A machine learning based hydrogen station compressor failure predictive maintenance decision system

Abstract

The present application relates to the technical field of machine learning, and particularly relates to a hydrogen station compressor failure predictive maintenance decision system based on machine learning. The present application obtains compressor historical operation data, hydrogen station business demand data and maintenance human resource data; different business characteristic time periods are divided according to the business demand data, and maintenance cost parameters are determined, a dynamic failure probability sequence is generated by using a machine learning model, failure risk parameters are calculated in combination with the business demand data and the maintenance cost parameters, and human resource evaluation parameters are obtained based on the human resource data; the maintenance cost parameters, the failure risk parameters and the human resource evaluation parameters are comprehensively calculated to obtain comprehensive loss parameters of each candidate maintenance window, and a window with the minimum comprehensive loss is selected as a recommended maintenance time window. The present application comprehensively considers the equipment failure risk, business load, maintenance cost and human resource constraints of the compressor, realizes optimal decision of the maintenance timing, and thus reduces commercial losses caused by unplanned shutdown.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and specifically to a machine learning-based predictive maintenance decision system for hydrogen refueling station compressor failures. Background Technology

[0002] Hydrogen-powered transportation is currently in a phase of commercial expansion globally, and the operational economics of hydrogen refueling stations, as the core infrastructure supporting the operation of hydrogen-powered vehicles, are of great concern. The operation of hydrogen refueling stations faces high capital and operating expenditures, with maintenance and downtime losses of the compressor system accounting for a significant portion of these expenditures. Hydrogen compressors require frequent start-stop cycles under extreme conditions to ensure rapid vehicle refueling, which can lead to frequent mechanical failures and unpredictable unexpected downtime. The compressor maintenance model directly impacts the profitability of hydrogen refueling stations; how to reduce maintenance costs while ensuring equipment reliability remains an unsolved problem in this field.

[0003] Traditional hydrogen refueling station compressor maintenance often employs fixed-cycle preventative maintenance or reactive repair models. At the equipment monitoring level, reliance on single-point sensor threshold alarms and manual inspections suffers from response delays and insufficient coverage, making it difficult to detect early signs of failure. At the operational decision-making level, the lack of a unified digital monitoring and predictive maintenance decision-making platform prevents dynamic assessment of compressor failure risks and hinders the comprehensive consideration of multi-dimensional constraints such as maintenance costs, workload, and personnel availability to determine the optimal maintenance timing. At the task execution level, traditional dispatch systems rely on simple rules when multiple stations are operating concurrently, easily leading to order-grabbing deadlocks. The lack of dynamic coordination between maintenance tasks, personnel scheduling, and spare parts inventory results in low operational efficiency, spare parts backlog, and commercial losses due to unplanned downtime. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures. The specific technical solution adopted is as follows: A machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures, comprising: a data acquisition module, used to acquire historical operating data of the compressor, business demand data of the hydrogen refueling station, and maintenance human resource data within a preset period; historical operating data includes the number of compressor start-ups and shutdowns, the time of each start-up and shutdown, and the compressor vibration spectrum; business demand data includes the predicted hydrogen refueling demand and the maximum hydrogen production capacity of the hydrogen refueling station within each preset period; maintenance human resource data includes the number of maintenance personnel and their availability; and a parameter analysis module, used to... Based on business demand data, each preset period is divided into different business characteristic time periods; the maintenance cost per unit time corresponding to different business characteristic time periods is obtained from the hydrogen refueling station maintenance records, serving as the maintenance cost parameter; historical operating data is analyzed using a machine learning model to generate a dynamic failure probability sequence for the compressor; based on the dynamic failure probability sequence, business demand data, and maintenance cost parameters, the accumulated failure risk parameter due to the compressor not performing maintenance operations is calculated; based on human resource data, human resource assessment parameters are obtained; the maintenance decision module is used to obtain the comprehensive loss parameter for the compressor to perform maintenance operations based on the maintenance cost parameter, failure risk parameter, and human resource assessment parameter; and a recommended maintenance time window is selected based on the comprehensive loss parameter.

