Data processing method and device, computer equipment, storage medium and program product

Abstract

The embodiment of the invention provides a data processing method and device, computer equipment, a storage medium and a program product. The data processing method comprises the steps of obtaining a historical business data sequence of a business, wherein the historical business data sequence comprises a plurality of business data of the business arranged according to time steps in a historical time range; fluctuation analysis is carried out on the historical business data sequence to obtain fluctuation characteristics of the business, and the fluctuation characteristics are used for representing the fluctuation of business data of the business along with the change of the time step; dependency analysis is carried out on the historical service data sequence to obtain dependency features of the service, and the dependency features are used for representing the dependency relationship between the service data of different time steps in the service; according to the fluctuation characteristics and the dependency characteristics, constructing prediction characteristics of the business; and performing prediction processing on the service according to the prediction features to obtain service data of the service in the future time step. By adopting the embodiment of the invention, the service data of the service in the future time step can be accurately predicted.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, a data processing apparatus, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In daily life, there are many businesses that are affected by time changes, such as financial business, product trading business, and multimedia on-demand business, etc. The business data of these businesses will change with time. These businesses affected by time changes usually have the need to predict future business data. Predicting the future business data of the business helps the business party to grasp the future development direction of the business and formulate relevant strategies for the future execution of the business. It can be seen that predicting future business data is of great significance to the smooth and smooth execution of the business. At present, there is an urgent need for a solution that can more accurately predict the future business data of the business. Summary of the invention

[0003] The embodiments of the present application provide a data processing method, apparatus, computer equipment, storage medium, and program product, which can more accurately predict the business data of a business at a future time step (i.e., can more accurately predict the future business data of a business).

[0004] On the one hand, an embodiment of the present application provides a data processing method, the data processing method comprising:

[0005] Acquire a historical business data sequence of a business, where the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range;

[0006] Perform fluctuation analysis on the historical business data sequence to obtain the business fluctuation characteristics. The fluctuation characteristics are used to characterize the volatility of the business data as the time step changes.

[0007] Perform dependency analysis on the historical business data sequence to obtain the business dependency features, which are used to characterize the dependency relationship between business data at different time steps in the business;

[0008] Build predictive features of the business based on volatility and dependency features;

[0009] The business is predicted based on the prediction features to obtain the business data of the business in the future time step.

[0010] Accordingly, an embodiment of the present application provides a data processing device, the data processing device comprising:

[0011] An acquisition unit, used to acquire a historical business data sequence of a business, where the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range;

[0012] A processing unit, used to perform fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business, and the fluctuation characteristics are used to characterize the volatility of the business data of the business as the time step changes;

[0013] A processing unit is used to perform dependency analysis on the historical business data sequence to obtain the dependency characteristics of the business, and the dependency characteristics are used to characterize the dependency relationship between business data at different time steps in the business;

[0014] The processing unit is also used to construct prediction features of the business based on the fluctuation features and the dependency features;

[0015] The processing unit is also used to perform prediction processing on the business according to the prediction characteristics to obtain the business data of the business in the future time step.

[0016] In one implementation, the processing unit is used to perform fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business, and is specifically used to perform the following steps:

[0017] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0018] Call the target fluctuation analysis model to predict business fluctuations based on the fluctuation data sequence, and obtain the fluctuation data of the business in the future time step;

[0019] The fluctuation data of the business in the future time step is determined as the fluctuation feature; the fluctuation feature is used to characterize the future volatility of the business data of the business as it changes with the time step.

[0020] In one implementation, the processing unit is used to perform fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business, and is specifically used to perform the following steps:

[0021] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0022] According to the preset time step, the fluctuation data sequence is split into multiple fluctuation data subsequences;

[0023] Determine observed fluctuation data and sample fluctuation data in each fluctuation data subsequence;

[0024] The target fluctuation analysis model is called to predict the predicted fluctuation data corresponding to the observed fluctuation data in each fluctuation data subsequence based on the sample fluctuation data in each fluctuation data subsequence;

[0025] Determine the volatility residual information corresponding to each volatility data subsequence according to the difference between the observed volatility data and the corresponding predicted volatility data in each volatility data subsequence;

[0026] The fluctuation residual information corresponding to each fluctuation data subsequence is determined as the fluctuation feature; the fluctuation feature is used to characterize the abnormal fluctuation caused by the business data of the business changing with the time step.

[0027] In one implementation, the processing unit is further configured to perform the following steps:

[0028] Obtaining an initial fluctuation analysis model and sample data, the sample data including training sample data and test sample data;

[0029] According to the training sample data, the model parameters of the initial fluctuation analysis model are estimated to obtain a first fluctuation analysis model;

[0030] Performing a fit check between the first fluctuation analysis model and the test sample data;

[0031] If the first fluctuation analysis model is compatible with the test sample data, the first fluctuation analysis model is determined as the target fluctuation analysis model.

[0032] In one implementation, the processing unit is further configured to perform the following steps:

[0033] If the first fluctuation analysis model is not compatible with the test sample data, a new fluctuation analysis model is obtained, and model parameters of the new fluctuation analysis model are estimated according to the training sample data to obtain a second fluctuation analysis model;

[0034] Performing a fit check between the second wave analysis model and the test sample data;

[0035] If the second fluctuation analysis model is compatible with the test sample data, the second fluctuation analysis model is determined as the target fluctuation analysis model;

[0036] If the second fluctuation analysis model does not match the test sample data, continue to update the fluctuation analysis model until the estimated fluctuation analysis model matches the test sample data, and determine the fluctuation analysis model that matches the test sample data as the target fluctuation analysis model;

[0037] The new fluctuation analysis model and the initial fluctuation analysis model are of different types; or, the new fluctuation analysis model and the initial fluctuation analysis model are of the same type, but have different levels of model complexity.

[0038] In one implementation, the processing unit is used to perform fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business, and is specifically used to perform the following steps:

[0039] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0040] According to the preset time step, the fluctuation data sequence is split into multiple fluctuation data subsequences;

[0041] Iteratively estimate the model parameters of the volatility analysis model according to multiple volatility data subsequences;

[0042] The model parameters obtained by iterative estimation are determined as fluctuation characteristics; the fluctuation characteristics are used to characterize the impact of historical volatility of business data of a business as it changes over time on future volatility.

[0043] In one implementation, the fluctuation data sequence is divided into N fluctuation data subsequences, and the iterative estimation refers to estimating the model parameters of the initial fluctuation analysis model N times, the N estimates correspond one-to-one to the N fluctuation data subsequences, and N is an integer greater than 1; the i-th estimation in the N estimates is based on the fluctuation analysis model obtained by the i-1-th estimation, and the model parameters obtained by the iterative estimation are the model parameters obtained by the N-th estimation, i is an integer greater than 1, and i is less than or equal to N;

[0044] The process of estimating the model parameters for the i-th time includes:

[0045] In the i-th fluctuation data subsequence among the N fluctuation data subsequences, the observed fluctuation data and the sample fluctuation data are determined;

[0046] The volatility analysis model obtained by the i-1th estimation is called to predict the predicted volatility data corresponding to the observed volatility data based on the sample volatility data, and the probability that the predicted volatility data is the observed volatility data is obtained;

[0047] Update the model parameters obtained by the i-1th estimation in the direction of maximizing the probability to obtain the model parameters obtained by the i-th estimation.

[0048] In one implementation, the historical business data sequence includes business data of T time steps within a historical time range, where T is an integer greater than 1; the dependency analysis is performed by calling a target dependency analysis model, where the target dependency analysis model includes T dependency analysis units, and the T dependency analysis units correspond to the business data of T time steps one by one; the processing unit is used to perform dependency analysis on the historical business data sequence, and when obtaining the dependency characteristics of the business, it is specifically used to execute the following steps:

[0049] Call each dependency analysis unit to perform dependency analysis on the business data of the corresponding time step and the business data of the previous time step of the corresponding time step, and obtain the dependency analysis result of each dependency analysis unit;

[0050] The dependency analysis result of the Tth dependency analysis unit among T dependency analysis units is determined as a dependency feature, and the dependency feature is used to characterize the dependency of the business data of the Tth time step on the business data of the preceding T-1 time steps of the Tth time step; or, the statistical result of the dependency analysis results of the T dependency analysis units is determined as the dependency feature, and the dependency feature is used to characterize the comprehensive dependency of the business data of each time step among the T time steps on the business data of the preceding time step of each time step.

[0051] In one implementation, the j-th dependency analysis unit among the T dependency analysis units is used to perform dependency analysis on the business data of the j-th time step and the business data of the j-1th time step preceding the j-th time step, to obtain the dependency analysis result of the j-th dependency analysis unit; the features of the business data of the j-1th time step preceding the j-1th dependency analysis unit are retained in the state information of the j-1th dependency analysis unit, and the state information of the j-1th dependency analysis unit and the dependency analysis result of the j-1th dependency analysis unit are input into the j-th dependency analysis unit; j is an integer greater than 1, and j is less than or equal to T;

[0052] The dependency analysis process of the j-th dependency analysis unit includes:

[0053] According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, forgetting analysis is performed on the state information of the j-1th dependency analysis unit to determine the first state information that needs to be retained in the state information of the j-1th dependency analysis unit;

[0054] Perform memory analysis on the j-th dependency analysis unit according to the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step, and determine the second state information that needs to be added to the state information of the j-th dependency analysis unit from the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step;

[0055] Combine the first state information and the second state information to obtain state information of the j-th dependency analysis unit;

[0056] According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the jth time step, the state information of the jth dependency analysis unit is output analyzed to obtain the dependency analysis result of the jth dependency analysis unit.

[0057] In one implementation, the processing unit is used to construct the prediction feature of the business according to the fluctuation feature and the dependency feature, and is specifically used to perform any of the following steps:

[0058] The fluctuation features and the dependent features are spliced ​​to obtain the prediction features;

[0059] Perform weighted averaging on the fluctuation characteristics and dependency characteristics to obtain the prediction characteristics;

[0060] Perform convolution processing on the fluctuation features and the dependent features to obtain the prediction features;

[0061] Perform feature fusion processing on the fluctuation feature and the dependent feature to obtain the prediction feature;

[0062] The fluctuation features and dependent features are cross-processed to obtain the prediction features.

[0063] In one implementation, the acquisition unit is further configured to perform the following steps:

[0064] Obtain data features of historical business data series;

[0065] The processing unit is used to construct the prediction characteristics of the business according to the fluctuation characteristics and the dependency characteristics, and is specifically used to perform the following steps:

[0066] Construct prediction features based on fluctuation features, dependency features, and data features;

[0067] Among them, the data characteristics include at least one of the following: statistical characteristics of historical business data sequences, and business indicator characteristics of historical business data sequences; statistical characteristics are used to reflect the comprehensive situation of each business data in the historical business data sequence; business indicator characteristics are used to reflect the business characteristics of the historical business data sequence under the business indicators.

[0068] In one implementation, the fluctuation feature includes a plurality of fluctuation feature data, and the dependency feature includes a plurality of dependency feature data; the processing unit is used to perform the following steps before constructing the prediction feature of the business according to the fluctuation feature and the dependency feature:

[0069] Obtaining a first contribution degree of each fluctuation feature data in the fluctuation feature to predicting the business data of the future time step, selecting the fluctuation feature data whose first contribution degree meets the contribution condition in the fluctuation feature, and obtaining the selected fluctuation feature;

[0070] Obtaining a second contribution degree of each dependent feature data in the dependent feature to predicting the business data of the future time step, selecting dependent feature data whose second contribution degree meets the contribution condition in the dependent feature, and obtaining the selected dependent feature;

[0071] The processing unit is used to construct the prediction characteristics of the business according to the fluctuation characteristics and the dependency characteristics, and is specifically used to perform the following steps:

[0072] Construct prediction features based on the post-selection fluctuation features and the post-selection dependency features.

[0073] In one implementation, the acquisition unit is used to acquire the historical business data sequence of the business, and is specifically used to perform the following steps:

[0074] Obtaining an initial business data sequence of the business;

[0075] Preprocessing the initial business data sequence to obtain a historical business data sequence;

[0076] The preprocessing includes at least one of the following: interpolating missing data in the initial business data sequence, detecting outlier data in the initial business data sequence, and performing outlier processing on the outlier data.

[0077] Accordingly, an embodiment of the present application provides a computer device, the computer device comprising:

[0078] a processor suitable for implementing a computer program;

[0079] A computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the above-mentioned data processing method.

[0080] Accordingly, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is read and executed by a processor of a computer device, the computer device executes the above-mentioned data processing method.

[0081] Accordingly, an embodiment of the present application provides a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the above-mentioned data processing method.

