A resource demand prediction method and device, computer equipment and storage medium

Abstract

This application relates to a resource demand forecasting method, apparatus, computer device, and storage medium. The method includes: acquiring a resource demand forecasting request for a target object, and determining a combination of forecasting fields for the target object based on the resource demand forecasting request; determining target knowledge entries matching the forecasting field combination from a resource forecasting knowledge base for the target object; acquiring structured and unstructured resource data of the target object according to the forecasting field combination, and determining fused resource data based on the structured and unstructured resource data; and forecasting the resource demand of the target object according to the resource demand forecasting request based on the target knowledge entries and the fused resource data, thereby obtaining a resource demand forecasting result. This method can comprehensively combine industry knowledge information and multi-source resource data, which is beneficial to improving the accuracy of resource demand forecasting.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a resource demand forecasting method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] Resource demand forecasting plays a crucial role in corporate management and strategic decision-making. Accurate resource budget forecasting can help companies rationally plan resource use, optimize resource allocation, reduce risks, and thus enhance their competitiveness and sustainable development capabilities.

[0003] With the rapid development of information technology, new technologies are constantly being introduced into the field of resource budget forecasting, such as resource budget forecasting based on machine learning, deep learning, large models, and knowledge graphs. However, existing forecasting methods suffer from low accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide a resource demand forecasting method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of resource demand forecasting in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a resource demand forecasting method, the method comprising:

[0006] Obtain a resource demand forecasting request for a target object, and determine a combination of forecasting fields for the target object based on the resource demand forecasting request;

[0007] Target knowledge entries that match the combination of prediction fields are determined from the resource prediction knowledge base for the target object; the target knowledge entries include industry knowledge information associated with the industry to which the target object belongs;

[0008] According to the predicted field combination, obtain the structured resource data and unstructured resource data of the target object, and determine the fused resource data based on the structured resource data and the unstructured resource data;

[0009] Based on the target knowledge entries and the fused resource data, the resource requirements of the target object are predicted according to the resource requirement prediction request, and the resource requirement prediction result is obtained.

[0010] In one embodiment, the step of obtaining structured and unstructured resource data of the target object according to the predicted field combination, and determining fused resource data based on the structured and unstructured resource data, includes:

[0011] Obtain the historical resource data for each resource demand field in the predicted field combination, and determine the structured and unstructured resource data included in the historical resource data for each resource demand field.

[0012] The structured resource data is vectorized to obtain first vector data, and the unstructured resource data is vectorized to obtain second vector data;

[0013] The first vector data and the second vector data are concatenated according to a preset concatenation method to obtain fused resource data.

[0014] In one embodiment, determining the target knowledge entry that matches the combination of prediction fields from the resource prediction knowledge base for the target object includes:

[0015] Obtain a resource prediction knowledge base for the target object; the resource prediction knowledge base includes at least one related knowledge entry associated with the industry to which the target object belongs;

[0016] For each of the associated knowledge entries, determine the semantic similarity between the associated knowledge entry and the combination of the prediction fields;

[0017] The target knowledge entry is determined from the at least one associated knowledge entry based on the semantic similarity of the combination of the at least one associated knowledge entry and the prediction field.

[0018] In one embodiment, determining the combination of prediction fields for the target object based on the resource demand prediction request includes:

[0019] The resource demand forecasting request is parsed to extract at least one key field for forecasting the resource demand of the target object;

[0020] A resource requirement field template for the target object is determined, and the at least one key field is mapped according to the resource requirement field template to obtain a predicted field combination.

[0021] In one embodiment, the method further includes:

[0022] In response to a visualization trigger operation for the resource demand forecast results, determine the visualization method for the resource demand forecast results;

[0023] The resource demand forecast results are visualized according to the visualization method described above.

[0024] In one embodiment, the method further includes:

[0025] In response to the result feedback operation for the resource demand forecast result, obtain the result feedback information for the resource demand forecast result;

[0026] Based on the feedback information, the resource demand prediction model used to predict the resource demand of the target object is updated to obtain the updated resource demand prediction model.

[0027] Secondly, this application also provides a resource demand forecasting device, the device comprising:

[0028] The prediction field determination module is used to obtain a resource demand prediction request for a target object, and determine a combination of prediction fields for the target object based on the resource demand prediction request.

[0029] The knowledge entry matching module is used to determine target knowledge entries that match the combination of prediction fields from the resource prediction knowledge base for the target object; the target knowledge entries include industry knowledge information associated with the industry to which the target object belongs;

[0030] The resource data fusion module is used to obtain structured and unstructured resource data of the target object according to the predicted field combination, and to determine the fused resource data based on the structured and unstructured resource data.

[0031] The resource demand prediction module is used to predict the resource demand of the target object based on the target knowledge item and the fused resource data, according to the resource demand prediction request, and obtain the resource demand prediction result.

[0032] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0033] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0034] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0035] The aforementioned resource demand forecasting method, apparatus, computer equipment, computer-readable storage medium, and computer program product acquire a resource demand forecasting request for a target object and determine a combination of forecasting fields for the target object based on the resource demand forecasting request; determine target knowledge entries that match the forecasting field combination from a resource forecasting knowledge base for the target object; the target knowledge entries include industry knowledge information related to the industry to which the target object belongs; acquire structured and unstructured resource data of the target object according to the forecasting field combination, and determine fused resource data based on the structured and unstructured resource data; and predict the resource demand of the target object according to the resource demand forecasting request based on the target knowledge entries and the fused resource data to obtain the resource demand forecasting result. By determining the forecasting field combination, matching target knowledge entries from the resource forecasting knowledge base, and simultaneously acquiring and fusing structured and unstructured resource data, the resource demand can be predicted based on the target knowledge entries and the fused resource data. This approach comprehensively combines industry knowledge information and multi-source resource data, which is beneficial for improving the accuracy and reliability of resource demand forecasting. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a diagram illustrating the application environment of a resource demand forecasting method in one embodiment.

[0038] Figure 2 This is a flowchart illustrating a resource demand forecasting method in one embodiment;

[0039] Figure 3 This is a flowchart illustrating the steps for determining the fused resource data in one embodiment;

[0040] Figure 4 An application architecture diagram for implementing a resource demand forecasting method in an application example;

[0041] Figure 5 This is a flowchart illustrating the process of processing structured and unstructured data in an application example.

