Resource allocation method, device and equipment

Abstract

The embodiments of this specification disclose a resource allocation method, device and equipment, which belongs to the field of machine learning technology. The scheme includes: determining business attributes related to available resources, and multiple micro time series corresponding to the macro time series to be predicted for describing the change of the business attributes; clustering the multiple micro time series to obtain multiple micro time series groups; predicting each of the micro time series groups respectively to obtain the predicted value of each of the micro time series groups; predicting the predicted value of the macro time series according to the predicted value of each of the micro time series groups; and allocating the available resources to the business corresponding to the business attributes according to the predicted value of the macro time series.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a resource allocation method, device and equipment. Background Art

[0002] With the development and popularization of Internet technology, many businesses are conducted online, such as SaaS office support business, ticket sales business, cloud disk business, cloud gaming business, and so on.

[0003] Some of these businesses are highly dynamic and require relatively real-time and sufficient computing resources. Time series can be used to describe the dynamic changes of these businesses. Time series, also known as time series, is a string of data indexed by time dimension, which describes the measurement value of a measured subject at each time point within a time range. For example, the number of online office workers, the number of concurrent ticket purchasing users, ticket sales, network disk usage load and other business attributes at different times.

[0004] Based on this, we need to allocate computing resources to the business in a more timely and reasonable manner in response to the dynamic changes in the business, so as to help achieve optimal resource allocation and reduce idle resources and waste. Summary of the invention

[0005] One or more embodiments of the present specification provide a resource allocation method, apparatus, device and storage medium to solve the following technical problems: It is necessary to allocate computing resources to the business more timely and reasonably in response to the dynamic variability of the business, so as to help achieve optimal allocation of resources and reduce idle and wasted resources.

[0006] To solve the above technical problems, one or more embodiments of this specification are implemented as follows:

[0007] One or more embodiments of this specification provide a resource allocation method, including:

[0008] Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute;

[0009] Clustering the multiple micro time series to obtain multiple micro time series groups;

[0010] Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups;

[0011] According to the predicted values ​​of each of the micro time series groups, the predicted values ​​of the macro time series are predicted;

[0012] The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0013] One or more embodiments of this specification provide a resource allocation device, including:

[0014] A micro-determination module determines the business attributes related to the available resources, and determines a plurality of micro-time series corresponding to the macro-time series to be predicted for describing the change of the business attributes;

[0015] A time series clustering module clusters the multiple micro time series to obtain multiple micro time series groups;

[0016] A first prediction module predicts each of the micro time series groups to obtain a prediction value of each of the micro time series groups;

[0017] A second prediction module predicts the predicted value of the macro time series according to the predicted value of each of the micro time series groups;

[0018] A resource allocation module allocates the available resources to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0019] One or more embodiments of this specification provide a resource allocation device, including:

[0020] at least one processor; and,

[0021] a memory communicatively connected to the at least one processor; wherein,

[0022] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0023] Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute;

[0024] Clustering the multiple micro time series to obtain multiple micro time series groups;

[0025] Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups;

[0026] According to the predicted values ​​of each of the micro time series groups, the predicted values ​​of the macro time series are predicted;

[0027] The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0028] One or more embodiments of this specification provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to:

[0029] Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute;

[0030] Clustering the multiple micro time series to obtain multiple micro time series groups;

[0031] Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups;

[0032] According to the predicted values ​​of each of the micro time series groups, the predicted values ​​of the macro time series are predicted;

[0033] The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0034] At least one of the above technical solutions adopted in one or more embodiments of this specification can achieve the following beneficial effects: time series prediction of macro business attributes related to available resources is performed to timely and prospectively grasp the dynamic changes of the business, and, instead of directly predicting the macro time series, the differences between the micro individuals (finer-grained attributes under the business attributes) involved in the macro time series are taken into account, and the micro time series corresponding to the micro individuals are clustered, so that micro time series with similar or homogeneous performances are clustered into the same micro time series group, and then more targeted predictions can be made for each group respectively, which can include natural predictions, as well as the application of The estimation of the effects brought about by some active intervention policies or measures, clustering allows natural predictions and effect estimates to be differentiated in each group, which is more in line with the actual situation of each group, thus helping to obtain more accurate prediction results. By predicting the micro time series based on clustering, the prediction value of the corresponding macro time series is determined, which also improves the prediction accuracy of the macro time series and can better support the prediction in scenarios where policies or measures are used for flexible intervention; resources are allocated accordingly based on the prediction results, which helps to allocate resources in a timely manner to businesses with dynamically increasing resource demands to support the normal operation of the business, improve the fault tolerance and execution efficiency of the business, and optimize resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A flowchart of a resource allocation method provided for one or more embodiments of this specification;

[0037] Figure 2 A schematic diagram of a resource allocation system provided for one or more embodiments of this specification;

[0038] Figure 3 One or more embodiments of this specification provide an application scenario in which Figure 1 Schematic diagram of the specific implementation process of the method;

[0039] FIG. 4(a), FIG. 4(b), and FIG. 4(c) are provided in one or more embodiments of this specification. In an application scenario Figure 2 Schematic diagram of the user interaction interface provided by the system;

[0040] Figure 5 A schematic diagram of the structure of a resource allocation device provided for one or more embodiments of this specification;

[0041] Figure 6 A schematic diagram of the structure of a resource allocation device provided in one or more embodiments of this specification. DETAILED DESCRIPTION

[0042] The embodiments of this specification provide a resource allocation method, apparatus, device and storage medium.