[0082] In an embodiment of the present application, after obtaining the historical business data sequence of the business, a fluctuation analysis can be performed on the historical business data sequence to obtain the fluctuation characteristics of the business, and a dependency analysis can be performed on the historical business data sequence to obtain the dependency characteristics of the business. Then, a prediction feature can be constructed based on the fluctuation characteristics and the dependency characteristics, and the business data of the business in the future time step can be predicted based on the prediction feature. Among them, the fluctuation analysis can analyze the historical business data sequence from the perspective of data fluctuation of the business data, and the fluctuation characteristics obtained by the fluctuation analysis can be used to characterize the volatility of the business data of the business as the time step changes; the dependency analysis can analyze the historical business data sequence from the perspective of the dependency relationship between business data at different time steps, and the dependency characteristics obtained by the dependency analysis can be used to characterize the dependency relationship between business data at different time steps in the business; by combining the fluctuation characteristics and the dependency characteristics, the business data of the business in the future time step can be predicted more accurately (that is, the future business data of the business can be predicted more accurately). BRIEF DESCRIPTION OF THE DRAWINGS

[0083] In order to more clearly illustrate the embodiments of the present application 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 of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0084] Figure 1 It is a schematic diagram of the architecture of a data processing system provided in an embodiment of the present application;

[0085] Figure 2 is a schematic diagram of future business data prediction provided by an embodiment of the present application;

[0086] Figure 3 It is a flowchart of a data processing method provided in an embodiment of the present application;

[0087] Figure 4 It is a flowchart of another data processing method provided in an embodiment of the present application;

[0088] Figure 5 It is a schematic diagram of a method for constructing a prediction feature provided in an embodiment of the present application;

[0089] Figure 6 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0090] Figure 7 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0091] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0092] In order to more clearly understand the technical solutions provided by the embodiments of the present application, the key terms involved in the embodiments of the present application are first introduced here.

[0093] (1) The embodiments of the present application involve services. The services mentioned in the embodiments of the present application may specifically refer to transactions whose execution process is affected by time changes; for example, financial services, product trading services, and multimedia on-demand services, etc., whose execution process is affected by time changes.

[0094] (2) The embodiments of the present application involve business data. The business data mentioned in the embodiments of the present application may specifically refer to the data generated during the execution of the business. The most intuitive manifestation of the impact of time changes on the execution of the business is that the business data generated during the execution of the business will change over time. For example, the business data generated during the execution of a financial business may be, for example, the price of a financial product, which may change over time; the business data generated during the execution of a product trading business may be, for example, the product trading volume, which may change over time; the business data generated during the execution of a multimedia on-demand business may be, for example, the multimedia on-demand volume, which may change over time.

[0095] Business data may include characteristic data under one or more business characteristics. For example, business data generated during the execution of a financial business may include the highest financial product price, the lowest financial product price, and the average financial product price, that is, the business data includes characteristic data under three business characteristics. For another example, business data generated during the execution of a product trading business may include product trading volume and total product trading amount, that is, the business data includes characteristic data under two business characteristics. For another example, business data generated during the execution of a multimedia on-demand business may include multimedia on-demand volume, that is, the business data includes characteristic data under one business characteristic.

[0096] (3) The embodiments of the present application relate to business data sequences. A business data sequence may specifically refer to a sequence of business data arranged in sequence according to the time sequence corresponding to a time step within a certain time range; wherein a time step may specifically refer to a statistical time unit of business data, and a time range may specifically refer to a time step, which is a time length composed of time steps. For example, the product transaction volume of a product transaction business in 7 days of a week is 100 pieces on Monday, 200 pieces on Tuesday, 128 pieces on Wednesday, 67 pieces on Thursday, 69 pieces on Friday, 154 pieces on Saturday, and 132 pieces on Sunday. Then, within the time range (a week), the business data is arranged in sequence according to the time sequence corresponding to the time step (day), and the business data sequence is [100, 200, 128, 67, 69, 154, 132].

[0097] Based on the above introduction to business, business data and business data sequence, in the execution process of business affected by time changes, there is usually a need to predict future business data; usually, the historical business data sequence of the business can be analyzed to predict the future business data of the business; historical business data refers to a sequence of business data arranged in the order of time corresponding to the time step within the historical time range, and future business data refers to the business data of the future time step. On this basis, the embodiment of the present application proposes a data processing method, which can accurately predict the future business data of the business based on the historical business data sequence of the business. Specifically, the data processing method analyzes the historical business data sequence from the data fluctuation angle of the business data and the data dependency angle of the business data; the historical business data sequence is analyzed from the data fluctuation angle of the business data, and the fluctuation characteristics obtained by the analysis can be used to characterize the volatility of the business data of the business with the change of the time step; the historical business data sequence is analyzed from the data dependency angle of the business data, and the dependency characteristics obtained by the analysis can be used to characterize the dependency relationship between the business data at different time steps in the business; thus, combining the fluctuation characteristics and the dependency characteristics to predict the future business data of the business can accurately predict the future business data of the business.

[0098] Taking financial business as an example, based on the data processing method provided in the embodiment of the present application, the historical financial product price series of the financial business in the past month can be analyzed from the perspective of data fluctuation of the business data and the perspective of data dependence of the business data, so as to more accurately predict the future financial product price of the financial business; the future financial product price of the financial business can be, for example, the financial product price of the next day, the financial product price of the next week, or the financial product price of the next month, that is, the embodiment of the present application does not limit the time step of the future time step, and the future time step can be one or more time steps in the future; in addition, the future financial product price of the financial business can be, for example, the specific numerical value of the financial product price of the future time step, or can be the price range of the financial product price of the future time step, and the embodiment of the present application does not limit the data format of the business data of the future time step.

[0099] Combine the following Figure 1 , the data processing system provided in the embodiment of the present application is introduced, and the data processing system is suitable for implementing the data processing method provided in the embodiment of the present application.

[0100] like Figure 1 As shown, the data processing system may include a data acquisition device 101 and a data processing device 102. The embodiment of the present application does not limit the connection method between the data acquisition device 101 and the data processing device 102. A direct communication connection can be established between the data acquisition device 101 and the data processing device 102 through wired communication, or an indirect communication connection can be established between the data acquisition device 101 and the data processing device 102 through wireless communication.

[0101] The data acquisition device 101 can be a terminal or a server, and the data processing device 102 can be a terminal or a server. The terminal mentioned in the embodiment of the present application may include but is not limited to any of the following: a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart watch, a smart home appliance, a smart car terminal and an aircraft, but is not limited to this. The server mentioned in the embodiment of the present application can be a separate physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, etc. The embodiment of the present application does not limit this.

[0102] In the data processing system composed of the data acquisition device 101 and the data processing device 102: the data acquisition device 101 can be used to collect the historical business data sequence of the business, and send the collected historical business data sequence to the data processing device 102. The data processing device 102 can be used to analyze the historical business data sequence and predict the future business data of the business. It should be noted that the process of the data acquisition device 101 collecting the historical business data sequence of the business and the data processing device 102 predicting the future business data of the business based on the historical business data sequence is cyclical and iterative, and the cyclical and iterative process is referred to as Figure 2 It is understood that after the data processing method is started, the historical business data sequence of the business is collected within the historical time range (for example, the past 1-30 days), and the business data of the business in the future time step (for example, the 31st day) is predicted based on the collected historical business data sequence. As time goes by, the historical time range will change, and the future time step will also change. It is necessary to collect the historical business data sequence of the business within the changed historical time range (for example, the past 2-31 days), and predict the business data of the business in the future time step (32nd day) after the change based on the collected historical business data sequence. This cycle is repeated continuously until the data processing method stops.

[0103] A fluctuation analysis model, a dependency analysis model and a data prediction model may be deployed in the data processing device 102; wherein the fluctuation analysis model may be used to analyze the historical business data sequence of the business from the perspective of data fluctuation of the business data, the dependency analysis model may be used to analyze the historical business data of the business from the perspective of data dependency of the business data, and the prediction model may be used to combine the fluctuation characteristics analyzed by the fluctuation analysis model and the dependency characteristics analyzed by the dependency analysis model to predict the business data of the business in future time steps.

[0104] Among them, the volatility analysis model refers to a model that can analyze the volatility of business data as it changes with time steps. For example, the volatility analysis model can be GARCH (Generalized Autoregressive Conditional Heteroskedasticity Model) or ARCH (Autoregressive Conditional Heteroskedasticity Model). The dependency analysis model refers to a model that can analyze the dependency relationship between business data that changes with time steps. For example, the dependency analysis model can be an LSTM (Long Short-Term Memory) model or an RNN (Recurrent Neural Network). The data prediction model refers to a model that can combine the volatility of business data as it changes with time steps and the dependency relationship between business data that changes with time steps to predict the business data of the business in the future time steps; when selecting a data prediction model, the data characteristics of the business data of the business that change with time steps can be considered, and a data prediction model that matches the data characteristics of the business data of the business that change with time steps can be selected. For example, the data prediction model can be any of the following machine learning models in the field of artificial intelligence: linear model, decision tree model and neural network model.

[0105] Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Machine learning (ML) is a multi-disciplinary cross-disciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all fields of artificial intelligence. Deep learning (DL) is a new research direction in the field of machine learning. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning by teaching.

[0106] exist Figure 1In the illustrated embodiment, the data acquisition device 101 and the data processing device 102 are separately configured as an example for introduction, that is, the data acquisition device 101 and the data processing device 102 are integrated into different devices, and the separately configured data acquisition device 101 and the data processing device 102 can collaboratively execute the data processing method; optionally, the data acquisition device 101 and the data processing device 102 can also be integrated into the same device, for example, the data acquisition device 101 and the data processing device 102 are integrated into the same terminal, and the data acquisition device 101 and the data processing device 102 are integrated into the same server. The embodiment of the present application does not limit the configuration of the data acquisition device 101 and the data processing device 102. It can be understood that, Figure 1 The data processing system of the illustrated embodiment is intended to more clearly illustrate the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided by the embodiment of the present application. A person of ordinary skill in the art can appreciate that with the evolution of the system architecture and the emergence of new business scenarios, the technical solution provided by the embodiment of the present application is equally applicable to similar technical problems.

[0107] The data processing method provided in the embodiment of the present application is described in more detail below with reference to the accompanying drawings.

[0108] The present application embodiment provides a data processing method, which introduces a fluctuation analysis process of a historical business data sequence and a dependency analysis process of a historical business data sequence. The data processing method can be executed by a computer device, which can be, for example, the above-mentioned Figure 1 The data processing device 102 in the data processing system shown in the figure may be, or may be, the data processing device 102 in the data processing system shown in the figure. Figure 1 The data processing system shown in FIG. 1 is an integrated device of a data acquisition device 101 and a data processing device 102. Figure 3 As shown, the data processing method may include but is not limited to the following steps S301 to S305:

[0109] S301, obtaining a historical business data sequence of a business, where the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range.

[0110] In step S201, the historical business data sequence of the business can be obtained by preprocessing the initial business data sequence of the business, that is, the initial business data sequence of the business can be obtained, and the initial business data sequence can include multiple initial business data arranged according to time steps within the historical time range. The initial business data sequence can be preprocessed to obtain the historical business data sequence.

[0111] Preprocessing may include but is not limited to at least one of the following: (1) interpolating missing data in the initial business data sequence; (2) detecting outlier data in the initial business data sequence and performing outlier processing on the outlier data. The following are introductions to these two preprocessing methods:

[0112] (1) Interpolation processing is performed on the missing data in the initial business data sequence: When one or more time steps in the initial business time sequence do not have corresponding initial business data, the initial business data corresponding to the time steps of these missing data can be considered as missing data. Interpolation processing can be performed on the missing data in the initial business data sequence. Interpolation processing refers to finding interpolation data that is as close as possible to the missing data as the missing data. Interpolation processing can include but is not limited to any of the following: linear interpolation and model interpolation; wherein linear interpolation refers to generating a linear function based on the mapping relationship between other time steps and the initial business data except the time step of the missing data, and then the time step of the missing data can be used as the input information of the linear function, the input information is calculated based on the linear function, and the output information of the linear function is used as the interpolation data of the missing data; model interpolation refers to establishing an interpolation model, using the initial business data corresponding to other time steps in the initial business data sequence except the time step of the missing data as the input information of the interpolation model, calling the interpolation model to predict the time step of the missing data based on the initial business data corresponding to other time steps, and obtaining the interpolation data corresponding to the time step of the missing data.

[0113] (2) Detecting outlier data in the initial business data sequence and performing outlier processing on the outlier data: Outlier data refers to initial business data in the initial business data sequence that is significantly different from other initial business data. Outlier processing can be performed on the outlier data in the initial business data sequence. The outlier processing can include any of the following: deletion processing and interpolation processing; wherein, deletion processing refers to deleting the outlier data in the initial business data sequence; interpolation processing refers to finding interpolation data with a smaller difference from other initial business data in the initial business data sequence to replace the outlier data. The interpolation processing of outlier data is similar to the interpolation processing of missing data mentioned above. For details, please refer to the interpolation processing of missing data mentioned above, which will not be described in detail here.

