An intelligent budget management method and system, an electronic device, and a storage medium

Abstract

The present application relates to the technical field of artificial intelligence, and more particularly to an intelligent budget management method and system, an electronic device and a storage medium. By fusing historical data and real-time data containing external environment scenarios, business progress and resource status, a prediction model is used to dynamically generate scenario feature weights and budget target prediction values, so that the budget target can adapt to external changes such as policies and competitors, significantly reducing prediction bias. At the same time, the decomposition model calculates the initial decomposition coefficients of each unit according to the resource status data and the scenario feature weights, realizing the linkage calibration of prediction and decomposition, making the budget index allocation match the actual resource carrying capacity of each unit, improving the scientificity of the decomposition scheme and the resource utilization efficiency, and improving the intelligent level of budget management.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent budget management method, system, electronic device, and storage medium. Background Technology

[0002] In the budget management of large enterprises such as telecom operators, budget targets typically cover multiple organizational levels, including company, branch, grid, regional, and professional groups, involving numerous indicators and various roles. Traditional budget target setting is often based on simple extrapolation of historical data, with budget targets decomposed into fixed proportions from top to bottom across multiple levels. This approach fails to perceive dynamic changes in the external environment or the actual resource capacity of each execution unit, resulting in significant discrepancies between budget targets and the actual operating environment. Furthermore, the decomposed indicators do not match the execution capabilities of lower-level units, leading to both resource waste and pressure to meet targets.

[0003] To address these issues, some companies have attempted to introduce data analytics tools or basic AI models to assist budget management. However, existing models often employ fixed feature weights, failing to dynamically adjust the importance of influencing factors based on changing scenarios. Furthermore, the data sources used are often limited, typically relying solely on historical budget data from within the company, resulting in insufficient responsiveness to sudden changes and inappropriate indicator allocation. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent budget management method, system, electronic device and storage medium to solve the above-mentioned technical problem.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: An intelligent budget management method, applied to a terminal device, the method comprising: acquiring historical data and real-time data of the current budget period; wherein, the historical data includes historical data of budget indicators and historical business data; the real-time data includes external environment scenario data, business progress data, and resource status data of each budget unit; performing feature engineering processing based on the historical data and the real-time data to generate feature vector data; inputting the feature vector data into a pre-trained prediction model to generate scenario feature weights and budget target prediction values, wherein the budget target prediction values ​​are budget indicator values ​​for the next budget period; calculating the initial decomposition coefficients of each budget unit through a pre-trained decomposition model based on the feature vector data and the scenario feature weights, wherein the initial decomposition coefficients of each budget unit represent the proportion of the budget target prediction values ​​allocated to the budget unit; and adjusting the budget indicator allocation of each budget unit based on the budget target prediction values, the initial decomposition coefficients of each budget unit, and the resource status data of each budget unit to generate a budget indicator allocation scheme for each budget unit.

[0006] The beneficial effects of this invention are as follows: By integrating historical data with real-time data including external environment scenarios, business progress, and resource status, a predictive model dynamically generates scenario feature weights and budget target predictions. This allows budget targets to adapt to external changes such as policies and competitors, significantly reducing prediction bias. Simultaneously, the decomposition model calculates the initial decomposition coefficients for each unit based on resource status data and scenario feature weights, achieving linked calibration between prediction and decomposition. This ensures that budget indicator allocation matches the actual resource carrying capacity of each unit, improving the scientific nature of the decomposition scheme and resource utilization efficiency, and enhancing the intelligence level of budget management.

[0007] Based on the above technical solution, the present invention can be further improved as follows.

[0008] Furthermore, the step of performing feature engineering processing based on the historical data and the real-time data to generate feature vector data includes: performing data cleaning and standardization processing on the historical data and the real-time data; performing scene labeling on the processed external environment scene data and extracting scene features; performing feature extraction processing based on the processed historical data and the processed real-time data to obtain multi-dimensional features, wherein the multi-dimensional features include at least the resource carrying capacity index and resource idle rate of each budget unit; and performing feature concatenation processing based on the scene features and the multi-dimensional features to generate feature vector data.