[0043] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0044] Some exemplary scenarios such as SaaS office support business, ticket sales business, cloud network disk business, cloud game business, etc. are given in the background technology. In practical applications, the above services can be regarded as a kind of commodity (the example is an online commodity, and this solution is also applicable to offline physical commodities). The service provider provides corresponding computing resources, publicity resources, human resources and other resources to maintain the service, and recovers costs and makes profits through users who sell commodities. This is a common model that is easier for the public to understand. Based on this, for ease of understanding, some of the following embodiments are mainly explained by taking the large scenario of commodity sales (for example, the forecast of retailer sales) as an example.

[0045] This application uses macro time series to describe the dynamic variability of the business, and then predicts the dynamic variability. Not only that, if the macro time series is predicted directly, usually only the historical information of the macro time series itself is considered. For example, only based on the historical sales of the retail industry, the future sales of retail sales are predicted, without considering the components of the retail industry, the historical sales of specific individual commodities, and the historical sales of commodities contain rich information, which helps to better predict the sales of the retail industry. In this case, it is easy to have the problem of eating from the same pot when allocating resources. Some local commodities that really need resources may not be supported by sufficient resources, so the effect of solving the problem is still not good in essence. In addition, it is often only predicted that the future value in the natural state will be used, without considering the trend of future sales when there is policy or means intervention (for example, increasing system maintenance time, ticket discounts, etc.), which will also affect the accuracy of resource allocation.

[0046] In response to the above problems, the industry is more concerned about macro time series. This solution does not directly predict macro time series, but indirectly predicts macro time series with the help of related micro time series. For example, if the time series of retailer sales is regarded as a macro time series, then its related micro time series can be the time series of sales of various commodities sold by the retailer. The time series of retailer sales is aggregated by the time series of these commodity sales. This example is also used for explanation below.

[0047] Micro time series contain a lot of information that helps to better predict macro time series. For example, the sales volume of different types of goods often show different patterns, some goods have a periodicity of regular weekly changes, some goods are more susceptible to serious impacts of holidays, and so on. Based on this, this scheme clusters micro time series with similar or homogeneous performance, so that for micro time series belonging to different groups after clustering, subsequent prediction actions that are more suitable for the group are performed separately. Not only that, when some policies or measures are imposed on some micro individuals, different micro individuals have different sensitivities to this, and the corresponding micro time series are affected differently, while the micro time series belonging to the same group may be more similarly affected, therefore, it helps to improve the accuracy of the prediction results of the macro time series obtained indirectly.

[0048] Based on this concept, it is further described in detail below.

[0049] Figure 1 A schematic diagram of a resource allocation method provided for one or more embodiments of this specification. The method can be applied to different business fields, such as ticket retail, cloud computing and cloud storage, financial business, medical business, e-commerce business, instant messaging business, robotics business, video service business, gaming business, official business, etc. The process can be executed by a computing device in the corresponding field (such as a risk control server corresponding to an e-commerce business), and some input parameters or intermediate results in the process allow manual intervention and adjustment to help improve accuracy.

[0050] Figure 1 The process in may include the following steps:

[0051] S102: Determine business attributes related to available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing changes in the business attributes.

[0052] In one or more embodiments of this specification, some available resources are currently available for some predetermined user service businesses. These are often performed online and provided through distributed systems. Therefore, these available resources typically include distributed computing resources. Business actions such as online business user interaction interface presentation, user request response, and online order processing all require computing resources to be executed, especially when the business action concurrency is high, and sufficient distributed computing resources need to be deployed to support the normal operation of the business.

[0053] The business attributes related to available resources mentioned in S102 include business attributes that can better reflect the demand for resources. There are two typical categories. One is the business attributes that reflect the concurrency of the user service business, such as the number of users online in real time, the number of concurrent user requests, etc. This type of attribute has stronger requirements for the timeliness of resource allocation; the other is the business attributes that reflect the cumulative degree of user service business, such as the retail sales of goods such as the number of tickets sold, the total number of non-duplicate user files in the cloud network disk, the total number of registered users, the total number of active users, etc. Although this type of attribute has reduced the timeliness requirements for resource allocation, it has a high guiding and reference value for the resource allocation strategy for a long period of time in the future, and helps to create a more stable and intelligent comprehensive resource allocation solution. This type of solution has better universality and is easy to transplant and reuse between different industries.

[0054] In one or more embodiments of the present specification, the macro time series can directly express the change of business attributes, and its value in the past period of time is known, and its value in the next period of time is to be predicted, and the value of the micro time series in the past period of time is known. In a more typical case, the corresponding multiple micro time series can be aggregated to obtain the macro time series. For example, the micro time series is a time series of individual sales data (such as sales volume or sales amount under a certain retailer) of a specified commodity as the measured subject, and the macro time series is a time series of overall sales data of all commodities within a specified range (for example, under the retailer). The specified commodity is included in all commodities, and the individual sales data is a component of the overall sales data. In such an example, the commodity is the measured subject itself or its component, and the sales data is a parameter to be measured specifically for the measured subject.