[0005] Furthermore, the machine learning model is a multivariate long short-term memory network.

[0006] Furthermore, based on business demand data, each preset period is divided into different business characteristic time periods, including: setting the preset period to 1 day, and dividing the preset period into 5 business characteristic time periods based on the hydrogen refueling volume at different times within the preset period in the hydrogen refueling station's historical records, namely: late night off-peak period, from 0:00 to 6:00, morning peak fast charging period from 6:00 to 9:00, daytime off-peak period from 9:00 to 17:00, evening peak fast charging period from 17:00 to 21:00, and nighttime off-peak period from 21:00 to 24:00.

[0007] Furthermore, the method for obtaining maintenance cost parameters includes: obtaining historical maintenance records and maintenance costs for different business characteristic periods within a preset period; and obtaining maintenance cost parameters according to the maintenance cost parameter calculation formula, which is shown below: In the formula, Indicates the first [number] within the preset period The compressor maintenance cost parameters within a specific business period; Indicates the first The number of preset unit times within each business characteristic period; Indicates the first Within the first business characteristic period The preset unit time to the first Repair costs between preset unit time periods; This represents the maximum maintenance cost recorded in the historical maintenance history of hydrogen refueling stations. Indicates the first A preset unit of time; Indicates the first A preset unit of time.

[0008] Furthermore, the method for obtaining the dynamic failure probability sequence includes: using a machine learning model to analyze historical operating data to generate a dynamic failure probability sequence of the compressor for all preset unit times within a preset period; wherein each value in the dynamic failure probability sequence represents the probability of a failure occurring within a preset unit time.

[0009] Furthermore, the method for obtaining fault risk parameters includes: obtaining the predicted hydrogen demand and the maximum hydrogen production capacity of the hydrogen refueling station for each preset unit time within a preset period; using the ratio of the predicted hydrogen demand to the maximum hydrogen production capacity for each preset unit time as the dynamic load saturation; obtaining candidate maintenance windows, the length of which is an integer multiple of the preset unit time; and obtaining fault risk parameters according to the fault risk parameter calculation formula, which is shown below: In the formula, This indicates the accumulated fault risk parameters resulting from the compressors in the candidate maintenance window not undergoing maintenance operations within a preset period; This indicates the number of business characteristic time periods spanned by the candidate repair window within a preset period; Indicates the first [number] within the preset period The compressor maintenance cost parameters within a specific business period; Indicates the candidate repair windows within the preset period; Indicates the first The overlapping area between the business characteristic time period and the candidate maintenance window; Indicates the first The probability of compressor failure within a preset unit of time; This indicates the number of hydrogen refueling stations. Predicted hydrogen addition per preset unit time period; This indicates the maximum hydrogen production capacity of the hydrogen refueling station; This represents an exponential function with the natural constant as its base.

[0010] Furthermore, the method for obtaining human resource evaluation parameters includes: calculating human resource evaluation parameters for each preset unit of time within a preset period, for different business characteristics, using the following calculation formula: In the formula, Indicates the first Human resource evaluation parameters within a preset unit of time; Indicates the first The number of professional maintenance personnel required within a preset unit of time; Indicates the first The number of maintenance personnel available within a preset unit of time; This represents a non-zero minimum constant, used to prevent the denominator from being 0, and is set to... ; This represents the logarithmic function with the natural constant as the base.

[0011] Furthermore, the method for obtaining the comprehensive loss parameter includes: for candidate maintenance windows within a preset period, calculating the comprehensive loss parameter according to the following formula: In the formula, Indicates the time sequence number of the preset unit within the preset period; Indicates the candidate repair window; This represents the overall loss parameters for the compressor when performing maintenance operations within the candidate maintenance window; Indicates the first Maintenance cost parameters for each preset unit of time within a business characteristic period; This represents the average value of human resources assessment parameters within the candidate maintenance window; This indicates that the accumulated fault risk parameters for repairs will not be executed within the candidate repair window; Indicates time The projected demand for hydrogen; Indicates the dimension conversion coefficient; This represents the maximum maintenance cost recorded in the historical maintenance history of hydrogen refueling stations.

