Task load fluctuation condition prediction method and device, equipment and storage medium thereof

Abstract

The invention belongs to the technical field of intelligent prediction, and relates to a task load fluctuation condition prediction method and device, equipment and a storage medium thereof. Core feature data influencing task load changes are decomposed through preprocessing, feature engineering and time sequence decomposition; fusing the core feature data to obtain a cross-modal core feature data fusion representation; and adjusting and updating the influence weight of the task load change to obtain a cross-modal task load fluctuation condition prediction model. In combination with historical multi-source heterogeneous data influencing task load change, a cross-modal task load fluctuation condition prediction model is constructed, so that the actual condition is better met when task load fluctuation condition prediction is actually performed subsequently. When the method is applied to a financial service scene which is easily influenced by multi-source heterogeneous data, the fluctuation change of the task load can be predicted more accurately.

Description

Technical Field

[0001] This application relates to the field of intelligent prediction technology and is applied to scenarios where the workload is increased or decreased, such as the prediction of the amount of claims workload within the next preset time period. It relates to a method, device, equipment and storage medium for predicting workload fluctuations. Background Technology

[0002] As offline businesses gradually shift to online processing, the volume of online processing will increase. Adding too many processing slots will lead to idle resources, while setting too few will result in untimely processing. Therefore, methods for predicting workload have emerged to assist the company's business monitoring department in monitoring workload.

[0003] However, traditional business volume forecasting methods often rely on single time-series models, such as AR / IMA-based or Prophet-based models, which struggle to effectively integrate external environmental data with internal business characteristics. For example, they fail to effectively combine social media sentiment, competitor activity, and weather conditions with the characteristics of financial business. This is particularly true in the financial industry, which is characterized by high volatility and the significant impact of unforeseen events. For instance, in auto insurance, natural disasters like typhoons and torrential rains can trigger a surge in claims, and during promotional periods, policy issuance may experience exponential growth. These factors contribute to higher forecasting errors under complex internal and external environments. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, device and storage medium for predicting workload fluctuations, so as to solve the problem of high workload prediction error under complex internal and external environments, thereby predicting more accurate workload fluctuation changes.

[0005] Firstly, embodiments of this application provide a method for predicting task volume fluctuations, employing the following technical solution:

[0006] A method for predicting workload fluctuations includes the following steps:

[0007] Acquire multi-source heterogeneous data that has historically influenced changes in task volume;

[0008] The multi-source heterogeneous data is preprocessed to perform quality control and integration, resulting in target data suitable for feature engineering.

[0009] Feature engineering is performed on the target data to obtain business rule feature data and spatiotemporal correlation feature data;

[0010] The time-series decomposition method is used to decompose the core feature data that affects the change in task volume from the business rule feature data and the spatiotemporal correlation feature data. The core feature data refers to the feature data that has an impact on the change in task volume that reaches a preset threshold.

[0011] By using a cross-modal fusion method, the core feature data is fused to obtain a cross-modal core feature data fusion representation;

[0012] According to the preset weight adjustment strategy, the weight adjustment and update of the cross-modal core feature data fusion representation is performed on the weight of the cross-modal task volume change, so as to obtain the cross-modal task volume fluctuation prediction model.

[0013] The system collects the latest multi-source heterogeneous data that affects changes in task volume in real time, and combines it with the cross-modal task volume fluctuation prediction model to predict the changes in task volume within a preset time period after the current time node.

[0014] Secondly, embodiments of this application also provide a device for predicting task volume fluctuations, which adopts the following technical solution:

[0015] A task load fluctuation prediction device, comprising:

[0016] The historical data acquisition module is used to acquire multi-source heterogeneous data that has historically influenced changes in task volume.

[0017] The target data acquisition module is used to perform quality control and integration processing on the multi-source heterogeneous data through preprocessing to obtain target data that can be used for feature engineering.

[0018] The feature engineering module is used to perform feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data;

[0019] The core feature data acquisition module is used to extract core feature data that affects the change in task volume from the business rule feature data and the spatiotemporal correlation feature data using a time-series decomposition method. The core feature data refers to feature data that affects the change in task volume to a preset threshold.

[0020] The core feature data fusion module is used to fuse the core feature data using a cross-modal fusion method to obtain a cross-modal core feature data fusion representation;

[0021] The task volume fluctuation prediction model acquisition module is used to adjust and update the weights of the cross-modal core feature data fusion representation based on a preset weight adjustment strategy to obtain the cross-modal task volume fluctuation prediction model.

[0022] The task volume change prediction module is used to collect the latest multi-source heterogeneous data affecting task volume changes in real time, and combine it with the cross-modal task volume fluctuation prediction model to predict the task volume changes within a preset time period after the current time node.

[0023] Thirdly, embodiments of this application also provide a computer device that adopts the technical solution described below:

[0024] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the task load fluctuation prediction method described above.

[0025] Fourthly, embodiments of this application also provide a computer-readable storage medium, which adopts the technical solutions described below:

[0026] A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the task load fluctuation prediction method described above.