[0009] Furthermore, the pre-trained prediction model includes an LSTM network, a scene-weight mapping sub-model, and an XGBoost model; the step of inputting the feature vector data into the pre-trained prediction model to generate scene feature weights and budget target prediction values ​​includes: extracting temporal latent features using the LSTM network based on the historical data; performing mapping processing using the scene-weight mapping sub-model based on the feature vector data to generate scene feature weights; and performing prediction processing using the XGBoost model based on the temporal latent features extracted by the LSTM network, the real-time data, and the scene feature weights to obtain the budget target prediction value.

[0010] Furthermore, the decomposition model includes a clustering module, a carrying capacity assessment sub-model, and a random forest regressor; the step of calculating the initial decomposition coefficients of each budget unit using the pre-trained decomposition model based on the feature vector data and the scene feature weights includes: dividing each budget unit into different potential categories using the clustering module; assessing the carrying capacity of each budget unit using the carrying capacity assessment sub-model based on the resource carrying capacity index of each budget unit in the feature vector data to obtain a carrying capacity score for each budget unit; and calculating the initial decomposition coefficients of each budget unit using the random forest regressor based on the potential category and carrying capacity score of each budget unit.

[0011] Furthermore, after calculating the initial decomposition coefficients of each budget unit using the random forest regressor based on the potential category and carrying capacity score of each budget unit, the method further includes: for each budget unit, comparing the carrying capacity score of the budget unit with a carrying capacity assessment threshold; when the carrying capacity score of the budget unit is lower than the carrying capacity assessment threshold, reducing the initial decomposition coefficients of the budget unit; wherein, the carrying capacity assessment threshold is dynamically adjusted based on the scene feature weights.

[0012] Furthermore, the step of adjusting the budget indicator allocation for each budget unit based on the predicted budget target value, the initial decomposition coefficients of each budget unit, and the resource status data of each budget unit to generate a budget indicator allocation scheme for each budget unit includes: receiving a manual adjustment value for the predicted budget target value, and generating a final budget target value based on the manual adjustment value; generating a decomposition template based on the final budget target value and the initial decomposition coefficients of each budget unit, and distributing the decomposition template to each budget unit; performing multi-level collaborative decomposition and cross-level resource scheduling processing based on the decomposition template to match the budget indicators of each budget unit with the resource status data of each budget unit, and generating an initial budget indicator allocation scheme for each budget unit; and receiving manual adjustment information for the initial budget indicator allocation scheme of each budget unit, and generating a budget indicator allocation scheme for each budget unit based on the manual adjustment information.

[0013] Furthermore, the method also includes: collecting decomposed execution data of each budget unit in real time during the execution of each budget indicator allocation scheme in each budget unit; using the decomposed execution data of each budget unit, employing the isolated forest algorithm combined with scenario-based dynamic thresholds for anomaly detection; when an anomaly is detected, identifying continuously occurring related scenarios through time series analysis to form a scenario chain, and tracing the cumulative impact of each related scenario on the anomaly based on the scenario chain to generate a scenario chain impact report; generating and pushing early warning information based on the scenario chain impact report, wherein the early warning information includes anomaly indicators, deviation values, scenario chain attribution conclusions, and resource scheduling suggestions.

[0014] Furthermore, the method also includes: acquiring execution effect data after abnormal intervention, wherein the execution effect data includes at least the magnitude of the decrease in deviation rate; evaluating in real time whether the magnitude of the decrease in deviation rate reaches a preset threshold, and periodically reviewing and evaluating multiple performance dimensions of budget management to obtain review and evaluation results; when the magnitude of the decrease in deviation rate reaches the preset threshold, updating the prediction model and the decomposition model, and storing the intervention plan and scheduling path corresponding to this intervention in a preset case library; when the review and evaluation results indicate that any performance dimension fails to meet the standard, triggering model adjustments for the prediction model and the decomposition model.