[0114] By preprocessing the initial business data sequence, the interpolation processing in the preprocessing can make each time step in the historical business data sequence obtained by preprocessing correspond to the business data, maintain the data integrity in the historical business data sequence, and improve the accuracy of subsequent future business data predictions; the outlier processing in the preprocessing can make the business data of each time step in the historical business data sequence relatively concentrated, avoid outlier data from misleading the subsequent prediction process, and further improve the accuracy of subsequent future business data predictions.

[0115] S302, performing fluctuation analysis on the historical business data sequence to obtain fluctuation characteristics of the business, where the fluctuation characteristics are used to characterize the fluctuation of the business data of the business as the time step changes.

[0116] In step S202, after acquiring the historical business data sequence, a fluctuation analysis may be performed on the historical business data sequence to obtain the fluctuation characteristics of the business. The fluctuation characteristics may be used to characterize the volatility of the business data of the business as the time step changes.

[0117] Specifically, volatility can include any of the following: (1) future volatility caused by the business data of a business changing with time steps; (2) abnormal volatility caused by the business data of a business changing with time steps; (3) the impact of historical volatility on future volatility caused by the business data of a business changing with time steps. The following introduces the volatility analysis process of the three types of volatility respectively:

[0118] (1) The volatility feature can be used to characterize the volatility of the business data of a business as it changes with time steps. Volatility refers to the future volatility of the business data of a business as it changes with time steps. In other words, the volatility feature can be used to further characterize the future volatility of the business data of a business as it changes with time steps.

[0119] In this case, the process of performing fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business may include: obtaining the fluctuation data sequence corresponding to the historical business data sequence, the fluctuation data sequence may include multiple fluctuation data of the business arranged according to time steps within the historical time range, the fluctuation data of any time step may be used to characterize the fluctuation of the business data of the corresponding time step, the fluctuation data of any time step may include the conditional variance of the business data of the corresponding time step and the square error of the business data of the corresponding time step; then, the target fluctuation analysis model may be called to perform fluctuation prediction on the business based on the fluctuation data sequence to obtain the fluctuation data of the business in the future time step, the fluctuation data of the future time step may be used to characterize the fluctuation of the business data in the future time step; the fluctuation data of the business in the future time step may be determined as the fluctuation characteristics; the fluctuation characteristics may be used to characterize the future volatility of the business data of the business as the time step changes.

[0120] Taking the target volatility analysis model as the GARCH model as an example, the GARCH model can be expressed as GARCH (p, q), and GARCH (p, q) can be expressed as the following formula 1:

[0121]

[0122] As shown in formula 1 above: represents the conditional variance at time step t, represents the conditional variance at time step tj; ∈ t-i (i=1, 2, ..., p) represents the square error of time step ti; p represents the order of the lagged square error, q represents the order of the lagged conditional variance; α i (i=1, 2, ..., p) represents the coefficient of the lagged square error; β j (j=1, 2, ..., j) represents the coefficient of the lagged conditional variance. For ease of understanding, taking the target volatility analysis model as GARCH (1, 1) as an example, GARCH (1, 1) can be expressed as the following formula 2:

[0123]

[0124] As shown in formula 2 above: represents the conditional variance at time step t, represents the conditional variance at time step t-1; ∈ t-1 represents the square error of time step t-1; α represents the coefficient of the lagged square error; β represents the coefficient of the lagged conditional variance. The target volatility analysis model is called to make volatility predictions for the business based on the volatility data sequence. Specifically, the volatility data sequence is used as the input information of the target volatility analysis model, and the target volatility analysis model is called to calculate the input information to obtain the conditional variance of time step t. The conditional variance at time step t is as the fluctuation data for future time steps.

[0125] (2) The fluctuation feature can be used to characterize the volatility of the business data of a business as it changes over time. Volatility refers to the abnormal fluctuation of the business data of a business as it changes over time. In other words, the fluctuation feature can be used to further characterize the abnormal fluctuation of the business data of a business as it changes over time.

[0126] In this case, the process of performing fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business may include: obtaining the fluctuation data sequence corresponding to the historical business data sequence, the fluctuation data sequence may include multiple fluctuation data of the business arranged according to time steps within the historical time range, the fluctuation data of any time step may be used to characterize the fluctuation of the business data of the corresponding time step, and the fluctuation data of any time step may include the conditional variance of the business data of the corresponding time step and the square error of the business data of the corresponding time step. The fluctuation data sequence may be split into multiple fluctuation data subsequences according to the preset time step, and the observed fluctuation data and the sample fluctuation data may be determined in each fluctuation data subsequence; the observed fluctuation data may be the fluctuation data of the last or last multiple time steps in the fluctuation data subsequence, and the sample fluctuation data may be the fluctuation data of other time steps except the last or last multiple time steps in the fluctuation data subsequence. Then, the target fluctuation analysis model can be called to predict the predicted fluctuation data corresponding to the observed fluctuation data in each fluctuation data subsequence based on the sample fluctuation data in each fluctuation data subsequence; specifically, the sample fluctuation data in each fluctuation data subsequence can be used as the input information of the target fluctuation analysis model, and the target fluctuation analysis model can be called to calculate the sample fluctuation data sequence to obtain the predicted fluctuation data corresponding to the observed fluctuation data. Then, the fluctuation residual information corresponding to each fluctuation data subsequence can be determined based on the difference between the observed fluctuation data and the corresponding predicted fluctuation data in each fluctuation data subsequence; the fluctuation residual information corresponding to each fluctuation data subsequence can be determined as a fluctuation feature; the fluctuation feature can be used to characterize the abnormal fluctuations caused by the business data of the business changing with the time step.

[0127] (3) The volatility feature can be used to characterize the volatility of the business data of a business as it changes over time. Volatility refers to the impact of the historical volatility of the business data of a business as it changes over time on the future volatility. In other words, the volatility feature can be used to further characterize the impact of the historical volatility of the business data of a business as it changes over time on the future volatility.

[0128] In this case, the process of performing fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business may include: obtaining the fluctuation data sequence corresponding to the historical business data sequence, the fluctuation data sequence includes multiple fluctuation data arranged according to time steps within the historical time range, the fluctuation data of any time step can be used to characterize the fluctuation of the business data of the corresponding time step, and the fluctuation data of any time step can include the conditional variance of the business data of the corresponding time step and the square error of the business data of the corresponding time step. The fluctuation data sequence can be split into multiple fluctuation data subsequences according to the preset time step. The model parameters of the fluctuation analysis model can be iteratively estimated based on the multiple fluctuation data subsequences, and the model parameters can specifically include the coefficient of the lagged conditional variance and the coefficient of the lagged square error; the coefficient of the lagged conditional variance can be used to characterize the influence of the historical conditional variance on the future conditional variance, the coefficient of the lagged square error can be used to characterize the influence of the historical square error on the future square error, and the future conditional variance and the future square error can be used to define future volatility. Then, the model parameters obtained by iterative estimation can be determined as fluctuation characteristics; the fluctuation characteristics can be used to characterize the influence of the historical volatility of the business data of the business over the time step on the future volatility.

[0129] Furthermore, iterative estimation refers to multiple estimations of the model parameters of the initial fluctuation analysis model, and the estimation method used can be the maximum likelihood estimation method. For ease of understanding, the number of iterative estimations can be expressed as N times, that is, iterative estimation refers to N estimates of the initial fluctuation analysis model, where N is an integer greater than 1; the fluctuation data sequence can be divided into N fluctuation data subsequences, and the N estimates correspond one-to-one to the N fluctuation data subsequences, that is, one fluctuation data subsequence can be used for one estimation. The following introduces the process of estimating the model parameters at any one of the N estimates (any estimate can be expressed as the i-th estimate, where i is an integer greater than 1, and i is less than or equal to N). The i-th estimate in the N estimates is based on the fluctuation analysis model obtained at the i-1-th estimate. The model parameters obtained by the iterative estimation are the model parameters obtained at the N-th estimate. The i-th process of estimating the model parameters can include:

[0130] In the i-th fluctuation data subsequence among the N fluctuation data subsequences, the observed fluctuation data and the sample fluctuation data are determined; the observed fluctuation data may be the fluctuation data of the last or last multiple time steps in the i-th fluctuation data subsequence, and the sample fluctuation data may be the fluctuation data of other time steps except the last or last multiple time steps in the i-th fluctuation data subsequence. The fluctuation analysis model obtained by the i-1th estimation may be called to predict the predicted fluctuation data corresponding to the observed fluctuation data based on the sample fluctuation data, and obtain the probability that the predicted fluctuation data is the observed fluctuation data; wherein, calling the fluctuation analysis model obtained by the i-1th estimation to predict the predicted fluctuation data corresponding to the observed fluctuation data based on the sample fluctuation data specifically means that the sample fluctuation data in the i-th fluctuation data subsequence is used as the input information of the fluctuation analysis model obtained by the i-1th estimation, and the fluctuation analysis model obtained by the i-1th estimation is called to calculate the sample fluctuation data, and obtain the probability that the predicted fluctuation data is the observed fluctuation data. Then, the model parameters obtained by the i-1th estimation may be updated in the direction of maximizing the probability, and the model parameters obtained by the i-th estimation may be obtained.

[0131] It should be noted that in the above-mentioned fluctuation analysis methods (1) and (2), the target fluctuation analysis model needs to be called for fluctuation prediction. The target fluctuation analysis model can be obtained by estimating the model parameters of the initial fluctuation analysis model and is a fluctuation analysis model that is compatible with the business data sequence of the business. Specifically, the initial fluctuation analysis model and sample data (the sample data here is specifically the first sample data) can be obtained. The first sample data can include a plurality of sample business data of the business arranged according to time steps within the historical time range. The sample data can be divided into a plurality of sample sub-data according to a preset time step. The plurality of sample sub-data can be divided into two types of sample sub-data, namely, training sample data and test sample data. That is, the sample data can include training sample data and test sample data. The model parameters of the initial fluctuation analysis model can be estimated based on the training sample data to obtain the first fluctuation analysis model. Then, the first fluctuation analysis model can be checked for compatibility with the test sample data. If the first fluctuation analysis model is compatible with the test sample data, the first fluctuation analysis model can be determined as the target fluctuation analysis model. If the first fluctuation analysis model does not match the test sample data, a new fluctuation analysis model can be obtained, and the model parameters of the new fluctuation analysis model can be estimated based on the training sample data to obtain a second fluctuation analysis model; an adaptation check can be performed between the second fluctuation analysis model and the test sample data; if the second fluctuation analysis model matches the test sample data, the second fluctuation analysis model can be determined as the target fluctuation analysis model; if the second fluctuation analysis model does not match the test sample data, the fluctuation analysis model can continue to be updated until the estimated fluctuation analysis model matches the test sample data, and the fluctuation analysis model that matches the test sample data can be determined as the target fluctuation analysis model.

[0132] In the process of estimating the model parameters of the initial fluctuation analysis model and performing the adaptation check between the estimated fluctuation analysis model and the test sample data, it is worth noting that when dividing the multiple sample sub-data into training sample data and test sample data, the embodiment of the present application does not limit the division method, and the division method may include but is not limited to any of the following: the retention method and the cross-validation method; wherein the retention method refers to directly dividing the multiple sample sub-data into two mutually exclusive sets, one set as the training sample data, and the other set as the test sample data; the cross-validation method refers to using multiple division methods (for example, different retention methods) to divide the multiple sample sub-data columns into training sample data and test sample data, and under each division method, the initial fluctuation analysis model can be iteratively trained based on the training sample data to obtain the trained fluctuation analysis model, and then, the trained fluctuation analysis model can be evaluated based on the test sample data, and the fluctuation analysis model estimated by the division method with the best evaluation result can be used as the first fluctuation analysis model. It is also worth noting that the process of estimating the model parameters of the initial fluctuation analysis model is similar to the above-mentioned iterative estimation process, and the details can be referred to the above-mentioned description of iterative estimation, which will not be repeated here.