[0055] It should be noted that it is not necessary for these multiple micro time series to be able to aggregate to obtain the macro time series, as long as these multiple micro time series have sufficient reference value for predicting the macro time series. For example, assuming that the macro time series and another macro time series belong to different business fields, but the change trends of the two are similar enough, the micro time series that can be aggregated to obtain the other macro time series can be used as the micro time series corresponding to the macro time series.

[0056] S104: Clustering the multiple micro time series to obtain multiple micro time series groups.

[0057] In one or more embodiments of the present specification, based on the similarity or homogeneity between micro time series, these multiple micro time series are divided into different categories through clustering, and each category is a micro time series group. The time series features based on clustering can be customized (for example, some key time points of concern such as weekends and holidays, for example, the scale of time series values, the global shape of the curve composed of time series, peak and valley distribution, etc.) to better learn the desired similarity or homogeneity.

[0058] Since the micro time series is the performance of a certain parameter related to the measured subject, when analyzing the micro time series, we can also refer to some other characteristics of the measured subject itself. For example, if the measured subject is a commodity, such as the type of commodity, the main audience type, the shelf life of the commodity, the purpose of the commodity, etc., these characteristics will have an impact on the performance of the parameter. The similarity or homogeneity between commodities is likely to be mapped to the corresponding micro time series.

[0059] S106: Predict each of the micro time series groups respectively to obtain a predicted value of each of the micro time series groups.

[0060] In one or more embodiments of the present specification, each micro time series group can be learned in the same prediction model, and the prediction is performed uniformly through the prediction model. However, according to the description of the above clustering process, each micro time series group has a large difference in characteristics, so the effect of unified prediction may not be good. Based on this, in order to improve the prediction accuracy, a separate prediction model can be constructed and trained for each micro time series group, which is used to predict the corresponding micro time series group. This helps to avoid the above same prediction model being too generalized and affecting the accuracy.

[0061] Furthermore, in order to reduce the amount of calculation, multiple micro time series in the same micro time series group can be aggregated first (for example, the values ​​of the corresponding time points are directly added or weighted superimposed, and then the micro time series obtained after aggregation are predicted. In this case, there are as many micro time series groups as there are (this is the minimum number, and the number can be increased according to demand, using more different micro time series to represent the same group) aggregated micro time series that need to be predicted, which effectively reduces the number of micro time series that need to be predicted, and because the sequence within the group The columns are similar or homogeneous. Therefore, the error caused by unified prediction within the group (after aggregation within the group, the aggregated time series is predicted) is smaller than that caused by separate prediction within the group (separate prediction of each micro time series within the group), and the impact on the global situation is small and acceptable. In addition, if the method of separate prediction within the group is adopted, it may be necessary to model the micro time series. However, due to the high magnitude and sparse data of micro time series, it is difficult to model and the effect is usually poor. If the method of unified prediction within the group is adopted, the aggregated time series can be modeled, which helps to solve problems such as data sparsity and facilitates modeling.

[0062] S108: Predicting the predicted value of the macro time series according to the predicted value of each of the micro time series groups.

[0063] In one or more embodiments of this specification, it is mentioned in the background technology that the industry usually imposes some policies or measures on micro-individuals in order to influence the macro time series. In practical applications, different micro-individuals will show different sensitivities under the same policies or measures. For example, under a 20% discount on the selling price, the sales of some daily necessities will increase sharply in the short term, while the sales increment of some foods will be limited.

[0064] This scheme refers to such policies or measures as active intervention factors. When predicting each micro-time series group, the impact of active intervention factors can be superimposed. For different groups or micro-individuals in different groups, the active intervention factors applied can be differentiated. Through effect estimation, the additional impact of active intervention factors can be predicted, thereby obtaining the post-effect prediction value of the micro-time series group. In this case, the prediction value is more valuable for the industry and can guide the flexible use of active intervention factors in decision-making.

[0065] Furthermore, in order to distinguish the influence of active intervention factors, in the prediction of step S106, the micro time series group can be predicted based on the main, conventional, general, and representative business conditions and environmental conditions as much as possible, and the prediction state that meets such conditions is called the natural prediction state. Based on this, the natural prediction state is first determined, in which at least no active intervention factors are applied, and in the natural prediction state, each micro time series group is predicted respectively to obtain the predicted value of each micro time series group as the natural prediction value, which is used to predict the predicted value of the macro time series.

[0066] In addition, it should be noted that the effect estimation mentioned in this step can also be performed in advance in step S106.

[0067] In one or more embodiments of the present specification, if the macro time series can be obtained by aggregating these micro time series groups, then the predicted values ​​of these micro time series groups can be aggregated accordingly to obtain the predicted values ​​of the macro time series. If these micro time series groups are only indirectly related to the macro time series, but not directly related, it is necessary to perform processing such as mapping and scaling on these micro time series groups, and then indirectly infer the predicted values ​​of the macro time series.

[0068] S110: Allocate the available resources to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0069] In one or more embodiments of this specification, the predicted value can reflect the subsequent changes in resource demand, and the allocated resources can be pre-set to have a specified positive correlation with the resource demand reflected by the prediction, and the available resources can be allocated for the business according to the positive correlation. For example, when it is predicted that the retail sales of goods will increase, more computing resources will be allocated in time to support the processing of goods orders; for another example, when it is predicted that the access volume of a local cloud network disk will decrease, the network disk storage space here will be temporarily recovered and allocated to other areas for use; and so on.

