 9 Taking connection by bus as an example.

[0145] The processor 91 may be a central processing unit (CPU). The processor 91 may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or combinations of the above types of chips.

[0146] The memory 92, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions / modules corresponding to the method for determining a target device in an embodiment of the present invention (for example, Figure 8 the acquisition module 10, prediction module 20, establishment module 30, solution module 40, and first determination module 50 shown). By running the non-transitory software programs, instructions, and modules stored in the memory 92, the processor 91 can execute various functional applications and data classification of the processor, that is, implement the method for determining a target device in the above method embodiments.

[0147] The memory 92 may include a program storage area and a data storage area. The program storage area may store an operating system and application programs required for at least one function. The data storage area may store data created by the processor 91 and the like. In addition, the memory 92 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory 92 may optionally include a memory remotely disposed relative to the processor 91, and these remote memories may be connected to the processor 91 through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0148] The one or more modules are stored in the memory 92 and, when executed by the processor 91, perform the method for determining a target device in the embodiment as Figures 1 to 3 shown.

[0149] Specific details of the above electronic device may be understood by referring to the corresponding relevant descriptions and effects in the embodiments of Figures 1 to 7 and will not be elaborated herein.

[0150] Those skilled in the art can understand that to implement all or part of the processes in the above method embodiments, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it may include the processes of the above method embodiments. Among them, the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above types of memories.

[0151] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many method flow improvements today can be regarded as direct improvements to hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented using a hardware entity module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by a user's programming of the device. Designers can program themselves to "integrate" a digital system onto a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip 2. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compilers used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL), and there is not just one type of HDL, but many types, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog 2. Those skilled in the art should also be aware that by simply performing some logical programming on the method flow using the above-mentioned several hardware description languages and programming it into an integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0152] Each embodiment in this specification is described in a progressive manner. For the parts that are the same or similar among the embodiments, reference can be made to each other, and each embodiment focuses on the differences from other embodiments.

[0153] The systems, devices, modules, or units described in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions.

[0154] For the convenience of description, when describing the above devices, they are divided into various units according to functions and described separately. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0155] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of certain parts of each embodiment of the present application.

[0156] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on.

[0157] The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0158] Although the present application is depicted through embodiments, those of ordinary skill in the art know that the present application has many variations and changes without departing from the spirit of the present application. It is hoped that the appended claims will cover these variations and changes without departing from the spirit of the present application.

Claims

1. A method for determining a target device, characterized in that Including: Obtaining characteristic data of multiple devices; Based on the characteristic data of the multiple devices, obtaining the predicted probability of each device having a fault within a predetermined time period starting from the characteristic data acquisition moment; Based on the predicted probability of each device in the multiple devices having a fault, the maintenance cost in the case of each device having a fault, and the maintenance cost of each device, establishing a total cost objective function with the maintenance trigger threshold as a variable; if the predicted probability corresponding to a device is greater than or equal to the maintenance trigger threshold, it means that the device should be maintained; Solving the total cost objective function, and taking the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold; Regarding the devices whose predicted probability of having a fault is greater than or equal to the target trigger threshold as target devices that need to be maintained; The solving the total cost objective function, and taking the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold includes: Taking 0 to 1 as the target interval, selecting a value at a predetermined interval within the target interval as the maintenance trigger threshold, substituting it into the total cost objective function, and obtaining the total cost objective function value; Selecting the maintenance trigger threshold corresponding to the minimum total cost objective function value as the reference value; Repeatedly executing the following steps until the currently determined reference value is equal to the previously determined reference value in the previous cycle: taking the interval with a predetermined length centered on the current reference value as the current target interval, selecting a value at a predetermined interval within the current target interval as the maintenance trigger threshold, substituting it into the total cost objective function, and obtaining the total cost objective function value, and selecting the maintenance trigger threshold corresponding to the minimum total cost objective function value as the current reference value; Taking the currently determined reference value as the target trigger threshold.