[0133] In addition, the process of performing a fit check between the estimated volatility analysis model and the test sample data may specifically include: the test sample data may include multiple test samples, and the observed volatility data and the sample volatility data may be determined in each test sample. The process of determining the observed volatility data and the sample volatility data is unified in the embodiments of the present application and will not be repeated here. The estimated volatility analysis model may be called to predict the predicted volatility data corresponding to the observed volatility data in each volatility data subsequence based on the sample volatility data in each test sample. Then, the volatility residual information corresponding to each test sample may be determined based on the difference between the observed volatility data and the corresponding predicted volatility data of each test sample, and the volatility residual information corresponding to each test sample constitutes a volatility residual sequence. Then, the volatility residual sequence can be analyzed. The analysis here can specifically include a normality test or an autocorrelation test on the volatility residual sequence; when the volatility residual sequence is tested for normality, if the volatility residual sequence conforms to the normal distribution, it can be determined that the estimated volatility analysis model is compatible with the test sample data, and if the volatility residual sequence does not conform to the normal distribution, it can be determined that the estimated volatility analysis model is not compatible with the test sample data; when the volatility residual sequence is tested for autocorrelation, if the volatility residual sequence has autocorrelation, it can be determined that the estimated volatility analysis model is compatible with the test sample data, and if the volatility residual sequence does not have autocorrelation, it can be determined that the estimated volatility analysis model is not compatible with the test sample data. By performing an adaptation check between the estimated volatility analysis model and the test sample data, the volatility analysis model can be more compatible with the business, and the volatility analysis model is more suitable for performing volatility analysis on the business data sequence of the business, which can improve the accuracy of volatility analysis, thereby improving the accuracy of predicting future business data of the business.

[0134] In addition, when the estimated volatility analysis model does not match the test sample data, the volatility analysis model needs to be updated, and the updating principle may include any of the following: First, the new volatility analysis model and the initial volatility analysis model are different types of volatility analysis models, for example, the initial volatility analysis model is an ARCH model, and the new volatility analysis model is a GARCH model. Second, the new volatility analysis model and the initial volatility analysis model are of the same type, but the model complexity is different; the complexity of the new volatility analysis model may be higher than that of the initial volatility analysis model, for example, the initial volatility analysis model is a GARCH (p, q) model, and the new volatility analysis model is a GARCH (p+1, q+1) model; or, the complexity of the new volatility analysis model may be lower than that of the initial volatility analysis model, for example, the initial volatility analysis model is a GARCH (p, q) model, and the new volatility analysis model is a GARCH (p-1, q-1) model.

[0135] S303, performing dependency analysis on the historical business data sequence to obtain business dependency features, where the dependency features are used to characterize the dependency relationship between business data at different time steps in the business.

[0136] In step S203, after the historical business data sequence is acquired, dependency analysis may be performed on the historical business data sequence to obtain dependency features of the business, and the dependency features may be used to characterize the dependency relationship between business data at different time steps in the business.

[0137] Dependency analysis can be performed by calling a target dependency analysis model. The historical business data sequence may include business data of T time steps within the historical time range. The target dependency analysis model may include T dependency analysis units. The T dependency analysis units correspond to the business data of T time steps one by one, and T is an integer greater than 1; that is, a dependency analysis unit in the target dependency analysis model can be used to analyze the business data of one time step, and each dependency analysis unit can be used to analyze the business data of the corresponding time step of each dependency analysis unit, and the dependency on the business data of the preceding time step of the corresponding time step. The preceding time step of the corresponding time step refers to the time step that is before the corresponding time step in time sequence.

[0138] The specific process of dependency analysis may include: calling each dependency analysis unit, performing dependency analysis on the business data of the corresponding time step and the business data of the previous time step of the corresponding time step, and obtaining the dependency analysis result of each dependency analysis unit. The dependency analysis result of each dependency analysis unit can be used to characterize the dependency of the business data of the corresponding time step of each dependency analysis unit on the business data of the previous time step of the corresponding time step. Then, the dependency analysis result of the Tth dependency analysis unit among the T dependency analysis units can be determined as a dependency feature. The Tth dependency analysis unit corresponds to the Tth time step. The dependency analysis result of the Tth dependency analysis unit can be used to characterize the dependency of the business data of the Tth time step on the business data of the previous T-1 time step of the Tth time step. That is to say, the dependency feature can be used to characterize the dependency of the business data of the Tth time step on the business data of the previous T-1 time step of the Tth time step. Alternatively, the statistical result of the dependency analysis results of T dependency analysis units can be determined as the dependency feature; the statistical result can be obtained by statistically processing the dependency analysis results of the T dependency analysis units, for example, the dependency analysis results of the T dependency analysis units can be averaged to obtain the statistical result, or for another example, the maximum dependency analysis result among the dependency analysis results of the T dependency analysis units can be selected as the statistical result, or for another example, the minimum dependency analysis result among the dependency analysis results of the T dependency analysis units can be selected as the statistical result; in this case, the dependency feature can be used to characterize the business data of each time step in the T time steps, and the comprehensive dependency of each time step on the business data of the previous time step.

[0139] In the above dependency analysis process, each dependency analysis unit in the dependency analysis model needs to be subjected to dependency analysis. The dependency analysis process of any dependency analysis unit in the dependency analysis model is introduced below. For ease of understanding, any dependency analysis unit in the dependency analysis model can be represented as the jth dependency analysis unit. The jth dependency analysis unit corresponds to the jth time step in the historical business data sequence. The jth dependency analysis unit can be used to perform dependency analysis on the business data of the jth time step and the business data of the j-1 time steps preceding the jth time step to obtain the dependency analysis result of the jth dependency analysis unit. The dependency analysis result of the jth dependency analysis unit can be used to characterize the business data of the jth time step and the dependency of the business data of the j-1 time steps preceding the jth time step. j is an integer greater than 1, and j is less than or equal to T.

[0140] The dependency analysis process of the j-th dependency analysis unit may specifically include: the features of the business data of the previous j-1 time step are retained in the state information of the j-1-th dependency analysis unit, and the state information of the j-1-th dependency analysis unit and the dependency analysis result of the j-1-th dependency analysis unit are input into the j-th dependency analysis unit. A forgetting analysis may be performed on the state information of the j-1-th dependency analysis unit based on the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step, to determine the first state information that needs to be retained in the state information of the j-1-th dependency analysis unit; the forgetting analysis here specifically refers to determining the forgetting weight based on the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step, and controlling the first state information that needs to be retained in the state information of the j-1-th dependency analysis unit through the forgetting weight. According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, the j-th dependency analysis unit can be subjected to memory analysis to determine the second state information that needs to be added to the state information of the j-th dependency analysis unit from the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step; the memory analysis here specifically refers to determining the memory weight according to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, and controlling the second state information that needs to be added to the state information of the j-th dependency analysis unit from the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step through the memory weight. The first state information and the second state information are combined to obtain the state information of the j-th dependency analysis unit; the combination here specifically refers to the first state information plus the second state information. According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, the state information of the j-th dependency analysis unit is output analyzed to obtain the dependency analysis result of the j-th dependency analysis unit; the output analysis here specifically refers to determining the output weight according to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, and controlling the information to be output in the state information of the j-th dependency analysis unit as the dependency analysis result of the j-th dependency analysis unit through the output weight.

[0141] Based on the above introduction to the dependency analysis process of the j-th dependency analysis unit, further analysis shows that the state information of the j-1-th dependency analysis unit memorizes the characteristics of the business data of the previous j-1 time step. In the j-th dependency analysis unit, through forgetting analysis, memory analysis and output analysis, the dependency relationship of the business data of the j-th time step on the business data of the previous j-1 time step is introduced into the state information of the j-1-th dependency analysis unit to update the state information of the j-1-th dependency analysis unit and obtain the state information of the j-th dependency analysis unit. In other words, the j-th dependency analysis unit analyzes the dependency relationship of the business data of the j-th time step on the business data of the previous j-1 time step by updating the state information of the j-1-th dependency analysis unit, and memorizes this dependency relationship in the state information of the j-th dependency analysis unit. The specific update process can be seen in the following formula 3:

[0142]

[0143] As shown in the above formula 3: C j Represents the status information of the j-th dependency analysis unit; f j represents the forget weight; C j-1 Represents the status information of the j-1th dependency analysis unit; i j represents the memory weight; represents the dependency analysis result of the j-1th dependency analysis unit and the learned features in the business data of the jth time step; The second state information represents the dependency analysis result of the j-1th dependency analysis unit and the state information of the jth dependency analysis unit in the business data of the jth time step.

[0144] The target dependency analysis model called by the dependency analysis can be obtained by training the initial dependency analysis model. The training process can specifically include: sample data (the sample data here is specifically the second sample data) used to train the initial dependency analysis model can be obtained. The sample data can include multiple sample business data arranged according to time steps within the historical time range. The data format of the sample data is [number of samples, time step, number of features]; wherein the number of samples refers to the number of time steps included in the sample data, the time step is the time step length used to divide the sample data into different sample sub-data, and the number of features refers to the number of feature data under the business feature contained in each sample business data. The sample data can be divided into multiple sample sub-data according to the time step, and the multiple sample sub-data can be divided into two types of sample sub-data, namely training sample data and test sample data. The training sample data can include multiple training samples, and the test sample data can include multiple test samples. The initial dependency analysis model can be iteratively trained based on the training sample data to obtain a trained dependency analysis model, and then the trained dependency analysis model can be tested and optimized based on the test sample data to obtain a target dependency analysis model.

[0145] Among them, the following four points need to be noted during the training process of the dependency analysis model: First, when dividing multiple sample sub-data into training sample data and test sample data, the embodiment of the present application does not limit the division method, and the division method may include but is not limited to any of the following: retention method and cross-validation method. Second, any training process in the iterative training can be described as follows: Determine the observed business data and sample business data in any training sample of the training sample data, call the dependency analysis model obtained in the previous training based on the sample business data, predict the observed business data, and obtain the predicted business data corresponding to the observed business data. Based on the difference between the observed business data and the predicted business data, determine the loss information, and update the model parameters of the dependency analysis model obtained in the previous training in the direction of minimizing the loss information to obtain the dependency analysis model obtained in the current training; wherein, the difference between the observed business data and the predicted business data can be reflected, for example, as mean square error, cross entropy loss, etc.; when updating the model parameters, the gradient descent algorithm, Newton method, etc. can be used. Third, test optimization may include but is not limited to any of the following: optimizing model parameters of the dependency analysis model (e.g., learning rate, batch size, etc.), increasing (or decreasing) the complexity of the dependency analysis model (e.g., increasing or decreasing the number of dependency analysis units), and optimizing the way loss information is constructed (e.g., L1 regularization (absolute value regularization) and L2 regularization (mean square error regularization)). Fourth, in order to improve the training efficiency and performance of the dependency analysis model, the sample data may also be normalized or standardized.

[0146] S304: Construct prediction features of the business based on the fluctuation features and the dependency features.

[0147] After performing fluctuation analysis on the historical business data series to obtain the fluctuation characteristics of the business, and performing dependency analysis on the historical business data series to obtain the fluctuation characteristics of the business, the prediction characteristics of the business can be constructed based on the fluctuation characteristics and dependency characteristics.

[0148] S305, performing prediction processing on the business according to the prediction characteristics to obtain business data of the business in the future time step.

[0149] After constructing the prediction features of the business according to the fluctuation characteristics and the dependency characteristics, the business can be predicted according to the prediction features to obtain the business data of the business in the future time step. The prediction processing may include but is not limited to any of the following: point prediction and interval prediction. Among them, point prediction means that the business data of the business in the future time step obtained by the prediction processing is a single value, and the single value can represent the central estimate of the business data that may appear in the future time step; for example, the predicted future financial product price of the financial business can be the specific financial product price value of the next day. Interval prediction means that the business data of the business in the future time step obtained by the prediction processing is an interval range, and the interval range can represent the possible numerical interval of the business data of the business in the future time step, and the probability of the business data of the business in the future time step appearing in the data interval is high; for example, the predicted future financial product price of the financial business can be the price interval range of the financial product price in the future time step.

[0150] In an embodiment of the present application, after obtaining the historical business data sequence of the business, a fluctuation analysis can be performed on the historical business data sequence to obtain the fluctuation characteristics of the business, and a dependency analysis can be performed on the historical business data sequence to obtain the dependency characteristics of the business. Then, a prediction feature can be constructed based on the fluctuation characteristics and the dependency characteristics, and the business data of the business in the future time step can be predicted based on the prediction feature. Among them, the fluctuation analysis can analyze the historical business data sequence from the perspective of data fluctuation of the business data, and the fluctuation characteristics obtained by the fluctuation analysis can be used to characterize the volatility of the business data of the business as the time step changes; the dependency analysis can analyze the historical business data sequence from the perspective of the dependency relationship between business data at different time steps, and the dependency characteristics obtained by the dependency analysis can be used to characterize the dependency relationship between business data at different time steps in the business; by combining the fluctuation characteristics and the dependency characteristics, the business data of the business in the future time step can be predicted more accurately (that is, the future business data of the business can be predicted more accurately).

