A method for predicting faults of single-opening devices based on deep neural networks

Abstract

The present invention discloses a method for predicting failure of an order opener based on a deep neural network, including: constructing an order complexity evaluation model to evaluate the complexity of an order; constructing an order opener hardware load prediction model to predict the hardware load data during the current order processing; constructing an order opener hardware failure risk evaluation model to determine whether the current order is a high-risk order; predicting the risk level of hardware failure of the order opener during the processing of continuous high-complexity orders, and determining failure risk orders; generating hardware resource allocation optimization measures, and optimizing the hardware resource allocation of the order opener in advance; obtaining the hardware load data after the hardware resource allocation optimization measures, and determining the effectiveness of the hardware resource allocation optimization measures. The present invention realizes more intelligent hardware management by dynamically adjusting the resource allocation strategy, greatly reduces the failure risk of the order opener in complex business scenarios, and improves the overall efficiency of order processing and the sustainable operation capability of the system.

Description

Technical Field

[0001] The present invention relates to the field of Internet information security, and in particular to a method for predicting failures of an order opener based on a deep neural network. Background Art

[0002] In the daily operation of modern enterprises, the order processing system has become one of the core links of business operation. Especially when processing high-frequency orders involving large amounts of data and complex operations, the performance and stability of the order opener are crucial to the overall operational efficiency of the enterprise. However, with the continuous increase in order complexity and business volume, the existing order opener system faces multiple challenges and problems when processing complex orders, which limits its performance in high-load scenarios. Existing order processing systems generally lack a scientific order complexity evaluation mechanism. Usually, the order processing complexity relies on manual experience or simple rules for judgment. This method is not only highly subjective, but also difficult to accurately evaluate the order's demand for system resources. When the system processes multiple complex orders at the same time, complexity judgments that have not been scientifically evaluated may lead to uneven resource allocation, insufficient resources for some key tasks, and excess resources for other tasks, thus causing system bottlenecks and delays. Insufficient hardware load prediction capabilities are also an important issue. In the process of high-load order processing, the hardware of the order opener, such as CPU, memory, and I / O devices, needs to withstand huge computing pressure. However, existing systems often rely only on real-time monitoring data and lack effective prediction of future hardware loads. When the system load is close to saturation, it is impossible to take measures in advance to schedule and optimize resources, which can easily lead to hardware overload and even the risk of system crash. For enterprises, such unforeseen failures not only affect the efficiency of order processing, but may also lead to business interruption and economic losses. In addition, the failure risk assessment method of the existing system is relatively simple, mainly relying on simple statistics and analysis of historical data, lacking in-depth understanding and modeling of the complex relationship between order characteristics, hardware load characteristics and failure risks. When faced with high-complexity orders, this traditional failure risk assessment method is difficult to effectively identify potential high-risk orders, resulting in untimely failure warnings and inadequate preventive measures, further exacerbating the failure risk of the system. More seriously, in the case of continuous processing of multiple high-complexity orders, the resource management and scheduling strategies of the existing system cannot effectively cope with the cumulative effect caused by the short time interval between orders. The continuous processing of multiple complex orders will cause the system hardware to experience the test of continuous high load in a short period of time, resulting in the gradual accumulation of load, and the difficulty of rapid adjustment of resource allocation, which may eventually cause hardware overload and failure. Traditional resource allocation methods usually set parameters statically and cannot be adjusted dynamically according to real-time load conditions, causing the system to show obvious lag and instability when facing rapidly changing load changes. In general, when processing high-complexity orders, existing order opening systems generally have problems such as inaccurate order complexity assessment, insufficient hardware load prediction, imperfect failure risk assessment methods, and insufficient resource allocation optimization capabilities. These problems not only limit the performance of the system under high load conditions, but also increase the risk of system failure, posing a serious threat to the company's operational efficiency and business continuity.Therefore, how to improve the stability and efficiency of the order opener in processing complex orders through scientific complexity assessment, accurate load forecasting, effective risk assessment and intelligent resource allocation has become a technical problem that companies urgently need to solve. Summary of the invention

[0003] The present invention solves the problems existing in the above-mentioned prior art and provides a method for predicting faults of a single-opener based on a deep neural network, which mainly includes:

[0004] Obtain order feature data through the order monitoring database, build an order complexity evaluation model, evaluate the complexity of orders, and set hardware resource allocation plans for orders of different complexities;

[0005] Based on the order feature data of historical high-complexity orders and the corresponding hardware load time series data, a hardware load prediction model for the order opener is constructed to predict the hardware load data during the current order processing period;

[0006] Based on the historical failure records of the order opener hardware failures and the hardware load characteristic data during the corresponding order processing period, a risk assessment model for the order opener hardware failure is constructed to determine whether the current order is a high-risk order;

[0007] If the time interval between consecutive high-complexity orders processed by the order opener is less than the preset time, the risk level of hardware failure of the order opener during the consecutive high-complexity order processing period is predicted based on the hardware load data during the consecutive high-complexity order processing period, and the failure risk orders are determined;

[0008] Generate hardware resource allocation optimization measures based on the predicted hardware load data during the current order processing period, and optimize the hardware resource allocation of the order opener in advance;

[0009] Obtain the hardware load data after the hardware resource allocation optimization measures are taken, judge the effectiveness of the hardware resource allocation optimization measures, and use the deep Q network algorithm to train the model to determine the hardware resource allocation optimization measures that meet different hardware load data.