[0027] Compared with the prior art, the embodiments of this application have the following main advantages:

[0028] The task volume fluctuation prediction method described in this application acquires multi-source heterogeneous data that historically influences task volume changes; preprocesses it to obtain target data suitable for feature engineering; performs feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data; uses a time-series decomposition method to extract core feature data influencing task volume changes from the business rule feature data and spatiotemporal correlation feature data; fuses the core feature data to obtain a cross-modal core feature data fusion representation; adjusts and updates the task volume change influence weights on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model; and collects the latest multi-source heterogeneous data influencing task volume changes in real time, combining it with the cross-modal task volume fluctuation prediction model to predict task volume changes within a preset time period after the current time point. By combining historical multi-source heterogeneous data influencing task volume changes to construct the cross-modal task volume fluctuation prediction model, subsequent task volume fluctuation predictions are more consistent with actual situations. Applying this method to financial business scenarios that are easily affected by multi-source heterogeneous data can more accurately predict task volume fluctuations. Attached Figure Description

[0029] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0031] Figure 2 This is a flowchart of an embodiment of a task volume fluctuation prediction method according to this application;

[0032] Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 201 shown;

[0033] Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 202 shown;

[0034] Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 204 shown;

[0035] Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 205 shown;

[0036] Figure 7 yes Figure 2 A flowchart of a specific embodiment of step 206 shown;

[0037] Figure 8 This is a schematic diagram of an embodiment of a task volume fluctuation prediction device according to this application;

[0038] Figure 9 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0040] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0041] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0042] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0043] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0044] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0045] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0046] It should be noted that the task volume fluctuation prediction method provided in this application embodiment is generally executed by a server, and correspondingly, a task volume fluctuation prediction device is generally set in the server.

[0047] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0048] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of a task volume fluctuation prediction method according to this application. The task volume fluctuation prediction method includes the following steps:

[0049] Step 201: Obtain multi-source heterogeneous data on historical changes in task volume.

[0050] In this embodiment, the multi-source heterogeneous data affecting changes in task volume refers to data that increases or decreases the task volume. This type of data may be due to internal factors in the business planning itself, or it may be due to external factors caused by force majeure. For example, when there are promotional activities during a certain period, the task volume will increase accordingly during that period; another example is that the New Year holiday or summer graduation season are peak seasons for car purchases, which will increase the volume of car sales and car insurance purchases; yet another example is that a month with many rainy days or icy days will result in more traffic accidents, which will increase the volume of claims. Furthermore, the spread of public opinion on social media will increase or decrease the corresponding business volume. These data, whether due to internal factors in the business planning itself or external factors caused by force majeure, constitute the multi-source heterogeneous data.

[0051] By acquiring multi-source heterogeneous data on historical changes in task volume, it is possible to subsequently filter out the core characteristic data affecting changes in task volume from this multi-source heterogeneous data. For internal data related to the business planning itself, it can be retrieved from the company's pre-set business repository; for external data related to force majeure events, data interfaces provided by third-party service providers can be used to retrieve the corresponding impact data, such as obtaining extreme weather forecast data from meteorological service interfaces or obtaining public opinion impact data from major social media platforms.

[0052] Step 202: The multi-source heterogeneous data is preprocessed to perform quality control and integration to obtain target data that can be used for feature engineering.

[0053] Specifically, by performing quality control and integration processing on the multi-source heterogeneous data, the final target data meets the requirements of feature engineering, avoiding the inconvenience or complexity of data processing caused by low-quality multi-source heterogeneous data.

[0054] In this embodiment, the preprocessing method for quality control and integration of the multi-source heterogeneous data includes data cleaning and format standardization.

[0055] Step 203: Perform feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data.

[0056] In this embodiment, feature engineering refers to converting the target data into data suitable for processing by a machine learning model. For example, extracting key information from the target data; or performing feature construction and transformation to obtain new data distributions. Feature engineering often uses statistics, algorithms, and domain knowledge to convert target data into feature representations that a computer model can understand.

[0057] Specifically, obtaining business rule feature data means obtaining relevant feature data expressing business rule situations contained in the target data by performing feature engineering on the target data. For example, the claim review rules when a vehicle accident occurs are generally prescriptive logical rules and feature data that do not change frequently. Obtaining spatiotemporal correlation feature data means obtaining spatiotemporally correlated feature data in the target data by performing feature engineering on the target data and combining it with time series information. For example, feature data that shows a clear upward or downward trend with certain changes over time, or feature data that changes or fluctuates periodically with changes in other data.

[0058] Step 204: Use the time-series decomposition method to extract the core feature data that affects the change in task volume from the business rule feature data and the spatiotemporal correlation feature data. The core feature data refers to the feature data that affects the change in task volume to a preset threshold.

[0059] In this embodiment, a temporal decomposition method is adopted. For example, the Prophet decomposition component is used to decompose the business rule feature data and the spatiotemporal correlation feature data to separate the core feature data. Alternatively, PCA dimensionality reduction decomposition method can be used to separate the low-dimensional core feature data.