[0042] Figure 6 A flowchart illustrating the process of generating prompts in an application example;

[0043] Figure 7 This is a flowchart illustrating the model optimization process in an application example.

[0044] Figure 8 A flowchart illustrating the process of updating the knowledge base in an application example;

[0045] Figure 9 This is a flowchart illustrating the process of performing hybrid training on a model in an application example.

[0046] Figure 10 This is a structural block diagram of a resource demand prediction device in one embodiment;

[0047] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0050] In related technologies, resource prediction is typically achieved through methods such as traditional machine learning, multi-source data fusion using deep learning, unstructured data analysis of large models, and knowledge graphs and rule engines. Among these:

[0051] Traditional machine learning-based prediction methods typically use algorithms such as linear regression, random forest, and support vector machines to model and predict structured resource data. These methods rely heavily on statistical feature engineering, requiring manual extraction of key indicators (such as mean and variance). They cannot handle unstructured data, such as text (e.g., strategic documents) or images, leading to biased prediction logic. Furthermore, their models have weak generalization ability and limited capacity to model complex nonlinear relationships, making them difficult to adapt to dynamic market environments.

[0052] Multi-source data fusion methods based on deep learning typically use LSTM (Long Short-Term Memory) and Transformer (a deep learning model based on self-attention mechanism) to process time-series data, partially integrating text summary information and fusing multi-source data through feature concatenation or attention mechanisms. However, these methods only extract text summaries and do not perform semantic modeling of the entire text. They lack depth in processing unstructured data, have low information utilization, and are not combined with industry knowledge bases (such as internal bank rules), resulting in a lack of professionalism in prediction results.

[0053] Unstructured data analysis based on large models typically utilizes large models such as GPT (Generative Pre-trained Transformer) and BERT (Bidirectional Encoder Representations from Transformers) to process unstructured data such as text and images, extract key information, and adapt them to specific domain tasks through fine-tuning or prompt word engineering. However, it does not deeply integrate with the data system of the target domain or object, lacks joint modeling of structured data, and the prediction results are difficult to guide actual budgeting. At the same time, the output is mostly labels or values, without forming an executable budgeting plan, and the ability to generate automated suggestions is weak.

[0054] Prediction methods based on knowledge graphs and rule engines typically involve constructing industry knowledge graphs, combining expert rules to optimize prediction results, and using rule engines to correct AI (Artificial Intelligence) predictions to ensure compliance with regulatory requirements. However, these methods only process structured data and do not integrate unstructured information (such as strategic documents), resulting in insufficient data fusion capabilities. Furthermore, they lack a user feedback-driven dynamic update mechanism, preventing continuous improvement in prediction capabilities.

[0055] The resource demand forecasting method provided in this application can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Users can initiate resource demand prediction requests for target objects through terminal 102. Server 104 can obtain these resource demand prediction requests and determine the prediction field combination for the target object based on the request. Server 104 then determines target knowledge entries from the resource prediction knowledge base for the target object that match the prediction field combination. Target knowledge entries include industry knowledge information related to the industry to which the target object belongs. Based on the prediction field combination, server 104 obtains structured and unstructured resource data for the target object, and determines fused resource data based on the structured and unstructured resource data. Server 104 predicts the resource demand of the target object based on the target knowledge entries and fused resource data, according to the resource demand prediction request, obtains the resource demand prediction result, and returns the resource demand prediction result to terminal 102.

[0056] Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0057] In one exemplary embodiment, such as Figure 2 As shown, a resource demand forecasting method is provided, which is then applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 202 to 208. Wherein:

[0058] Step 202: Obtain the resource demand forecasting request for the target object, and determine the forecasting field combination for the target object based on the resource demand forecasting request.

[0059] The target object refers to the object for which resource demand forecasting needs to be performed. The target object can be a specific entity, such as a company, project, or product; or it can be an abstract concept, such as a business area. For example, when forecasting a company's funding needs, the target object is that company.

[0060] A resource demand forecasting request refers to a request initiated by a user or system to predict the future resource needs of a target entity. Resources refer to quantifiable assets, specifically any of the following: digital resources, physical resources, or virtual resources. Digital resources are physical resources that have been digitized and can be stored in a digital resource account. Virtual resources can include, but are not limited to, at least one of various virtual assets, funds, stocks, bonds, virtual avatar products, virtual recharge cards, and game equipment. Physical resources can be tangible items that a user can own, specifically including, but not limited to, houses, vehicles, electronic products, toys, handicrafts, or autographed photos. Resource demand forecasting refers to predicting the amount of resources needed, including the scale of resource demand and the forecast of the demand gap. For example, predicting that a department's funding needs for the next quarter will be 5 million yuan.

[0061] A forecasting field set refers to the set of fields determined based on a resource demand forecasting request and used to predict resource needs. These fields can be various factors or indicators related to the resource needs of the target object. For example, for a company forecasting its financial resource needs, the forecasting field set may include, but is not limited to, the forecasting scope, the forecasting object, the forecasting dimension, and the cash flow gap.

[0062] For example, when a user has a resource demand forecasting need, they can initiate a resource demand forecasting request for a target object through a terminal. For instance, a user can input "Analyze the funding needs of object A in the next quarter" or "Forecast the material shortage in a market segment of region X" through the terminal. The terminal can then generate a funding demand budget request based on the user's input and send it to the server. The server can obtain the resource demand forecasting request for the target object and determine the forecasting field combination based on the request. For example, the server can extract "Target object = object A, Time range = next quarter, Resource type = funding, Forecasting dimension = demand" from "Analyze the funding needs of object A in the next quarter" to obtain the forecasting field combination.

[0063] Step 204: Identify target knowledge entries from the resource prediction knowledge base for the target object that match the combination of prediction fields.

[0064] The resource forecasting knowledge base refers to a database used to store knowledge related to resource forecasting. This knowledge base can include general rules, empirical formulas, historical data patterns, and typical cases related to resource demand forecasting for different industries and target objects. This information can be stored in a specific database in the form of knowledge entries. In practice, each knowledge entry can be determined through expert experience summaries, historical data analysis, industry research reports, etc. Target knowledge entries refer to specific knowledge entries selected from the resource forecasting knowledge base that match the combination of forecasting fields. Target knowledge entries can include industry knowledge information related to the industry to which the target object belongs. In practice, there can be one or more target knowledge entries, and they can also include other useful information related to the forecasting fields. For example, if the target object is a company in the financial industry, the target knowledge entries can include the impact of macroeconomic indicators of the industry on capital demand, resource demand patterns of similar companies within the industry, etc.