[0010] Furthermore, the order feature data is obtained through the order monitoring database, an order complexity evaluation model is constructed, the complexity of the order is evaluated, and hardware resource allocation schemes for orders of different complexities are set, including:

[0011] Through the real-time monitoring system, the hardware load data of the order opener when processing orders is recorded and marked with timestamps. The hardware load data includes CPU utilization, memory read and write rates, and I / O throughput; the hardware load data with timestamps are integrated to form time series data, and the time series data of the hardware load is stored in the order monitoring database; through the order monitoring database, the order feature data of each order is obtained, and the order complexity of each order is marked, and the obtained order feature data is cleaned, including removing null values ​​and outlier processing. The order feature data includes the order processing steps, data processing volume, number of concurrent operations, and processing time. The complexity includes high, medium, and low; based on the cleaned order feature data, the decision tree algorithm is used for model training, and an order complexity evaluation model is constructed to evaluate the complexity of the order; according to the complexity level of the order, hardware resource allocation plans for orders of different complexities are set, and according to the order complexity evaluation results, the pre-set resource allocation plan is called.

[0012] Furthermore, the order feature data of historical high-complexity orders and the corresponding hardware load time series data are used to construct an order opener hardware load prediction model to predict the hardware load data during the current order processing period, including:

[0013] The order feature data of the order currently processed by the order opener is obtained, and the order complexity evaluation model is used to evaluate the complexity of the order currently processed by the order opener; if the order currently processed by the order opener is identified as a high-complexity order, the order feature data of historical high-complexity orders and the corresponding hardware load time series data are obtained through the order monitoring database; based on the order feature data of historical high-complexity orders and the corresponding hardware load time series data, a long short-term memory network is used for model training to build a hardware load prediction model for the order opener; the hardware load data during the current order processing period is obtained through the real-time monitoring system, and combined with the order feature data of the current order, the hardware load prediction model of the order opener is used to predict the hardware load data during the current order processing period.

[0014] Furthermore, the method of constructing a risk assessment model for hardware failure of the order opener based on the historical failure records of the order opener hardware failure and the hardware load characteristic data during the corresponding order processing period to determine whether the current order is a high-risk order includes:

[0015] Through the order monitoring database, the historical hardware failure records of the order opener and the time series data of the hardware load during the corresponding order processing period are obtained; based on the time series data of the hardware load during the order processing period, the hardware load fluctuation amplitude factor, the hardware load average value, the hardware load peak value, the order processing time and the time when the hardware load exceeds the preset load threshold during the order processing period are calculated, and the hardware load characteristic data during the corresponding order processing period are constructed; based on the historical hardware failure records of the order opener and the hardware load characteristic data during the corresponding order processing period, the decision tree algorithm is used to train the model and construct a hardware failure risk assessment model for the order opener; based on the predicted hardware load data during the current order processing period, the hardware load characteristic data during the current order processing period is extracted, and the hardware failure risk assessment model for the order opener is used to predict the risk level of the hardware failure of the order opener during the current order processing period, and the risk level includes high and low; if the risk level of the hardware failure of the order opener during the current order processing period is high, the current order is judged to be a high-risk order; the predicted hardware load data during the high-risk order processing period is obtained, and the time point when the hardware load exceeds the preset first load threshold is determined.

[0016] The method also includes calculating a hardware load fluctuation amplitude factor during the order processing period based on the time series data of the hardware load during the order processing period.

[0017] The step of calculating the hardware load fluctuation amplitude factor during the order processing period according to the time series data of the hardware load during the order processing period specifically includes:

[0018] Obtain the time series data of the hardware load during the order processing period, divide the time series data of the hardware load according to the preset time window, and calculate the average value of the hardware load of different hardware load categories. The hardware load categories include CPU utilization, memory read and write rate Mem, and I / O throughput; according to the hardware load fluctuation amplitude factor formula Calculate the hardware load fluctuation factor F during the current order processing period w (t), where N is the number of data points in the time window, j represents the hardware load category, and x i,j (t) represents the j-th type of hardware load value at the ith moment, μ j (t) is the average value of the j-th type of load in the time window, |x i,j (t)-x i-1,j (t)| is the adjacent difference of load value, which is multiplied by the logarithmic term to emphasize the severity of fluctuation between adjacent moments.

[0019] Furthermore, if the time interval between consecutive high-complexity orders processed by the order opener is less than a preset time, the risk level of hardware failure of the order opener during the consecutive high-complexity order processing is predicted according to the hardware load data during the consecutive high-complexity order processing, and the failure risk order is determined, including:

[0020] If the time interval between consecutive processing of high-complexity orders by the order opener is less than the preset time, the order feature data of the current order and the order feature data of the first n high-complexity orders are obtained; the risk level of hardware failure of the order opener during the processing of the first n consecutive high-complexity orders is determined according to the risk assessment model of hardware failure of the order opener, and the risk level of hardware failure of the order opener during the processing of the current order is predicted, and the current order is the n+1th order of the consecutive high-complexity orders; if the risk levels of hardware failure of the order opener during the processing of the first n consecutive high-complexity orders and the current order are both low, then according to the hardware load prediction model of the order opener, Predict the hardware load data during the processing of continuous high-complexity orders; obtain the hardware load characteristic data during the processing of continuous high-complexity orders based on the hardware load data during the processing of continuous high-complexity orders, and use the order opener hardware failure risk assessment model to predict the risk level of the order opener hardware failure during the processing of continuous high-complexity orders; if the risk level of the order opener hardware failure during the processing of continuous high-complexity orders is high, judge the current order as a failure risk order, and determine the time point when the hardware load in the failure risk order exceeds the preset second load threshold based on the obtained hardware load data during the processing of continuous high-complexity orders.