[0060] Step 205: Using a cross-modal fusion method, the core feature data is fused to obtain a cross-modal core feature data fusion representation.

[0061] In this embodiment, fusing the core feature data includes fusing temporal feature data, contextual feature data, and core feature data that affect changes in task volume.

[0062] Step 206: According to the preset weight adjustment strategy, the weight adjustment and update of the cross-modal core feature data fusion representation is performed to adjust the weight of the cross-modal task volume change, so as to obtain the cross-modal task volume fluctuation prediction model.

[0063] Specifically, when extracting temporal feature data and contextual feature data, Transformer neural networks and CNN convolutional neural networks are used respectively. In step 205, when fusing temporal feature data, contextual feature data, and core feature data affecting task volume changes, an XGBoost meta-learner can be used to perform cross-modal fusion of the core feature data affecting task volume changes, the temporal feature data, and the contextual feature data to obtain the cross-modal core feature data fusion representation. Based on a preset weight adjustment strategy, the weights affecting task volume changes in the cross-modal core feature data fusion representation are adjusted and updated to obtain a cross-modal task volume fluctuation prediction model, including the data extracted using the aforementioned methods.

[0064] The Prophet decomposition component, Transformer neural network, and CNN convolutional neural network are fused to obtain a cross-modal task load fluctuation prediction model. For example, the initial weights of the Prophet decomposition component, Transformer neural network, and CNN convolutional neural network are set to 0.4, 0.1, and 0.5, respectively. If a certain feature data is both time-series feature data and core feature data, then the weight corresponding to its fused feature representation is 0.4 + 0.1 = 0.5. As another example, if a certain feature data is both contextual feature data and core feature data, then the weight corresponding to its fused feature representation is 0.4 + 0.5 = 0.9.

[0065] Step 207: Collect the latest multi-source heterogeneous data affecting the change in task volume in real time, and combine it with the cross-modal task volume fluctuation prediction model to predict the change in task volume within a preset time period after the current time node.

[0066] Specifically, the system acquires the latest multi-source heterogeneous data affecting task volume changes in real time. Then, it uses the cross-modal task volume fluctuation prediction model to acquire feature data, identifying the time-series feature weights, core feature weights, and context-specific feature weights contained in the corresponding feature data. Combining the time-series feature weights, core feature weights, and context-specific feature weights contained in the corresponding feature data, as well as a preset task volume change reference form, the system predicts the task volume changes within a preset time length after the current time node. The preset task volume change reference form contains the corresponding task volume changes within a preset time length after the current time node under different time-series feature weights, core feature weights, and context-specific feature weights.

[0067] In this embodiment, multi-source heterogeneous data influencing historical changes in task volume are acquired. Preprocessing yields target data suitable for feature engineering. Feature engineering is performed on the target data to obtain business rule feature data and spatiotemporal correlation feature data. A time-series decomposition method is used to extract core feature data influencing task volume changes from the business rule feature data and spatiotemporal correlation feature data. The core feature data is fused to obtain a cross-modal core feature data fusion representation. The weights affecting task volume changes are adjusted and updated on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model. The latest multi-source heterogeneous data influencing task volume changes is collected in real time and combined with the cross-modal task volume fluctuation prediction model to predict task volume changes within a preset time period after the current time point. By combining historical multi-source heterogeneous data influencing task volume changes to construct a cross-modal task volume fluctuation prediction model, the prediction of task volume fluctuations becomes more consistent with actual conditions. Applying this method to financial business scenarios susceptible to multi-source heterogeneous data can more accurately predict task volume fluctuations.

[0068] Continue to refer to Figure 3 , Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 201 shown includes:

[0069] Step 301: Analyze the preset table of factors affecting changes in task volume to obtain all influencing factors;

[0070] In this embodiment, the preset task volume change influencing factor form integrates external business influencing factors and internal business influencing factors that affect task volume changes.

[0071] Step 302: Identify the data acquisition methods for each of the influencing factors;

[0072] In this embodiment, the different sources of external and internal business influencing factors result in different methods for obtaining the factor data for each factor. For example, factor data for internal business influencing factors can be obtained directly from the company's internal data platform, while factor data for external business influencing factors needs to be obtained in conjunction with a third-party service platform.

[0073] Step 303: Based on the acquisition method, collect historical factor data that affect changes in task volume.

[0074] For example, collecting data from last year's flood season.

[0075] In this embodiment, after performing the step of parsing the preset task volume change influencing factor form to obtain all influencing factors, the method further includes: identifying the business external influencing factors and business internal influencing factors among the influencing factors according to the preset classification results.

[0076] Continue to refer to Figure 4 , Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 202 shown includes:

[0077] Step 401: An outlier detection algorithm is used to detect invalid and outlier data in the multi-source heterogeneous data.

[0078] Step 402: Remove the invalid data.