[0015] To address the aforementioned technical problems, the present invention also provides an intelligent budget management system, comprising: The data acquisition module is used to acquire historical data and real-time data for the current budget period; wherein, the historical data includes historical data of budget indicators and historical business data; the real-time data includes external environment scenario data, business progress data, and resource status data of each budget unit; The feature processing module is used to perform feature engineering processing based on the historical data and the real-time data to generate feature vector data; The prediction module is used to input the feature vector data into a pre-trained prediction model to generate scene feature weights and budget target prediction values, wherein the budget target prediction values ​​are budget indicator values ​​for the next budget period. The decomposition module is used to calculate the initial decomposition coefficients of each budget unit based on the feature vector data and the scene feature weights through a pre-trained decomposition model. The initial decomposition coefficients of each budget unit represent the proportion of the budget target prediction value allocated to the budget unit. The allocation module is used to adjust the allocation of budget indicators for each budget unit based on the predicted budget target value, the initial decomposition coefficient of each budget unit, and the resource status data of each budget unit, and to generate a budget indicator allocation scheme for each budget unit.

[0016] To address the aforementioned technical problems, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent budget management method as described above.

[0017] To address the aforementioned technical problems, the present invention also provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute an intelligent budget management method as described above. Attached Figure Description

[0018] Figure 1 This is a flowchart of an intelligent budget management method according to the present invention; Figure 2 This is a schematic diagram of an intelligent budget management system according to the present invention; Figure 3 This is a schematic diagram of an electronic device according to the present invention. Detailed Implementation

[0019] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0020] This invention addresses the multi-level, multi-scenario, and highly dynamic characteristics of operator budget management by proposing a full-link solution that includes dual-model linkage calibration, cross-level resource scheduling, scenario chain attribution, and real-time feedback optimization. It constructs an intelligent budget management system that better meets the business needs of operators and overcomes the core deficiencies of existing technologies in dynamic response, global collaboration, closed-loop optimization, and accurate attribution.

[0021] Example 1 like Figure 1 As shown, this embodiment provides an intelligent budget management method applied to a terminal device. The method includes: S101. Obtain historical data and real-time data for the current budget period; wherein, the historical data includes historical data of budget indicators and historical business data; the real-time data includes external environment scenario data, business progress data, and resource status data of each budget unit.

[0022] Historical data includes historical data of budget indicators over a previous period (such as budget indicator allocation value, breakdown value, and completion value) and historical business data (such as user growth rate and revenue completion rate).

[0023] Real-time data includes external environment scenario data (such as policy documents, competitor dynamics, regional GDP, industry prosperity, etc.), business progress data (actual completion value in the current period), and resource status data of each budget unit (such as manpower availability, channel coverage density, and resource idle status of each grid).

[0024] The aforementioned historical and real-time data can be collected through ETL tools, API interfaces, or web crawlers.

[0025] S102. Perform feature engineering processing based on the historical data and the real-time data to generate feature vector data.

[0026] S103. Input the feature vector data into the pre-trained prediction model to generate scene feature weights and budget target prediction values, wherein the budget target prediction values ​​are budget indicator values ​​for the next budget period.

[0027] S104. Based on the feature vector data and the scene feature weights, calculate the initial decomposition coefficients of each budget unit using a pre-trained decomposition model. The initial decomposition coefficients of each budget unit represent the proportion of the budget target prediction value allocated to the budget unit.

[0028] S105. Based on the predicted budget target value, the initial decomposition coefficient of each budget unit, and the resource status data of each budget unit, adjust the budget indicator allocation of each budget unit to generate a budget indicator allocation scheme for each budget unit.

[0029] Existing methods generally rely on linear extrapolation of historical data, and have not established a structured association mechanism between external variables such as policies, markets and supply chains and budget items. When encountering non-steady-state events such as sudden policy changes and competitive product impacts, target parameters cannot be dynamically adjusted, which causes the budget target to seriously deviate from the actual operating environment. In special scenarios, the deviation rate often exceeds 30%, losing its guiding significance.

[0030] The method provided by the present invention fuses historical data and real-time data including external environment scenarios, business progress and resource status, and uses a prediction model to dynamically generate scenario feature weights and budget target prediction values, enabling budget targets to adapt to external changes such as policies and competitive products, and significantly reducing prediction deviations. Meanwhile, a decomposition model calculates the initial decomposition coefficient of each unit based on resource status data and scenario feature weights, realizing linkage calibration of prediction and decomposition, matching budget indicator allocation with the actual resource carrying capacity of each unit, improving the scientificity of decomposition schemes and resource utilization efficiency, and enhancing the intelligent level of budget management.

