Cost prediction method and system based on deep learning

Through deep learning technology combining feature extraction and multiple trend analysis models, the shortcomings of traditional time series prediction in prepaid scenarios are solved, and the high-precision cloud cost prediction for the entire scenario is achieved, and the accuracy of prediction of prepaid and postpaid scenarios is improved. sex.

WO2024244951A1PCT designated stage expired Publication Date: 2024-12-05ALIBABA CLOUD COMPUTING CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2024/092775
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-29
Filing Date
2024-05-13
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Traditional time series prediction algorithms cannot effectively predict prepaid scenarios for cloud computing, and their accuracy is not enough to meet the high-sensitive data needs of enterprises for cloud costs, making it difficult for enterprises to trust and apply in production environments.

Method used

Using a cost prediction method based on deep learning, by obtaining time information, user information, cloud product information and historical cost trend information, extracting features and determining the target model in multiple preset trend analysis models, combining consumption mode information to predict, and using BIDA, APTN, and ARIMA algorithms process different types of historical cost trend information and introduce expert experience models to improve prediction accuracy in prepaid scenarios.

Benefits of technology

It realizes cloud-native cost prediction with high accuracy in all scenarios, improves prediction accuracy, especially in prepaid scenarios, and reaches 69% WMAPE prediction accuracy in covering prepaid and postpaid scenarios. Beyond the performance of traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024092775_05122024_PF_FP_ABST
    Figure CN2024092775_05122024_PF_FP_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of cloud computing. Disclosed is a cost prediction method based on deep learning. The method comprises: acquiring a data source in a preset time period, the data source comprising time information, user information, cloud product information, historical cost trend information and consumption mode information; obtaining a feature tensor according to the time information, the user information and the cloud product information; according to the historical cost trend information, determining a target trend analysis model from among a plurality of preset trend analysis models; inputting the feature tensor into the target trend analysis model to obtain a trend prediction result; determining a consumption prediction result according to the consumption mode information; and obtaining a cost and budget prediction result according to the trend prediction result, the consumption prediction result and the prediction time information. According to the present invention, an all-scenario prediction algorithm service can be provided, the prediction precision is improved, and high-precision personalized cloud-native all-scenario deep-learning-based cost prediction is realized.
Need to check novelty before this filing date? Find Prior Art

Description

A cost prediction method and system based on deep learning

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on May 29, 2023, with application number 202310619035.2 and application name “A cost prediction method and system based on deep learning”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present invention belongs to the field of cloud computing technology, and in particular relates to a cost prediction method and system based on deep learning. Background Art

[0003] While cloud computing brings increased agility and efficiency to enterprise IT infrastructure, improper or over-allocated cloud resources are common. Data from Flexera's 2022 State of the Cloud report shows that enterprises waste an average of 32% of their cloud spending after migrating to the cloud, making cloud costs difficult to control.

[0004] To solve the problem of cloud cost overruns, companies often observe and control expenses through budget management and cost analysis. Intelligent cost forecasting provides a future perspective for budget management and cost analysis, allowing companies to simulate future consumption cost trends based on historical consumption cost trends, thereby better planning future cloud expenditures.

[0005] Traditional time series forecasting algorithms typically focus solely on reasoning about historical time series patterns without incorporating business information, such as the characteristics of cloud-native services. Therefore, they cannot guarantee the accuracy of their forecasts. Furthermore, traditional time series forecasting algorithms are only applicable to postpaid consumption scenarios and are ineffective in prepaid scenarios, which are more random. Therefore, they cannot guarantee comprehensive forecast results. In common time series algorithm competitions, the winning algorithm often achieves a WMAPE forecasting accuracy of only around 60%, which is ineffective for highly sensitive data such as enterprise costs. The lack of prepaid scenarios also leads to a lack of trust among enterprise customers and a lack of application in production environments for traditional time series forecasting algorithms.