[0065] For example, the server can determine the target knowledge entry that matches the combination of prediction fields from the resource prediction knowledge base for the target object. For instance, the server can extract the prediction field combination "target object = object A, time range = next quarter, resource type = funds, prediction dimension = demand amount" based on "analyze the funding needs of object A in the next quarter". It can then retrieve related knowledge entries from the resource prediction knowledge base, such as the historical funding needs data of object A in the corresponding quarter (e.g., Q3), the funding allocation standards of similar projects in the same industry in Q3 of the same year, the market financing interest rate documents for Q3, and project-related strategic planning documents, to determine at least one target knowledge entry that matches the combination of prediction fields.

[0066] Step 206: According to the combination of predicted fields, obtain the structured and unstructured resource data of the target object, and determine the fused resource data based on the structured and unstructured resource data.

[0067] Structured resource data refers to data with a defined structure and format. It can be stored in a relational database and presented in tabular form. Structured resource data typically has fixed fields and types; for example, it may be at least one of the following: basic enterprise information (such as company name, registered capital, number of employees), financial data (such as data from cash flow statements, balance sheets, and profit and loss statements). Unstructured resource data refers to data without a fixed structure and format. It can exist in the form of text, images, audio, video, etc. For example, it may be at least one of the following: strategic documents, news reports, industry research reports, user comments on social media, product manuals, etc. Integrated resource data refers to data obtained by combining structured and unstructured resource data.

[0068] For example, a server can obtain structured and unstructured resource data of a target object based on a combination of prediction fields. For instance, the server can extract structured data such as cash flow statements and budget execution details for object A within a certain time range, as well as unstructured data such as the market environment analysis report for the next quarter, from a database targeting the target object, based on a prediction field combination of "target object = object A, time range = next quarter, resource type = funds, prediction dimension = demand". Then, the server can use data fusion technology to merge the structured and unstructured resource data to obtain fused resource data. For example, the server can extract, transform, and correlate relevant information from the structured and unstructured resource data, such as merging corporate financial data (structured) with information about corporate strategic adjustments in news reports (unstructured) to obtain fused resource data.

[0069] Step 208: Based on the target knowledge items and integrated resource data, predict the resource demand of the target object according to the resource demand prediction request, and obtain the resource demand prediction result.

[0070] The resource demand forecasting result refers to the resource demand result of the target object obtained after analyzing and processing the target knowledge items and integrated resource data according to the resource demand forecasting request. In specific implementation, the resource demand forecasting result can be at least one of the following: a specific numerical value (such as the predicted resource quantity, amount, etc.), a trend chart (showing the trend of resource demand changes over time or other factors), or a probability distribution (representing the probability of different resource demand scenarios occurring).

[0071] For example, a server can predict the resource needs of a target object based on target knowledge items and integrated resource data, according to a resource demand prediction request. For instance, the server can input historical financial data (structured), market environment analysis reports for the next quarter (unstructured), and the funding allocation standards for similar projects in the same industry for the next quarter (target knowledge items) into a resource demand prediction model. The server then uses this model to predict the resource needs of the target object, obtaining the resource demand prediction result. Specifically, the resource demand prediction model can be implemented using at least one of linear regression models, time series models, or machine learning models (such as decision trees and neural networks). By establishing a mapping relationship between integrated resource data, target knowledge items, and the resource demand prediction result, the prediction result is obtained. For example, the resource demand prediction model can dynamically allocate the weights of each structured and unstructured resource data point through an attention mechanism, and combine this with the funding allocation standards for similar projects in the same industry for the next quarter to obtain the funding demand scale of object A in the next quarter, thus obtaining the resource demand prediction result.

[0072] In the aforementioned resource demand forecasting method, a resource demand forecasting request for a target object is obtained, and a combination of forecasting fields for the target object is determined based on the request. Target knowledge entries matching the forecasting field combination are then identified from the resource forecasting knowledge base for the target object. These target knowledge entries include industry knowledge information related to the target object's industry. Structured and unstructured resource data for the target object are acquired according to the forecasting field combination, and fused resource data is determined based on this data. Finally, based on the target knowledge entries and the fused resource data, the resource demand of the target object is predicted according to the resource demand forecasting request, resulting in a resource demand forecasting result. By determining the forecasting field combination, matching target knowledge entries from the resource forecasting knowledge base, and simultaneously acquiring and fusing structured and unstructured resource data, the method ultimately forecasts resource demand based on the target knowledge entries and the fused resource data. This approach comprehensively integrates industry knowledge information and multi-source resource data, which helps improve the accuracy and reliability of resource demand forecasting.

[0073] In one embodiment, such as Figure 3 As shown, according to the combination of prediction fields, structured and unstructured resource data of the target object are obtained, and based on the structured and unstructured resource data, the fused resource data is determined, including:

[0074] Step 302: Obtain the historical resource data for each resource demand field in the prediction field combination, and determine the structured and unstructured resource data included in the historical resource data for each resource demand field.

[0075] The resource requirement field refers to the fields included in the prediction field combination. For example, in the example above, the time range, resource type, and prediction dimension are resource requirement fields. Historical resource data refers to the field values ​​related to the resource requirement field within a past time period. Historical resource data can reflect various information related to the resource requirements of the target object during historical periods. For example, for predicting a company's funding needs, historical resource data may include, but is not limited to, the company's monthly operating revenue, cost expenditure, accounts receivable, and accounts payable data over the past few years.

[0076] For example, the server can obtain historical resource data for each resource requirement field in the prediction field combination. For instance, the server can communicate with a target database storing relevant resource data of the target object via MCP (Multi-Channel Protocol) to obtain the historical resource data for each resource requirement field from the target database. Furthermore, the server can determine the structured and unstructured resource data included in the historical resource data for each resource requirement field. For example, the target database may include a structured database for storing structured resource data and a vector database for storing unstructured resource data. The server can determine the structured resource data from the structured database and the unstructured resource data from the vector database, respectively.