[0070] pass Figure 1The method is to make time series forecasts for the macro business attributes related to available resources, so as to grasp the dynamic changes of the business in a timely and forward-looking manner. In addition, instead of directly predicting the macro time series, the differences between the micro individuals (the finer-grained attributes under the business attributes) involved in the macro time series are taken into account, and the micro time series corresponding to the micro individuals are clustered, so that the micro time series with similar or homogeneous performance are clustered into the same micro time series group, and then more targeted forecasts can be made for each group. The forecast can include natural forecasts, as well as some proactive intervention policies or measures to bring about Clustering enables natural prediction and effect estimation to be differentiated in each group, which is more in line with the actual situation of each group, thus helping to obtain more accurate prediction results. By predicting the micro time series based on clustering, the predicted value of the corresponding macro time series is determined, which also improves the prediction accuracy of the macro time series and can better support the prediction in scenarios where policies or measures are used for flexible intervention. Resources are allocated accordingly based on the prediction results, which helps to allocate resources to the business with dynamically increasing resource demand in a timely manner to support the normal operation of the business, improve the fault tolerance and execution efficiency of the business, and optimize resource allocation.

[0071] based on Figure 1 This specification also provides some specific implementation plans and extension plans of the method, which will be described below.

[0072] In one or more embodiments of the present specification, in order to better adapt to the different flexible requirements for applying active intervention factors in practical applications, the effect estimation is selectively performed at the granularity of micro time series during prediction. The active intervention factor to be applied to the measured subject of the macro time series is first determined. If the macro time series is aggregated by multiple micro time series, and the application is specific to the micro time series, the active intervention factor corresponding to the measured subject of one or more micro time series is determined, so as to estimate the effect of the corresponding micro time series according to the active intervention factor.

[0073] Of course, the granularity can be finer, for example, the active intervention factor applied to each measured subject of the micro time series in a specific local time interval. Take the sales volume of any commodity as an example: assuming that the sales volume to be predicted for the next month is to be directly given a 20% discount on the price of the commodity, this is the active intervention factor of the granularity of the micro time series, which is effective for the next month or even longer; assuming that the sales volume to be predicted for the next month is to be given a 10% discount on the price of the commodity in the first week, no discount on the price of the commodity in the second week, and a 20% discount on the price of the commodity in the third week, then the 10% discount and the 20% discount are both active intervention factors of the granularity of the local time interval.

[0074] Furthermore, based on the active intervention factors, the effect of the micro time series group is estimated to estimate the incremental value that the active intervention factors can bring to the micro time series group. Based on the incremental value and the predicted value of the micro time series group (i.e., the natural predicted value mentioned above), the predicted value of the macro time series is predicted.

[0075] There are two issues to consider when estimating the effect. First, different individuals react differently to the same intervention. Second, it is impossible to simultaneously observe the results of an individual being intervened and the results of the individual not being intervened. Therefore, it is not easy to accurately calculate the amount of change caused by the intervention on the individual. In the case of an intervention with a positive effect, the amount of change is the incremental value mentioned above.

[0076] The first problem can be solved by performing predictions for each group or even the sequences within each group through homogeneous clustering. The second problem can be solved by further constructing a counterfactual control group within the micro-time series group. Specifically, one or more micro-time series can be selected within the micro-time series group as intervention sequences, in order to apply active intervention factors to the measured subjects of the intervention sequences, and other micro-time series within the micro-time series group can be selected as non-intervention sequences, in order not to apply active intervention factors to the measured subjects of the non-intervention sequences. In this way, a counterfactual control group is constructed based on the intervention sequence and the non-intervention sequence for effect estimation. Since the micro-individuals corresponding to the sequences within the same micro-time series group also have similarity and homogeneity, they can be approximately regarded as the same micro-individual, which is equivalent to observing the states of the micro-individual applying and not applying active intervention factors at the same time, which helps to more accurately determine the incremental value when estimating the effect.

[0077] In one or more embodiments of this specification, the natural state is mentioned above, and some covariate vectors can be used to define the natural state. In addition, for different business scenarios, additional covariate vectors are required to define some factors that need special attention when predicting in the business scenario. For example, when predicting sales data, covariate vectors can be used to represent the distribution of holidays, the distribution of commodity supply channels, domestic and foreign popular trends, and so on.

[0078] Based on this idea, we can determine the time interval corresponding to each micro time series group as the first time interval (a period of history), and determine the overall time interval composed of the first time interval and the second time interval after the first time interval (one end of the future time), determine the covariate vector corresponding to each micro time series group in the overall time interval, and make predictions based on each micro time series group and the corresponding covariate vector to obtain the predicted value of each micro time series group in the second time interval.

[0079] Furthermore, if the influence of the covariate vector is considered during prediction, it is considered that the multiple determined micro time series are also affected by the corresponding covariate vector. As mentioned earlier, the micro time series is the numerical performance of the measured subject (a micro individual) on a certain parameter. Therefore, in essence, the measured subject itself is the factor that most affects the micro time series, and the remaining factors are secondary factors, especially the effects that may change over time are not stable. Taking this into consideration, in order to focus more on the measured subject itself, the influence of the covariate vector can be eliminated and then the micro time series can be clustered. Specifically, for example, the sequence characteristic vectors of multiple micro time series can be determined (for example, reflecting the shape of the sequence curve, the numerical scale, etc.), and the sequence characteristic vector can be corrected according to the sub-vector of the covariate vector corresponding to the sequence characteristic vector in the first time interval to obtain the stable state vector of the measured subject corresponding to the micro time series (taking commodities as an example, its stable state is determined by some inherent characteristics of the commodities themselves, such as function, selling price, cost, type, generation, etc.). According to the stable state vector, multiple micro time series can be clustered. This will help to more accurately discover similarity or homogeneity relationships.