[0021] Furthermore, the hardware resource allocation optimization measures are generated according to the predicted hardware load data during the current order processing period, and the hardware resource allocation of the order opener is optimized in advance, including:

[0022] Based on the predicted hardware load data during the processing of high-risk orders and the hardware load data during the processing of failure risk orders, hardware resource allocation optimization measures are generated. The hardware resource allocation optimization measures include CPU time slice allocation, memory partition adjustment, and I / O queue priority scheduling. Based on the hardware resource allocation optimization measures and the time point when the hardware load in the high-risk orders exceeds the preset first load threshold, and the time point when the hardware load in the failure risk orders exceeds the preset second load threshold, the hardware resource allocation of the order opener is optimized in advance.

[0023] Furthermore, the hardware load data after the hardware resource allocation optimization measures are obtained, the effectiveness of the hardware resource allocation optimization measures is judged, and the deep Q network algorithm is used for model training to determine the hardware resource allocation optimization measures that meet different hardware load data, including:

[0024] Through the real-time monitoring system, the hardware load data after the hardware resource allocation optimization measures are obtained, the hardware load characteristic data during the current order processing period is extracted, and the obtained data is filtered, denoised and standardized using data cleaning and preprocessing technology; based on the preprocessed hardware load data, the order opener hardware failure risk assessment model is used to determine the risk level of the order opener hardware failure during the current order processing period, and to judge the effectiveness of the hardware resource allocation optimization measures; if the effectiveness of the hardware resource allocation optimization measures is lower than expected, the hardware resource allocation optimization measures are adjusted until the effectiveness of the hardware resource allocation optimization measures reaches the expected level, and the adjustment of the hardware resource allocation optimization measures includes reallocating CPU resources, increasing memory partitions and optimizing the priority of I / O operations; the results of each adjustment of the hardware resource allocation optimization measures, the effectiveness evaluation results and the hardware load data during the order processing period after the hardware resource allocation optimization measures are recorded, and the deep Q network algorithm is used for model training to determine the hardware resource allocation optimization measures that meet different hardware load data.

[0025] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0026] The present invention provides a method for predicting failures of an order opener based on a deep neural network. By accurately evaluating the complexity of an order, the present invention can achieve more reasonable resource allocation before order processing, thereby avoiding resource waste and the occurrence of processing bottlenecks. Secondly, the present invention can predict the hardware load during order processing in advance, thereby effectively preventing the risk of overload and ensuring the stability of the processing process. By identifying and warning high-risk orders, measures can be taken in advance to reduce the possibility of system downtime and business interruption due to hardware failure. In the case of continuously processing multiple complex orders, the present invention can optimize resource allocation in advance to avoid failures caused by untimely resource scheduling. The present invention achieves more intelligent hardware management by dynamically adjusting the resource allocation strategy, greatly reducing the failure risk of the order opener in complex business scenarios, and improving the overall efficiency of order processing and the sustainable operation capability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of a method for predicting failure of a billing device based on a deep neural network of the present invention;

[0028] Figure 2 It is a schematic diagram of a method for predicting failure of a billing device based on a deep neural network according to the present invention;

[0029] Figure 3 It is another schematic diagram of a method for predicting faults of an opener based on a deep neural network according to the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] like Figure 1-3 In this embodiment, a method for predicting failure of a billing device based on a deep neural network may specifically include:

[0032] Step S101, obtain order feature data through the order monitoring database, build an order complexity evaluation model, evaluate the complexity of the order, and set hardware resource allocation plans for orders of different complexities.

[0033] Through the real-time monitoring system, the hardware load data of the order opener when processing orders is recorded and marked with timestamps. The hardware load data includes CPU utilization, memory read and write rate, and I / O throughput. The hardware load data with timestamps is integrated to form time series data, and the time series data of the hardware load is stored in the order monitoring database. Through the order monitoring database, the order feature data of each order is obtained, and the order complexity of each order is marked. The obtained order feature data is cleaned, including removing null values ​​and outlier processing. The order feature data includes the processing steps of the order, the data processing volume, the number of concurrent operations, and the processing time. The complexity includes high, medium, and low. According to the cleaned order feature data, the decision tree algorithm is used for model training, and the order complexity evaluation model is constructed to evaluate the complexity of the order. According to the complexity level of the order, the hardware resource allocation plan for orders of different complexity is set, and according to the order complexity evaluation results, the pre-set resource allocation plan is called.