[0077] In an optional embodiment, when transmitting data with the target database, the server can use TLS 1.3 (Transport Layer Security 1.3) for end-to-end encryption and combine it with the AES-256-GCM (Advanced Encryption Standard 256 Galois / Counter Mode) encryption algorithm to encrypt the data to be transmitted, ensuring secure data transmission.

[0078] Step 304: Vectorize the structured resource data to obtain the first vector data, and vectorize the unstructured resource data to obtain the second vector data.

[0079] The first type of vector data refers to the data obtained after vectorizing structured resource data. Since structured data typically has explicit numerical characteristics, the vectorization process converts these numerical characteristics into vector form. For example, a structured dataset containing multiple numerical features (such as sales revenue, cost, and profit) can be represented by combining the feature values ​​of each structured resource data into a vector. For instance, the feature vector of a structured resource data can be represented as [sales revenue value, cost value, profit value]. The second type of vector data refers to the data obtained after vectorizing unstructured resource data. Because unstructured resource data comes in various forms, different methods can be used to vectorize different forms of unstructured resource data. For example, for text data, word embedding techniques can be used to convert each word into a vector, and then the entire text can be represented as a vector through averaging or weighting. For image data, convolutional neural networks can be used to extract feature vectors from the image.

[0080] For example, the server can vectorize structured resource data. Specifically, it can standardize the structured resource data of the target object and then convert it into a fixed-dimensional numerical vector using a feature embedding algorithm to obtain the first vector data. The server can also vectorize unstructured resource data. For instance, it can use Sentence-BERT (a BERT-based sentence embedding model) to semantically encode the unstructured resource data of the target object, generating a high-dimensional semantic vector containing contextual information to obtain the second vector data.

[0081] In an optional embodiment, before vectorizing the structured resource data, the server can perform missing value completion and outlier handling for any missing structured resource data. For example, the server can use KNN (K-Nearest Neighbor) interpolation or time series interpolation to fill in missing data in the structured resource data. In addition, the server can also use Z-Score standardization to identify outliers in the structured resource data. For example, data points exceeding 3σ (3 times the standard deviation) can be identified as outliers. Then, the server can use Winsorization to truncate the outliers in the structured resource data, ensuring the integrity, accuracy, and reasonable distribution of the structured resource data, thereby providing high-quality data support for subsequent vectorization and predictive modeling.

[0082] Furthermore, the server can convert the structured resource data, after missing value completion and outlier handling, into a standardized report in JSON (JavaScript Object Notation) format. For example, the processed structured resource data can be represented as {"Time":"yyyy-mm","Cash Flow":123456}. The server can then vectorize the processed structured resource data to obtain the first vector data.

[0083] Step 306: The first vector data and the second vector data are concatenated according to a preset concatenation method to obtain fused resource data.

[0084] The preset concatenation method refers to the concatenation method predetermined when concatenating the first vector data and the second vector data. For example, the concatenation method may include, but is not limited to, at least one of the following: direct concatenation (simply connecting the two vectors together) and weighted concatenation (assigning different weights according to the importance of the vector data before concatenation).

[0085] For example, the server can determine the concatenation method of the first vector data and the second data vector, such as direct concatenation; the server can directly concatenate the first vector data and the second vector data to obtain fused resource data, such as [first vector data, second vector data]. In some other embodiments, the server can also concatenate the first vector data and the second vector data according to a determined fusion weight, such as the concatenated fused resource data can be represented as [A × first vector data + B × second vector data] (where A is the fusion weight of the first vector data and B is the fusion weight of the second vector data). In specific implementation, the fusion weights of the first vector data and the second vector data can be determined based on the data volume, data dimension, etc. of the first vector data and the second vector data.

[0086] In this embodiment, historical resource data for each resource demand field in the prediction field combination is obtained separately, and structured resource data and unstructured resource data are determined. Then, the structured and unstructured resource data are vectorized and fused to obtain fused resource data. This can realize the integration of different types of resource data, so as to provide a richer, more accurate and unified data foundation for the subsequent prediction process.

[0087] In one embodiment, determining target knowledge entries that match the combination of prediction fields from a resource prediction knowledge base targeting a target object includes:

[0088] Obtain a resource prediction knowledge base for the target object; the resource prediction knowledge base includes at least one related knowledge entry associated with the industry to which the target object belongs; for each related knowledge entry, determine the semantic similarity between the related knowledge entry and the prediction field combination; based on the semantic similarity between the at least one related knowledge entry and the prediction field combination, determine the target knowledge entry from the at least one related knowledge entry.

[0089] Among them, related knowledge entries refer to individual knowledge entries included in the resource prediction knowledge base. Each knowledge entry may contain specific knowledge information related to the industry to which the target object belongs. This knowledge information may include, but is not limited to, general rules, empirical formulas, historical data patterns, and typical cases within the industry. For example, in the financial industry, related knowledge entries may include, but are not limited to, the relationship between macroeconomic indicators (such as GDP growth rate and interest rate level) and the target market index; as another example, in the e-commerce industry, related knowledge entries may include, but are not limited to, the impact patterns of different seasons and promotional activities on product sales volume.

[0090] Semantic similarity is an indicator used to measure the degree of similarity between two pieces of data at the semantic level. For example, in this embodiment, semantic similarity can be used to measure the semantic proximity between the associated knowledge item and the predicted field combination. In specific implementation, semantic similarity can be determined using a word vector model-based method. For example, the word vector model-based method can convert the associated knowledge item and the predicted field combination into vector form, and then determine the cosine similarity between the two to obtain the semantic similarity between the associated knowledge item and the predicted field combination. Semantic similarity can also be determined using a knowledge graph-based method, by analyzing the position and relationship between the associated knowledge item and the predicted field combination in the knowledge graph.