[0151] The present application embodiment provides a data processing method, which introduces the process of constructing a prediction feature based on a fluctuation feature and a dependency feature. The data processing method can be executed by a computer device, such as the above-mentioned Figure 1 The data processing device 102 in the data processing system shown in the figure may be, or may be, the data processing device 102 in the data processing system shown in the figure. Figure 1 The data processing system shown in FIG. 1 is an integrated device of a data acquisition device 101 and a data processing device 102. Figure 4 As shown, the data processing method may include but is not limited to the following steps S401 to S406:

[0152] S401, obtaining an initial service data sequence of a service.

[0153] S402, preprocessing the initial business data sequence to obtain a historical business data sequence.

[0154] In the embodiment of the present application, the execution process of step S401 and step S402 is the same as the above Figure 3 The execution process of step S301 in the embodiment shown is the same, and the details can be found in the above Figure 3 The specific description of step S301 in the illustrated embodiment will not be repeated here.

[0155] S403, performing fluctuation analysis on the historical business data sequence to obtain fluctuation characteristics of the business, where the fluctuation characteristics are used to characterize the fluctuation of the business data of the business as the time step changes.

[0156] In the embodiment of the present application, the execution process of step S403 is the same as the above Figure 3 The execution process of step S302 in the embodiment shown is the same, and the details can be found in the above Figure 3 The specific description of step S302 in the illustrated embodiment will not be repeated here.

[0157] S404, performing dependency analysis on the historical business data sequence to obtain dependency features of the business, where the dependency features are used to characterize the dependency relationship between business data at different time steps in the business.

[0158] In the embodiment of the present application, the execution process of step S404 is the same as the above Figure 3 The execution process of step S303 in the embodiment shown is the same, and the details can be found in the above Figure 3 The specific description of step S303 in the illustrated embodiment will not be repeated here.

[0159] S405: Construct prediction features of the business based on the fluctuation features and the dependency features.

[0160] In step S405, after the fluctuation characteristics of the business are obtained through the fluctuation analysis and the dependency characteristics of the business are obtained through the dependency analysis, the prediction characteristics of the business can be constructed according to the fluctuation characteristics and the dependency characteristics. Figure 5 As shown, the prediction feature can be constructed in any of the following ways, including but not limited to:

[0161] (1) The fluctuation feature and the dependent feature are concatenated to obtain the prediction feature; the concatenation process refers to the first-order connection of the fluctuation feature and the dependent feature. The concatenation process is suitable for the case where there is no obvious correlation between the fluctuation feature and the dependent feature. Through the concatenation process, the feature dimension can be expanded to provide more feature information. (2) The fluctuation feature and the dependent feature are weighted averaged to obtain the prediction feature; the weighted average process refers to assigning weights to the fluctuation feature and the dependent feature, for example, assigning fluctuation weights to the fluctuation feature and assigning dependency weights to the dependency feature. Then, the fluctuation feature can be weighted based on the fluctuation weight to obtain the weighted fluctuation feature, and the dependency feature can be weighted based on the dependency weight to obtain the weighted dependency weight. The weighted fluctuation feature and the weighted dependency feature can be summed to obtain the prediction feature; the fluctuation weight and the dependency weight can be set according to the importance of the fluctuation feature and the dependent feature. For example, the weight assigned to the feature with high importance is greater than the weight assigned to the feature with low importance; the weighted average process can balance different features and highlight the influence of important features. (3) Convolution processing is performed on the fluctuation feature and the dependency feature to obtain the prediction feature; the feature construction method of the convolution processing can be used to extract the local correlation between the fluctuation feature and the dependency feature, and the prediction feature can be used to characterize the correlation feature between the fluctuation feature and the dependency feature. (4) Feature fusion processing is performed on the fluctuation feature and the dependency feature to obtain the prediction feature; feature fusion processing can refer to taking the fluctuation feature and the dependency feature as the input information of the feature learning network, calling the feature learning network to comprehensively learn the fluctuation feature and the dependency feature, performing feature fusion in the process of feature learning, and the output information of the feature learning network is the prediction feature; the feature construction method of the feature fusion processing can comprehensively consider the impact of different features on the prediction of future business data. (5) Feature cross processing is performed on the fluctuation feature and the dependency feature to obtain the prediction feature; feature cross processing can include any of the following: multiplying the fluctuation feature and the dependency feature, and determining the difference between the fluctuation feature and the dependency feature.

[0162] In the process of constructing a prediction, in addition to considering the fluctuation characteristics and dependency characteristics, the data characteristics of the historical business data series can also be introduced. Specifically, the data characteristics of the historical business data series can be obtained. In this case, the process of constructing the prediction characteristics of the business based on the fluctuation characteristics and dependency characteristics can include: constructing the prediction characteristics based on the fluctuation characteristics, dependency characteristics and data characteristics, and the construction method can include any of the following: splicing processing, weighted average processing, convolution processing, feature fusion processing and feature cross processing. The data characteristics can include at least one of the following: statistical characteristics of the historical business data series (for example, the mean, variance, skewness, kurtosis, etc. of the historical business data series) and business indicator characteristics of the historical business data series; wherein the statistical characteristics of the historical business data series can be used to reflect the comprehensive situation of each business data in the historical business data series; the business indicator characteristics of the historical business data series can be used to reflect the business characteristics of the historical business data series under the business indicators (for example, moving average, relative strength index). In this way, when constructing predictive features for predicting future business data of a business, in addition to considering the volatility of the business data of the business as it changes over time, and the dependency between business data at different time steps of the business, the statistical characteristics of the historical business data series and the business indicator characteristics of the historical business data series can also be considered. This allows the future business data of the business to be predicted from a more comprehensive perspective, which can improve the accuracy of predicting the future business data of the business.

[0163] In addition, the fluctuation feature may include multiple fluctuation feature data, and the dependent feature may include multiple dependent feature data. Before constructing the prediction feature of the business based on the fluctuation feature and the dependent feature, if the number of fluctuation feature data in the fluctuation feature is large, the first contribution degree of each fluctuation feature data in the fluctuation feature to the business data of the future time step can be obtained, and the fluctuation feature data with the first contribution degree that meets the contribution condition is selected in the fluctuation feature to obtain the selected fluctuation feature, and then the prediction feature of the business can be constructed based on the selected fluctuation feature and the dependent feature. If the number of dependent feature data in the dependent feature is large, the second contribution degree of each dependent feature data in the dependent feature to the business data of the future time step can be obtained, and the dependent feature data with the second contribution degree that meets the contribution condition is selected in the dependent feature to obtain the selected dependent feature, and then the prediction feature of the business can be constructed based on the fluctuation feature and the selected dependent feature. If the number of fluctuation feature data in the fluctuation feature and the number of dependent feature data in the dependent feature are both large, the first contribution degree of each fluctuation feature data in the fluctuation feature to the business data of the predicted future time step can be obtained, and the fluctuation feature data whose first contribution degree meets the contribution condition can be selected in the fluctuation feature to obtain the selected fluctuation feature; the second contribution degree of each dependent feature data in the dependent feature to the business data of the predicted future time step can be obtained, and the dependent feature data whose second contribution degree meets the contribution condition can be selected in the dependent feature to obtain the selected dependent feature; then, the prediction feature of the business can be constructed according to the selected fluctuation feature and the selected dependent feature. In this way, feature data with a large contribution degree to the predicted future business data of the business can be screened out from the fluctuation feature and / or the dependent feature to predict the future business data of the business, and the future business data can be predicted. When the data prediction model is used to predict the future business data of the business, the complexity of the model prediction can be reduced, and the risk of overfitting during model prediction can be reduced.

[0164] In the above feature selection process, the method for determining the first contribution degree of the fluctuation feature data in the fluctuation feature to the business data for predicting the future time step, and the method for determining the second contribution degree of the dependent feature data in the dependent feature to the business data for predicting the future time step are the same. Here, taking the fluctuation feature as an example, the method for determining the first contribution degree of the fluctuation feature data in the fluctuation feature is introduced. The method for determining the second contribution degree of the dependent feature data in the dependent feature can refer to the method for determining the first contribution degree of the fluctuation feature data in the fluctuation feature, and will not be repeated here. The method for determining the first contribution degree of the fluctuation feature data in the fluctuation feature may include any of the following: filtering method, packaging method, and embedding method. Among them, the filtering method may include a variance filtering method and a correlation filtering method; the variance filtering method refers to using the variance of each fluctuation feature data as the first contribution degree of each fluctuation feature data to the business data predicted in the future time step. The smaller the variance of the fluctuation feature data, the smaller the contribution degree of the fluctuation feature data to the business data predicted in the future time step. The larger the variance of the fluctuation feature data, the greater the contribution degree of the fluctuation feature data to the business data predicted in the future time step. The correlation filtering method refers to calculating the correlation between each fluctuation feature data and other fluctuation feature data. The stronger the correlation of the fluctuation feature data with other fluctuation feature data, the greater the contribution degree of the fluctuation feature data to the business data predicted in the future time step. The weaker the correlation of the fluctuation feature data with other fluctuation feature data, the smaller the contribution degree of the fluctuation feature data to the business data predicted in the future time step. The packaging method is similar to the embedding method, and both directly use the machine learning model to predict the contribution degree of each fluctuation feature data to the business data predicted in the future time step.

[0165] In addition, the fluctuation feature may include multiple fluctuation feature data, and the dependent feature may include multiple dependent feature data. Before constructing the prediction feature of the business according to the fluctuation feature and the dependent feature, if the data scales of the fluctuation feature data in the fluctuation feature are different, the fluctuation feature may be subjected to feature scaling processing to reduce the scale differences of each fluctuation feature data in the fluctuation feature, and then the prediction feature of the business may be constructed according to the fluctuation feature and the dependent feature after feature scaling. If the data scales of the dependent feature data in the dependent feature are different, the dependent feature may be subjected to feature scaling processing to reduce the scale differences of each dependent feature data in the dependent feature, and then the prediction feature of the business may be constructed according to the fluctuation feature and the dependent feature after feature scaling. If the data scales of the fluctuation feature data in the fluctuation feature are different, and the data scales of the dependent feature data in the dependent feature are also different, the fluctuation feature may be subjected to feature scaling processing to reduce the scale differences of each fluctuation feature data in the fluctuation feature, and the dependent feature may be subjected to feature scaling processing to reduce the scale differences of each dependent feature data in the dependent feature, and then the prediction feature of the business may be constructed according to the fluctuation feature after feature scaling and the dependent feature after feature scaling. Feature scaling methods may include any of the following: minimum-maximum scaling, standardization, and normalization. Through feature scaling processing, the scale differences of each fluctuation feature data in the fluctuation feature and / or the scale differences of each dependent feature data in the dependent feature can be reduced. When the data prediction model is used to predict the future business data of the business, the prediction performance of the model can be improved.

[0166] S406, performing prediction processing on the business according to the prediction characteristics to obtain business data of the business in the future time step.

[0167] After constructing the prediction features of the business according to the fluctuation characteristics and the dependency characteristics, the business can be predicted according to the prediction features to obtain the business data of the business in the future time step. The prediction process can be to call the data prediction model to predict the business according to the prediction features. When selecting the data prediction model, the data characteristics of the business data of the business can be considered to change with the time step, and the data prediction model that matches the data characteristics of the business data of the business can be selected. The data prediction model can be any of the following machine learning models in the field of artificial intelligence: linear model, decision tree model and neural network model. For example, by observing the historical business data sequence, it is found that the data characteristics of the business data of the business change with the time step are: there may be a linear relationship between the historical business data sequence and the future business data, or through simple data transformation (for example, logarithmic transformation, polynomial transformation, etc.), there may be a linear relationship between the historical business data sequence and the future business data, then a linear model can be selected as the data prediction model; if a linear model is selected as the data prediction model, it is necessary to train and determine the linear parameters in the linear model. An exemplary linear model is as follows Formula 4:

[0168] y=β0+β1x1+β2x2+…+β n x n Formula 4

[0169] As shown in the above formula 4, β0, β1, …, β n represents the linear parameter of the linear model, and n represents the data dimension of the historical business data sequence, that is, the number of business data included in the historical business data sequence.

[0170] For example, by observing the historical business data sequence, it is found that the data characteristics of the business data of the business change with time steps are: there is no linear relationship and complex interactive relationship between the historical business data sequence and the future business data, and there are categorical variables in the historical business data sequence, then the decision tree model can be selected as the data prediction model. The decision tree model can divide the data into different areas through a series of questions and give a simple prediction in each area (such as the average value of the target in the area).

[0171] The training process of the data prediction model can be roughly described as follows: an initial data prediction model can be obtained, sample data for training the initial data prediction model can be obtained (the sample data here is specifically the third sample data), and the sample data can include multiple sample business data of the business arranged according to the time step within the historical time range. The sample data can be divided into multiple sample sub-data according to the preset time step, and the multiple sample sub-data can be divided into two types of sample sub-data, namely training sample data and test sample data. The training sample data may include multiple training samples, and the test sample data may include multiple test samples. The initial data prediction model can be iteratively trained based on the training sample data to obtain a trained data prediction model, and then the trained data prediction model can be tested and optimized based on the test sample data to obtain a target data prediction model, which can be used to predict the future business data of the business. Among them, when the trained data prediction model is tested and optimized based on the test sample data, the evaluation index may include at least one of the following: mean square error, accuracy, and AUC (a model evaluation index), etc.