[0034] For example, in an order opening system, the real-time monitoring system records the hardware load data of the order opening device when processing multiple orders, and obtains the hardware load data of multiple orders at different time points on January 1, 2020, including that when processing order A, the CPU utilization rate is 60%, the memory read and write rate is 280MB / s, and the I / O throughput is 110MB / s. When processing order B, the CPU utilization rate is 75%, the memory read and write rate is 350MB / s, and the I / O throughput is 140MB / s. When processing order C, the CPU utilization rate is 50%, the memory read and write rate is 250MB / s, and the I / O throughput is 100MB / s. When processing order D, the CPU utilization rate is 80%, the memory read and write rate is 370MB / s, and the I / O throughput is 150MB / s. When processing order E, the CPU utilization rate is 65%, the memory read and write rate is 300MB / s, and the I / O throughput is 120MB / s. The hardware load data of the order opener during multiple orders are integrated into time series data and stored in the order monitoring database. Through the order monitoring database, the order feature data of the five orders are obtained, and the complexity of each order is marked according to the pre-defined standards. Among them, the processing steps of order A are 5 steps, the data processing volume is 500MB, the number of concurrent operations is 3, and the processing time is 15 minutes. The complexity is marked as medium. The processing steps of order B are 7 steps, the data processing volume is 700MB, the number of concurrent operations is 5, and the processing time is 25 minutes. The complexity is marked as high. The processing steps of order C are 4 steps, the data processing volume is 400MB, the number of concurrent operations is 2, and the processing time is 10 minutes. The complexity is marked as low. The processing steps of order D are 6 steps, the data processing volume is 650MB, the number of concurrent operations is 4, and the processing time is 20 minutes. The complexity is marked as high. The processing steps of order E are 5 steps, the data processing volume is 550MB, the number of concurrent operations is 3, and the processing time is 18 minutes. The complexity is marked as medium. After data cleaning, null values ​​are removed and outliers are processed, the decision tree algorithm is used to train the model of the cleaned order feature data, build an order complexity evaluation model, and identify the complexity of new orders. According to the complexity level of the order, set the hardware resource allocation plan for orders of different complexity. For high-complexity orders, configure high-performance hardware resources, including increasing CPU time slice allocation, increasing memory partitions, and increasing the priority of I / O queues. For medium-complexity orders, configure medium-performance hardware resources, and adjust the CPU time slice, memory partition, and I / O queue priority to a moderate level. For low-complexity orders, configure lower-performance hardware resources, reduce CPU time slices, reduce memory partitions, and lower the priority of I / O operations. During each order processing, call the pre-set resource allocation plan based on the order complexity evaluation results to allocate and adjust system hardware resources in real time.

[0035] Step S102, based on the order feature data of historical high-complexity orders and the corresponding hardware load time series data, a hardware load prediction model for the order opener is constructed to predict the hardware load data during the current order processing period.

[0036] The order feature data of the order currently processed by the order opener is obtained, and the order complexity evaluation model is used to evaluate the complexity of the order currently processed by the order opener. If the order currently processed by the order opener is identified as a high-complexity order, the order feature data of the historical high-complexity orders and the corresponding hardware load time series data are obtained through the order monitoring database. Based on the order feature data of the historical high-complexity orders and the corresponding hardware load time series data, the long short-term memory network is used for model training to build the order opener hardware load prediction model. The hardware load data during the current order processing period is obtained through the real-time monitoring system, and combined with the order feature data of the current order, the hardware load prediction model of the order opener is used to predict the hardware load data during the current order processing period.

[0037] For example, the order feature data of the currently processed order F is obtained. The processing steps of the order are 8 steps, the data processing volume is 800MB, the number of concurrent operations is 6, and the processing time is expected to be 30 minutes. The order complexity evaluation model is used to evaluate that order F is a high-complexity order. Since order F was identified as a high-complexity order, the feature data and corresponding hardware load time series data of five historical high-complexity orders were retrieved through the order monitoring database. The data of these historical orders are as follows: the processing steps of order G are 7 steps, the data processing volume is 750MB, the number of concurrent operations is 5, and the processing time is 28 minutes. The hardware load time series data includes the average CPU utilization rate of 70% during the processing period, the average memory read and write rate is 320MB / s, and the average I / O throughput is 130MB / s. The processing steps of order H are 9 steps, the data processing volume is 900MB, the number of concurrent operations is 7, and the processing time is 35 minutes. The hardware load time series data includes the average CPU utilization rate of 75%, the average memory read and write rate is 350MB / s, and the I / O throughput is 140MB / s. The processing steps of order I are The processing steps of order J are 8 steps, the data processing volume is 600MB, the number of concurrent operations is 8, and the processing time is 30 minutes. The hardware load data includes an average CPU utilization of 75%, an average memory read and write rate of 300MB / s, and an average I / O throughput of 100MB / s. The processing steps of order J are 8 steps, the data processing volume is 900MB, the number of concurrent operations is 5, and the processing time is 35 minutes. The hardware load data includes an average CPU utilization of 75%, an average memory read and write rate of 350MB / s, and an average I / O throughput of 120MB / s. The processing steps of order K are 7 steps, the data processing volume is 800MB, the number of concurrent operations is 9, and the processing time is 45 minutes. The hardware load data includes an average CPU utilization of 78%, an average memory read and write rate of 380MB / s, and an average I / O throughput of 150MB / s. According to the characteristic data of these historical high-complexity orders and the corresponding hardware load time series data, the long short-term memory network was used for model training to build a hardware load prediction model for the order opener. During the processing of order F, the actual hardware load data of order F, including CPU utilization, memory read and write rate, and I / O throughput, were obtained through the real-time monitoring system. Combined with the order characteristic data of order F, these data were input into the previously trained hardware load prediction model of the order opener, and the hardware load data of order F during the processing period were predicted, including CPU utilization will reach 72%, memory read and write rate will reach 340MB / s, and I / O throughput will reach 135MB / s.

[0038] Step S103, based on the historical failure records of the order opener hardware failures and the hardware load characteristic data during the corresponding order processing period, a risk assessment model for the order opener hardware failure is constructed to determine whether the current order is a high-risk order.