[0031] Optionally, in an embodiment, said performing feature engineering processing according to said historical data and said real-time data to generate feature vector data includes: performing data cleaning and standardization processing on said historical data and said real-time data; performing scenario labeling on the processed external environment scenario data to extract scenario features; performing feature extraction processing according to the processed historical data and the processed real-time data to obtain multi-dimensional features, wherein said multi-dimensional features at least include the resource carrying capacity index and resource idle rate of each budget unit; performing feature splicing processing according to said scenario features and said multi-dimensional features to generate feature vector data.

[0032] Specifically, data cleaning mainly removes null values and abnormal values; standardization processing mainly includes normalizing data and unifying indicator calibers. Then, scenario labeling is performed on the processed external environment scenario data to extract scenario features. In this embodiment, the BERT-NLP algorithm is used to perform scenario labeling on the processed external environment scenario data, which mainly labels scenario tags for policy-related data and competitive product-related data therein (such as policy favorable, competitive product impact, resource shortage), and extracts scenario features. Scenario features include scenario tags, as well as policy impact cycle, competitive product impact intensity, scenario occurrence sequence, etc.

[0033] Meanwhile, the resource carrying capacity index and resource idle rate of each budget unit are calculated according to resource status data and business historical data, providing core evaluation indicators for the decomposition model and resource scheduling. Specifically, the resource carrying capacity index is represented as human resources × channels / historical indicator completion rate, and the resource idle rate is represented as idle resource volume / total resource volume. In addition to the resource carrying capacity index and resource idle rate, other features can be extracted and constructed according to actual requirements to jointly form multi-dimensional features. Feature vector data is obtained based on scenario features and multi-dimensional features.

[0034] Optionally, in an embodiment, the pre-trained prediction model includes an LSTM network, a scene-weight mapping sub-model, and an XGBoost model; the step of inputting the feature vector data into the pre-trained prediction model to generate scene feature weights and budget target prediction values ​​includes: extracting temporal latent features using the LSTM network based on the historical data; performing mapping processing using the scene-weight mapping sub-model based on the feature vector data to generate scene feature weights; and performing prediction processing using the XGBoost model based on the temporal latent features extracted by the LSTM network, the real-time data, and the scene feature weights to obtain the budget target prediction value.

[0035] In this embodiment, the pre-trained prediction model is trained in the following manner: Three years of historical data were collected as the training dataset, including historical budget indicator data, historical business data, historical external environment scenario data, and historical resource status data. Specifically, historical budget indicator data mainly includes monthly / quarterly allocation values, breakdown values, and completion values. Historical business data mainly includes user growth rate, revenue achievement rate, etc. Historical external environment scenario data mainly includes policy documents (such as extracted keywords "5G subsidy," "package discounts," etc.) and competitor dynamics (such as new store openings, promotional activities, package prices), etc. Historical resource status data mainly includes historical manpower and channel coverage density for each budget unit.

[0036] For the above data, the same steps of data cleaning, standardization, scene feature extraction, feature extraction and processing, and feature concatenation are used to construct input features, which correspond to real labels, i.e., the actual completed values ​​of the next period.

[0037] LSTM networks are used to extract temporal latent features from historical budget indicator sequences. Past budget indicator sequences are used as input to the LSTM, which outputs a fixed-dimensional latent state vector representing the sequence's trend, seasonality, and periodicity. During training, the LSTM is jointly trained with a subsequent XGBoost model, and the loss function can be either the mean squared error or the mean absolute percentage error between predicted and true values.

[0038] The scene-weight mapping sub-model can employ a lightweight neural network (such as a 2-3 layer fully connected network), taking feature vector data as input and outputting weight vectors for each relevant feature. This sub-model is trained using labeled data. The specific construction method for the labeled data is as follows: The historical time window is divided into multiple scenario periods (such as policy benefit periods, competitor price war periods, and routine periods), obtained through business rules or clustering. For each scenario period, a set of feature weights is found using grid search or Bayesian optimization methods to minimize the prediction bias of the XGBoost model within that scenario period; this set of weights is the optimal weight for that scenario. The feature vectors of the scenario period and the corresponding optimal weight vectors are used as training samples to train the scenario-weight mapping sub-model.