[0012] This invention offers the following advantages: Traditional post-construction maintenance or periodic upkeep of hydrogen refueling station compressors lacks comprehensive analysis of compressor equipment status, hydrogen refueling demand, and maintenance human resources, leading to increased maintenance costs and workload. Therefore, this invention collects historical operating data of the compressor, business demand data of the hydrogen refueling station, and maintenance human resource data within a preset period. Since the level of business activity varies significantly across different time periods within the preset period, resulting in drastically different downtime losses, the preset period is divided into multiple business characteristic periods with different business characteristics based on the business demand data. To quantify the unit-time economic cost of maintenance operations during different business characteristic periods, the maintenance cost per unit time corresponding to different business characteristic periods is obtained from the hydrogen refueling station's maintenance records and used as a maintenance cost parameter. Since fixed-cycle preventative maintenance alone cannot respond promptly to dynamic changes in compressor status, dynamic prediction of the compressor's future failure probability is necessary. Furthermore, in existing technologies, compressor failures are often set as fixed maintenance penalty costs without considering the differences in downtime losses between different operating modes such as fast charging and slow charging at hydrogen refueling stations. Therefore, a dynamic risk cost assessment model is constructed. Based on the dynamic failure probability sequence, business demand data, and maintenance cost parameters, the accumulated failure risk parameters of the compressor due to non-performance of maintenance operations are calculated. Based on the maintenance cost parameters, failure risk parameters, and human resource assessment parameters, the comprehensive loss parameters of the compressor performing maintenance operations are obtained, and the optimal maintenance time window is found through this indicator. Based on the comprehensive loss parameters, a recommended maintenance time window is selected. This invention comprehensively considers the equipment failure risk, business load, maintenance cost, and human resource constraints of the hydrogen refueling station compressor to achieve optimal decision-making on the timing of compressor maintenance, thereby reducing commercial losses caused by unplanned downtime. Attached Figure Description

[0013] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a block diagram of a machine learning-based predictive maintenance decision system for hydrogen refueling station compressor failures, provided as an embodiment of the present invention. Detailed Implementation

[0015] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a machine learning-based predictive maintenance decision system for hydrogen refueling station compressors proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] The following description, in conjunction with the accompanying drawings, details the specific scheme of the machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressors provided by this invention.

[0018] Please see Figure 1 This invention illustrates a machine learning-based predictive maintenance decision system for hydrogen refueling station compressor failures, provided by an embodiment of the present invention. The system includes a data acquisition module 101, a parameter analysis module 102, and a maintenance decision module 103. Specific steps include: Data acquisition module 101: acquiring historical operating data of the compressor, business demand data of the hydrogen refueling station, and maintenance human resource data within a preset period; historical operating data includes the number of compressor start-ups and shutdowns, the time of each start-up and shutdown, and the compressor vibration spectrum; business demand data includes the predicted hydrogen refueling demand and the maximum hydrogen production capacity of the hydrogen refueling station within each preset period; maintenance human resource data includes the number of maintenance personnel and their availability status.

[0019] Traditional post-event maintenance or periodic upkeep of hydrogen refueling station compressors lacks a comprehensive analysis of the compressor's equipment status, hydrogen refueling demand, and maintenance human resources, which can lead to increased maintenance costs and workload. Therefore, in this embodiment of the invention, historical operating data of the compressor, business demand data of the hydrogen refueling station, and maintenance human resources data are collected within a preset period.

[0020] First, in one embodiment of the present invention, the preset period is set to 1 day. It should be noted that the preset period can be set by the user and is not limited here.

[0021] To assess the health status of the compressor and predict equipment failures, historical operating data of the compressor is collected, specifically including: the number of times the compressor started and stopped during historical operation, the corresponding time points and duration of each start and stop, and the compressor vibration spectrum. This data originates from the compressor's control system and vibration sensors and is continuously recorded in the form of historical operating logs.