[0091] For example, the server can acquire a resource prediction knowledge base for a target object. For instance, the server can retrieve a resource prediction knowledge base comprised of industry cases, market regulations, historical prediction rules, etc., related to the target object's industry from a vector database associated with the target object. For each associated knowledge entry in the resource prediction knowledge base, the server can determine the semantic similarity between the associated knowledge entry and the prediction field combination. For example, the server can vectorize the prediction field combination (e.g., "target object = object A, time range = next quarter, resource type = funds, prediction dimension = demand") and each associated knowledge entry using the Sentence-BERT model, obtaining a combination vector for the prediction field combination and an entry vector for each associated knowledge entry. The server can then determine the cosine similarity between the combination vector and each entry vector to obtain the semantic similarity between each associated knowledge entry and the prediction field combination. After obtaining the semantic similarity between each associated knowledge entry and the prediction field combination, the server can compare these semantic similarities to determine the target knowledge entry from at least one associated knowledge entry. For example, the server can determine the top N associated knowledge entries with the highest semantic similarity or those with semantic similarity greater than a preset similarity threshold as the target knowledge entry.

[0092] In this embodiment, by acquiring a resource prediction knowledge base for the target object, determining the semantic similarity between each associated knowledge item in the resource prediction knowledge base and the prediction field combination, and then determining the target knowledge item based on the semantic similarity, valuable information for resource prediction can be quickly and accurately filtered out, which is beneficial to improving the accuracy and efficiency of the subsequent prediction process.

[0093] In one embodiment, determining a combination of forecast fields for a target object based on a resource demand forecasting request includes:

[0094] The resource demand forecasting request is parsed to extract at least one key field for forecasting the resource demand of the target object; a resource demand field template for the target object is determined, and the at least one key field is mapped according to the resource demand field template to obtain the forecast field combination.

[0095] Key fields refer to the fields extracted from the resource demand forecasting request that are related to the resource demand of the target object. For example, in a resource demand forecasting request to "analyze the funding needs of object A in the next quarter," key fields such as target object, time range, resource type, and forecasting dimension can be extracted. A resource demand field template is a predefined structural framework used to standardize the fields required for resource demand forecasting. The resource demand field template can include a series of standard fields and their attributes related to resource demand, determined based on the characteristics of the target object and the requirements of resource demand forecasting. For example, in the above funding demand forecasting, in addition to target object, time range, resource type, and forecasting dimension, the resource demand field template can also include fields such as basic information of the target object and the forecasting department. Field mapping is the process of matching and associating the extracted key fields with the corresponding fields in the resource demand field template. Through field mapping, key fields can be integrated according to the specifications of the resource demand field template, allowing the key fields to form a structured data set.

[0096] For example, the server can use natural language processing, regular expression matching, and other methods to parse the resource demand prediction request and extract at least one key field for predicting the resource demand of the target object. For instance, the server can use a natural language processing model to perform semantic parsing on the request "analyze the funding needs of object A in the next quarter" and combine it with regular expressions to extract key fields such as "target object = object A, time range = next quarter, resource type = funding, prediction dimension = demand amount". The server can determine a resource requirement field template for a target object. For example, based on industry-standard budget forecasting specifications and internal enterprise system field definitions, the server can pre-define a field framework including core dimensions such as target object, time range, resource type, forecast dimension, basic information of the target object, and forecasting department to obtain a resource requirement field template. The server can map at least one key field according to the resource requirement field template. For example, the server can map the extracted "target object = object A" to the "target object" field in the template, "next quarter" to the "time dimension," "funds" to the "resource type," and "demand quantity" to the forecast dimension. Furthermore, if the resource requirement forecasting request does not extract basic information of the target object, forecasting department, or other information, the server can match the corresponding field information using a decision tree model or pre-defined rules to obtain a combination of forecast fields. After obtaining the combination of forecast fields, the server can also verify each combination of forecast fields and its corresponding field values ​​to ensure the completeness of the forecast fields, the legality of the field values, and logical consistency, meeting the data input specifications for resource requirement forecasting.

[0097] In an optional embodiment, the user can also directly trigger resource demand prediction through the terminal's interactive interface. For example, the user can upload form data through the terminal. The server can directly extract key fields from the form data. Furthermore, the server can perform data validation and completion, as well as outlier detection, on the extracted key fields to obtain a predicted field combination. For example, the server can predict missing fields using a decision tree model and identify abnormal inputs using an isolated forest algorithm to ensure the completeness of the predicted fields, the legality of the field values, and logical consistency.

[0098] In this embodiment, after parsing the resource demand forecasting request and extracting the key fields, the key fields are mapped according to the resource demand field template to obtain the forecast field combination. This can standardize the data format and unify the processing standards, which is conducive to further improving the accuracy and efficiency of resource demand forecasting.

[0099] In one embodiment, the resource demand forecasting method further includes:

[0100] In response to a visualization trigger action for the resource demand forecast results, determine the visualization method for the resource demand forecast results; and visualize the resource demand forecast results according to the visualization method.

[0101] In this context, visualization triggering operations refer to user-initiated actions that initiate the visualization process of resource demand forecast results. These operations can be performed on the terminal's interactive interface by clicking a specific button (such as the "Visualization Display" button), selecting visualization options from a menu, or entering specific commands. Visualization methods refer to the way resource demand forecast results are presented in intuitive forms such as graphics, charts, text, or reports. Different visualization methods can be used for different data characteristics and user needs; for example, a bar chart can be used to compare the resource demand of different categories or at different time points, while a line chart can be used to analyze the trend of resource demand changes over time.

[0102] For example, after a user obtains resource demand forecast results through a terminal, if there is a need for further analysis of these results, the user can perform a visualization trigger operation through the terminal's interactive interface via dialogue or button triggering. The server can respond to the visualization trigger operation for the resource demand forecast results and determine the visualization method accordingly. For instance, the server can determine the visualization method (e.g., a natural language command to display monthly fund distribution in a dialogue, or a trend chart analysis function selection triggered by a button) based on the user's visualization trigger operation (e.g., a user inputting a natural language command to display monthly fund distribution in a dialogue, or a button triggering a trend chart analysis function selection), combined with the type of resource demand forecast results (e.g., scale data, time series distribution, structural proportion). Optionally, the server can also determine the visualization method based on the user's selection. The server can visualize the resource demand forecast results according to the visualization method. For example, the server can display the monthly fund demand forecast data for object A in the next quarter as a time series trend using a line chart, or present the proportion of fund demand components using a pie chart, and generate an interactive visualization page with data annotations, which is then pushed to the user's terminal for viewing.