[0039] Through the order monitoring database, the historical hardware failure records of the order opener and the time series data of the hardware load during the corresponding order processing period are obtained. According to the time series data of the hardware load during the order processing period, the hardware load fluctuation amplitude factor, the hardware load average value, the hardware load peak value, the order processing time, and the time when the hardware load exceeds the preset load threshold during the order processing period are calculated, and the hardware load characteristic data during the corresponding order processing period are constructed. According to the historical hardware failure records of the order opener and the hardware load characteristic data during the corresponding order processing period, the decision tree algorithm is used for model training to construct the hardware failure risk assessment model of the order opener. According to the predicted hardware load data during the current order processing period, the hardware load characteristic data during the current order processing period is extracted, and the hardware failure risk assessment model of the order opener is used to predict the risk level of the hardware failure of the order opener during the current order processing period, and the risk level includes high and low. If the risk level of the hardware failure of the order opener during the current order processing period is high, the current order is judged to be a high-risk order. The predicted hardware load data during the high-risk order processing period is obtained, and the time point when the hardware load exceeds the preset first load threshold is determined.

[0040] Exemplarily, through the order monitoring database, the historical hardware failure records of the order opener and the hardware load time series data during the corresponding order processing are obtained, and a high-complexity order L in the past is obtained. The order experienced two hardware failures during the processing period. The corresponding hardware load time series data includes that the average value of CPU utilization during the processing period is 80%, the peak value reaches 95%, the average value of memory read and write rate is 400MB / s, the peak value reaches 450MB / s, and the average value of I / O throughput is 150MB / s, and the peak value reaches 170MB / s. In addition, the processing time of order L is 40 minutes, of which the hardware load exceeds the preset load threshold for 15 minutes. Based on these data, the hardware load fluctuation amplitude factor of order L is calculated, and the hardware load characteristic data during the processing of order L is constructed by combining the hardware load average value, hardware load peak value, order processing time and the time when the hardware load exceeds the preset load threshold of order L. Based on the historical hardware failure records obtained and the hardware load characteristic data during the corresponding order processing period, the decision tree algorithm is used for model training to construct a hardware failure risk assessment model for the order opener. When processing the current order L, based on the predicted hardware load data, the hardware load characteristic data during the processing of order L is extracted, and the hardware load fluctuation factor of order L is 0.721, the predicted average value of CPU utilization is 85%, and the peak value will reach 98%, the predicted average value of memory read and write rate is 420MB / s, and the peak value will reach 460MB / s, and the predicted average value of I / O throughput is 160MB / s, and the peak value will reach 180MB / s. Using the constructed hardware failure risk assessment model, it is predicted that the hardware failure risk level of order L during the processing period is high, so order L is judged as a high-risk order. The predicted hardware load data during the processing of order L is obtained, and it is determined that the hardware load will exceed the preset first load threshold at multiple time points during the processing, such as at the 20th and 35th minutes of processing. Therefore, these time points require special attention and resource management to prevent potential hardware failures.

[0041] Wherein, the hardware load fluctuation amplitude factor during the order processing period is calculated according to the time series data of the hardware load during the order processing period.

[0042] Obtain the time series data of the hardware load during the order processing period, divide the time series data of the hardware load according to the preset time window, and calculate the average value of the hardware load of different hardware load categories. The hardware load categories include CPU utilization, memory read and write rate Mem, and I / O throughput. According to the hardware load fluctuation amplitude factor formula Calculate the hardware load fluctuation factor F during the current order processing period w (t), where N is the number of data points in the time window, j represents the hardware load category, and x i,j(t) represents the j-th type of hardware load value at the ith moment, μ j (t) is the average value of the j-th type of load in the time window, |x i,j (t)-x i-1,j (t)| is the adjacent difference of load value, which is multiplied by the logarithmic term to emphasize the severity of fluctuation between adjacent moments.

[0043] Exemplarily, the time series data of the hardware load during order processing is obtained, including specific data of CPU utilization, memory read and write rate Mem, and I / O throughput. These data are divided according to the preset 5-minute time window, and the average values ​​of different hardware load categories are calculated. It is found that within a certain time window, the average CPU utilization is 70%, the average memory read and write rate is 320MB / s, and the average I / O throughput is 140MB / s. Within this time window, the hardware load data at the i-th moment is recorded, of which the CPU utilization is 80%, the memory read and write rate is 330MB / s, and the I / O throughput is 155MB / s. The CPU utilization at the i-1th moment is 78%, the memory read and write rate is 325MB / s, and the I / O throughput is 153MB / s. According to the hardware load fluctuation amplitude factor formula Calculate the hardware load fluctuation factor F during the current order processing period w (t) is 0.12, which means that the fluctuation of hardware load during the processing of order Q is relatively stable, where N is the number of data points in the time window, j represents the hardware load category, and x i,j (t) represents the j-th type of hardware load value at the ith moment, μ j (t) is the average value of the j-th type of load in the time window, |x i,j (t)-x i-1,j (t)| is the adjacent difference of load value, which is multiplied by the logarithmic term to emphasize the severity of fluctuation between adjacent moments.

[0044] Step S104, if the time interval between consecutive high-complexity orders processed by the order opener is less than a preset time, the risk level of hardware failure of the order opener during the consecutive high-complexity order processing is predicted based on the hardware load data during the consecutive high-complexity order processing, and the failure risk order is determined.