[0079] Step 403: Correct the abnormal data using time series interpolation or contextual compensation methods;

[0080] Step 404: After the removal and correction operations, the multi-source heterogeneous data is obtained and integrated according to the time series method to obtain the target data.

[0081] In this embodiment, the step of performing feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data specifically includes: performing feature engineering on the factor data corresponding to the internal business influencing factors in the target data to obtain business rule feature data, wherein the factor data corresponding to the internal business influencing factors includes business promotion activity data and business lifecycle data; specifically, performing feature engineering on the business promotion activity data and business lifecycle data to identify the business rule feature data respectively matched by the business promotion activity data and business lifecycle data; and performing feature engineering on the factor data corresponding to the external business influencing factors in the target data to obtain spatiotemporal correlation feature data, wherein the factor data corresponding to the external business influencing factors includes regional disaster early warning data and public opinion topic discussion data; specifically, performing feature engineering on the regional disaster early warning data and public opinion topic discussion data to identify the spatiotemporal correlation feature data respectively possessed by the regional disaster early warning data and public opinion topic discussion data.

[0082] Continue to refer to Figure 5 , Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 204 shown includes:

[0083] Step 501: Use the Prophet decomposition component to decompose the business rule feature data and the spatiotemporal correlation feature data respectively, and decompose them into trend feature data, periodic feature data and festival effect feature data.

[0084] Specifically, for example, the New Year holiday or summer graduation season are peak seasons for vehicle purchases, leading to an increase in vehicle sales and car insurance purchases. By decomposing car insurance purchase data, we can identify cyclical characteristics related to vehicle purchases. Another example is the significant increase in e-commerce transaction volume during the annual 6.18 shopping festival, Valentine's Day, or 11.11 shopping festival. By decomposing business data, we can identify holiday-related effects.

[0085] Step 502: By using the data thresholds pre-set for the impact of changes in task volume, select feature data exceeding the corresponding data thresholds from the trend feature data, periodic feature data, and holiday effect feature data, and use them as the core feature data affecting changes in task volume.

[0086] Specifically, in step 502, based on the data thresholds pre-set for the impact of changes in task volume, relatively important feature data are selected from the trend feature data, periodic feature data, and holiday effect feature data as the core feature data affecting changes in task volume.

[0087] Continue to refer to Figure 6 , Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 205 shown includes:

[0088] Step 601: Obtain time-series feature data extracted from the trend feature data and the periodic feature data, wherein a Transformer neural network is used to extract the time-series feature data from the target data;

[0089] Specifically, the serialization encoding mechanism of the Transformer neural network is used to extract temporal feature data from the target data.

[0090] Step 602: Obtain contextual feature data extracted from the festival effect feature data, wherein a CNN convolutional neural network is used to extract the contextual feature data from the target data;

[0091] Specifically, using a CNN (Convolutional Neural Network) can fully incorporate contextual information and extract more comprehensive contextual feature data from the target data.

[0092] Step 603: Use the XGBoost meta-learner to perform cross-modal fusion of the core feature data that affects the change in task volume, the temporal feature data, and the contextual feature data to obtain the cross-modal core feature data fusion representation.

[0093] Continue to refer to Figure 7 , Figure 7 yes Figure 2 A flowchart of a specific embodiment of step 206 shown includes:

[0094] Step 701: Obtain the initial weights corresponding to the time-series feature data, the contextual feature data, and the core feature data affecting the change in task volume, respectively;

[0095] Specifically, the initial weights can be obtained directly from the initial weights set for the Prophet decomposition component, the Transformer neural network, and the CNN convolutional neural network, and these initial weights can be used as the initial weights for the corresponding temporal feature data, the contextual feature data, and the core feature data that affects the change in task volume.

[0096] Step 702: Adjust the initial weights according to the cross-modal core feature data fusion representation results to obtain the task quantity change impact weights corresponding to the adjusted cross-modal core feature data fusion representation.

[0097] Specifically, for example, if the cross-modal core feature data fusion representation result corresponding to feature data 'a' corresponds to temporal feature data, contextual feature data, and the core feature data affecting task volume changes, then the task volume change impact weight corresponding to feature data 'a' is 0.4 + 0.1 + 0.5 = 1. Using this method, the task volume change impact weights corresponding to all feature data are calculated. Following the example in this application, there will be 7 possible combinations: 0.4, 0.1, 0.5, 0.4 + 0.1, 0.1 + 0.5, 0.4 + 0.5, and 0.4 + 0.1 + 0.5. These represent, respectively, that the core feature data fusion representation result corresponding to the current feature data corresponds to only one of the temporal feature data, contextual feature data, and the core feature data affecting task volume changes, as well as two and one cases of full correspondence. Subsequently, corresponding node weight adjustments are made based on the task volume change impact weights corresponding to different feature data.

[0098] Step 703: Based on the adjusted cross-modal core feature data fusion representation corresponding to the weights of task volume change impact, integrate the Prophet decomposition component, Transformer neural network and CNN convolutional neural network to obtain a cross-modal task volume fluctuation prediction model.