[0103] In this embodiment, the corresponding visualization method is determined by responding to the user's visualization trigger operation for the resource demand prediction results, and the resource demand prediction results are visualized according to the determined visualization method. This can meet the visualization needs of different scenarios and users, and present the resource demand prediction results intuitively, making it easier for users to quickly understand and analyze the resource demand prediction results.

[0104] In one embodiment, the resource demand forecasting method further includes:

[0105] In response to the feedback operation on the resource demand forecast results, obtain the feedback information on the resource demand forecast results; based on the feedback information, update the resource demand forecasting model used to forecast the resource demand of the target object, and obtain the updated resource demand forecasting model.

[0106] The result feedback operation is an action initiated by the user to provide feedback on the resource demand forecast results. This operation can be performed by clicking specific feedback buttons (such as "Accurate," "Inaccurate," "Questionable," "Accept," "Ignore," etc.) on the terminal's interactive interface, filling out a feedback form, or entering text comments. The result feedback information refers to the specific content expressed by the user through this operation. This information can be a simple binary evaluation (such as "Accurate" or "Inaccurate") or a detailed textual description, including an analysis of deviations in the resource demand forecast results, factors affecting forecast accuracy, and the differences between actual and forecasted demand. For example, the result feedback information could be that the predicted funding demand was 20% higher than the actual demand, mainly due to inadequate staffing.

[0107] For example, after a user obtains the resource demand forecast result through a terminal, if they have any opinions on the result, they can perform a feedback operation through the terminal's interactive interface via dialogue or button triggering. The server can respond to the feedback operation and obtain the feedback information. For example, the server can receive text feedback from the user via dialogue, such as "The predicted funding demand is 20% higher than the actual demand," or structured feedback information such as specific adjustment values ​​and deviation reason tags submitted via feedback buttons, or positive or negative feedback submitted via feedback buttons (such as clicking the "accept" or "ignore" button). The server can update the resource demand forecasting model used to forecast the resource demand of the target object based on the feedback information, obtaining an updated resource demand forecasting model. For example, the server can label the feedback information as supervised samples, combine it with the original forecast data (integrated resource data and target knowledge items) to incrementally fine-tune the resource demand forecasting model, or adjust the fusion weights of multi-source data (such as increasing the weight of certain knowledge items) to obtain an updated resource demand forecasting model.

[0108] In some optional embodiments, the server can also periodically execute the data pipeline through Apache Airflow (Apache workflow scheduling platform), clean the result feedback information and store it into the vector database after generating new related knowledge entries, so as to update the resource prediction knowledge base.

[0109] In this embodiment, by obtaining the result feedback information of the resource demand prediction results and updating the resource demand prediction model based on the result feedback information, the model can be continuously optimized, which is conducive to further improving the accuracy of the resource demand prediction of the target object.

[0110] In one application example, a resource demand forecasting method is provided, such as... Figure 4As shown, this method can be applied to architectures that include an input layer, a large model processing layer, a feedback layer, and a data tuning layer. Specifically, the method in this example can include the following procedures:

[0111] For the input layer, which serves as a multimodal interaction entry point, resource demand prediction can be initiated through dialogue interaction or button clicks.

[0112] Taking a company's funding gap for the next quarter as an example, in a dialogue interaction, when a user inputs "analyze the funding gap of headquarters for the next quarter" (i.e., a resource demand forecast request) through the terminal, the input layer can use natural language parsing to parse the resource demand forecast request. For example, the input layer can use a bidirectional Transformer model, such as BERT or RoBERTa (Robustly Optimized BERT Pretraining Approach), combined with historical dialogue records to perform context-aware semantic analysis. Afterward, the input layer can use TF-IDF (Term Frequency-Inverse Document Frequency) combined with rule matching to extract key fields from the context information (such as "funding forecast", "next quarter", "gap", etc.) and map the extracted key fields to resource forecast field templates (such as time range, department, forecast dimension). Afterwards, the input layer can use a rule engine to validate the validity of each resource data (i.e., field value) in the resource prediction field template (e.g., the time range cannot exceed 365 days from the current date). Furthermore, the input layer can dynamically complete the field values ​​of certain prediction fields based on contextual information. For example, if the user does not specify a department, the input layer can call a collaborative filtering algorithm to predict the default department, or make predictions based on the user's historical operation records. The resource prediction field template is a combined JSON parameter format. The specific content is combined based on the current user's unit, the input content, and the user's backend data to generate a parameter format for subsequent processing. For example, the resource prediction field template includes a department. If the user does not enter a department, it needs to be automatically completed, which the subsequent prediction engine will need. There are multiple ways to complete the resource prediction field template. If the customer has never made a prediction before, it will be completed based on the current user's unit and department. If the customer's registered department is Department A, then the parameter will be completed as Department A. If the user has specified a prediction for a certain department multiple times, then the record will be searched from the user's historical data and completed. The mapping completion method for the time range parameter in the resource prediction field template is to complete it by predicting the time range of the next quarter. The automatic completion of the prediction dimension is mapped into the resource situation.

[0113] For example, when a user enters a resource demand forecast request to "analyze the funding gap of the headquarters in the next quarter", the input layer can output a resource forecast field template as {"Time Range":"Next Quarter","Department":"Headquarters","Forecast Dimension":"Cash Flow Gap"}.

[0114] In the button-triggered mode, users can click the "Predict" button on the terminal's interactive page and submit form data (such as department, time range, and prediction dimension). The input layer can predict missing fields using a decision tree model (e.g., if the user hasn't selected a department, the input layer can set the department field to "Headquarters"). The input layer can also use the Isolation Forest algorithm to identify abnormal inputs (e.g., time ranges exceeding system support). Afterward, the input layer can invoke reinforcement learning strategies to dynamically adjust parameter priorities (e.g., if the user frequently selects "Department A," the prediction results for Department A will be displayed first by default), thus obtaining the resource prediction field template. The form data is from the resource planning page; after the button is clicked, the resource planning data from the currently viewed page can be directly submitted.

[0115] For the large model processing layer, multimodal data fusion and knowledge base integration can be achieved.