[0022] Business demand data can be used to assess downtime losses and avoid peak hydrogen refueling periods. Hydrogen refueling demand varies significantly across different time periods within a preset period. In reality, during the daytime, high-frequency hydrogen refueling demand occurs in fast-charging mode, and downtime for maintenance would result in substantial hydrogen sales losses and customer churn; while maintenance during nighttime or daytime downtime in slow-charging mode results in minimal commercial losses. Therefore, data reflecting future hydrogen refueling demand is needed so that the system can differentiate the downtime costs at different times. The collected business demand data includes the predicted hydrogen refueling demand for each preset unit of time within the preset period and the maximum hydrogen production capacity of the refueling station. In one embodiment of this invention, the predicted hydrogen refueling demand can be obtained from the refueling station's sales module, which generates the predicted hydrogen refueling demand for each preset unit of time based on historical refueling records, pre-booked orders, and seasonal factors. The maximum hydrogen production capacity of the refueling station is determined by its hardware configuration and is a fixed value.

[0023] Even if the system determines that a certain time period is the optimal time for maintenance, the maintenance task cannot be performed if there are insufficient qualified and available maintenance personnel during that time. Therefore, it is necessary to collect maintenance human resource data, specifically including: the number of maintenance personnel and their availability status. In one embodiment of the present invention, the number of maintenance personnel and their availability status can be obtained from a human resource management system, which is used to reflect the total number of qualified professionals with compressor maintenance qualifications and their scheduling.

[0024] Parameter Analysis Module 102: Divides each preset period into different business characteristic time periods based on business demand data; obtains the preset unit time period maintenance cost corresponding to different business characteristic time periods based on hydrogen refueling station maintenance records, as a maintenance cost parameter; analyzes historical operating data using a machine learning model to generate a dynamic failure probability sequence for the compressor; calculates the accumulated failure risk parameter due to the compressor not performing maintenance operations based on the dynamic failure probability sequence, business demand data, and maintenance cost parameter; and obtains human resource evaluation parameters based on human resource data.

[0025] Within a preset period, the level of business activity varies significantly across different time slots, resulting in drastically different downtime losses. For example, fast-charging stations face a surge in demand for rapid hydrogen refueling during the day, making downtime a substantial business loss; conversely, refueling demand is minimal at night, rendering downtime for maintenance insignificant. Without differentiating between time slots, the system will fail to recognize the significant losses during peak periods, potentially leading to erroneous decisions to perform maintenance during peak times. Therefore, it is necessary to divide the preset period into multiple time slots with distinct business characteristics based on business demand data.

[0026] Preferably, in one embodiment of the present invention, each preset period is divided into different business characteristic time periods according to business demand data, including: setting the preset period to 1 day, and dividing the preset period into 5 business characteristic time periods according to the hydrogen refueling volume at different times within the preset period in the hydrogen refueling station's historical records, namely: late night off-peak period, from 0:00 to 6:00, morning peak fast charging period from 6:00 to 9:00, daytime off-peak period from 9:00 to 17:00, evening peak fast charging period from 17:00 to 21:00, and nighttime off-peak period from 21:00 to 24:00.

[0027] In order to quantify the unit time economic cost of maintenance operations during different business characteristic periods, in this embodiment of the invention, the maintenance cost per unit time corresponding to different business characteristic periods is obtained from the hydrogen refueling station maintenance records, and used as the maintenance cost parameter.

[0028] Preferably, in one embodiment of the present invention, the method for obtaining maintenance cost parameters includes: obtaining historical maintenance records and maintenance costs within different business characteristic time periods within a preset period; and obtaining maintenance cost parameters according to a maintenance cost parameter calculation formula, the maintenance cost parameter calculation formula being as follows: In the formula, Indicates the first [number] within the preset period The compressor maintenance cost parameters within a specific business period; Indicates the first The number of preset unit times within each business characteristic period; Indicates the first Within the first business characteristic period The preset unit time to the first Repair costs between preset unit time periods; This represents the maximum maintenance cost recorded in the historical maintenance history of hydrogen refueling stations. Indicates the first A preset unit of time; Indicates the first A preset unit of time.