[0045] If the time interval between consecutive processing of high-complexity orders by the order opener is less than the preset time, the order feature data of the current order and the order feature data of the first n high-complexity orders are obtained. According to the order opener hardware failure risk assessment model, the risk level of the order opener hardware failure during the processing of the first n orders of consecutive high-complexity orders is determined, and the risk level of the order opener hardware failure during the processing of the current order is predicted, and the current order is the n+1th order of consecutive high-complexity orders. If the risk level of the order opener hardware failure during the processing of the first n orders of consecutive high-complexity orders and the current order is low, the hardware load data during the processing of consecutive high-complexity orders is predicted according to the order opener hardware load prediction model. According to the hardware load data during the processing of consecutive high-complexity orders, the hardware load feature data during the processing of consecutive high-complexity orders is obtained, and the hardware failure risk assessment model of the order opener is used to predict the risk level of the order opener hardware failure during the processing of consecutive high-complexity orders. If the risk level of hardware failure of the order opener during the processing of continuous high-complexity orders is high, the current order is judged to be a failure risk order, and based on the obtained hardware load data during the processing of continuous high-complexity orders, the time point when the hardware load in the failure risk order exceeds the preset second load threshold is determined.

[0046] For example, the order opener is processing order X, which is a high-complexity order, and before it, the order opener has just processed a high-complexity order W. The processing time interval between the two orders is less than 10 minutes, so the order feature data of order X is obtained, including 8 processing steps, 800MB of data processing volume, 6 concurrent operations, and 30 minutes of processing time, and the order feature data of order W is obtained, including 7 processing steps, 750MB of data processing volume, 5 concurrent operations, and 28 minutes of processing time. Then, according to the order opener hardware failure risk assessment model, the order opener hardware failure risk during the processing of order W is judged, and the hardware failure risk during the processing of order W is judged to be low. According to the same model, the hardware failure risk during the current processing of order X is predicted, and the judgment result is also low risk. Since the hardware failure risk of the previous order W and the current order X are both low, the hardware load data during the continuous processing of these high-complexity orders are predicted according to the order opener hardware load prediction model. It is predicted that during the processing of order X, the CPU utilization rate will reach 85%, the memory read and write rate will be 350MB / s, and the I / O throughput will be 160MB / s. Based on these predicted data, the hardware load characteristic data during the continuous processing of high-complexity orders are extracted, and the hardware failure risk assessment model of the order opener is used to re-evaluate the hardware failure risk during this period. If the hardware failure risk during the continuous processing of these orders is high after re-evaluation, the current order X is judged to be a failure risk order, and based on the obtained hardware load data during the continuous high-complexity order processing, it is determined that the time point when the hardware load of order X exceeds the preset second load threshold is the 15th minute during the processing of order X. Therefore, this time point requires special attention and resource management to prevent potential hardware failures. If after re-evaluation, it is found that the risk level of hardware failure during the continuous processing of these orders is low, and the subsequent continuous processing of order Y is a high-complexity order, and the hardware failure risk level during the processing of order Y is also predicted to be low according to the hardware load prediction model of the order opener, then the hardware load data during the continuous processing of these high-complexity orders is predicted according to the hardware load prediction model of the order opener, and the hardware failure risk assessment model of the order opener is used to re-evaluate the hardware failure risk during the period to determine whether the current order Y is a failure risk order.

[0047] Step S105, generating hardware resource allocation optimization measures based on the predicted hardware load data during the current order processing period, and optimizing the hardware resource allocation of the order opener in advance.

[0048] Based on the predicted hardware load data during the processing of high-risk orders and the hardware load data during the processing of fault-risk orders, hardware resource allocation optimization measures are generated, including CPU time slice allocation, memory partition adjustment, and priority scheduling of I / O queues. Based on the hardware resource allocation optimization measures and the time point when the hardware load in high-risk orders exceeds the preset first load threshold, and the time point when the hardware load in fault-risk orders exceeds the preset second load threshold, the hardware resource allocation of the order opener is optimized in advance.

[0049] For example, in an order processing system, it is predicted that order O is a high-risk order and order P is a failure risk order, wherein during the processing of order O, it is predicted that at the 12th minute, the CPU utilization rate will reach 90%, exceeding the preset first load threshold of 85%. During the processing of order P, it is predicted that at the 15th minute, the memory read and write rate will reach 480MB / s, exceeding the preset second load threshold of 450MB / s. Therefore, based on the obtained hardware load data during the processing of high-risk orders and the hardware load data during the processing of failure risk orders, hardware resource allocation optimization measures are generated, including shortening the CPU time slice of order O from the original 40ms to 20ms to reduce the single task processing load of the CPU, adjusting the memory partition, and allocating more memory to key tasks during the processing of order P to ensure stability at high memory read and write rates. According to these optimization measures, when order O is processed to the 10th minute, the CPU time slice is adjusted in advance so that the load can be effectively dispersed when the CPU utilization reaches the peak in the 12th minute. When order P is processed to the 13th minute, memory resources are reallocated to prevent potential hardware failures when the memory read and write rates increase in the 15th minute.

[0050] Step S106, obtaining hardware load data after the hardware resource allocation optimization measures are implemented, determining the effectiveness of the hardware resource allocation optimization measures, and using a deep Q network algorithm to perform model training to determine hardware resource allocation optimization measures that meet different hardware load data.