[0006] A similar prediction solution in the existing technology is AWS cost prediction. The disadvantage of AWS cost prediction is its low prediction accuracy of only 45%. Another similar solution in the existing technology is traditional time series prediction (commonly used in CMP cloud management platforms). However, traditional time series prediction generally only supports predictions for postpaid parts, and does not support predictions for prepaid parts and resource packages.

[0007] Summary of the Invention

[0008] The purpose of the present invention is to provide a cost prediction method and system based on deep learning, which can provide full-scene prediction algorithm services, improve prediction accuracy, and realize high-precision cloud-native full-scene deep learning cost prediction for thousands of people.

[0009] To achieve the above objectives, the present invention provides a cost prediction method based on deep learning, comprising:

[0010] Obtaining a data source for a predetermined time period, the data source including time information, user information, cloud product information, historical cost trend information, and consumption pattern information;

[0011] Obtaining a feature tensor according to the time information, the user information, and the cloud product information;

[0012] Determining a target trend analysis model from a plurality of preset trend analysis models based on the historical cost trend information;

[0013] Inputting the feature tensor into the target trend analysis model to obtain a trend prediction result;

[0014] Determining a consumption forecast result based on the consumption pattern information;

[0015] Cost and budget forecast results are obtained based on the trend forecast results, the consumption forecast results and the forecast time information.

[0016] Furthermore, the step of obtaining a feature tensor based on the time information, the user information, and the cloud product information includes:

[0017] Performing feature extraction on the time information, the user information, and the cloud product information to obtain time features, user features, and product features;

[0018] The time features, user features, and product features are fused to obtain a feature tensor.

[0019] Furthermore, the step of determining a target trend analysis model from a plurality of preset trend analysis models based on the historical cost trend information includes:

[0020] performing trend analysis on the historical expense trend information in sequence to determine a trend type of the historical expense information;

[0021] According to the trend type, a target model for processing the historical expense trend information is determined from a plurality of preset trend analysis models.

[0022] Furthermore, the trend type includes a non-periodic noise type, a periodic noise type and a stable type;

[0023] According to the trend type, a target model for processing the historical cost trend information is determined from a plurality of preset trend analysis models, including:

[0024] For the non-periodic noise type historical cost trend information, select a BIDA calculation model from a plurality of preset trend analysis models for processing;

[0025] Selecting an APTN calculation model from a plurality of preset trend analysis models to process the periodic noise type historical cost trend information;

[0026] For the stable historical cost trend information, an ARIMA calculation model from a plurality of preset trend analysis models is selected for processing.

[0027] Furthermore, the consumption mode information includes prepaid information and postpaid information.

[0028] The step of determining the consumption forecast result according to the consumption pattern information includes:

[0029] Determine stable renewal instances based on preset screening criteria;

[0030] Determining the renewal duration and renewal fee of the stable renewal instance within the forecast period based on the preset forecast duration;

[0031] The consumption forecast result is determined based on the renewal duration and renewal fee.

[0032] Furthermore, the step of obtaining cost and budget forecast results based on the trend forecast results, the consumption forecast results and the forecast time information includes:

[0033] Fusion of the trend forecast results and the consumption forecast results to obtain forecast information;

[0034] The cost and budget forecast results are obtained according to the forecast information and the forecast time information.

[0035] The present invention also provides a cost prediction system based on deep learning, comprising:

[0036] An information input module is used to obtain a data source for a predetermined time period, wherein the data source includes time information, user information, cloud product information, historical cost trend information, and consumption pattern information;

[0037] A feature fusion module, configured to obtain a feature tensor based on the time information, the user information, and the cloud product information;

[0038] a prediction analysis module, configured to input the feature tensor into the target trend analysis model to obtain a trend prediction result, and determine a consumption prediction result based on the consumption pattern information;

[0039] The forecast output module is used to obtain cost forecast results and budget forecast results based on the trend forecast results, the consumption forecast results and the forecast time information.