[0029] In the formula for calculating maintenance cost parameters, through Calculate the first The maintenance cost per unit time is preset within each business characteristic period, and then normalized for the first... The average of the normalized maintenance costs for all preset unit times within a specific business characteristic time period is used to obtain the first... The maintenance cost parameter of the compressor within a specific business period. The maintenance cost parameter is a dimensionless constant between 0 and 1. The larger the value, the higher the economic cost of maintenance during that period.

[0030] It should be noted that, It is calculated in advance from the historical maintenance data accumulated in the system backend, a technical method well known to those skilled in the art, and will not be elaborated here.

[0031] In one embodiment of the present invention, the preset unit time is set to 1 hour. It should be noted that the preset unit time can be set freely and is not limited here.

[0032] Fixed-cycle preventative maintenance alone cannot respond promptly to dynamic changes in compressor status. Therefore, it is necessary to dynamically predict the future failure probability of the compressor so that the system can detect risks in advance. Thus, in this embodiment of the invention, a machine learning model is used to analyze historical operating data to generate a dynamic failure probability sequence for the compressor.

[0033] Preferably, in one embodiment of the present invention, the machine learning model is a multivariate long short-term memory network (LSTM). As a deep learning model suitable for time-series data, this model can learn the inherent patterns of fault evolution from the compressor's historical operating logs and output the probability of faults occurring in future time periods. It should be noted that this model receives time-stamped historical operating logs over a long period to learn the dynamic trends in the equipment's health status and output future predictions.

[0034] Preferably, in one embodiment of the present invention, the method for obtaining the dynamic fault probability sequence includes: analyzing historical operating data using a machine learning model to generate a dynamic fault probability sequence of the compressor for all preset unit times within a preset period; wherein each value in the dynamic fault probability sequence represents the probability of a fault occurring within a preset unit time. This probability value is a scalar between 0 and 1. In this embodiment of the present invention, since the preset unit time is 1 hour, the historical operating data aggregated hourly, i.e., the number of compressor start-stop cycles, cumulative running time, and average vibration spectrum per hour, is input into the model. The model outputs the fault probability value for each hour within 24 hours, forming a probability sequence of length 24.

[0035] In existing technologies, compressor failures are often set as a fixed maintenance penalty cost, without considering the differences in downtime losses under different operating modes such as fast charging and slow charging at hydrogen refueling stations. Therefore, in this embodiment of the invention, a dynamic risk cost assessment model is constructed to calculate the accumulated failure risk parameters of the compressor when maintenance operations are not performed, based on a dynamic failure probability sequence, business demand data, and maintenance cost parameters.

[0036] Preferably, in one embodiment of the present invention, the method for obtaining fault risk parameters includes: obtaining the predicted hydrogen demand and the maximum hydrogen production capacity of the hydrogen refueling station in each preset unit time within a preset period; using the ratio of the predicted hydrogen demand to the maximum hydrogen production capacity in each preset unit time as the dynamic load saturation; obtaining candidate maintenance windows, the length of which is an integer multiple of the preset unit time; and obtaining fault risk parameters according to the fault risk parameter calculation formula, which is shown below: In the formula, This indicates the accumulated fault risk parameters resulting from the compressors in the candidate maintenance window not undergoing maintenance operations within a preset period; This indicates the number of business characteristic time periods spanned by the candidate repair window within a preset period; Indicates the first [number] within the preset period The compressor maintenance cost parameters within a specific business period; Indicates the candidate repair windows within the preset period; Indicates the first The overlapping area between the business characteristic time period and the candidate maintenance window; Indicates the first The probability of compressor failure within a preset unit of time; This indicates the number of hydrogen refueling stations. Predicted hydrogen addition per preset unit time period; This indicates the maximum hydrogen production capacity of the hydrogen refueling station; This represents an exponential function with the natural constant as its base.