[0099] Specifically, the Prophet decomposition component, Transformer neural network, and CNN convolutional neural network are integrated, and the weight of the task load change impact corresponding to each feature data is obtained based on the fusion representation result of the cross-modal core feature data, thereby obtaining a cross-modal task load fluctuation prediction model.

[0100] In this embodiment, multi-source heterogeneous data influencing historical changes in task volume are acquired. Preprocessing yields target data suitable for feature engineering. Feature engineering is performed on the target data to obtain business rule feature data and spatiotemporal correlation feature data. A time-series decomposition method is used to extract core feature data influencing task volume changes from the business rule feature data and spatiotemporal correlation feature data. The core feature data is fused to obtain a cross-modal core feature data fusion representation. The weights affecting task volume changes are adjusted and updated on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model. The latest multi-source heterogeneous data influencing task volume changes is collected in real time and combined with the cross-modal task volume fluctuation prediction model to predict task volume changes within a preset time period after the current time point. By combining historical multi-source heterogeneous data influencing task volume changes to construct a cross-modal task volume fluctuation prediction model, the prediction of task volume fluctuations becomes more consistent with actual conditions. Applying this method to financial business scenarios susceptible to multi-source heterogeneous data can more accurately predict task volume fluctuations.

[0101] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. 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 that knowledge to obtain optimal results.

[0102] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0103] In this embodiment, multi-source heterogeneous data influencing historical changes in task volume are acquired. Preprocessing yields target data suitable for feature engineering. Feature engineering is performed on the target data to obtain business rule feature data and spatiotemporal correlation feature data. A time-series decomposition method is used to extract core feature data influencing task volume changes from the business rule feature data and spatiotemporal correlation feature data. The core feature data is fused to obtain a cross-modal core feature data fusion representation. The weights affecting task volume changes are adjusted and updated on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model. The latest multi-source heterogeneous data influencing task volume changes is collected in real time and combined with the cross-modal task volume fluctuation prediction model to predict task volume changes within a preset time period after the current time point. By combining historical multi-source heterogeneous data influencing task volume changes to construct a cross-modal task volume fluctuation prediction model, the prediction of task volume fluctuations becomes more consistent with actual conditions. Applying this method to financial business scenarios susceptible to multi-source heterogeneous data can more accurately predict task volume fluctuations.

[0104] Further reference Figure 8 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of a task load fluctuation prediction device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0105] like Figure 8 As shown, the task volume fluctuation prediction device 800 described in this embodiment includes: a historical data acquisition module 801, a target data acquisition module 802, a feature engineering module 803, a core feature data acquisition module 804, a core feature data fusion module 805, a task volume fluctuation prediction model acquisition module 806, and a task volume change prediction module 807. Wherein:

[0106] The historical data acquisition module 801 is used to acquire multi-source heterogeneous data that has historically affected changes in task volume.

[0107] The target data acquisition module 802 is used to perform quality control and integration processing on the multi-source heterogeneous data through preprocessing to obtain target data that can be used for feature engineering.

[0108] Feature engineering module 803 is used to perform feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data;

[0109] The core feature data acquisition module 804 is used to extract core feature data that affects the change in task volume from the business rule feature data and the spatiotemporal correlation feature data using a time-series decomposition method. The core feature data refers to feature data that affects the change in task volume to a preset threshold.

[0110] The core feature data fusion module 805 is used to fuse the core feature data using a cross-modal fusion method to obtain a cross-modal core feature data fusion representation;

[0111] The task volume fluctuation prediction model acquisition module 806 is used to adjust and update the weight of the cross-modal core feature data fusion representation according to a preset weight adjustment strategy to obtain the cross-modal task volume fluctuation prediction model.

[0112] The task volume change prediction module 807 is used to collect the latest multi-source heterogeneous data affecting task volume changes in real time, and combine it with the cross-modal task volume fluctuation prediction model to predict the task volume changes within a preset time period after the current time node.

[0113] This application acquires multi-source heterogeneous data on historical impacts on task volume changes; preprocesses it to obtain target data suitable for feature engineering; performs feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data; uses a time-series decomposition method to extract core feature data affecting task volume changes from the business rule feature data and spatiotemporal correlation feature data; fuses the core feature data to obtain a cross-modal core feature data fusion representation; adjusts and updates the task volume change impact weights on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model; and collects the latest multi-source heterogeneous data affecting task volume changes in real time, combining it with the cross-modal task volume fluctuation prediction model to predict task volume changes within a preset time period after the current time point. By combining historical multi-source heterogeneous data on task volume changes to construct a cross-modal task volume fluctuation prediction model, subsequent predictions of task volume fluctuations are more consistent with actual conditions. Applying this method to financial business scenarios that are easily affected by multi-source heterogeneous data can more accurately predict task volume fluctuations.