[0039] In this embodiment, the scene-weight mapping sub-model is implemented using a rule-based scenario weighting (RSW) model. This model generates scene feature weights through the following steps: External environmental factors are quantified into monitorable indicators, such as policy risk index, competitor price change rate, and supply chain stability level, and each indicator is assigned a status level. Business experts define several typical scenario combinations (e.g., high policy risk + competitor price war), and configure adjustment coefficients for each budget item for each scenario (e.g., marketing expense coefficient 0.7, emergency reserve coefficient 1.5). These scenario templates and corresponding adjustment coefficient tables are pre-stored in the rule base.

[0040] External environment data for the current budget cycle is collected in real time, and the weighted similarity between the current scenario and each preset scenario is calculated (e.g., using weighted Hamming distance). If the similarity between the current scenario and a preset scenario exceeds a preset threshold, the adjustment coefficient corresponding to that scenario is directly used as the feature weight. If the current scenario is a mixture of multiple scenarios (i.e., the similarity with multiple scenarios is high but does not exceed the threshold), the adjustment coefficients of multiple scenarios are linearly fused according to the similarity ratio to generate a comprehensive scenario feature weight.

[0041] The input to the XGBoost model consists of three parts: the temporal latent features output by the LSTM, business progress data from real-time data, and external environment scene data. During training, the XGBoost model uses the feature weights output by the scene-weight mapping sub-model to weight the input features (multiplying by the corresponding feature weights when calculating the split gain).

[0042] The LSTM, scene-weight mapping sub-model, and XGBoost model can be fine-tuned end-to-end as a whole, or trained separately and then fixed. After training, the model parameters are saved to a file or database for loading during online inference. New data can be used monthly or quarterly to incrementally train the model, updating the scene-weight mapping and XGBoost parameters.

[0043] During the prediction model's operation, the budget indicator sequence from historical data is input into an LSTM to extract temporal latent features. The feature vector data is then input into a scene-weight mapping sub-model, which outputs scene feature weights.

[0044] For example, if a competitor offers a low-priced package in a certain area, the model automatically increases the dynamic weight of the competitor from 20% to 35%, while reducing the weight of historical growth rate, thus reducing the FTTR user growth prediction deviation rate from 25% to 8%.

[0045] Then, the time-series latent features output by LSTM, business progress data and scenario data in real-time data, along with scenario feature weights, are input into the XGBoost model. The XGBoost model uses scenario feature weights to weight the input features (or as conditional inputs) and outputs the budget target prediction value for the next budget period (e.g., 12,500 FTTR users).

[0046] Optionally, in an embodiment, the decomposition model includes a clustering module, a carrying capacity assessment sub-model, and a random forest regressor; the step of calculating the initial decomposition coefficients of each budget unit using the pre-trained decomposition model based on the feature vector data and the scene feature weights includes: dividing each budget unit into different potential categories using the clustering module; assessing the carrying capacity of each budget unit using the carrying capacity assessment sub-model based on the resource carrying capacity index of each budget unit in the feature vector data to obtain a carrying capacity score for each budget unit; and calculating the initial decomposition coefficients of each budget unit using the random forest regressor based on the potential category and carrying capacity score of each budget unit.

[0047] In this embodiment, the clustering module employs the K-means algorithm to divide each budget unit into different potential categories, such as high-potential areas, stable areas, and development areas. Clustering can be performed based on the feature vector of each budget unit, which mainly includes historical business growth rate, average resource carrying capacity index, and historical completion rate standard deviation.

[0048] The carrying capacity assessment sub-model can use a regression model (such as linear regression, support vector regression, or a small neural network). The input is the resource carrying capacity index of the unit, and the output is the carrying capacity score (such as a continuous value from 1 to 10, where a higher score indicates a stronger ability to complete the indicator).

[0049] For each historical period and each unit, its resource carrying capacity index is calculated as an input feature, and its actual completion rate (completed value / issued value) is calculated and normalized to a score of 1-10 as a label. Mean squared error is used as the loss function, and a regression model is trained using gradient descent.