[0037] In the formula for calculating the failure risk parameter, when the hydrogen refueling station is in a low-operation period, i.e., the hydrogen refueling station's [number]th [period]... Predicted hydrogen addition per preset unit time period Approaching 0, at this point When the value approaches 1, the maintenance risk is determined solely by the probability of failure and the maintenance cost; when the hydrogen refueling station enters its peak operating period, that is... Approaching 1, at this point By drastically amplifying the probability of minor faults, the fault risk parameter increases dramatically, enabling the system to automatically avoid scheduling repairs within that candidate repair window during subsequent decision-making; and Note that it is necessary to calculate the first... The preset number of time units that intersect with candidate maintenance windows for each business characteristic time period, and for each intersecting preset time unit... Add; for each business characteristic time period spanned by the candidate repair window within the preset period. The parameters are summed to obtain the accumulated fault risk parameters for compressors that did not perform maintenance operations within the candidate maintenance window over a preset period. The larger the candidate maintenance window, the greater the risk of failure. The larger the value, the greater the accumulated fault risk parameter of the compressor not performing maintenance operations within the candidate maintenance window during the preset period.

[0038] It should be noted that, Obtained by predictions from machine learning models. Provided by the sales module. These are fixed parameters for hydrogen refueling stations, all of which are known to those skilled in the art and will not be described in detail here.

[0039] Besides equipment risks and business losses, the number and scheduling of professional maintenance personnel are also key factors determining whether maintenance tasks can be performed. Therefore, in this embodiment of the invention, human resource evaluation parameters are obtained based on human resource data.

[0040] Preferably, in one embodiment of the present invention, the method for obtaining human resource evaluation parameters includes: calculating human resource evaluation parameters for each preset unit time period included in time periods with different business characteristics within a preset period, wherein the calculation formula is as follows: In the formula, Indicates the first Human resource evaluation parameters within a preset unit of time; Indicates the first The number of professional maintenance personnel required within a preset unit of time; Indicates the first The number of maintenance personnel available within a preset unit of time; This represents a non-zero minimum constant, used to prevent the denominator from being 0, and is set to... ; This represents the logarithmic function with the natural constant as the base.

[0041] In the formula for calculating human resource evaluation parameters, when there are sufficient available maintenance personnel... Approaching 0, at this point The value approaches 1 when maintenance personnel are in short supply. As it increases, we can use the logarithmic function to... Perform smoothing.

[0042] Maintenance Decision Module 103: Based on maintenance cost parameters, fault risk parameters, and human resource assessment parameters, obtain comprehensive loss parameters for the compressor to perform maintenance operations; select a recommended maintenance time window based on the comprehensive loss parameters.

[0043] Based on maintenance cost parameters, failure risk parameters, and human resource assessment parameters, the comprehensive loss parameters for performing compressor maintenance operations are obtained. The optimal maintenance time window is then identified using these comprehensive loss parameters.

[0044] Preferably, in one embodiment of the present invention, the method for obtaining the comprehensive loss parameter includes: for candidate maintenance windows within a preset period, calculating the comprehensive loss parameter according to the following formula: In the formula, Indicates the time sequence number of the preset unit within the preset period; Indicates the candidate repair window; This represents the overall loss parameters for the compressor when performing maintenance operations within the candidate maintenance window; Indicates the first Maintenance cost parameters for each preset unit of time within a business characteristic period; This represents the average value of human resources assessment parameters within the candidate maintenance window; This indicates that the accumulated fault risk parameters for repairs will not be executed within the candidate repair window; Indicates time The projected demand for hydrogen; The dimension conversion coefficient is set to the average selling price of hydrogen in this embodiment of the invention. This represents the maximum maintenance cost recorded in the historical maintenance history of hydrogen refueling stations.