[0040] Furthermore, the prediction analysis module includes:

[0041] A BIDA calculation model for processing non-periodic noise information in the historical cost trend information;

[0042] An APTN calculation model for processing periodic noise information in the historical cost trend information;

[0043] An ARIMA calculation model is used to process the stable information in the historical cost trend information;

[0044] The expert experience model groups and processes the consumption mode information based on renewal rules.

[0045] Furthermore, the forecast output module includes a budget management interface and a cost analysis interface, the budget management interface is used to display the budget forecast result, and the cost analysis interface is used to display the cost forecast result.

[0046] The present invention also provides a computer device, which includes a processor and a memory; wherein the memory stores computer program instructions; when the computer program is executed by the processor, the above-mentioned deep learning-based cost prediction method is implemented.

[0047] The present invention also provides a computer-readable storage medium, which includes instructions. When the instructions are run on a computer, the computer executes a program that implements the above-mentioned deep learning-based cost prediction method.

[0048] Compared with the existing technology, according to the deep learning-based cost prediction method and system of the present invention, the data source includes time information, user information, cloud product information, historical cost trend information and consumption method information. The effective information in the data source is divided into two categories and five subcategories, namely four types of business attributes: user information, cloud product information, historical consumption trends, and consumption methods, and the natural attributes of time information; among them, time information, user information, and cloud product information can be used to extract time features, user features, and product features through data mining; effective features can efficiently distinguish the cloud-native time series of different businesses, and realize the differentiation of thousands of people through feature means.

[0049] In addition, the cost forecasting method and system based on deep learning of the present invention can also select the BIDA calculation model from multiple preset trend analysis models to process the non-periodic noise type historical cost trend information. The BIDA algorithm has the characteristics of two-way learning and can more effectively deal with the prediction of non-periodic noise type; for the historical cost trend information of periodic noise type, select the APTN calculation model from multiple preset trend analysis models to process it. The APTN algorithm has a periodic detection component and can effectively deal with the prediction of periodic noise type; for the stable historical cost trend information, select the ARIMA calculation model from multiple preset trend analysis models to process it. The use of the ARIMA algorithm can save computing resources; reasonable trend grouping is performed for different historical cost trend information to realize the adaptation of various algorithms to local conditions.

[0050] At the same time, the expert experience model groups and processes the consumption method information based on the renewal rules. The renewal rule expert model can achieve 100% prediction accuracy for the time series that meet the model requirements, which can improve the prediction accuracy of the prepaid scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] FIG1 is a flow chart of a cost prediction method based on deep learning according to an embodiment of the present invention;

[0052] FIG2 is a schematic diagram of a cost prediction method based on deep learning according to an embodiment of the present invention;

[0053] FIG3 is a schematic diagram of a classification process of historical cost trend information according to an embodiment of the present invention;

[0054] FIG4 is a schematic diagram of a BIDA algorithm model according to an embodiment of the present invention;

[0055] FIG5 is a schematic diagram of an APTN algorithm model according to an embodiment of the present invention;

[0056] FIG6 is a schematic diagram of a cost analysis daily forecast application according to an embodiment of the present invention;

[0057] FIG7 is a schematic diagram of a monthly cost analysis forecast application according to an embodiment of the present invention;

[0058] FIG8 is a schematic diagram of daily and monthly forecast results of budget management according to an embodiment of the present invention;

[0059] FIG9 is a schematic diagram of prediction results of a pre-real-time analysis dashboard according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The specific embodiments of the present invention are described in detail below, but it should be understood that the protection scope of the present invention is not limited by the specific embodiments.

[0061] Unless expressly stated otherwise, throughout the specification and claims, the term "comprise" or variations such as "include" or "comprising", etc., will be understood to include the stated elements or components but not to exclude other elements or other components.