[0050] A random forest regressor is used to output initial decomposition coefficients based on the unit's potential category and carrying capacity score. For each historical period, given the total budget target and the actual decomposition coefficients of each unit, training samples are constructed. Sample features include the unit's potential category, carrying capacity score, and other auxiliary features (such as historical completion rate and seasonality factors). The sample label is the actual decomposition coefficient of that unit within that period. Based on the constructed training samples, the random forest regressor is trained.

[0051] Optionally, in an embodiment, after calculating the initial decomposition coefficients of each budget unit using the random forest regressor based on the potential category and carrying capacity score of each budget unit, the method further includes: for each budget unit, comparing the carrying capacity score of the budget unit with a carrying capacity assessment threshold; when the carrying capacity score of the budget unit is lower than the carrying capacity assessment threshold, reducing the initial decomposition coefficients of the budget unit; wherein the carrying capacity assessment threshold is dynamically adjusted based on the scene feature weights.

[0052] The carrying capacity assessment threshold is an empirical parameter used to determine whether to reduce the decomposition coefficient. An initial carrying capacity assessment threshold can be set in advance through calibration or other methods, along with dynamic adjustment rules. During operation, the carrying capacity assessment threshold is dynamically adjusted based on the scenario feature weights. The scenario weights output by the prediction model are synchronized to the decomposition model in real time, serving as scenario correction factors for the carrying capacity assessment. For example, when a scenario weight is ≥30%, the decomposition model automatically lowers the resource carrying capacity score threshold for that area (e.g., from 5 points to 4 points) to prioritize scenario adaptability.

[0053] In a specific embodiment, in a certain area, the weight of the 5G policy subsidy scenario reaches 40%. The decomposition model automatically reduces the carrying capacity score threshold of the area from 5 points to 4 points. The original score of grid A was 4 points (due to insufficient manpower). After the adjustment, there is no need to reduce the decomposition coefficient. This not only adapts to the policy dividend scenario but also avoids resource misallocation. The prediction deviation rate is further reduced from 8% to 5%.

[0054] Using the training method described above, the decomposition model can reasonably generate initial decomposition coefficients based on the resource carrying capacity and potential category of each budget unit, and can dynamically adjust the threshold according to the scene feature weights to avoid assigning excessively high indicators to units with insufficient resources.

[0055] During the decomposition model's operation, the clustering module categorizes each budget unit into potential zones such as high-potential areas, stable areas, and development areas based on characteristics like historical growth rates and resource carrying capacity. Then, the carrying capacity assessment sub-model takes the resource carrying capacity index of each unit as input and outputs a carrying capacity score. The random forest regressor outputs the initial decomposition coefficients for each unit based on the potential category and carrying capacity score. The carrying capacity score of each unit is compared to a carrying capacity assessment threshold; if the score is lower than the threshold, the initial decomposition coefficient of that unit is reduced, and the reduced portion is allocated to other units.

[0056] In this embodiment, when the score is less than 5 points, the decomposition coefficient is automatically reduced (by 5%-8% for every point lower), and the remaining capacity is allocated to high-potential areas with scores greater than 8 points. For example, grid A has a carrying capacity score of 4 points (insufficient manpower), with an initial decomposition coefficient of 0.2, which is automatically reduced to 0.16; grid B has a score of 9 points (sufficient resources), and the coefficient is increased from the original 0.18 to 0.22, achieving global balance.

[0057] In this embodiment, a function is defined to map the scene feature weights output by the prediction model to the carrying capacity assessment threshold adjustment amount in the decomposition model for the linkage between the prediction model and the decomposition model. On the validation set, the parameters of the two models are fixed, and only the parameters in the linkage rules are adjusted. The goal is to ensure that the overall comprehensive performance meets the standard, i.e., the prediction deviation rate is ≤10% and the resource matching degree is ≥85%. If the standard is not met, the model structure is adjusted.

[0058] Currently, multi-level budget decomposition mostly adopts a top-down, step-by-step manual approach. Higher levels cannot know the actual resource capacity of lower levels during allocation, and cross-level resource adjustments are not supported. Resources can only be allocated within the same level, and gaps cannot be dynamically filled from redundant units or upper-level elastic pools. This results in a resource-task matching rate of less than 50%, heavily relying on manual coordination, leading to low efficiency and a high error rate.