2. The method according to claim 1, characterized in that, The total cost objective function with the maintenance trigger threshold as a variable includes at least one of the following: Among them, K is the number of devices with a predicted probability greater than or equal to the maintenance trigger threshold, and pay k is the maintenance cost in case of device failure; where K is the number of devices with a predicted probability greater than or equal to the maintenance trigger threshold, pay k is the repair cost in the case of device failure, and n is the maximum number of failures corresponding to non-zero predicted probabilities; Among them, 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a prediction probability greater than or equal to the maintenance trigger threshold, and pay k is the repair cost in the case of device failure; Among them, 0 < p k < 1 and p k ≥ λ, where λ is the maintenance trigger threshold; α is a predetermined coefficient, K is the number of devices with a predicted probability greater than or equal to the maintenance trigger threshold, pay k is the repair cost in case of device failure, and n is the maximum number of failures corresponding to non-zero predicted probabilities.

3. The method according to claim 1, wherein The obtaining the predicted probability of each device having a fault within a predetermined time period starting from the characteristic data acquisition moment based on the characteristic data of the multiple devices includes: Inputting the characteristic data of the multiple devices into a pre-established prediction model; Taking the output of the prediction model as the predicted probability of each device having a fault within a predetermined time period starting from the characteristic data acquisition moment.

4. The method according to claim 3, characterized in that The prediction model is trained in the following manner: Obtaining sample pairs of multiple devices, where the sample pairs include characteristic data and marked values, and the marked values are used to mark whether the device has a fault within a predetermined time period starting from the characteristic data acquisition moment; Training the prediction model based on the sample pairs of the multiple devices.

5. The method according to claim 3, characterized in that In the case where the prediction model is a lightGBM model, the data set selected on the leaf node to be split in the lightGBM model includes: Calculating the sum of the mutual distances between all data in the data set on each leaf node; Taking the data set with the maximum mutual distance between the data as the data set on the leaf node to be split.

6. The method according to claim 4, wherein In the case where the prediction model is a lightGBM model, splitting the first data set on the leaf node according to the following method to obtain the second data set on the leaf node: Calculate the sum of distances between each data in the first data set and the rest of the data; Take the data with the largest sum of distances from the rest of the data as the data in the second data set; Repeat the following steps until the difference is negative: Calculate the mean of the distances between each data point in the first data set and the rest of the data in the first data set as the first data, and calculate the mean of the distances from each data point in the first data set to all data points in the second data set as the second data, and calculate the difference between the first data and the second data; Partition the data point with the largest difference into the second data set.

7. The method according to claim 4, characterized in that Before training the lightGBM model based on the sample pairs of the multiple devices, it further includes: Input the sample pairs into the lightGBM model; Determine the number of times each type of feature data is split as a leaf node; Filter out multiple types of feature data according to the number of splits; Correspondingly, when training the lightGBM model, use the sample pairs composed of the filtered multiple types of feature data and the labeled values to train the lightGBM model.

8. A determining device for a target device, characterized in that, It includes: An acquisition module, including acquiring the feature data of multiple devices; A prediction module, including obtaining the predicted probability of each device having a failure within a predetermined time period starting from the feature data acquisition time according to the feature data of the multiple devices; A construction module, used to construct a total cost objective function with the maintenance trigger threshold as a variable according to the predicted probabilities of each device in the multiple devices having a failure, the maintenance costs in the case of each device failure, and the maintenance costs of each device; if the predicted probability corresponding to a device is greater than or equal to the maintenance trigger threshold, it means that the device should be maintained; A solution module, used to solve the total cost objective function, and take the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold; A first determination module, used to take the devices with the predicted probability of failure greater than or equal to the target trigger threshold as the target devices to be maintained; The solving of the total cost objective function, and taking the maintenance trigger threshold when the total cost objective function takes the minimum value as the target trigger threshold includes: Using 0 to 1 as the target interval, select a value at a predetermined interval within the target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value; Select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the reference value; Repeat the following steps in a loop until the currently determined reference value is equal to the previously determined reference value in the previous loop: use the interval with a predetermined length centered on the current reference value as the current target interval, select a value at a predetermined interval within the current target interval as the maintenance trigger threshold, substitute it into the total cost objective function, and obtain the total cost objective function value, and select the maintenance trigger threshold corresponding to the minimum total cost objective function value as the current reference value; Take the currently determined reference value as the target trigger threshold.

9. An electronic device, characterized in that, It includes: A memory and a processor, which are communicatively connected to each other. Computer instructions are stored in the memory, and the processor realizes the steps of the method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer storage medium, characterized in that, The computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the method according to any one of claims 1 to 7 are realized.

Images