[0114] In this embodiment, the historical data acquisition module 801 includes an influence factor analysis unit, an acquisition method identification unit, and a factor data acquisition unit. Wherein:

[0115] The impact factor analysis unit is used to analyze a preset table of impact factors for changes in task volume and obtain all impact factors.

[0116] The acquisition method identification unit is used to identify the acquisition method of factor data for each of the influencing factors.

[0117] The factor data acquisition unit is used to acquire historical factor data that influences changes in task quantity based on the acquisition method described above.

[0118] In this embodiment, the target data acquisition module 802 further includes: a detection unit, a rejection unit, a correction unit, and an integration unit. Wherein:

[0119] The detection unit is used to detect invalid and abnormal data in the multi-source heterogeneous data using an outlier detection algorithm.

[0120] The rejection unit is used to reject the invalid data.

[0121] The correction unit is used to correct the abnormal data using time series interpolation or context-related compensation methods.

[0122] The integration unit is used to obtain the multi-source heterogeneous data after the elimination and correction operations, and integrate it in a time series manner to obtain the target data.

[0123] In this embodiment, the feature engineering module 803 includes an internal factor data feature engineering unit and an external factor data feature engineering unit. Wherein:

[0124] The internal factor data feature engineering unit is used to perform feature engineering processing on the factor data corresponding to the internal business influence factors in the target data to obtain business rule feature data. The factor data corresponding to the internal business influence factors includes business promotion activity data and business life cycle data. Specifically, it includes: performing feature engineering processing on the business promotion activity data and business life cycle data to identify the business rule feature data that the business promotion activity data and business life cycle data respectively hit.

[0125] The external factor data feature engineering unit is used to perform feature engineering processing on the factor data corresponding to the business external influencing factors in the target data to obtain spatiotemporal correlation feature data. The factor data corresponding to the business external influencing factors includes regional disaster early warning data and public opinion topic discussion data. Specifically, it includes: performing feature engineering processing on the regional disaster early warning data and public opinion topic discussion data to identify the spatiotemporal correlation feature data possessed by the regional disaster early warning data and public opinion topic discussion data respectively.

[0126] In this embodiment, the core feature data acquisition module 804 includes a Prophet decomposition unit and a core feature data filtering unit. Wherein:

[0127] The Prophet decomposition unit is used to decompose the business rule feature data and the spatiotemporal correlation feature data using the Prophet decomposition component, respectively, to decompose trend feature data, periodic feature data and festival effect feature data.

[0128] The core feature data filtering unit is used to filter feature data exceeding the corresponding data thresholds from the trend feature data, periodic feature data, and holiday effect feature data, respectively, based on the data thresholds set in advance for the impact of changes in task volume, as the core feature data affecting changes in task volume.

[0129] In this embodiment, the core feature data fusion module 805 includes a Transformer neural network extraction unit, a CNN convolutional neural network extraction unit, and an XGBoost meta-learner fusion unit. Wherein:

[0130] The Transformer neural network extraction unit is used to obtain time-series feature data extracted from the trend feature data and the periodic feature data, wherein the Transformer neural network is used to extract the time-series feature data from the target data;

[0131] A CNN convolutional neural network extraction unit is used to obtain contextual feature data extracted from the festival effect feature data, wherein a CNN convolutional neural network is used to extract contextual feature data from the target data;

[0132] The XGBoost meta-learner fusion unit is used to perform cross-modal fusion of the core feature data affecting the change of task volume, the temporal feature data, and the contextual feature data using the XGBoost meta-learner to obtain the cross-modal core feature data fusion representation.

[0133] In this embodiment, the task volume fluctuation prediction model acquisition module 806 includes an initial weight acquisition unit, a weight adjustment unit, and a processing network integration unit. Wherein:

[0134] An initial weight acquisition unit is used to acquire the initial weights corresponding to the time-series feature data, the contextual feature data, and the core feature data that affects the change in task volume, respectively.

[0135] The weight adjustment unit is used to adjust the initial weights according to the cross-modal core feature data fusion representation result, so as to obtain the weights of the task quantity change impact corresponding to the cross-modal core feature data fusion representation after adjustment.

[0136] The processing network integration unit is used to integrate the Prophet decomposition component, Transformer neural network and CNN convolutional neural network based on the task load change impact weights corresponding to the adjusted cross-modal core feature data fusion representation, to obtain a cross-modal task load fluctuation prediction model.

[0137] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0138] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0139] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 9 , Figure 9 This is a basic structural block diagram of the computer device in this embodiment.

[0140] The computer device 9 includes a memory 9a, a processor 9b, and a network interface 9c that are interconnected via a system bus. It should be noted that... Figure 9Only a computer device 9 with component memory 9a, processor 9b, and network interface 9c is shown. However, it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0141] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0142] The memory 9a includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 9a may be an internal storage unit of the computer device 9, such as the hard disk or memory of the computer device 9. In other embodiments, the memory 9a may also be an external storage device of the computer device 9, such as a plug-in hard disk, smart memory card (SMC), secure digital (SD) card, flash memory card, etc. Of course, the memory 9a may include both internal storage units and external storage devices of the computer device 9. In this embodiment, the memory 9a is typically used to store the operating system and various application software installed on the computer device 9, such as computer-readable instructions for a task load fluctuation prediction method. In addition, the memory 9a can also be used to temporarily store various types of data that have been output or will be output.