[0062] As shown in FIG1 and FIG2 , a cost prediction method based on deep learning according to a preferred embodiment of the present invention includes:

[0063] Step S1, obtaining a data source for a predetermined time period, wherein the data source includes time information, user information, cloud product information, historical cost trend information, and consumption method information; wherein the data source refers to the source historical data used for Alibaba Cloud time series prediction training, such as billing data; time information refers to time-related information attached to a time series, such as whether a certain time point is a holiday, the day of the week at a certain time point, etc.; user information refers to user-related information attached to a time series, such as the user's industry, region, etc.; cloud product information refers to cloud product-related information attached to a time series, such as the category of cloud products, the billing method of cloud products, etc.; historical cost trend information refers to trend fluctuation information of a time series itself, such as potential trend, potential periodicity, etc.; consumption method information refers to whether a time series is postpaid or prepaid information;

[0064] Among them, step S1 includes: extracting features of the time information, the user information and the cloud product information respectively by means of data mining to obtain time features, user features and product features, and performing feature fusion on the time features, user features and product features to obtain a feature tensor.

[0065] The attribute features of the time information, the user information, and the cloud product information may be extracted respectively. The obtained time features may be used to represent the time attributes when the consumption occurs, as shown in the following table:

[0066] The obtained user features can be used to represent the attribute features of the user who generates the consumption behavior, as shown in the following table:

[0067] The obtained product features can be used to represent the attribute information of the consumed product or service, as shown in the following table:

[0068] Step S2: obtaining a feature tensor based on the time information, the user information, and the cloud product information;

[0069] Step S3, based on the historical expense trend information, determines a target trend analysis model from a plurality of preset trend analysis models; wherein step S3 includes: sequentially performing trend analysis on the historical expense trend information to determine the trend type of the historical expense information; based on the trend type, determining a target trend analysis model for processing the historical expense trend information from a plurality of preset trend analysis models; the trend types include non-periodic noise type, periodic noise type, and stable type. Based on the trend type, determining a target model for processing the historical expense trend information from a plurality of preset trend analysis models includes: selecting a BIDA calculation model from a plurality of preset trend analysis models for processing the non-periodic noise type historical expense trend information; selecting an APTN calculation model from a plurality of preset trend analysis models for processing the periodic noise type historical expense trend information; and selecting an ARIMA calculation model from a plurality of preset trend analysis models for processing the stable type historical expense trend information.

[0070] The business attributes of historical consumption trends are key to deep learning algorithms. Different deep learning algorithms have their own strengths and weaknesses, and excel at handling time series data for different business types. Therefore, before inputting time series data into the algorithm model, appropriate trend grouping should be performed to ensure that each algorithm is tailored to the specific needs of the application.

[0071] In step S4, the feature tensor is input into the target trend analysis model to obtain a trend prediction result. In this embodiment of the present invention, time series are classified according to their trends into aperiodic noise type, periodic noise type, and stable type. The deep learning algorithm corresponding to the aperiodic noise type is BiDA, which is a modification of Seq2Seq and LSTM and features bidirectional learning, making it more effective for predicting aperiodic noise types. FIG4 shows a schematic diagram of the BiDA algorithm model. BiDA (Bi-Directional Attention Network) is a type of Seq2Seq network architecture and an encoder-decoder network based on an attention mechanism. BiDA is a global multivariate model (trained by combining all time series and capable of predicting multiple time-related variables) and is non-probabilistic. BiDA requires observations within a historical observation window and supports multi-step predictions. For example, if the observation window is T and the number of prediction steps is N, the model requires observations from the previous T time intervals to obtain predictions for the next time intervals T+1 to T+N.

[0072] The deep learning algorithm for periodic noise is APTN, which is also an improvement on Seq2Seq and LSTM. It features a periodicity detection component, making it effective for predicting periodic noise. Figure 5 shows the schematic diagram of the APTN algorithm model. Based on the BiDA network architecture, APTN (Attention-based Periodic-Temporal Neural Network) adds a periodic encoder to decompose the periodicity of time series. The encoding is then decoded using an attention mechanism. Compared to BiDA, APTN can more accurately capture the periodicity that occurs in time series history and reflect this periodicity in predictions.