[0080] In one or more embodiments of the present specification, if the macro time series is aggregated by multiple micro time series, the predicted value of the macro time series is obtained by aggregating the predicted value of the micro time series. Specifically, in the case of at least partially applying the active intervention factor, if there is a corresponding incremental value in the micro time series group, the natural predicted value of the micro time series group and the corresponding incremental value are aggregated to obtain the predicted value of the micro time series group after the effect (that is, the active intervention factor is applied) as the predicted value to be aggregated of the micro time series group. If there is no corresponding incremental value in the micro time series group, the natural predicted value of the micro time series group is used as the predicted value to be aggregated of the micro time series group, and the predicted values ​​to be aggregated of these multiple micro time series groups are aggregated to predict the predicted value after the effect of the macro time series. It can be seen that this prediction method of processing separately and then aggregating after processing takes into account both accuracy and efficiency, meets the actual needs of the industry, and can also better support the industry to flexibly intervene in policies or means.

[0081] Figure 2A schematic diagram of the framework of a resource allocation system provided for one or more embodiments of this specification. The system uses a distributed data warehouse as the storage infrastructure of its related data sources, and the storage infrastructure includes macro time series, micro time series, and covariate vectors for time series, etc. In this system, components with different functions are modularized into three core modules, namely, a clustering module, a natural prediction module, and an effect module, as well as several auxiliary modules such as a preprocessing module. In order to facilitate user operation, a graphical user interface is provided for the components involved in these modules, based on which users can complete time series prediction tasks by simply integrating different components according to their own needs. These three core modules are introduced in detail.

[0082] For the clustering module:

[0083] Assume z i,1:T =[z i,1 ,...,z i,T ] is the microscopic time series of microscopic individual i, represents the value of microscopic individual i at time t. The corresponding macroscopic time series is denoted as z 1:T =[z1,...,z T ], which is aggregated from m micro time series. For example, the value of the macro time series at time t is The prediction of macro time series is essentially the prediction of its future values ​​at one or more time points.

[0084] The clustering module can learn homogeneous micro-time series groups. Its goal is to cluster m micro-time series into K homogeneous micro-time series groups, denoted as kth i After the micro time series in a homogeneous micro time series group are aggregated, they are recorded as Can be used represents the corresponding micro time series group, where k i ∈{1,...,K}, K≤m is a hyperparameter that can be customized or learned based on business experience.

[0085] The system includes a variety of clustering algorithms, such as MixSeq and MixARMA. For example, MixSeq assumes that the micro time series in the homogeneous group obey the same probability distribution, while the time series from different groups obey different probability distributions. By introducing an end-to-end mixed model, the optimal clustering is studied from the perspective of resource allocation. The clustered components come from a series of sequence-to-sequence models. Under the assumption of this mixed model, the variance of the mixture is not less than the expected value of the variance of the individual components. This means that after predicting each micro time series separately, the predicted values ​​are aggregated, which is more accurate than directly predicting the macro time series.

[0086] For the natural prediction module:

[0087] Assume x t For z t The corresponding covariate vector. t The prediction is specifically based on the given historical data z 1,T and the covariate vector x 1:T+δ In the case of future data z T+1,T+δ That is, the learning model p(z T+1:T+δ |z 1:T ,x 1,T+δ ; Θ), Θ represents some other parameters used in the model.

[0088] By considering the corresponding micro time series, we can improve the T+1,T+δ For example, a corresponding model is trained for each micro time series group, denoted as express The corresponding covariate vector.

[0089] The system includes a variety of prediction models, such as Deepar, TGCN, Transformer, etc. Aggregating the natural prediction values ​​of each micro time series group can obtain the prediction value of the macro time series, for example,

[0090] For effect modules:

[0091] Assuming τ i ∈{0,1} to indicate whether the microscopic time series of microscopic individual i is subject to active intervention factors, τ i =1 means that τ is applied. i =0 means no application. Δz i,T+1:T+δ =[Δz i,T+1 ,...,Δz i,T+δ ] represents the effect (i.e. the above-mentioned increment value) on the microscopic time series of i due to the active intervention factor applied to i from time T from time T to time T+1. Then, the effect of the macroscopic time series is The goal of effect estimation is to transform That is, the effect caused by the active intervention factor applied to the kth micro time series group.

[0092] The system includes a variety of effect estimation models that can be used to learn the above effects, such as tree-based CABTree, neural network-based AutoSeqs Effect, and meta-algorithm MetaDeep Effect.

[0093] Combined with the natural prediction results, the overall prediction results after applying the active intervention factor can be obtained, that is:

[0094] in, is the number of micro time series to which active intervention factors are applied in the kth micro time series group.

[0095] Based on the above description, one or more embodiments of this specification further provide: Figure 1 The specific implementation flow chart of the method is as follows: Figure 3 As shown, the process is, for example, Figure 2 to be executed by the system in .

[0096] exist Figure 3 In the process, it is assumed that the goal is to predict the future sales trend of retailers (i.e., macro time series). In terms of active intervention factors, there are two situations: not issuing discount coupons for goods, and issuing discount coupons for goods.