[0051] Through the real-time monitoring system, the hardware load data after the hardware resource allocation optimization measures are obtained, the hardware load characteristic data during the current order processing period is extracted, and the obtained data is filtered, denoised and standardized using data cleaning and preprocessing technology. According to the preprocessed hardware load data, the hardware failure risk assessment model of the order opener is used to determine the risk level of the hardware failure of the order opener during the current order processing period, and to judge the effectiveness of the hardware resource allocation optimization measures. If the effectiveness of the hardware resource allocation optimization measures is lower than expected, the hardware resource allocation optimization measures are adjusted until the effectiveness of the hardware resource allocation optimization measures reaches the expected level. Adjusting the hardware resource allocation optimization measures includes reallocating CPU resources, increasing memory partitions, and optimizing the priority of I / O operations. The results of each adjustment of the hardware resource allocation optimization measures, the effectiveness evaluation results, and the hardware load data during the order processing period after the hardware resource allocation optimization measures are recorded, and the deep Q network algorithm is used for model training to determine the hardware resource allocation optimization measures that meet different hardware load data.

[0052] For example, in an order processing system, when processing order Z, the real-time monitoring system obtains the hardware load data after the hardware resource allocation optimization measures are taken, and obtains that during the order processing, the CPU utilization rate is 85%, the memory read and write rate is 420MB / s, and the I / O throughput is 170MB / s. The hardware load characteristic data during the processing of order Z is extracted, and the data is cleaned and preprocessed, including noise removal and standardization. According to the preprocessed hardware load data, the fault risk degree during the processing of order Z is evaluated using the order opener hardware failure risk assessment model. The evaluation results show that after the optimization measures are applied, the hardware failure risk of order Z is still high. Therefore, it is judged that the effectiveness of the current hardware resource allocation optimization measures is lower than expected, and the optimization measures need to be further adjusted, including reallocating CPU resources, further reducing the CPU time slice to reduce the high load state of the CPU, increasing memory partitions, ensuring that the memory can withstand higher read and write rate requirements, and optimizing the priority of I / O operations to ensure that high-priority I / O tasks can be processed in time. After these adjustments, the hardware failure risk is evaluated again, and it is found that the failure risk has been significantly reduced, achieving the expected effect. During this process, the adjustment results of each hardware resource allocation optimization measure are recorded in detail, including the changes in CPU utilization, memory usage, I / O throughput before and after optimization, and the corresponding effectiveness evaluation results. The deep Q network algorithm is used to train the model of these recorded data and determine the most suitable hardware resource allocation optimization measures under different hardware load conditions, thereby improving the stability and efficiency of the system in future order processing.

[0053] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) to form a technical solution.

Claims

1. A method for predicting failure of a single-opener based on a deep neural network, characterized in that: The method comprises: Obtain order feature data through the order monitoring database, build an order complexity evaluation model, evaluate the complexity of orders, and set hardware resource allocation plans for orders of different complexities; Based on the order feature data of historical high-complexity orders and the corresponding hardware load time series data, a hardware load prediction model for the order opener is constructed to predict the hardware load data during the current order processing period; Based on the historical failure records of the order opener hardware failures and the hardware load characteristic data during the corresponding order processing period, a risk assessment model for the order opener hardware failure is constructed to determine whether the current order is a high-risk order; If the time interval between consecutive high-complexity orders processed by the order opener is less than the preset time, the risk level of hardware failure of the order opener during the consecutive high-complexity order processing period is predicted based on the hardware load data during the consecutive high-complexity order processing period, and the failure risk orders are determined; Generate hardware resource allocation optimization measures based on the predicted hardware load data during the current order processing period, and optimize the hardware resource allocation of the order opener in advance; Obtain the hardware load data after the hardware resource allocation optimization measures are taken, judge the effectiveness of the hardware resource allocation optimization measures, and use the deep Q network algorithm to train the model to determine the hardware resource allocation optimization measures that meet different hardware load data.

2. The method according to claim 1, wherein: The order feature data is obtained through the order monitoring database, an order complexity evaluation model is constructed, the complexity of the order is evaluated, and hardware resource allocation schemes for orders of different complexity are set, including: Through the real-time monitoring system, the hardware load data of the order opener when processing orders is recorded and marked with timestamps. The hardware load data includes CPU utilization, memory read and write rates, and I / O throughput; the hardware load data with timestamps are integrated to form time series data, and the time series data of the hardware load is stored in the order monitoring database; through the order monitoring database, the order feature data of each order is obtained, and the order complexity of each order is marked, and the obtained order feature data is cleaned, including removing null values ​​and outlier processing. The order feature data includes the order processing steps, data processing volume, number of concurrent operations, and processing time. The complexity includes high, medium, and low; based on the cleaned order feature data, the decision tree algorithm is used for model training, and an order complexity evaluation model is constructed to evaluate the complexity of the order; according to the complexity level of the order, hardware resource allocation plans for orders of different complexities are set, and according to the order complexity evaluation results, the pre-set resource allocation plan is called.

3. The method according to claim 1, wherein: The order feature data of the historical high-complexity orders and the corresponding hardware load time series data are used to construct an order opener hardware load prediction model to predict the hardware load data during the current order processing period, including: Obtain order feature data of the order currently processed by the order opener, and use the order complexity evaluation model to evaluate the complexity of the order currently processed by the order opener; If the order currently processed by the order opener is identified as a high-complexity order, the order feature data of historical high-complexity orders and the corresponding hardware load time series data are obtained through the order monitoring database; based on the order feature data of historical high-complexity orders and the corresponding hardware load time series data, a long short-term memory network is used for model training to build an order opener hardware load prediction model; the hardware load data during the current order processing period is obtained through the real-time monitoring system, and combined with the order feature data of the current order, the order opener hardware load prediction model is used to predict the hardware load data during the current order processing period.