[0073] Stable time series are relatively easy to predict, so the ARIMA algorithm is used to conserve computing resources. This algorithm primarily adjusts three parameters, p, d, and q, to ​​find the optimal forecast for each time series. p determines the order of autoregression, d determines the order of differencing of the original time series to ensure stationarity, and q determines the order of moving average calculations performed on the white noise series. This algorithm can only perform single-step forecasts and cannot input features. Therefore, it is only suitable for series with short forecast horizons and strong time series regularity.

[0074] The feature tensor is input into the target trend analysis model to obtain a trend prediction result. The trend prediction result can represent the trend of consumption behavior of various types of users.

[0075] Step S5 determines a consumption forecast based on the consumption pattern information, where the consumption forecast refers to future consumption amounts. Step S5 includes: determining stable renewal instances based on preset screening criteria; determining the renewal duration and renewal fee for the stable renewal instances within the predicted time period based on a preset prediction duration; and determining the consumption forecast based on the renewal duration and renewal fee. The consumption pattern information includes prepaid and postpaid information. Among stable renewal instances, some cloud product instances are regularly renewed, for example, incurring a renewal fee on January 1st of each year and zero consumption at other times. Time series observed in the business can be broadly categorized into four types based on their different patterns: aperiodic noise, periodic noise, periodic, and stationary. Each type of time series is suitable for different algorithms. Only by inputting each time series into the most appropriate algorithm can optimal accuracy and efficiency be achieved. Upstream in the algorithmic process, each time series is determined to belong to one of the four categories based on its historical trends, and similar data is grouped into a subset. The judgment criteria are the rate of change of the mean and variance of the time series within a given window, as well as the determination of periodicity in the Auto-ARIMA component. The classification process is shown in Figure 3. From a business perspective, the time series data forms of different business types also have certain commonalities. For example, amortized amount data has less noise, while payable amount prepaid data has stronger periodicity. Therefore, each business also has a main algorithm used. The main algorithms in each business are as follows:

[0076] Among them, the Prophet model is used to process the information with strong periodicity and weak noise in the historical cost trend information. The expression of the Prophet algorithm is as follows:

[0077] The parameters in the above formula are explained as follows:

[0078] g(t): the trend function of the model, which learns the basic trend of the model;

[0079] h(t): holiday function of the model, which is used to learn the impact of holidays on trends;

[0080] s(t): periodic function of the model, learning the periodicity of the model;

[0081] et: random error term, which makes the model have a certain degree of randomness;

[0082] The Prophet algorithm determines forecast results by fitting growth trends, seasonal trends, and the impact of holidays. By decomposing the time series and outputting separate forecasts, Prophet often better fits the trend and seasonality of the time series. Furthermore, by incorporating the impact of holidays on forecasts, Prophet is more flexible in handling special dates than ARIMA.

[0083] In some embodiments, in order to improve the prediction accuracy of prepaid scenarios, an expert experience model based on renewal rules is specially introduced. The basic logic of using renewal rules for prediction is: for some prepaid instances, if the historical renewal method of the instance is relatively stable in time and cost, the historical renewal method can be used to predict the time and cost of future renewal of the instance.

[0084] Step S6, obtaining the cost and budget forecast results based on the trend forecast results, the consumption forecast results and the forecast time information; wherein, step S6 includes fusing the trend forecast results and the consumption forecast results to obtain forecast information; obtaining the cost and budget forecast results based on the forecast information and the forecast time information; in this step, only the deep learning forecast results and the expert model forecast results are fused: if a time series has only the deep learning forecast result or the expert model forecast result, then the only result is taken; if there are both deep learning forecast results and expert model forecast results, then the accuracy of the two results is calculated, and the one with higher accuracy is taken as the final result.