[0097] First, obtain the micro time series corresponding to the retailer's sales time series, specifically the historical sales trends of each product under the retailer, assuming a total of 10 million products.

[0098] According to the historical sales trends of these commodities, the historical sales trends of these 10 million commodities are clustered and grouped into 10 groups (i.e. the above-mentioned micro time series groups) through the designated functional components of the time series clustering module. The time series within each group are aggregated respectively, and 10 corresponding grouped and aggregated time series are obtained.

[0099] The 10 grouped time series are processed by the specified functional components of the natural prediction module to obtain the future sales of the 10 grouped time series under the natural state (at least satisfying the condition that no discount coupons are issued), that is, the natural prediction value corresponding to each group.

[0100] The natural prediction values ​​of the 10 grouped time series are added together to obtain the natural prediction value of the macro time series, which is the natural sales trend of the retailer.

[0101] The 10 grouped time series are processed through the specified functional components of the effect estimation module to obtain the future sales increment values ​​of the 10 grouped time series under the influence of discount coupons.

[0102] The natural prediction value and incremental value of each grouping time series are added together to obtain the post-effect prediction value of each grouping time series, that is, the future sales volume of each grouping time series under the influence of discount coupons.

[0103] The future sales volume of these 10 grouped time series under the influence of discount coupons is added up to predict the future sales volume of the retailer when issuing discount coupons.

[0104] For the convenience of user use, a graphical user interface including multiple optional features and multiple optional components can be displayed. In response to user operations, features are selected from the multiple optional features for the above-mentioned clustering, and components are selected from the multiple optional components for the above-mentioned clustering, natural prediction and effect estimation, respectively. A workflow is formed according to the selected components, and the workflow is executed to predict the predicted value of the macro time series. In this way, users can flexibly select algorithms and models to form a workflow by selecting components to perform their desired time series prediction tasks.

[0105] More intuitively, one or more embodiments of this specification also provide: Figure 2 The schematic diagram of the user interaction interface provided by the system is shown in Figure 4(a), Figure 4(b), and Figure 4(c). Still taking the prediction of retailer sales as an example, the retailer sales is a macro time series, and the sales of retail goods is the corresponding micro time series.

[0106] First, clustering is required to group the micro time series corresponding to commodities with similar sales trends (e.g., patterns, scales) into the same group. Figure 4(a) shows the interface for the micro time series clustering task. The entire workflow starts with the data component, which loads samples of commodity sales time series in the data warehouse. The MixSeq component in the clustering module (other components for clustering can also be selected) performs model training and outputs the trained model to the inference component. The inference component infers the clustering results of each micro time series. When the user clicks on the right side of the component, the corresponding configuration form interface is provided, and the user can set it as needed, such as selecting sample features and model output paths. Through the processing of the clustering module, the commodity sales time series is divided into K groups.

[0107] Figure 4(b) shows how the natural prediction module performs natural prediction on time series in combination with the clustering module (shown briefly). For each group obtained by clustering, the time series in each group are aggregated and preprocessed through the data preprocessing component, and the time series are trained through the Transformer component of the natural prediction module. The inference component uses the trained model to perform natural prediction on each group to obtain the natural prediction values ​​of the K groups of commodity sales time series. The data used to train the model is input into the Transformer component in the workflow, and the model and the data used for prediction are input into the inference component in the workflow.

[0108] Figure 4(c) shows how the effect module estimates the effect of the time series in combination with the clustering module and the natural prediction module (shown briefly). The clustering results are input into the natural prediction module and the effect module respectively. The AutoSeqs Effect component is selected as an example in the effect module. The processing method also includes the AutoSeqs Effect component for model training and output model. The inference component uses the model for inference to obtain the effect estimation result. The natural prediction results and the effect estimation results are aggregated to obtain the final prediction result, that is, the future trend of the retailer's sales volume.

[0109] Based on the same idea, one or more embodiments of this specification also provide devices and apparatuses corresponding to the above methods, such as Figure 5 , Figure 6 shown.

[0110] Figure 5 A schematic diagram of a resource allocation device provided in one or more embodiments of this specification, the device comprising:

[0111] A micro-determination module 502 determines a plurality of micro-time series corresponding to the macro-time series to be predicted;

[0112] A time series clustering module 504 clusters the plurality of micro time series to obtain a plurality of micro time series groups;

[0113] A first prediction module 506 predicts each of the micro time series groups to obtain a predicted value of each of the micro time series groups;

[0114] A second prediction module 508 predicts the predicted value of the macro time series according to the predicted values ​​of each of the micro time series groups;

[0115] The resource allocation module 510 allocates the available resources to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0116] Optionally, the available resources include business processing resources;

[0117] The micro-determination module 502 determines the user service business that can be executed using the business processing resources;

[0118] A service attribute reflecting the concurrency degree and / or accumulation degree of the user service service is determined as a service attribute related to the available resources.

[0119] Optionally, the user service business includes an online commodity sales business, and the business processing resources include distributed computing resources.

[0120] Optionally, the second prediction module 508 includes an effect estimation module 5082;

[0121] The effect estimation module 5082 determines the active intervention factor to be applied to the measured subject of the macro time series;

[0122] According to the active intervention factor, the effect of the micro time series group is estimated to estimate the incremental value that the active intervention factor can bring to the micro time series group;

[0123] The second prediction module 508 predicts the predicted value of the macro time series according to the incremental value and the predicted value of the micro time series group.