[0116] like Figure 5As shown, the large model processing layer can achieve real-time data communication between the treasury database (target database) and the large model system (resource demand forecasting model) through the MCP protocol. During data communication, end-to-end encryption can be performed using the TLS 1.3 protocol, combined with the AES-256-GCM encryption algorithm to protect data transmission. Then, the large model processing layer can obtain structured resource data (such as monthly cash flow statements and balance sheets) from the treasury database, and unstructured resource data (such as strategic documents and news texts) and resource forecasting knowledge bases (such as industry trends, forecasting experience, and case studies) from the vector database. After obtaining the structured and unstructured resource data, the large model processing layer can process the structured and unstructured resource data separately. For structured resource data, the large model processing layer can use KNN interpolation or time series interpolation to fill in missing data, and identify outliers (such as data points exceeding the 3σ range) in the structured resource data through Z-Score normalization, and use Winsorization to truncate outliers. After that, the large model processing layer can convert the cleaned data into a standardized report in JSON format (e.g., {"time":"yyyy-mm","cash flow":123456}). For unstructured resource data, the large model processing layer can use Sentence-BERT to convert strategic documents and news text into vector representations (e.g., [0.12,0.45,...,0.87]). Finally, the large model processing layer can employ a large model with 70B parameters, combined with mixed precision training to improve computational efficiency. Specifically, the large model processing layer can concatenate structured report data (first vector data) and unstructured vector data (second vector data) into a mixed input (e.g., [report vector, text vector], i.e., fused resource data), and dynamically allocate weights for different data sources (i.e., first vector data and second vector data) through an attention mechanism. At the same time, the large model processing layer uses the Faiss (Facebook AISimilarity Search) vector library to achieve approximate nearest neighbor search, matching knowledge base entries (i.e., target knowledge entries, such as "market expansion case in XX region") through cosine similarity, and injecting the retrieved target knowledge entries (e.g., "market expansion case in XX region in year X") into the intermediate layer of the large model to enhance the professionalism of the prediction logic.

[0117] like Figure 6As shown, the large model processing layer can also dynamically generate prompts by combining user input parameters with vector database content. For example, when a user inputs "XX reads market expansion", the large model processing layer can retrieve "market changes in XX region in year X" from the knowledge base and generate prompts, namely "based on user strategic document A and industry dynamics B, predict the funding needs of XX market in the next quarter", and obtain the corresponding resource demand forecast request, so that the large model processing layer can perform subsequent resource demand forecasts on this basis.

[0118] The feedback layer can display resource demand prediction results across multiple scenarios. Specifically, the feedback layer can provide visual feedback on the resource demand prediction results through interactive dialogue or button clicks.

[0119] For example, when using interactive visualization, the feedback layer can use a T5 model to generate natural language summaries (e.g., "The expected cash flow deficit for next quarter is XX million yuan; it is recommended to prioritize adjusting the budget of Department A"). Then, the feedback layer can display key data (e.g., |time|predicted value|actual value|difference rate|) in Markdown format, and the terminal's interactive interface supports clicking "details" to view the data source. Alternatively, when using button-based visualization, visualization charts, forms, and pop-ups can be used, such as generating interactive line charts (showing quarterly cash flow trends) and heatmaps (showing departmental cash flow distribution), or dynamically rendering tabular data through functional components.

[0120] For the data tuning layer, model optimization can be driven by user feedback.

[0121] like Figure 7As shown, users can provide positive / negative feedback on the model's output (i.e., resource demand prediction results) through the interactive interface on the terminal. For example, users can click the "Accept" or "Ignore" button in the interactive interface, and the data optimization layer can record feedback labels (e.g., {"Prediction ID":"12345","Feedback":"Positive"}). If the feedback is "negative," the data optimization layer can call SHAP (SHapley Additive exPlanations) to analyze and locate the cause of prediction bias (e.g., "User strategy document was not correctly parsed") to obtain result feedback information. Furthermore, the data optimization layer can merge user feedback data (i.e., result feedback information) with historical data, incrementally train the resource demand prediction model through the PyTorch Lightning framework, and use the AdamW (Adaptive Moment Estimation with Weight Decay) optimizer to dynamically adjust the learning rate (e.g., increasing the learning rate by 50% when there is a lot of negative feedback).

[0122] Furthermore, such as Figure 8 As shown, the data optimization layer can also update the resource prediction knowledge base. For example, it can use Apache Airflow to periodically execute data pipelines, clean up new feedback content (such as "market changes in XX region") and store it in a vector database. It can also use Sentence-BERT to revectorize new knowledge and update the knowledge base index using Faiss's IVF-PQ (Inverted File with Product Quantization, a hybrid method of inverted index and product quantization).

[0123] Optionally, this example can also dynamically adjust the context weight based on the confidence level of the user input (such as keyword matching degree) when determining key fields (e.g., when the confidence level is >80%, the weight of historical dialogue can be increased to 70% when determining key fields). At the same time, targeted prompt words can be generated by tracking the dialogue status through multiple rounds of dialogue interaction (e.g., when the user asks "Why is there a large funding gap in Department A?").

[0124] Optionally, when updating the resource prediction knowledge base, the update method of the resource prediction knowledge base can be defined (such as "when user feedback contains 'changes in normative documents', the knowledge base update is automatically triggered"). At the same time, real-time synchronization between user feedback and the resource prediction knowledge base can be achieved through a message queue with a delay of <1 second.

[0125] Optional, such as Figure 9As shown, in the large model prediction stage, FP16 (Floating Point 16) can be used to accelerate the calculation, while the key layers retain the accuracy of FP32 (Floating Point 32) to balance speed and accuracy.

[0126] Using the method described in this example, context-aware prompt generation technology enables joint modeling of structured and unstructured information, improving prediction accuracy by over 30%. Furthermore, real-time knowledge base updates driven by user feedback ensure that the prediction logic is synchronized with market changes, reducing prediction response time to within 5 minutes. The entire process, from data input to budget suggestion generation, is fully automated, reducing manual intervention and improving the efficiency of enterprise financial management.

[0127] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to 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 embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0128] Based on the same inventive concept, this application also provides a resource demand forecasting apparatus for implementing the resource demand forecasting method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more resource demand forecasting apparatus embodiments provided below can be found in the limitations of the resource demand forecasting method described above, and will not be repeated here.