[0085] During the actual prediction process, the historical renewal patterns of all instances must first be screened. The screening criteria include whether each renewal has the same length, the number of historical renewals, and the day of the month when the renewal occurred. Once the screening criteria are determined, the data analysis results must be used to determine the range of values ​​within which each condition is considered stable. For example, product prices fluctuate over time, so if multiple historical renewals have the same length but different fees, the instance should be considered stable. Stable renewal instances generally exhibit the following behavior patterns:

[0086] The renewal interval and duration are the same each time;

[0087] Renew on the same day every month, with the same renewal duration;

[0088] Renewal occurs every month, but the renewal date and interval are not fixed each time, and the renewal amount remains the same;

[0089] After screening stable instances, we need to extract each instance's renewal pattern. Then, based on the desired forecast duration (e.g., three months, six months), we calculate the expected renewal times and fees for each instance based on historical renewal rules. After generating a daily forecast for each instance, we can aggregate it to product and monthly forecasts.

[0090] The deep learning-based cost prediction method of the present invention divides valid information from the data source into two major categories and five subcategories, namely, four types of business attributes: user information, cloud product information, historical consumption trends, and consumption methods, and the natural attribute of time information. Time information, user information, and cloud product information can be extracted through data mining to extract time features, user features, and product features. Effective features can efficiently distinguish the cloud-native time series of different businesses, achieving personalized differentiation through feature-based methods. The method can process the historical cost trend information of the non-periodic noise type and the periodic noise type using different trend analysis models from multiple preset trend analysis models, thereby improving the prediction accuracy of prepaid scenarios.

[0091] The embodiment of the present invention uses the mined time features, user features, and product features as the input of the deep learning algorithm, and enters the deep learning model to assist in the training and prediction of the model. First, the three features are fused to form a feature tensor. Only features in the form of tensors can be parameterized by the deep learning algorithm and thus integrated into the model. The three features will play a role in distinguishing the environment and guiding trends in the training and prediction of the model, so that the model can adapt to the data distribution of cloud-native predictions, thereby improving the accuracy of the prediction. At the same time, the business attributes of cloud-native consumption methods are reasonably utilized to establish an expert model for cloud product renewal rules. The renewal rule expert model can achieve 100% prediction accuracy for time series that meet the requirements of the model. Finally, the deep learning prediction results and the expert model prediction results are fused to produce the final prediction results.

[0092] The prediction accuracy of the embodiment of the present invention can achieve 100% prediction accuracy in scenarios that conform to the expert experience model. In full scenarios covering prepaid and postpaid, it has a WMAPE prediction accuracy of 69%, which is greater than the average WMAPE prediction accuracy of 55% for predicting leagues and the 45% prediction accuracy of AWS.

[0093] Figures 6 to 9 show application display examples of the budget management interface and cost analysis interface of the user center. The embodiment of the present invention can predict and output forecasts for the next 90 days and the next 12 months. Figure 6 shows that in cost analysis, when the time granularity is selected as day, the daily forecast results are displayed using a dashed box graphic; Figure 7 shows that in cost analysis, when the time granularity is selected as month, the monthly forecast results are displayed using a dashed box graphic; Figure 8 shows that in budget management, when the time granularity is selected as day or month, historical reference data and forecast results for future days or months are displayed to guide budget preparation; Figure 9 shows that in the budget-actual analysis dashboard, the cumulative actual and forecast sums and future forecast amounts are displayed to analyze budget execution and predict the overspending date of the budget.

[0094] This embodiment of the present invention can process the non-periodic noise-type, periodic noise-type, and stable historical cost trend information using the BIDA, APTN, and ARIMA calculation models, respectively. This allows for reasonable trend grouping of different historical cost trend information, enabling the tailoring of various algorithms to local conditions. Furthermore, the expert experience model groups and processes the consumption pattern information based on renewal rules. The renewal rule expert model can achieve 100% prediction accuracy for time series that meet the model's requirements, improving prediction accuracy in prepaid scenarios.