[0143] In some embodiments, the processor 9b may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 9b is typically used to control the overall operation of the computer device 9. In this embodiment, the processor 9b is used to execute computer-readable instructions stored in the memory 9a or to process data, for example, to execute computer-readable instructions for the task load fluctuation prediction method described above.

[0144] The network interface 9c may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 9 and other electronic devices.

[0145] The computer device proposed in this embodiment belongs to the field of intelligent prediction technology and is applied in scenarios where intelligent retrieval of response text is performed during customer question-and-answer or customer service dialogue. This application acquires multi-source heterogeneous data that historically influences task volume changes; through preprocessing, it obtains target data suitable for feature engineering; it performs feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data; it uses a time-series decomposition method to extract core feature data influencing task volume changes from the business rule feature data and spatiotemporal correlation feature data; it fuses the core feature data to obtain a cross-modal core feature data fusion representation; it adjusts and updates the task volume change influence weights on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model; it collects the latest multi-source heterogeneous data influencing task volume changes in real time, and combines it with the cross-modal task volume fluctuation prediction model to predict the task volume changes within a preset time period after the current time node. By combining historical multi-source heterogeneous data influencing task volume changes to construct a cross-modal task volume fluctuation prediction model, subsequent predictions of task volume fluctuations are more consistent with actual conditions. Applying this method to financial business scenarios that are easily affected by multi-source heterogeneous data can more accurately predict fluctuations in task volume.

[0146] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by a processor to cause the processor to perform the steps of the task load fluctuation prediction method described above.

[0147] The computer-readable storage medium proposed in this embodiment belongs to the field of intelligent prediction technology and is applied in scenarios of intelligent retrieval of response text during customer question-and-answer or customer service dialogue. This application acquires multi-source heterogeneous data that historically affects task volume changes; through preprocessing, it obtains target data suitable for feature engineering; it performs feature engineering on the target data to obtain business rule feature data and spatiotemporal correlation feature data; it uses a time-series decomposition method to extract core feature data affecting task volume changes from the business rule feature data and spatiotemporal correlation feature data; it fuses the core feature data to obtain a cross-modal core feature data fusion representation; it adjusts and updates the task volume change impact weights on the cross-modal core feature data fusion representation to obtain a cross-modal task volume fluctuation prediction model; it collects the latest multi-source heterogeneous data affecting task volume changes in real time, and combines it with the cross-modal task volume fluctuation prediction model to predict the task volume changes within a preset time period after the current time node. By combining historical multi-source heterogeneous data affecting task volume changes to construct a cross-modal task volume fluctuation prediction model, subsequent predictions of task volume fluctuations are more consistent with actual situations. Applying this method to financial business scenarios that are easily affected by multi-source heterogeneous data can more accurately predict fluctuations in task volume.

[0148] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0149] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to make the disclosure of this application more thorough and comprehensive. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application. Software tools or components not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

Claims

1. A method of predicting a workload fluctuation situation, characterized by, The method comprises the following steps: obtaining multi-source heterogeneous data of historical influence on task quantity change; controlling and integrating the multi-source heterogeneous data through preprocessing to obtain target data for feature engineering; performing feature engineering on the target data to obtain business rule feature data and space-time correlation feature data; adopting a time series decomposition method to decompose core feature data influencing task quantity change from the business rule feature data and the space-time correlation feature data, wherein the core feature data refers to feature data influencing task quantity change by a preset threshold; fusing the core feature data by a cross-modal fusion method to obtain cross-modal core feature data fusion representation; performing task quantity change influence weight adjustment and update on the cross-modal core feature data fusion representation according to a preset weight adjustment strategy to obtain a cross-modal task quantity fluctuation situation prediction model; real-time collection of the latest multi-source heterogeneous data of influence on task quantity change, combination of the cross-modal task quantity fluctuation situation prediction model, and prediction of task quantity change within a preset time length after a current time node.

2. The task fluctuation situation prediction method according to claim 1, characterized by, The step of obtaining multi-source heterogeneous data of historical influence on task quantity change specifically comprises: analyzing a preset task quantity change influence factor table to obtain all influence factors; identifying factor data acquisition methods corresponding to all influence factors respectively; based on the acquisition methods, collecting historical factor data of influence on task quantity change.

3. The task fluctuation situation prediction method according to claim 1, characterized by, The step of controlling and integrating the multi-source heterogeneous data through preprocessing to obtain target data for feature engineering specifically comprises: detecting invalid data and abnormal data in the multi-source heterogeneous data by an outlier detection algorithm; performing a rejection operation on the invalid data; performing correction processing on the abnormal data by a time series interpolation method or a context correlation compensation method; obtaining the multi-source heterogeneous data after the rejection operation and the correction processing, integrating the multi-source heterogeneous data in a time series manner to obtain the target data.