[0059] Optionally, in an embodiment, the step of adjusting the budget indicator allocation of each budget unit based on the predicted budget target value, the initial decomposition coefficients of each budget unit, and the resource status data of each budget unit to generate a budget indicator allocation scheme for each budget unit includes: receiving a manual adjustment value for the predicted budget target value, and generating a final budget target value based on the manual adjustment value; generating a decomposition template based on the final budget target value and the initial decomposition coefficients of each budget unit, and distributing the decomposition template to each budget unit; performing multi-level collaborative decomposition and cross-level resource scheduling processing based on the decomposition template to match the budget indicators of each budget unit with the resource status data of each budget unit, and generating an initial budget indicator allocation scheme for each budget unit; and receiving manual adjustment information for the initial budget indicator allocation scheme of each budget unit, and generating a budget indicator allocation scheme for each budget unit based on the manual adjustment information.

[0060] Specifically, the system can receive manual adjustments (no more than 10%) from marketing personnel to the projected budget target, resulting in the final budget target value. Then, based on the final budget target value and the initial decomposition coefficients for each unit, a decomposition template is generated. This template includes the budget indicators, resource matching degree, and resource gap indications for each unit, and is then distributed to all levels. For example, if a grid has a matching degree of 92% and a resource gap of 0.5 people, cross-regional dispatching for supplementation is recommended.

[0061] Subsequently, a multi-level collaborative decomposition is executed. The specific process includes: the upper level (e.g., the company) issues initial targets to the lower level (e.g., branch offices), and the lower level reports resource gaps based on its own resource status. Cross-level resource scheduling is triggered, prioritizing resource allocation from redundant units at the same level. If none are available, resources are allocated from the upper level's resource pool, while ensuring that scheduling costs do not exceed scheduling benefits (scheduling benefits = target completion rate improvement × target weight). The decomposition coefficients are simultaneously adjusted. This process of issuing, reporting, scheduling, and adjusting is repeated until the targets at each level match the resource status. Finally, administrators are allowed to manually adjust the allocation plan, record the reasons for the adjustments, and output the final budget target allocation plan.

[0062] Among them, underutilized resources at each level (such as redundant manpower and idle channels) are automatically collected to form a global resource pool, and the availability of resources and scheduling costs are updated in real time.

[0063] For example, the 5G user development target of grid C had a 30% gap (less than 2 people). One full-time staff member was transferred from the branch's resource pool, and support was dispatched from the adjacent grid D (with 1 redundant person). The decomposition coefficient was increased from 0.15 to 0.21, and the resource-target matching degree was improved from 65% to 93%.

[0064] Current anomaly detection relies on post-event manual audits, making it impossible to identify data fluctuations in real time during budget breakdown or execution. Even when anomalies are detected, they can only pinpoint the surface symptoms, failing to provide accurate attribution by combining contextual information such as the behavior of homogeneous groups and the path of continuous impact. More importantly, it cannot automatically generate intervention strategies, nor can it incorporate effective intervention experience into the rule base, leading to recurring problems and response cycles exceeding 48 hours.

[0065] Optionally, in the embodiments, the method further includes: collecting decomposed execution data of each budget unit in real time during the execution of each budget indicator allocation scheme in each budget unit; using the decomposed execution data of each budget unit, performing anomaly detection by employing the isolated forest algorithm combined with scenario-based dynamic thresholds; when an anomaly is detected, identifying continuously occurring related scenarios through time-series analysis to form a scenario chain, and tracing the cumulative impact of each related scenario on the anomaly based on the scenario chain to generate a scenario chain impact report; generating and pushing early warning information based on the scenario chain impact report, wherein the early warning information includes anomaly indicators, deviation values, scenario chain attribution conclusions, and resource scheduling suggestions.