[0172] In an embodiment of the present application, after obtaining the historical business data sequence of the business, a fluctuation analysis can be performed on the historical business data sequence to obtain the fluctuation characteristics of the business, and a dependency analysis can be performed on the historical business data sequence to obtain the dependency characteristics of the business. Then, a prediction feature can be constructed based on the fluctuation characteristics and the dependency characteristics, and the business data of the business in the future time step can be predicted based on the prediction feature. Among them, the fluctuation analysis can analyze the historical business data sequence from the perspective of data fluctuation of the business data, and the fluctuation characteristics obtained by the fluctuation analysis can be used to characterize the volatility of the business data of the business as the time step changes; the dependency analysis can analyze the historical business data sequence from the perspective of the dependency relationship between business data at different time steps, and the dependency characteristics obtained by the dependency analysis can be used to characterize the dependency relationship between business data at different time steps in the business; by combining the fluctuation characteristics and the dependency characteristics, the business data of the business in the future time step can be predicted more accurately (that is, the future business data of the business can be predicted more accurately).

[0173] The method of the embodiment of the present application is described in detail above. In order to facilitate better implementation of the above scheme of the embodiment of the present application, the device of the embodiment of the present application is provided below accordingly.

[0174] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a data processing device provided in an embodiment of the present application. The data processing device may be provided in a computer device provided in an embodiment of the present application. For example, the computer device may be Figure 1 The data processing device 102 in the data processing system shown may be Figure 1 The data processing system shown is an integrated device of a data acquisition device 101 and a data processing device 102. Figure 6 The data processing device shown may be a computer program running in a computer device, which may be used to execute Figure 3 or Figure 4 Some or all of the steps in the method embodiment shown. Figure 6 , the data processing device may include the following units:

[0175] An acquisition unit 601 is used to acquire a historical business data sequence of a business, where the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range;

[0176] Processing unit 602, used to perform fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business, where the fluctuation characteristics are used to characterize the fluctuation of the business data of the business as the time step changes;

[0177] Processing unit 602, used to perform dependency analysis on the historical business data sequence to obtain business dependency features, where the dependency features are used to characterize the dependency relationship between business data at different time steps in the business;

[0178] The processing unit 602 is further used to construct a prediction feature of the business according to the fluctuation feature and the dependency feature;

[0179] The processing unit 602 is further used to perform prediction processing on the business according to the prediction characteristics to obtain business data of the business in the future time step.

[0180] In one implementation, the processing unit 602 is used to perform fluctuation analysis on the historical service data sequence to obtain the fluctuation characteristics of the service, and is specifically used to perform the following steps:

[0181] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0182] Call the target fluctuation analysis model to predict business fluctuations based on the fluctuation data sequence, and obtain the fluctuation data of the business in the future time step;

[0183] The fluctuation data of the business in the future time step is determined as the fluctuation feature; the fluctuation feature is used to characterize the future volatility of the business data of the business as it changes with the time step.

[0184] In one implementation, the processing unit 602 is used to perform fluctuation analysis on the historical service data sequence to obtain the fluctuation characteristics of the service, and is specifically used to perform the following steps:

[0185] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0186] According to the preset time step, the fluctuation data sequence is split into multiple fluctuation data subsequences;

[0187] Determine observed fluctuation data and sample fluctuation data in each fluctuation data subsequence;

[0188] The target fluctuation analysis model is called to predict the predicted fluctuation data corresponding to the observed fluctuation data in each fluctuation data subsequence based on the sample fluctuation data in each fluctuation data subsequence;

[0189] Determine the volatility residual information corresponding to each volatility data subsequence according to the difference between the observed volatility data and the corresponding predicted volatility data in each volatility data subsequence;

[0190] The fluctuation residual information corresponding to each fluctuation data subsequence is determined as the fluctuation feature; the fluctuation feature is used to characterize the abnormal fluctuation caused by the business data of the business changing with the time step.

[0191] In one implementation, the processing unit 602 is further configured to perform the following steps:

[0192] Obtaining an initial fluctuation analysis model and sample data, the sample data including training sample data and test sample data;

[0193] According to the training sample data, the model parameters of the initial fluctuation analysis model are estimated to obtain a first fluctuation analysis model;

[0194] Performing a fit check between the first fluctuation analysis model and the test sample data;

[0195] If the first fluctuation analysis model is compatible with the test sample data, the first fluctuation analysis model is determined as the target fluctuation analysis model.

[0196] In one implementation, the processing unit 602 is further configured to perform the following steps:

[0197] If the first fluctuation analysis model is not compatible with the test sample data, a new fluctuation analysis model is obtained, and model parameters of the new fluctuation analysis model are estimated according to the training sample data to obtain a second fluctuation analysis model;

[0198] Performing a fit check between the second wave analysis model and the test sample data;

[0199] If the second fluctuation analysis model is compatible with the test sample data, the second fluctuation analysis model is determined as the target fluctuation analysis model;

[0200] If the second fluctuation analysis model does not match the test sample data, continue to update the fluctuation analysis model until the estimated fluctuation analysis model matches the test sample data, and determine the fluctuation analysis model that matches the test sample data as the target fluctuation analysis model;

[0201] The new fluctuation analysis model and the initial fluctuation analysis model are of different types; or, the new fluctuation analysis model and the initial fluctuation analysis model are of the same type, but have different levels of model complexity.

[0202] In one implementation, the processing unit 602 is used to perform fluctuation analysis on the historical service data sequence to obtain the fluctuation characteristics of the service, and is specifically used to perform the following steps:

[0203] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0204] According to the preset time step, the fluctuation data sequence is split into multiple fluctuation data subsequences;

[0205] Iteratively estimate the model parameters of the volatility analysis model according to multiple volatility data subsequences;

[0206] The model parameters obtained by iterative estimation are determined as fluctuation characteristics; the fluctuation characteristics are used to characterize the impact of historical volatility of business data of a business as it changes over time on future volatility.

[0207] In one implementation, the fluctuation data sequence is divided into N fluctuation data subsequences, and the iterative estimation refers to estimating the model parameters of the initial fluctuation analysis model N times, the N estimates correspond one-to-one to the N fluctuation data subsequences, and N is an integer greater than 1; the i-th estimation in the N estimates is based on the fluctuation analysis model obtained by the i-1-th estimation, and the model parameters obtained by the iterative estimation are the model parameters obtained by the N-th estimation, and i is an integer greater than 1, and i is less than or equal to N;

[0208] The process of estimating the model parameters for the i-th time includes:

[0209] In the i-th fluctuation data subsequence among the N fluctuation data subsequences, the observed fluctuation data and the sample fluctuation data are determined;

[0210] The volatility analysis model obtained by the i-1th estimation is called to predict the predicted volatility data corresponding to the observed volatility data based on the sample volatility data, and the probability that the predicted volatility data is the observed volatility data is obtained;

[0211] Update the model parameters obtained by the i-1th estimation in the direction of maximizing the probability to obtain the model parameters obtained by the i-th estimation.

[0212] In one implementation, the historical business data sequence includes business data of T time steps within a historical time range, where T is an integer greater than 1; the dependency analysis is performed by calling a target dependency analysis model, where the target dependency analysis model includes T dependency analysis units, and the T dependency analysis units correspond to the business data of T time steps in a one-to-one manner; the processing unit 602 is used to perform dependency analysis on the historical business data sequence, and when obtaining the dependency characteristics of the business, it is specifically used to execute the following steps:

[0213] Call each dependency analysis unit to perform dependency analysis on the business data of the corresponding time step and the business data of the previous time step of the corresponding time step, and obtain the dependency analysis result of each dependency analysis unit;

[0214] The dependency analysis result of the Tth dependency analysis unit among T dependency analysis units is determined as a dependency feature, and the dependency feature is used to characterize the dependency of the business data of the Tth time step on the business data of the preceding T-1 time steps of the Tth time step; or, the statistical result of the dependency analysis results of the T dependency analysis units is determined as the dependency feature, and the dependency feature is used to characterize the comprehensive dependency of the business data of each time step among the T time steps on the business data of the preceding time step of each time step.

[0215] In one implementation, the j-th dependency analysis unit among the T dependency analysis units is used to perform dependency analysis on the business data of the j-th time step and the business data of the j-1th time step preceding the j-th time step, to obtain the dependency analysis result of the j-th dependency analysis unit; the features of the business data of the j-1th time step preceding the j-1th dependency analysis unit are retained in the state information of the j-1th dependency analysis unit, and the state information of the j-1th dependency analysis unit and the dependency analysis result of the j-1th dependency analysis unit are input into the j-th dependency analysis unit; j is an integer greater than 1, and j is less than or equal to T;

[0216] The dependency analysis process of the j-th dependency analysis unit includes:

[0217] According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, forgetting analysis is performed on the state information of the j-1th dependency analysis unit to determine the first state information that needs to be retained in the state information of the j-1th dependency analysis unit;

[0218] Perform memory analysis on the j-th dependency analysis unit according to the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step, and determine the second state information that needs to be added to the state information of the j-th dependency analysis unit from the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step;

[0219] Combine the first state information and the second state information to obtain state information of the j-th dependency analysis unit;

[0220] According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the jth time step, the state information of the jth dependency analysis unit is output analyzed to obtain the dependency analysis result of the jth dependency analysis unit.

[0221] In one implementation, the processing unit 602 is configured to construct a prediction feature of a service according to the fluctuation feature and the dependency feature, and is specifically configured to perform any of the following steps:

[0222] The fluctuation features and the dependent features are spliced ​​to obtain the prediction features;

[0223] Perform weighted averaging on the fluctuation characteristics and dependency characteristics to obtain the prediction characteristics;

[0224] Perform convolution processing on the fluctuation features and the dependent features to obtain the prediction features;

[0225] Perform feature fusion processing on the fluctuation feature and the dependent feature to obtain the prediction feature;

[0226] The fluctuation features and dependent features are cross-processed to obtain the prediction features.

[0227] In one implementation, the obtaining unit 601 is further configured to perform the following steps:

[0228] Obtain data features of historical business data series;

[0229] The processing unit 602 is used to construct the prediction feature of the business according to the fluctuation feature and the dependency feature, and is specifically used to perform the following steps:

[0230] Construct prediction features based on fluctuation features, dependency features, and data features;

[0231] Among them, the data characteristics include at least one of the following: statistical characteristics of historical business data sequences, and business indicator characteristics of historical business data sequences; statistical characteristics are used to reflect the comprehensive situation of each business data in the historical business data sequence; business indicator characteristics are used to reflect the business characteristics of the historical business data sequence under the business indicators.

[0232] In one implementation, the fluctuation feature includes a plurality of fluctuation feature data, and the dependency feature includes a plurality of dependency feature data; the processing unit 602 is used to perform the following steps before constructing the prediction feature of the business according to the fluctuation feature and the dependency feature:

[0233] Obtaining a first contribution degree of each fluctuation feature data in the fluctuation feature to predicting the business data of the future time step, selecting the fluctuation feature data whose first contribution degree meets the contribution condition in the fluctuation feature, and obtaining the selected fluctuation feature;

[0234] Obtaining a second contribution degree of each dependent feature data in the dependent feature to predicting the business data of the future time step, selecting dependent feature data whose second contribution degree meets the contribution condition in the dependent feature, and obtaining the selected dependent feature;

[0235] The processing unit 602 is used to construct the prediction feature of the business according to the fluctuation feature and the dependency feature, and is specifically used to perform the following steps:

[0236] Construct prediction features based on the post-selection fluctuation features and the post-selection dependency features.

[0237] In one implementation, when the acquisition unit 601 is used to acquire the historical service data sequence of a service, it is specifically used to perform the following steps:

[0238] Obtaining an initial business data sequence of the business;

[0239] Preprocessing the initial business data sequence to obtain a historical business data sequence;

[0240] The preprocessing includes at least one of the following: interpolating missing data in the initial business data sequence, detecting outlier data in the initial business data sequence, and performing outlier processing on the outlier data.