4. The task fluctuation situation prediction method according to claim 1, characterized by, The step of performing feature engineering on the target data to obtain business rule feature data and space-time correlation feature data specifically comprises: performing feature engineering processing on factor data corresponding to business internal influence factors in the target data to obtain business rule feature data, wherein the factor data corresponding to the business internal influence factors comprises business promotion activity data and business life cycle data; performing feature engineering processing on factor data corresponding to business external influence factors in the target data to obtain space-time correlation feature data, wherein the factor data corresponding to the business external influence factors comprises regional disaster warning data and public opinion topic discussion data.

5. The task fluctuation situation prediction method according to claim 4, characterized by, The step of performing feature engineering on the factor data corresponding to the business internal influence factors in the target data to obtain business rule feature data specifically comprises: performing feature engineering processing on the business promotion activity data and the business life cycle data to identify business rule feature data respectively hit by the business promotion activity data and the business life cycle data; The step of performing feature engineering processing on the factor data corresponding to the external influence factor of the target data to obtain spatio-temporal correlation feature data specifically includes: The regional disaster warning data and the public opinion topic discussion data are subjected to feature engineering processing to identify the spatio-temporal correlation feature data possessed by the regional disaster warning data and the public opinion topic discussion data respectively.

6. The task fluctuation situation prediction method according to claim 1, characterized by, The step of decomposing the core feature data affecting the task quantity change from the business rule feature data and the spatio-temporal correlation feature data by using a time series decomposition method specifically includes: The business rule feature data and the spatio-temporal correlation feature data are respectively decomposed by using a Prophet decomposition component to decompose trend feature data, periodic feature data and holiday effect feature data; The feature data exceeding the corresponding data threshold is selected from the trend feature data, the periodic feature data and the holiday effect feature data as the core feature data affecting the task quantity change by setting the data threshold in advance.

7. The task fluctuation situation prediction method according to claim 6, characterized by, The step of obtaining the cross-modal core feature data fusion representation by fusing the core feature data by using a cross-modal fusion manner specifically includes: The time series feature data extracted for the trend feature data and the periodic feature data is obtained, wherein the time series feature data in the target data is extracted by using a Transformer neural network; The context feature data extracted for the holiday effect feature data is obtained, wherein the context feature data in the target data is extracted by using a CNN convolutional neural network; The core feature data affecting the task quantity change, the time series feature data and the context feature data are fused by using an XGBoost meta-learner to obtain the cross-modal core feature data fusion representation; The step of obtaining the cross-modal task quantity fluctuation situation prediction model by adjusting and updating the weight of the task quantity change of the cross-modal core feature data fusion representation according to a preset weight adjustment strategy specifically includes: The initial weights corresponding to the time series feature data, the context feature data and the core feature data affecting the task quantity change are obtained; The initial weights are adjusted according to the cross-modal core feature data fusion representation result to obtain the task quantity change influence weight corresponding to the cross-modal core feature data fusion representation after adjustment; Based on the task quantity change influence weight corresponding to the cross-modal core feature data fusion representation after adjustment, the Prophet decomposition component, the Transformer neural network and the CNN convolutional neural network are integrated to obtain the cross-modal task quantity fluctuation situation prediction model.

8. A task fluctuation situation prediction device characterized by comprising: It includes: a historical data acquisition module for acquiring historical multi-source heterogeneous data affecting the task quantity change; a target data acquisition module for performing quality control and integration processing on the multi-source heterogeneous data by a preprocessing manner to obtain target data capable of being subjected to feature engineering; The feature engineering module is configured to perform feature engineering on the target data to obtain business rule feature data and space-time correlation feature data. The core feature data obtaining module is configured to decompose core feature data affecting task quantity change from the business rule feature data and the space-time correlation feature data by using a time series decomposition method, wherein the core feature data refers to feature data affecting task quantity change by a preset threshold. The core feature data fusion module is configured to fuse the core feature data by using a cross-modal fusion manner to obtain cross-modal core feature data fusion representation. The task quantity fluctuation situation prediction model obtaining module is configured to perform task quantity change influence weight adjustment update on the cross-modal core feature data fusion representation according to a preset weight adjustment strategy to obtain a cross-modal task quantity fluctuation situation prediction model. The task quantity change situation prediction module is configured to collect latest multi-source heterogeneous data affecting task quantity change in real time, and predict task quantity change situations within a preset time length after a current time node in combination with the cross-modal task quantity fluctuation situation prediction model.

9. A computer device, comprising: The computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by the processor to implement the steps of the task quantity fluctuation situation prediction method in any one of claims 1 to 7.

10. A computer readable storage medium characterized by The computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by the processor to implement the steps of the task quantity fluctuation situation prediction method in any one of claims 1 to 7.

Images