[0095] This embodiment of the present invention leverages cloud-native features and introduces deep learning algorithms, balancing both business and algorithmic approaches. It also extracts an expert model for renewal rules in prepaid scenarios, achieving high-precision predictions across all scenarios, including prepaid and postpaid. In scenarios consistent with the expert experience model, it achieves 100% prediction accuracy. Across all prepaid and postpaid scenarios, it boasts a WMAPE prediction accuracy of 69%, exceeding the league average of 55% and AWS's 45%.

[0096] An embodiment of the present invention further provides a cost prediction system based on deep learning, comprising:

[0097] An information input module is used to obtain a data source for a predetermined time period, wherein the data source includes time information, user information, cloud product information, historical cost trend information, and consumption pattern information;

[0098] A feature fusion module, configured to obtain a feature tensor based on the time information, the user information, and the cloud product information;

[0099] a prediction analysis module for inputting the feature tensor into the target trend analysis model to obtain a trend prediction result, and determining a consumption prediction result based on the consumption pattern information; the prediction analysis module includes: a BIDA calculation model for processing non-periodic noise-type information in the historical cost trend information; an APTN calculation model for processing periodic noise-type information in the historical cost trend information; an ARIMA calculation model for processing stable information in the historical cost trend information; and an expert experience model for grouping and processing the consumption pattern information based on renewal rules;

[0100] A forecast output module is used to obtain cost forecast results and budget forecast results based on the trend forecast results, the consumption forecast results and the forecast time information; wherein, the forecast output module includes a budget management interface and a cost analysis interface, the budget management interface is used to display the budget forecast results, and the cost analysis interface is used to display the cost forecast results.

[0101] The deep learning-based cost prediction system in the embodiment of the present invention can extract time features, user features, and product features through data mining. Effective features can efficiently distinguish the cloud-native time series of different businesses, and achieve differentiation of thousands of people through feature means. It can also select the BIDA calculation model from multiple preset trend analysis models to process the non-periodic noise-type historical cost trend information; select the APTN calculation model from multiple preset trend analysis models to process the periodic noise-type historical cost trend information; select the ARIMA calculation model from multiple preset trend analysis models to process the stable historical cost trend information; and perform reasonable trend grouping for different historical cost trend information to achieve the adaptation of various algorithms to local conditions. At the same time, the expert experience model groups and processes the consumption method information based on the renewal rules. The renewal rule expert model can achieve 100% prediction accuracy for time series that meet the model requirements, which can improve the prediction accuracy of prepaid scenarios.

[0102] An embodiment of the present invention further provides a computer device, comprising a processor and a memory; wherein the memory stores computer program instructions; and the processor implements the above method when the computer program is executed by the processor.

[0103] An embodiment of the present invention also provides a computer-readable storage medium, which includes instructions. When the instructions are run on a computer, the computer implements the cost prediction method based on deep learning when executing the instructions.

[0104] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0105] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0106] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0108] The foregoing descriptions of specific exemplary embodiments of the present invention are for purposes of illustration and description. These descriptions are not intended to limit the invention to the precise forms disclosed, and it is apparent that many variations and modifications are possible in light of the foregoing teachings. The exemplary embodiments have been selected and described for the purpose of explaining the specific principles of the invention and their practical application, thereby enabling those skilled in the art to realize and utilize a variety of exemplary embodiments of the invention and various options and modifications. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A cost prediction method based on deep learning, It is characterized in that include: Acquire a data source for a predetermined time period, the data source including time information, user information, cloud product information, historical cost trend information, and consumption mode information; Obtaining a feature tensor according to the time information, the user information, and the cloud product information; Determining a target trend analysis model from among a plurality of preset trend analysis models according to the historical cost trend information; Inputting the feature tensor into the target trend analysis model to obtain a trend prediction result; Determining consumption forecast results according to the consumption mode information; Cost and budget forecast results are obtained based on the trend forecast results, the consumption forecast results and the forecast time information.