[0045] In the formula for calculating the comprehensive loss parameter, where, It is the first The maintenance cost parameters of each preset unit time period are used to accumulate the maintenance cost parameters of their respective business characteristic periods when the candidate maintenance window spans different business characteristic periods. This serves as an amplification factor for the average value of human resource assessment parameters within the candidate maintenance window. When there is a severe shortage of personnel... Larger, leading to As the number of available slots increases, the system will tend to avoid scheduling maintenance during periods of high staffing levels; the longer the period without scheduled maintenance, the larger the potential maintenance window, and the less likely the system will execute maintenance-accumulated fault risk parameters. The larger the value, the larger the overall loss parameter; This indicates the amount of hydrogen sales lost during the downtime. The larger the value, the less likely it is that the repair time window should be set in the current candidate repair window.

[0046] Based on the comprehensive loss parameters, a recommended maintenance time window is selected. In one embodiment of the present invention, the specific steps include: the system iterates through all possible candidate maintenance windows within a preset period with a preset unit time as the step size, calculates the corresponding comprehensive loss parameters for each window, and selects the candidate maintenance window that minimizes the comprehensive loss parameters as the recommended maintenance time window.

[0047] This completes the predictive maintenance decision-making for compressor failure at the hydrogen refueling station.

[0048] In summary, the following steps are taken: Historical operating data of the compressor, business demand data of the hydrogen refueling station, and maintenance human resource data are obtained within a preset period. Historical operating data includes the number of compressor start-ups and shutdowns, the duration of each start-up and shutdown, and the compressor vibration spectrum. Business demand data includes the predicted hydrogen demand and the maximum hydrogen production capacity of the refueling station for each preset period. Maintenance human resource data includes the number of maintenance personnel and their availability. Each preset period is divided into different business characteristic time periods based on the business demand data. Maintenance costs per unit time corresponding to different business characteristic time periods are obtained from the hydrogen refueling station's maintenance records and used as maintenance cost parameters. A machine learning model is used to analyze the historical operating data to generate a dynamic failure probability sequence for the compressor. Based on the dynamic failure probability sequence, business demand data, and maintenance cost parameters, the accumulated failure risk parameters due to the compressor not performing maintenance operations are calculated. Human resource assessment parameters are obtained based on the human resource data. A comprehensive loss parameter for performing compressor maintenance operations is obtained based on the maintenance cost parameter, failure risk parameter, and human resource assessment parameter. A recommended maintenance time window is selected based on the comprehensive loss parameter.