[0241] According to another embodiment of the present application, Figure 6 The various units in the data processing device shown can be respectively or all combined into one or several other units to constitute, or some of the units (some) can also be split into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the function of a unit can also be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of the present application, the data processing device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0242] According to another embodiment of the present application, the program can be executed by running on a general computing device such as a computer including a central processing unit (CPU), a random access memory medium (RAM), a read-only memory medium (ROM), and other processing elements and storage elements. Figure 3 or Figure 4 A computer program for each step involved in part or all of the method shown in the figure is used to construct a computer program for each step involved in part or all of the method shown in the figure. Figure 6 The data processing device shown in the embodiment of the present application is implemented by the data processing device. The computer program can be recorded on a computer-readable storage medium, for example, and loaded into the above-mentioned computing device through the computer-readable storage medium and run therein.

[0243] In an embodiment of the present application, after obtaining the historical business data sequence of the business, a fluctuation analysis can be performed on the historical business data sequence to obtain the fluctuation characteristics of the business, and a dependency analysis can be performed on the historical business data sequence to obtain the dependency characteristics of the business. Then, a prediction feature can be constructed based on the fluctuation characteristics and the dependency characteristics, and the business data of the business in the future time step can be predicted based on the prediction feature. Among them, the fluctuation analysis can analyze the historical business data sequence from the perspective of data fluctuation of the business data, and the fluctuation characteristics obtained by the fluctuation analysis can be used to characterize the volatility of the business data of the business as the time step changes; the dependency analysis can analyze the historical business data sequence from the perspective of the dependency relationship between business data at different time steps, and the dependency characteristics obtained by the dependency analysis can be used to characterize the dependency relationship between business data at different time steps in the business; by combining the fluctuation characteristics and the dependency characteristics, the business data of the business in the future time step can be predicted more accurately (that is, the future business data of the business can be predicted more accurately).

[0244] Based on the above method and device embodiments, the present application embodiment provides a computer device. Figure 7 , Figure 7 It is a structural diagram of a computer device provided in an embodiment of the present application. Figure 7 The computer device shown includes at least a processor 701, an input interface 702, an output interface 703, and a computer-readable storage medium 704. The processor 701, the input interface 702, the output interface 703, and the computer-readable storage medium 704 may be connected via a bus or other means.

[0245] The computer-readable storage medium 704 may be stored in a memory of a computer device. The computer-readable storage medium 704 is used to store a computer program. The computer program includes computer instructions. The processor 701 is used to execute the computer program stored in the computer-readable storage medium 704. The processor 701 (or CPU (Central Processing Unit)) is the computing core and control core of the computer device, which is suitable for implementing a computer program, and is specifically suitable for loading and executing a computer program to implement a corresponding method flow or a corresponding function.

[0246] The embodiment of the present application also provides a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the computer device. In addition, a computer program suitable for being loaded and executed by a processor is also stored in the storage space. It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (Non-Volatile Memory), such as at least one disk storage; optionally, it can also be at least one computer-readable storage medium located away from the aforementioned processor.

[0247] The computer device may be, for example, Figure 1 The data processing device 102 in the data processing system shown may be, or may be Figure 1 The integrated device of the data acquisition device 101 and the data processing device 102 in the data processing system shown in the figure can be loaded and executed by the processor 701. The computer program stored in the computer readable storage medium 704 can be implemented to achieve the above-mentioned Figure 3 or Figure 4 In a specific implementation, the computer program in the computer-readable storage medium 704 is loaded by the processor 701 and executes the following steps:

[0248] Acquire a historical business data sequence of a business, where the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range;

[0249] Perform fluctuation analysis on the historical business data sequence to obtain the business fluctuation characteristics. The fluctuation characteristics are used to characterize the volatility of the business data as the time step changes.

[0250] Perform dependency analysis on historical business data sequences to obtain business dependency features, which are used to characterize the dependency relationship between business data at different time steps in the business;

[0251] Build predictive features of the business based on volatility and dependency features;

[0252] The business is predicted based on the prediction features to obtain the business data of the business in the future time step.

[0253] In one implementation, the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to perform fluctuation analysis on the historical business data sequence and obtain the fluctuation characteristics of the business, specifically for performing the following steps:

[0254] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0255] Call the target fluctuation analysis model to predict business fluctuations based on the fluctuation data sequence, and obtain the fluctuation data of the business in the future time step;

[0256] The fluctuation data of the business in the future time step is determined as the fluctuation feature; the fluctuation feature is used to characterize the future volatility of the business data of the business as it changes with the time step.

[0257] In one implementation, the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to perform fluctuation analysis on the historical business data sequence and obtain the fluctuation characteristics of the business, specifically for performing the following steps:

[0258] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0259] According to the preset time step, the fluctuation data sequence is split into multiple fluctuation data subsequences;

[0260] Determine observed fluctuation data and sample fluctuation data in each fluctuation data subsequence;

[0261] The target fluctuation analysis model is called to predict the predicted fluctuation data corresponding to the observed fluctuation data in each fluctuation data subsequence based on the sample fluctuation data in each fluctuation data subsequence;

[0262] Determine the volatility residual information corresponding to each volatility data subsequence according to the difference between the observed volatility data and the corresponding predicted volatility data in each volatility data subsequence;

[0263] The fluctuation residual information corresponding to each fluctuation data subsequence is determined as the fluctuation feature; the fluctuation feature is used to characterize the abnormal fluctuation caused by the business data of the business changing with the time step.

[0264] In one implementation, the computer program in the computer-readable storage medium 704 is loaded by the processor 701 and is further used to perform the following steps:

[0265] Obtaining an initial fluctuation analysis model and sample data, the sample data including training sample data and test sample data;

[0266] According to the training sample data, the model parameters of the initial fluctuation analysis model are estimated to obtain a first fluctuation analysis model;

[0267] Performing a fit check between the first fluctuation analysis model and the test sample data;

[0268] If the first fluctuation analysis model is compatible with the test sample data, the first fluctuation analysis model is determined as the target fluctuation analysis model.

[0269] In one implementation, the computer program in the computer-readable storage medium 704 is loaded by the processor 701 and is further used to perform the following steps:

[0270] If the first fluctuation analysis model is not compatible with the test sample data, a new fluctuation analysis model is obtained, and model parameters of the new fluctuation analysis model are estimated according to the training sample data to obtain a second fluctuation analysis model;

[0271] Performing a fit check between the second wave analysis model and the test sample data;

[0272] If the second fluctuation analysis model is compatible with the test sample data, the second fluctuation analysis model is determined as the target fluctuation analysis model;

[0273] If the second fluctuation analysis model does not match the test sample data, continue to update the fluctuation analysis model until the estimated fluctuation analysis model matches the test sample data, and determine the fluctuation analysis model that matches the test sample data as the target fluctuation analysis model;

[0274] The new fluctuation analysis model and the initial fluctuation analysis model are of different types; or, the new fluctuation analysis model and the initial fluctuation analysis model are of the same type, but have different levels of model complexity.

[0275] In one implementation, the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to perform fluctuation analysis on the historical business data sequence and obtain the fluctuation characteristics of the business, specifically for performing the following steps:

[0276] Obtain a fluctuation data sequence corresponding to a historical business data sequence, where the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within a historical time range;

[0277] According to the preset time step, the fluctuation data sequence is split into multiple fluctuation data subsequences;

[0278] Iteratively estimate the model parameters of the volatility analysis model according to multiple volatility data subsequences;

[0279] The model parameters obtained by iterative estimation are determined as fluctuation characteristics; the fluctuation characteristics are used to characterize the impact of historical volatility of business data of a business as it changes over time on future volatility.

[0280] In one implementation, the fluctuation data sequence is divided into N fluctuation data subsequences, and the iterative estimation refers to estimating the model parameters of the initial fluctuation analysis model N times, the N estimates correspond one-to-one to the N fluctuation data subsequences, and N is an integer greater than 1; the i-th estimation in the N estimates is based on the fluctuation analysis model obtained by the i-1-th estimation, and the model parameters obtained by the iterative estimation are the model parameters obtained by the N-th estimation, and i is an integer greater than 1, and i is less than or equal to N;

[0281] The process of estimating the model parameters for the i-th time includes:

[0282] In the i-th fluctuation data subsequence among the N fluctuation data subsequences, the observed fluctuation data and the sample fluctuation data are determined;

[0283] The volatility analysis model obtained by the i-1th estimation is called to predict the predicted volatility data corresponding to the observed volatility data based on the sample volatility data, and the probability that the predicted volatility data is the observed volatility data is obtained;

[0284] Update the model parameters obtained by the i-1th estimation in the direction of maximizing the probability to obtain the model parameters obtained by the i-th estimation.

[0285] In one implementation, the historical business data sequence includes business data of T time steps within a historical time range, where T is an integer greater than 1; the dependency analysis is performed by calling a target dependency analysis model, where the target dependency analysis model includes T dependency analysis units, and the T dependency analysis units correspond to the business data of T time steps in a one-to-one manner; the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to perform dependency analysis on the historical business data sequence, and when the dependency characteristics of the business are obtained, it is specifically used to perform the following steps:

[0286] Call each dependency analysis unit to perform dependency analysis on the business data of the corresponding time step and the business data of the previous time step of the corresponding time step, and obtain the dependency analysis result of each dependency analysis unit;

[0287] The dependency analysis result of the Tth dependency analysis unit among T dependency analysis units is determined as a dependency feature, and the dependency feature is used to characterize the dependency of the business data of the Tth time step on the business data of the preceding T-1 time steps of the Tth time step; or, the statistical result of the dependency analysis results of the T dependency analysis units is determined as the dependency feature, and the dependency feature is used to characterize the comprehensive dependency of the business data of each time step among the T time steps on the business data of the preceding time step of each time step.

[0288] In one implementation, the j-th dependency analysis unit among the T dependency analysis units is used to perform dependency analysis on the business data of the j-th time step and the business data of the j-1th time step preceding the j-th time step, to obtain the dependency analysis result of the j-th dependency analysis unit; the features of the business data of the j-1th time step preceding the j-1th dependency analysis unit are retained in the state information of the j-1th dependency analysis unit, and the state information of the j-1th dependency analysis unit and the dependency analysis result of the j-1th dependency analysis unit are input into the j-th dependency analysis unit; j is an integer greater than 1, and j is less than or equal to T;

[0289] The dependency analysis process of the j-th dependency analysis unit includes:

[0290] According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, forgetting analysis is performed on the state information of the j-1th dependency analysis unit to determine the first state information that needs to be retained in the state information of the j-1th dependency analysis unit;

[0291] Perform memory analysis on the j-th dependency analysis unit according to the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step, and determine the second state information that needs to be added to the state information of the j-th dependency analysis unit from the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step;

[0292] Combine the first state information and the second state information to obtain state information of the j-th dependency analysis unit;

[0293] According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the jth time step, the state information of the jth dependency analysis unit is output analyzed to obtain the dependency analysis result of the jth dependency analysis unit.

[0294] In one implementation, when the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to construct the prediction characteristics of the business according to the fluctuation characteristics and the dependency characteristics, it is specifically used to perform any of the following steps:

[0295] The fluctuation features and the dependent features are spliced ​​to obtain the prediction features;

[0296] Perform weighted averaging on the fluctuation characteristics and dependency characteristics to obtain the prediction characteristics;

[0297] Perform convolution processing on the fluctuation features and the dependent features to obtain the prediction features;

[0298] Perform feature fusion processing on the fluctuation feature and the dependent feature to obtain the prediction feature;

[0299] The fluctuation features and dependent features are cross-processed to obtain the prediction features.

[0300] In one implementation, the computer program in the computer-readable storage medium 704 is loaded by the processor 701 and is further used to perform the following steps:

[0301] Obtain data features of historical business data series;

[0302] When the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to construct the prediction characteristics of the business according to the fluctuation characteristics and the dependency characteristics, it is specifically used to perform the following steps:

[0303] Construct prediction features based on fluctuation features, dependency features, and data features;

[0304] Among them, the data characteristics include at least one of the following: statistical characteristics of historical business data sequences, and business indicator characteristics of historical business data sequences; statistical characteristics are used to reflect the comprehensive situation of each business data in the historical business data sequence; business indicator characteristics are used to reflect the business characteristics of the historical business data sequence under the business indicators.

[0305] In one implementation, the fluctuation feature includes a plurality of fluctuation feature data, and the dependency feature includes a plurality of dependency feature data; before the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to construct the prediction feature of the business according to the fluctuation feature and the dependency feature, it is also used to perform the following steps:

[0306] Obtaining a first contribution degree of each fluctuation feature data in the fluctuation feature to predicting the business data of the future time step, selecting the fluctuation feature data whose first contribution degree meets the contribution condition in the fluctuation feature, and obtaining the selected fluctuation feature;

[0307] Obtaining a second contribution degree of each dependent feature data in the dependent feature to predicting the business data of the future time step, selecting dependent feature data whose second contribution degree meets the contribution condition in the dependent feature, and obtaining the selected dependent feature;

[0308] When the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to construct the prediction characteristics of the business according to the fluctuation characteristics and the dependency characteristics, it is specifically used to perform the following steps:

[0309] Construct prediction features based on the post-selection fluctuation features and the post-selection dependency features.