4. The method according to claim 1, wherein: The method of constructing a risk assessment model for hardware failure of an order opener based on the historical failure records of the order opener hardware failure and the hardware load characteristic data during the corresponding order processing period to determine whether the current order is a high-risk order includes: Obtain the historical hardware failure records of the order opener and the time series data of the hardware load during the corresponding order processing period through the order monitoring database; calculate the hardware load fluctuation amplitude factor, hardware load average value, hardware load peak value, order processing time and the time when the hardware load exceeds the preset load threshold during the order processing period based on the time series data of the hardware load during the order processing period, and construct the hardware load characteristic data during the corresponding order processing period; Based on the historical hardware failure records of the order opener and the hardware load characteristic data during the corresponding order processing period, the decision tree algorithm is used for model training to build an order opener hardware failure risk assessment model; based on the predicted hardware load data during the current order processing period, the hardware load characteristic data during the current order processing period is extracted, and the order opener hardware failure risk assessment model is used to predict the risk level of the order opener hardware failure during the current order processing period, and the risk level includes high and low; If the risk level of hardware failure of the order opener during the current order processing is high, the current order is judged to be a high-risk order; The predicted hardware load data during the high-risk order processing period is obtained, and a time point when the hardware load exceeds a preset first load threshold is determined.

5. The method according to claim 4, wherein: The step of calculating the hardware load fluctuation amplitude factor during the order processing period according to the time series data of the hardware load during the order processing period includes: Obtain the time series data of the hardware load during the order processing period, divide the time series data of the hardware load according to the preset time window, and calculate the average value of the hardware load of different hardware load categories. The hardware load categories include CPU utilization, memory read and write rate Mem, and I / O throughput; according to the hardware load fluctuation amplitude factor formula Calculate the hardware load fluctuation factor F during the current order processing period w (t), where N is the number of data points in the time window, j represents the hardware load category, and x i,j (t) represents the j-th type of hardware load value at the ith moment, μ j (t) is the average value of the j-th load in the time window, |x i,j (t)-x i-1,j (t)| is the adjacent difference of load value, which is multiplied by the logarithmic term to emphasize the severity of fluctuation between adjacent moments.

6. The method according to claim 1, wherein: If the time interval between consecutive high-complexity orders processed by the order opener is less than a preset time, the risk level of hardware failure of the order opener during the consecutive high-complexity order processing is predicted based on the hardware load data during the consecutive high-complexity order processing, and the failure risk order is determined, including: If the time interval between consecutive processing of high-complexity orders by the order opener is less than the preset time, the order feature data of the current order and the order feature data of the first n high-complexity orders are obtained; According to the opener hardware failure risk assessment model, determine the opener hardware failure risk level during the processing of the first n orders of continuous high-complexity orders, and predict the opener hardware failure risk level during the processing of the current order, where the current order is the n+1th order of continuous high-complexity orders; if the opener hardware failure risk levels during the processing of the first n orders of continuous high-complexity orders and the current order are both low, predict the hardware load data during the processing of continuous high-complexity orders according to the opener hardware load prediction model; obtain the hardware load characteristic data during the processing of continuous high-complexity orders according to the hardware load data during the processing of continuous high-complexity orders, and use the opener hardware failure risk assessment model to predict the opener hardware failure risk level during the processing of continuous high-complexity orders; If the risk level of hardware failure of the order opener during the processing of continuous high-complexity orders is high, the current order is judged to be a failure risk order, and based on the obtained hardware load data during the processing of continuous high-complexity orders, the time point when the hardware load in the failure risk order exceeds the preset second load threshold is determined.

7. The method according to claim 1, wherein: The hardware resource allocation optimization measures are generated according to the predicted hardware load data during the current order processing period, and the hardware resource allocation of the order opener is optimized in advance, including: Based on the predicted hardware load data during the processing of high-risk orders and the hardware load data during the processing of failure risk orders, hardware resource allocation optimization measures are generated. The hardware resource allocation optimization measures include CPU time slice allocation, memory partition adjustment, and I / O queue priority scheduling. Based on the hardware resource allocation optimization measures and the time point when the hardware load in the high-risk orders exceeds the preset first load threshold, and the time point when the hardware load in the failure risk orders exceeds the preset second load threshold, the hardware resource allocation of the order opener is optimized in advance.

8. The method according to claim 1, wherein: The hardware load data after the hardware resource allocation optimization measures are obtained, the effectiveness of the hardware resource allocation optimization measures is judged, and the model training is performed using a deep Q network algorithm to determine the hardware resource allocation optimization measures that meet different hardware load data, including: Through the real-time monitoring system, the hardware load data after the hardware resource allocation optimization measures are obtained, the hardware load characteristic data during the current order processing period is extracted, and the obtained data is filtered, denoised and standardized using data cleaning and preprocessing technology; based on the preprocessed hardware load data, the order opener hardware failure risk assessment model is used to determine the risk level of the order opener hardware failure during the current order processing period, and to judge the effectiveness of the hardware resource allocation optimization measures; if the effectiveness of the hardware resource allocation optimization measures is lower than expected, the hardware resource allocation optimization measures are adjusted until the effectiveness of the hardware resource allocation optimization measures reaches the expected level, and the adjustment of the hardware resource allocation optimization measures includes reallocating CPU resources, increasing memory partitions and optimizing the priority of I / O operations; the results of each adjustment of the hardware resource allocation optimization measures, the effectiveness evaluation results and the hardware load data during the order processing period after the hardware resource allocation optimization measures are recorded, and the deep Q network algorithm is used for model training to determine the hardware resource allocation optimization measures that meet different hardware load data.

Images