2. The cost prediction method based on deep learning according to claim 1, It is characterized in that The step of obtaining a feature tensor according to the time information, the user information and the cloud product information includes: Extracting features from the time information, the user information, and the cloud product information to obtain time features, user features, and product features; The time features, user features and product features are fused to obtain a feature tensor.

3. The cost prediction method based on deep learning according to claim 1, It is characterized in that The step of determining a target trend analysis model from among a plurality of preset trend analysis models according to the historical cost trend information comprises: Performing trend analysis on the historical expense trend information in sequence to determine the trend type of the historical expense information; According to the trend type, a target model for processing the historical expense trend information is determined from a plurality of preset trend analysis models.

4. The cost prediction method based on deep learning according to claim 3, It is characterized in that The trend types include non-periodic noise type, periodic noise type and stable type; According to the trend type, a target model for processing the historical cost trend information is determined from a plurality of preset trend analysis models, including: For the non-periodic noise type historical cost trend information, a BIDA calculation model from a plurality of preset trend analysis models is selected for processing; For the historical cost trend information of the periodic noise type, an APTN calculation model from among a plurality of preset trend analysis models is selected for processing; For the stable historical cost trend information, an ARIMA calculation model from among a plurality of preset trend analysis models is selected for processing.

5. The cost prediction method based on deep learning according to claim 1, It is characterized in that The consumption method information includes prepaid information and postpaid information. The step of determining the consumption forecast result according to the consumption mode information comprises: Determine stable renewal instances based on preset screening conditions; Determining the renewal duration and renewal fee of the stable renewal instance within the forecast time period according to the preset forecast duration; The consumption forecast result is determined based on the renewal duration and the renewal fee.

6. The cost prediction method based on deep learning according to claim 1, It is characterized in that The step of obtaining cost and budget forecast results according to the trend forecast results, the consumption forecast results and the forecast time information comprises: Merging the trend prediction result and the consumption prediction result to obtain prediction information; The cost and budget forecast results are obtained according to the forecast information and the forecast time information.

7. A cost prediction system based on deep learning, It is characterized in that include: An information input module, used to obtain a data source for a predetermined time period, wherein the data source includes time information, user information, cloud product information, historical cost trend information, and consumption mode information; A feature fusion module, used to obtain a feature tensor according to the time information, the user information and the cloud product information; A prediction analysis module, used for inputting the feature tensor into the target trend analysis model to obtain a trend prediction result, and determining a consumption prediction result according to the consumption mode information; The prediction output module is used to obtain cost prediction results and budget prediction results according to the trend prediction results, the consumption prediction results and the prediction time information.

8. The cost prediction system based on deep learning according to claim 7, It is characterized in that The prediction analysis module includes: A BIDA calculation model is used to process non-periodic noise type information in the historical cost trend information; An APTN calculation model is used to process periodic noise type information in the historical cost trend information; ARIMA calculation model, used to process the stable information in the historical cost trend information; The expert experience model groups and processes the consumption mode information based on renewal rules.

9. The cost prediction system based on deep learning according to claim 7, It is characterized in that The forecast output module includes a budget management interface and a cost analysis interface. The budget management interface is used to display the budget forecast result, and the cost analysis interface is used to display the cost forecast result.

10. A computer device, It is characterized in that The computer device includes a processor and a memory; wherein the memory stores computer program instructions; and the processor, when the computer program is executed by the processor, implements the deep learning-based cost prediction method as described in any one of claims 1 to 5.

11. A computer-readable storage medium, It is characterized in that The computer-readable storage medium includes instructions, which, when executed on a computer, enable the computer to implement the deep learning-based cost prediction method as described in any one of claims 1 to 5.

Citation Information

Patent Citations

  • Heterogeneous cloud service oriented resource configuration method and device

    CN110198244A

  • Cloud cost prediction method and system based on service awareness

    CN112767027A

  • Data monitoring method, device and equipment

    CN113934615A

  • Cost prediction method and system based on deep learning

    CN116862548A

  • Determining and implementing a feasilbe resource optimization plan for public cloud consumption

    US20210240539A1