[0310] In one implementation, when the computer program in the computer-readable storage medium 704 is loaded and executed by the processor 701 to obtain the historical service data sequence of the service, it is specifically used to perform the following steps:

[0311] Obtaining an initial business data sequence of the business;

[0312] Preprocessing the initial business data sequence to obtain a historical business data sequence;

[0313] The preprocessing includes at least one of the following: interpolating missing data in the initial business data sequence, detecting outlier data in the initial business data sequence, and performing outlier processing on the outlier data.

[0314] In an embodiment of the present application, after obtaining the historical business data sequence of the business, a fluctuation analysis can be performed on the historical business data sequence to obtain the fluctuation characteristics of the business, and a dependency analysis can be performed on the historical business data sequence to obtain the dependency characteristics of the business. Then, a prediction feature can be constructed based on the fluctuation characteristics and the dependency characteristics, and the business data of the business in the future time step can be predicted based on the prediction feature. Among them, the fluctuation analysis can analyze the historical business data sequence from the perspective of data fluctuation of the business data, and the fluctuation characteristics obtained by the fluctuation analysis can be used to characterize the volatility of the business data of the business as the time step changes; the dependency analysis can analyze the historical business data sequence from the perspective of the dependency relationship between business data at different time steps, and the dependency characteristics obtained by the dependency analysis can be used to characterize the dependency relationship between business data at different time steps in the business; by combining the fluctuation characteristics and the dependency characteristics, the business data of the business in the future time step can be predicted more accurately (that is, the future business data of the business can be predicted more accurately).

[0315] According to one aspect of the present application, a computer program product is provided, the computer program product comprising a computer program, the computer program being stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the data processing method provided in the above-mentioned various optional manners.

[0316] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A data processing method, characterized in that: include: Acquire a historical business data sequence of a business, wherein the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range; Performing fluctuation analysis on the historical business data sequence to obtain fluctuation characteristics of the business, wherein the fluctuation characteristics are used to characterize the fluctuation of the business data of the business as the time step changes; Performing dependency analysis on the historical business data sequence to obtain dependency features of the business, wherein the dependency features are used to characterize the dependency relationship between business data at different time steps in the business; Constructing a prediction feature of the business according to the fluctuation feature and the dependency feature; The business is predicted according to the prediction feature to obtain business data of the business in the future time step.

2. The method according to claim 1, characterized in that The performing fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business includes: Acquire a fluctuation data sequence corresponding to the historical business data sequence, wherein the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within the historical time range; Calling the target fluctuation analysis model to perform fluctuation prediction on the business based on the fluctuation data sequence to obtain fluctuation data of the business at the future time step; The fluctuation data of the business at the future time step is determined as the fluctuation feature; the fluctuation feature is used to characterize the future volatility generated by the business data of the business changing with the time step.

3. The method according to claim 1, characterized in that The performing fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business includes: Acquire a fluctuation data sequence corresponding to the historical business data sequence, wherein the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within the historical time range; According to a preset time step, splitting the fluctuation data sequence into a plurality of fluctuation data subsequences; Determining observed fluctuation data and sample fluctuation data in each of the fluctuation data subsequences; Calling the target fluctuation analysis model to predict the predicted fluctuation data corresponding to the observed fluctuation data in each of the fluctuation data subsequences based on the sample fluctuation data in each of the fluctuation data subsequences; Determine the fluctuation residual information corresponding to each of the fluctuation data subsequences according to the difference between the observed fluctuation data and the corresponding predicted fluctuation data in each of the fluctuation data subsequences; The fluctuation residual information corresponding to each of the fluctuation data subsequences is determined as the fluctuation feature; the fluctuation feature is used to characterize the abnormal fluctuation caused by the business data of the business changing with time steps.

4. The method according to claim 2 or 3, characterized in that The method further comprises: Acquire an initial fluctuation analysis model and sample data, wherein the sample data includes training sample data and test sample data; According to the training sample data, model parameters of the initial fluctuation analysis model are estimated to obtain a first fluctuation analysis model; Performing a compatibility check between the first fluctuation analysis model and the test sample data; If the first fluctuation analysis model is compatible with the test sample data, the first fluctuation analysis model is determined as the target fluctuation analysis model.

5. The method according to claim 4, characterized in that The method further comprises: If the first fluctuation analysis model is not compatible with the test sample data, a new fluctuation analysis model is obtained, and model parameters of the new fluctuation analysis model are estimated according to the training sample data to obtain a second fluctuation analysis model; Performing a compatibility check between the second fluctuation analysis model and the test sample data; If the second fluctuation analysis model is compatible with the test sample data, determining the second fluctuation analysis model as the target fluctuation analysis model; If the second fluctuation analysis model is not compatible with the test sample data, continue to update the fluctuation analysis model until the estimated fluctuation analysis model is compatible with the test sample data, and determine the fluctuation analysis model compatible with the test sample data as the target fluctuation analysis model; The new fluctuation analysis model and the initial fluctuation analysis model are fluctuation analysis models of different types; or, the new fluctuation analysis model and the initial fluctuation analysis model are of the same type but have different model complexity.

6. The method according to claim 1, characterized in that The performing fluctuation analysis on the historical business data sequence to obtain the fluctuation characteristics of the business includes: Acquire a fluctuation data sequence corresponding to the historical business data sequence, wherein the fluctuation data sequence includes a plurality of fluctuation data of the business arranged according to time steps within the historical time range; According to a preset time step, splitting the fluctuation data sequence into a plurality of fluctuation data subsequences; Iteratively estimating model parameters of a fluctuation analysis model according to the plurality of fluctuation data subsequences; The model parameters obtained by iterative estimation are determined as the fluctuation characteristics; the fluctuation characteristics are used to characterize the impact of the historical volatility of the business data of the business changing with time steps on the future volatility.

7. The method according to claim 6, characterized in that The fluctuation data sequence is divided into N fluctuation data subsequences, and the iterative estimation refers to estimating the model parameters of the initial fluctuation analysis model N times, and the N estimates correspond to the N fluctuation data subsequences one by one, and N is an integer greater than 1; The i-th estimation in the N-th estimation is performed based on the fluctuation analysis model obtained in the i-1-th estimation, the model parameters obtained by the iterative estimation are the model parameters obtained in the N-th estimation, i is an integer greater than 1, and i is less than or equal to N; The process of estimating the model parameters for the i-th time includes: In the i-th fluctuation data subsequence among the N fluctuation data subsequences, determining the observed fluctuation data and the sample fluctuation data; Calling the fluctuation analysis model obtained by the (i-1)th estimation to predict the predicted fluctuation data corresponding to the observed fluctuation data based on the sample fluctuation data, and obtaining the probability that the predicted fluctuation data is the observed fluctuation data; The model parameters obtained by the (i-1)th estimation are updated in the direction of maximizing the probability to obtain the model parameters obtained by the (i)th estimation.

8. The method according to claim 1, characterized in that The historical business data sequence includes business data of T time steps within the historical time range, where T is an integer greater than 1; the dependency analysis is performed by calling a target dependency analysis model, where the target dependency analysis model includes T dependency analysis units, and the T dependency analysis units correspond to the business data of the T time steps in a one-to-one manner; performing dependency analysis on the historical business data sequence to obtain dependency features of the business includes: Calling each of the dependency analysis units to perform dependency analysis on the business data of the corresponding time step and the business data of the previous time step of the corresponding time step, to obtain the dependency analysis result of each of the dependency analysis units; The dependency analysis result of the T-th dependency analysis unit among the T dependency analysis units is determined as the dependency feature, and the dependency feature is used to characterize the dependency of the business data of the T-th time step on the business data of the preceding T-1 time steps of the T-th time step; or, the statistical result of the dependency analysis results of the T dependency analysis units is determined as the dependency feature, and the dependency feature is used to characterize the comprehensive dependency of the business data of each time step among the T time steps on the business data of the preceding time step of each time step.

9. The method according to claim 8, characterized in that The j-th dependency analysis unit among the T dependency analysis units is used to perform dependency analysis on the business data of the j-th time step and the business data of the j-1th time step preceding the j-th time step, to obtain the dependency analysis result of the j-th dependency analysis unit; the features of the business data of the j-1th time step preceding the j-1th dependency analysis unit are retained in the state information of the j-1th dependency analysis unit, and the state information of the j-1th dependency analysis unit and the dependency analysis result of the j-1th dependency analysis unit are input into the j-th dependency analysis unit; j is an integer greater than 1, and j is less than or equal to T; The dependency analysis process of the j-th dependency analysis unit includes: Performing forgetting analysis on the state information of the j-1th dependency analysis unit according to the dependency analysis result of the j-1th dependency analysis unit and the business data of the j-th time step, and determining first state information that needs to be retained in the state information of the j-1th dependency analysis unit; Performing a memory analysis on the j-th dependency analysis unit according to the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step, and determining the second state information that needs to be added to the state information of the j-th dependency analysis unit from the dependency analysis result of the j-1-th dependency analysis unit and the business data of the j-th time step; Combining the first state information and the second state information to obtain state information of the j-th dependency analysis unit; According to the dependency analysis result of the j-1th dependency analysis unit and the business data of the jth time step, the state information of the jth dependency analysis unit is output analyzed to obtain the dependency analysis result of the jth dependency analysis unit.

10. The method according to claim 1, characterized in that The constructing the prediction feature of the business according to the fluctuation feature and the dependency feature includes any one of the following: Performing splicing processing on the fluctuation feature and the dependency feature to obtain the prediction feature; Performing weighted average processing on the fluctuation feature and the dependency feature to obtain the prediction feature; Performing convolution processing on the fluctuation feature and the dependency feature to obtain the prediction feature; Performing feature fusion processing on the fluctuation feature and the dependency feature to obtain the prediction feature; The fluctuation feature and the dependency feature are subjected to feature cross processing to obtain the prediction feature.

11. The method according to claim 1, characterized in that The method further comprises: Acquire data features of the historical business data sequence; The constructing the prediction feature of the business according to the fluctuation feature and the dependency feature includes: constructing the prediction feature according to the fluctuation feature, the dependency feature and the data feature; Among them, the data characteristics include at least one of the following: statistical characteristics of the historical business data sequence, and business indicator characteristics of the historical business data sequence; the statistical characteristics are used to reflect the comprehensive situation of each business data in the historical business data sequence; the business indicator characteristics are used to reflect the business characteristics of the historical business data sequence under the business indicators.

12. The method according to claim 1, characterized in that The fluctuation feature includes a plurality of fluctuation feature data, and the dependency feature includes a plurality of dependency feature data; Before constructing the prediction feature of the business according to the fluctuation feature and the dependency feature, the method further includes: Obtaining a first contribution degree of each fluctuation feature data in the fluctuation feature to predicting the business data of the future time step, selecting fluctuation feature data whose first contribution degree satisfies a contribution condition in the fluctuation feature, and obtaining a selected fluctuation feature; Obtaining a second contribution degree of each dependent feature data in the dependent feature to predicting the business data of the future time step, and selecting dependent feature data whose second contribution degree satisfies a contribution condition in the dependent feature to obtain the selected dependent feature; The constructing the prediction feature of the business according to the fluctuation feature and the dependency feature includes: The prediction feature is constructed according to the selected fluctuation feature and the selected dependency feature.

13. The method according to claim 1, characterized in that The acquiring of the historical business data sequence of the business includes: Acquire an initial service data sequence of the service; Preprocessing the initial service data sequence to obtain the historical service data sequence; The preprocessing includes at least one of the following: interpolating missing data in the initial business data sequence, detecting outlier data in the initial business data sequence, and performing outlier processing on the outlier data.

14. A data processing device, characterized in that: include: An acquisition unit, used to acquire a historical business data sequence of a business, wherein the historical business data sequence includes a plurality of business data of the business arranged according to time steps within a historical time range; A processing unit, used for performing fluctuation analysis on the historical business data sequence to obtain fluctuation characteristics of the business, wherein the fluctuation characteristics are used to characterize the fluctuation of the business data of the business as the time step changes; The processing unit is further used to perform dependency analysis on the historical business data sequence to obtain dependency features of the business, where the dependency features are used to characterize the dependency relationship between business data at different time steps in the business; The processing unit is further used to construct a prediction feature of the business according to the fluctuation feature and the dependency feature; The processing unit is further used to perform prediction processing on the business according to the prediction characteristics to obtain business data of the business in the future time step.

15. A computer device, characterized in that: The computer device comprises: a processor suitable for implementing a computer program; A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is suitable for being loaded by the processor and executing the data processing method according to any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the data processing method according to any one of claims 1 to 13.

17. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the data processing method according to any one of claims 1 to 13 is implemented.

Images