[0049] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0050] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures, characterized in that, The system includes: a data acquisition module for acquiring historical operating data of the compressor, business demand data of the hydrogen refueling station, and maintenance human resource data within a preset period; the historical operating data includes the number of compressor start-ups and shutdowns, the time of each start-up and shutdown, and the compressor vibration spectrum; the business demand data includes the predicted hydrogen demand and the maximum hydrogen production capacity of the hydrogen refueling station within each preset period; the maintenance human resource data includes the number of maintenance personnel and their availability; a parameter analysis module for dividing each preset period into different business characteristic time periods based on the business demand data; obtaining the maintenance cost per unit time corresponding to different business characteristic time periods based on the hydrogen refueling station maintenance records, as a maintenance cost parameter; analyzing the historical operating data using a machine learning model to generate a dynamic failure probability sequence of the compressor; and analyzing the dynamic failure probability sequence, the business demand data, and the... The maintenance cost parameter is used to calculate the accumulated fault risk parameter due to the compressor not performing maintenance operations; based on the human resource data, a human resource assessment parameter is obtained; the maintenance decision module is used to obtain a comprehensive loss parameter for the compressor performing maintenance operations based on the maintenance cost parameter, the fault risk parameter, and the human resource assessment parameter; and select a recommended maintenance time window based on the comprehensive loss parameter; the method for obtaining the fault risk parameter includes: obtaining the predicted hydrogen demand and the maximum hydrogen production capacity of the hydrogen refueling station in each preset unit time within a preset period; using the ratio of the predicted hydrogen demand to the maximum hydrogen production capacity in each preset unit time as the dynamic load saturation; obtaining a candidate maintenance window, the length of which is an integer multiple of the preset unit time; and obtaining the fault risk parameter according to the fault risk parameter calculation formula, which is shown below: In the formula, This indicates the accumulated fault risk parameters resulting from the compressors in the candidate maintenance window not undergoing maintenance operations within a preset period; This indicates the number of business characteristic time periods spanned by the candidate repair window within a preset period; Indicates the first [number] within the preset period The compressor maintenance cost parameters within a specific business period; Indicates the candidate repair windows within the preset period; Indicates the first The overlapping area between the business characteristic time period and the candidate maintenance window; Indicates the first The probability of compressor failure within a preset unit of time; This indicates the number of hydrogen refueling stations. Predicted hydrogen addition per preset unit time period; This indicates the maximum hydrogen production capacity of the hydrogen refueling station; The term represents an exponential function with a base of a natural constant. The method for obtaining the human resource evaluation parameters includes: within the preset period, calculating the human resource evaluation parameters for each preset unit of time within time periods of different business characteristics, using the following calculation formula: In the formula, Indicates the first Human resource evaluation parameters within a preset unit of time; Indicates the first The number of professional maintenance personnel required within a preset unit of time; Indicates the first The number of maintenance personnel available within a preset unit of time; This represents a non-zero minimum constant, used to prevent the denominator from being 0, and is set to... ; The logarithmic function is represented with the natural constant as the base; the method for obtaining the comprehensive loss parameter includes: for candidate maintenance windows within a preset period, calculating the comprehensive loss parameter according to the following formula: In the formula, This indicates the time sequence number of the preset unit within the preset period; Indicates the candidate repair window; This represents the overall loss parameters for the compressor when performing maintenance operations within the candidate maintenance window; Indicates the first The maintenance cost parameters for each preset unit of time within a business characteristic period; This represents the average value of the human resources assessment parameters within the candidate maintenance window; This indicates that the accumulated fault risk parameters will not be executed within the candidate repair window; Indicates time The predicted hydrogen demand; Indicates the dimension conversion coefficient; This represents the maximum maintenance cost recorded in the historical maintenance history of hydrogen refueling stations.

2. The machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures according to claim 1, characterized in that, The machine learning model is a multivariate long short-term memory network.

3. The machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures according to claim 1, characterized in that, Based on the business demand data, each preset period is divided into different business characteristic time periods, including: setting the preset period to 1 day, and dividing the preset period into 5 business characteristic time periods based on the hydrogen refueling volume at different times within the preset period in the hydrogen refueling station's historical records, namely: late night off-peak period, from 0:00 to 6:00, morning peak fast charging period from 6:00 to 9:00, daytime off-peak period from 9:00 to 17:00, evening peak fast charging period from 17:00 to 21:00, and nighttime off-peak period from 21:00 to 24:

00.

4. The machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures according to claim 1, characterized in that, The method for obtaining the maintenance cost parameter includes: obtaining historical maintenance records and maintenance costs within different business characteristic periods of a preset period; and obtaining the maintenance cost parameter according to the maintenance cost parameter calculation formula, which is shown below: In the formula, Indicates the first [number] within the preset period The compressor maintenance cost parameters within a specific business period; Indicates the first The number of preset unit times within each business characteristic period; Indicates the first Within the first business characteristic period The preset unit time to the first Repair costs between preset unit time periods; This represents the maximum maintenance cost recorded in the historical maintenance history of hydrogen refueling stations. Indicates the first A preset unit of time; Indicates the first A preset unit of time.

5. A machine learning-based predictive maintenance decision-making system for hydrogen refueling station compressor failures according to claim 4, characterized in that, The method for obtaining the dynamic fault probability sequence includes: analyzing the historical operating data using a machine learning model to generate a dynamic fault probability sequence of the compressor for all preset unit times within the preset period; wherein each value in the dynamic fault probability sequence represents the probability of a fault occurring within the preset unit time.

Images