[0124] Optionally, the first prediction module 506 includes a natural prediction module 5062;

[0125] The natural prediction module 5062 determines a natural prediction state, in which at least the active intervention factor is not applied;

[0126] In the natural prediction state, each of the micro time series groups is predicted respectively to obtain the predicted value of each of the micro time series groups, which is used as the natural prediction value to predict the predicted value of the macro time series.

[0127] Optionally, the first prediction module 506 determines a time interval corresponding to each of the micro time series groups as a first time interval, and determines an overall time interval consisting of the first time interval and a second time interval after the first time interval.

[0128] Determine the covariate vector corresponding to each of the micro time series groups in the overall time interval;

[0129] Prediction is performed based on each of the micro time series groups and the corresponding covariate vector to obtain a predicted value of each of the micro time series groups in the second time interval.

[0130] Optionally, the time series clustering module 504 determines the sequence feature vectors of the plurality of micro time series respectively;

[0131] According to the subvector of the covariate vector corresponding to the sequence feature vector in the first time interval, the sequence feature vector is corrected to obtain a stable state vector of the measured subject corresponding to the microscopic time series;

[0132] The plurality of micro time series are clustered according to the stable state vector.

[0133] Optionally, the macro time series is aggregated from the multiple micro time series;

[0134] The effect estimation module 5082 determines the active intervention factor applied to the measured subjects of one or more of the micro time series respectively, so as to perform the effect estimation on the corresponding micro time series according to the active intervention factor.

[0135] Optionally, the first prediction module 506 selects one or more micro time series in the micro time series group as intervention sequences, before the effect estimation module 5082 performs effect estimation on the micro time series group according to the active intervention factor, to apply the active intervention factor to the measured subject of the intervention sequence;

[0136] Treating other microscopic time series in the microscopic time series group as non-intervention series so as not to apply the active intervention factor to the measured subject of the non-intervention series;

[0137] A counterfactual control group is constructed based on the intervention sequence and the non-intervention sequence so that the effect estimation module 5082 can perform the effect estimation.

[0138] Optionally, the second prediction module 508, if the micro time series group has a corresponding incremental value, aggregates the natural prediction value and the corresponding incremental value of the micro time series group to obtain a post-effect prediction value of the micro time series group as the prediction value to be aggregated of the micro time series group;

[0139] If there is no corresponding incremental value for the micro time series group, the natural prediction value of the micro time series group is used as the prediction value to be aggregated for the micro time series group;

[0140] The predicted values ​​to be aggregated of the multiple micro time series groups are aggregated to predict the post-effect predicted value of the macro time series.

[0141] Optionally, a workflow processing module 512 is also included;

[0142] The workflow processing module 512 displays a graphical user interface including a plurality of optional features and a plurality of optional components;

[0143] In response to a user operation, selecting a feature from the plurality of optional features for use in the clustering;

[0144] In response to a user operation, components for the clustering, the natural prediction, and the effect estimation are selected from the plurality of optional components, and a workflow is constructed according to the selected components;

[0145] The workflow is executed to predict the predicted value of the macro time series.

[0146] Optionally, the micro time series is a time series of individual sales data of a specified commodity as the measured subject, the macro time series is a time series of overall sales data of all commodities within a specified range, and the active intervention factor is a specified commodity discount strategy.

[0147] Figure 6 A schematic diagram of a resource allocation device provided for one or more embodiments of this specification, the device comprising:

[0148] at least one processor; and,

[0149] a memory communicatively connected to the at least one processor; wherein,

[0150] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0151] Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute;

[0152] Clustering the multiple micro time series to obtain multiple micro time series groups;

[0153] Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups;

[0154] According to the predicted values ​​of each of the micro time series groups, the predicted values ​​of the macro time series are predicted;

[0155] The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0156] The processor and the memory may communicate with each other via a bus, and the device may further include an input / output interface for communicating with other devices.

[0157] Based on the same idea, one or more embodiments of this specification also provide a non-volatile computer storage medium corresponding to the above method, storing computer executable instructions, wherein the computer executable instructions are configured as follows:

[0158] Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute;

[0159] Clustering the multiple micro time series to obtain multiple micro time series groups;

[0160] Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups;

[0161] According to the predicted values ​​of each of the micro time series groups, the predicted values ​​of the macro time series are predicted;

[0162] The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

[0163] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0164] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0165] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0166] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0167] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification may be in the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the embodiments of this specification may be in the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0168] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented 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 processor, 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 generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0169] These computer program instructions may also be stored in a computer-readable memory capable of directing 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 an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0170] 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, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0171] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0172] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0173] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0174] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0175] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0176] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0177] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0178] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A resource allocation method, comprising: Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute; Clustering the multiple micro time series to obtain multiple micro time series groups; Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups; According to the predicted values ​​of each of the micro time series groups, the predicted value of the macro time series is predicted, specifically including: determining the active intervention factor to be applied to the measured subject of the macro time series, and estimating the effect of the micro time series group according to the active intervention factor to estimate the incremental value that the active intervention factor can bring to the micro time series group, and predicting the predicted value of the macro time series according to the incremental value and the predicted value of the micro time series group; The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

2. The method of claim 1, wherein the available resources include business processing resources; The determining of the service attributes related to the available resources specifically includes: Determining a user service business that can be executed using the business processing resources; A service attribute reflecting the concurrency degree and / or accumulation degree of the user service service is determined as a service attribute related to the available resources.