[0129] In one exemplary embodiment, such as Figure 10 As shown, a resource demand forecasting device is provided, comprising: a forecasting field determination module 1002, a knowledge item matching module 1004, a resource data fusion module 1006, and a resource demand forecasting module 1008, wherein:

[0130] The prediction field determination module 1002 is used to obtain a resource demand prediction request for a target object and determine a combination of prediction fields for the target object based on the resource demand prediction request.

[0131] The knowledge item matching module 1004 is used to determine the target knowledge items that match the combination of prediction fields from the resource prediction knowledge base for the target object; the target knowledge items include industry knowledge information related to the industry to which the target object belongs;

[0132] The resource data fusion module 1006 is used to obtain structured and unstructured resource data of the target object according to the combination of prediction fields, and to determine the fused resource data based on the structured and unstructured resource data.

[0133] The resource demand prediction module 1008 is used to predict the resource demand of the target object based on the target knowledge items and integrated resource data, according to the resource demand prediction request, and obtain the resource demand prediction result.

[0134] In an optional embodiment, the prediction field determination module 1002 is further configured to parse the resource demand prediction request, extract at least one key field for predicting the resource demand of the target object, determine a resource demand field template for the target object, and map at least one key field according to the resource demand field template to obtain a prediction field combination.

[0135] In an optional embodiment, the knowledge item matching module 1004 is further configured to acquire a resource prediction knowledge base for the target object; the resource prediction knowledge base includes at least one associated knowledge item related to the industry to which the target object belongs; for each associated knowledge item, the semantic similarity between the associated knowledge item and the prediction field combination is determined; and the target knowledge item is determined from the at least one associated knowledge item based on the semantic similarity between the at least one associated knowledge item and the prediction field combination.

[0136] In an optional embodiment, the resource data fusion module 1006 is further configured to acquire the historical resource data of each resource demand field in the prediction field combination, and determine the structured resource data and unstructured resource data included in the historical resource data of each resource demand field; vectorize the structured resource data to obtain first vector data, and vectorize the unstructured resource data to obtain second vector data; and concatenate the first vector data and the second vector data according to a preset concatenation method to obtain fused resource data.

[0137] In an optional embodiment, the resource demand forecasting apparatus further includes a visualization module, configured to determine a visualization method for the resource demand forecasting results in response to a visualization trigger operation for the resource demand forecasting results; and to visualize the resource demand forecasting results according to the visualization method.

[0138] In an optional embodiment, the resource demand forecasting device further includes a forecast result feedback module, which is used to obtain result feedback information on the resource demand forecast result in response to a result feedback operation on the resource demand forecast result; and update the resource demand forecasting model used to forecast the resource demand of the target object based on the result feedback information to obtain an updated resource demand forecasting model.

[0139] Each module in the aforementioned resource demand forecasting device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0140] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores field data, structured resource data, unstructured resource data, and a resource prediction knowledge base. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a resource demand prediction method.

[0141] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0142] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the resource requirement prediction methods of the above embodiments.

[0143] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the resource requirement prediction methods of the above embodiments.

[0144] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the resource requirement prediction methods of the above embodiments.

[0145] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0146] 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 a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0147] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0148] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A resource demand forecasting method, characterized in that, The method includes: Obtain a resource demand forecasting request for a target object, and determine a combination of forecasting fields for the target object based on the resource demand forecasting request; Target knowledge entries that match the combination of prediction fields are determined from the resource prediction knowledge base for the target object; the target knowledge entries include industry knowledge information associated with the industry to which the target object belongs; According to the predicted field combination, obtain the structured resource data and unstructured resource data of the target object, and determine the fused resource data based on the structured resource data and the unstructured resource data; Based on the target knowledge entries and the fused resource data, the resource requirements of the target object are predicted according to the resource requirement prediction request, and the resource requirement prediction result is obtained.

2. The method according to claim 1, characterized in that, The step of obtaining structured and unstructured resource data of the target object according to the predicted field combination, and determining fused resource data based on the structured and unstructured resource data, includes: Obtain the historical resource data for each resource demand field in the predicted field combination, and determine the structured and unstructured resource data included in the historical resource data for each resource demand field. The structured resource data is vectorized to obtain first vector data, and the unstructured resource data is vectorized to obtain second vector data; The first vector data and the second vector data are concatenated according to a preset concatenation method to obtain fused resource data.

3. The method according to claim 1, characterized in that, The step of determining the target knowledge entry that matches the combination of prediction fields from the resource prediction knowledge base for the target object includes: Obtain a resource prediction knowledge base for the target object; the resource prediction knowledge base includes at least one related knowledge entry associated with the industry to which the target object belongs; For each of the associated knowledge entries, determine the semantic similarity between the associated knowledge entry and the combination of the prediction fields; The target knowledge entry is determined from the at least one associated knowledge entry based on the semantic similarity of the combination of the at least one associated knowledge entry and the prediction field.

4. The method according to claim 1, characterized in that, The step of determining the combination of prediction fields for the target object based on the resource demand prediction request includes: The resource demand forecasting request is parsed to extract at least one key field for forecasting the resource demand of the target object; A resource requirement field template for the target object is determined, and the at least one key field is mapped according to the resource requirement field template to obtain a predicted field combination.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: In response to a visualization trigger operation for the resource demand forecast results, determine the visualization method for the resource demand forecast results; The resource demand forecast results are visualized according to the visualization method described above.

6. The method according to any one of claims 1 to 4, characterized in that, The method further includes: In response to the result feedback operation for the resource demand forecast result, obtain the result feedback information for the resource demand forecast result; Based on the feedback information, the resource demand prediction model used to predict the resource demand of the target object is updated to obtain the updated resource demand prediction model.

7. A resource demand forecasting device, characterized in that, The device includes: The prediction field determination module is used to obtain a resource demand prediction request for a target object, and determine a combination of prediction fields for the target object based on the resource demand prediction request. The knowledge entry matching module is used to determine target knowledge entries that match the combination of prediction fields from the resource prediction knowledge base for the target object; the target knowledge entries include industry knowledge information associated with the industry to which the target object belongs; The resource data fusion module is used to obtain structured and unstructured resource data of the target object according to the predicted field combination, and to determine the fused resource data based on the structured and unstructured resource data. The resource demand prediction module is used to predict the resource demand of the target object based on the target knowledge item and the fused resource data, according to the resource demand prediction request, and obtain the resource demand prediction result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images