3. According to the method as claimed in claim 2, the user service business includes an online commodity sales business, and the business processing resources include distributed computing resources.

4. The method according to claim 1, wherein the predicting of each of the micro time series groups to obtain the predicted value of each of the micro time series groups specifically comprises: determining a natural prediction state, in which at least the active intervention factor is not applied; In the natural prediction state, each of the micro time series groups is predicted respectively to obtain the predicted value of each of the micro time series groups, which is used as the natural prediction value to predict the predicted value of the macro time series.

5. The method according to claim 1, wherein the predicting of each of the micro time series groups to obtain the predicted value of each of the micro time series groups specifically comprises: Determine a time interval corresponding to each of the micro time series groups as a first time interval, and determine an overall time interval consisting of the first time interval and a second time interval after the first time interval, Determine the covariate vector corresponding to each of the micro time series groups in the overall time interval; Prediction is performed based on each of the micro time series groups and the corresponding covariate vector to obtain a predicted value of each of the micro time series groups in the second time interval.

6. The method according to claim 5, wherein clustering the plurality of micro time series comprises: Determine the sequence feature vectors of each of the plurality of micro time series; According to the subvector of the covariate vector corresponding to the sequence feature vector in the first time interval, the sequence feature vector is modified to obtain a stable state vector of the measured subject corresponding to the microscopic time series; The plurality of micro time series are clustered according to the stable state vector.

7. The method according to claim 1, wherein the macro time series is aggregated from the multiple micro time series; The determining of the active intervention factor to be applied to the measured subject of the macro time series specifically includes: Determine the active intervention factor applied to the measured subject of one or more of the micro time series respectively, so as to estimate the effect of the corresponding micro time series according to the active intervention factor.

8. The method according to claim 1, before estimating the effect of the micro time series group according to the active intervention factor, the method further comprises: Selecting one or more micro time series in the micro time series group as intervention sequences, so as to apply the active intervention factor to the measured subject of the intervention sequence; Treating other microscopic time series in the microscopic time series group as non-intervention series so as not to apply the active intervention factor to the measured subject of the non-intervention series; A counterfactual control group is constructed based on the intervention sequence and the non-intervention sequence to perform the effect estimation.

9. The method according to claim 4, wherein the predicted value of the macro time series is predicted based on the incremental value and the predicted value of the micro time series group, specifically comprising: If the micro time series group has a corresponding incremental value, the natural prediction value of the micro time series group and the corresponding incremental value are aggregated to obtain the post-effect prediction value of the micro time series group as the prediction value to be aggregated of the micro time series group; If there is no corresponding incremental value for the micro time series group, the natural prediction value of the micro time series group is used as the prediction value to be aggregated for the micro time series group; The predicted values ​​to be aggregated of the multiple micro time series groups are aggregated to predict the post-effect predicted value of the macro time series.

10. The method according to claim 4, before predicting and obtaining the predicted value of the macro time series, the method further comprises: Presenting a graphical user interface including a plurality of optional features and a plurality of optional components; In response to a user operation, selecting a feature from the plurality of optional features for use in the clustering; In response to a user operation, components for the clustering, the natural prediction, and the effect estimation are selected from the plurality of optional components, and a workflow is constructed according to the selected components; The workflow is executed to predict the predicted value of the macro time series.

11. The method according to any one of claims 1 to 10, wherein the micro time series is a time series of individual sales data of a specified commodity as a measured subject, the macro time series is a time series of overall sales data of all commodities within a specified range, and the active intervention factor is a specified commodity discount strategy.

12. A resource allocation device, comprising: A micro-determination module determines the business attributes related to the available resources, and determines a plurality of micro-time series corresponding to the macro-time series to be predicted for describing the change of the business attributes; A time series clustering module clusters the multiple micro time series to obtain multiple micro time series groups; A first prediction module predicts each of the micro time series groups to obtain a prediction value of each of the micro time series groups; The second prediction module predicts the predicted value of the macro time series according to the predicted value of each of the micro time series groups, specifically including: determining the active intervention factor to be applied to the measured subject of the macro time series, and estimating the effect of the micro time series group according to the active intervention factor to estimate the incremental value that the active intervention factor can bring to the micro time series group, and predicting the predicted value of the macro time series according to the incremental value and the predicted value of the micro time series group; A resource allocation module allocates the available resources to the business corresponding to the business attribute according to the predicted value of the macro time series.

13. A resource allocation device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Determine a business attribute related to the available resources, and determine a plurality of micro time series corresponding to a macro time series to be predicted for describing a change in the business attribute; Clustering the multiple micro time series to obtain multiple micro time series groups; Predicting each of the micro time series groups respectively to obtain predicted values ​​of each of the micro time series groups; According to the predicted values ​​of each of the micro time series groups, the predicted value of the macro time series is predicted, specifically including: determining the active intervention factor to be applied to the measured subject of the macro time series, and estimating the effect of the micro time series group according to the active intervention factor to estimate the incremental value that the active intervention factor can bring to the micro time series group, and predicting the predicted value of the macro time series according to the incremental value and the predicted value of the micro time series group; The available resources are allocated to the business corresponding to the business attribute according to the predicted value of the macro time series.

Images