[0066] Specifically, each budget unit executes the budget according to the final allocation plan. During execution, decomposed execution data (actual completed values) is collected and broken down in real time. An isolation forest algorithm, combined with scenario-based dynamic thresholds (e.g., the threshold is relaxed to 20% during periods of favorable policy), is used for anomaly detection to avoid false positives and false negatives. When an anomaly is detected (e.g., the deviation between actual completed values ​​and budget targets exceeds a threshold), time-series analysis is used to identify consecutively occurring related scenarios (e.g., favorable policy → competitor counterattack → resource shortage), forming a scenario chain, and tracing the cumulative impact weight of each scenario on the anomaly. When an anomaly occurs, not only is the current scenario analyzed, but the cumulative impact of preceding scenarios in the scenario chain is also traced, generating a scenario chain impact report. Based on the report, early warning information is pushed to relevant responsible persons, including anomaly indicators, deviation values, attribution conclusions, and resource scheduling suggestions.

[0067] For example, if a certain region's decomposition value is abnormal, the path can be traced back to the following scenario chain: policy subsidies (T-1 month, impact weight 20%) → competitor price reduction (first week of T month, impact weight 45%) → manpower reassignment (second week of T month, impact weight 35%). Based on this, a precise intervention plan can be generated to first supplement manpower and then adjust the strategy to deal with competitors, increasing the anomaly resolution rate from 70% to 95%.

[0068] In this embodiment, dynamic thresholds are established for key indicators (such as labor costs, material consumption, and user growth) of each budget unit (e.g., grid, package area). The specific implementation is as follows: For a key indicator of a budget unit in period t, its dynamic upper limit threshold is calculated by the following formula: ; in, and These are the mean and standard deviation of the indicator over the past N periods, respectively; k is the risk sensitivity coefficient, a configurable parameter, typically ranging from 1.5 to 3.0. The larger the k value, the wider the threshold band and the higher the tolerance for anomalies. This is an amplification factor adjusted according to business attributes and seasonal factors. For example, for highly volatile businesses (such as marketing campaigns), f takes a value greater than 1; for stable operation and maintenance businesses, f takes a value close to 1; at the same time, seasonal factors can be used to adjust f according to the month (such as peak promotional seasons).

[0069] Furthermore, the thresholds are further adjusted based on current scenario characteristics (such as favorable policies and competition from competitors). For example, the thresholds can be temporarily relaxed during periods of favorable policies or peak promotional seasons; and appropriately tightened during periods of resource scarcity or high risk.

[0070] Optionally, in an embodiment, the method further includes: acquiring execution effect data after abnormal intervention, wherein the execution effect data includes at least the deviation rate reduction magnitude; evaluating in real time whether the deviation rate reduction magnitude reaches a preset threshold, and periodically reviewing and evaluating multiple performance dimensions of budget management to obtain review and evaluation results; when the deviation rate reduction magnitude reaches the preset threshold, updating the prediction model and the decomposition model, and storing the intervention plan and scheduling path corresponding to this intervention in a preset case library; when the review and evaluation results indicate that any performance dimension fails to meet the standard, triggering model adjustment of the prediction model and the decomposition model.

[0071] Through a dual-track mechanism combining real-time evaluation and monthly review, model performance and intervention strategies are continuously improved. Specifically, the execution effect data also includes the reduction in deviation rate after the intervention plan is implemented, the time to fill resource gaps, and the accuracy of scenario chain attribution. The monthly review dimensions include prediction accuracy (scenario-based prediction deviation rate not exceeding 10%), decomposition fit (resource-indicator matching degree not less than 85%), intervention efficiency (minor anomaly handling time not exceeding 1 hour, severe anomaly not exceeding 24 hours), and anomaly recurrence rate (not exceeding 8%).

[0072] When the deviation rate decreases by a preset threshold (e.g., 40%) after intervention, the parameters and linkage calibration rules of the scenario-weight mapping sub-model are updated in real time, and the optimal intervention strategy and scheduling scheme in the scenario are solidified as the model default rules.

[0073] When the monthly review finds that any dimension fails to meet the standard, the model is automatically adjusted. For example, if the prediction accuracy is not up to standard, additional scenario training data is added; if the decomposition fit is not up to standard, the carrying capacity assessment parameters are optimized; and if the intervention efficiency is not up to standard, high-priority cases are added to the case library.

[0074] Cases with excellent intervention results (e.g., a deviation rate reduction of 40% or more) are categorized and stored. Each case is labeled with a scenario chain tag, adjustment plan, resource scheduling path, and effect data. The case library allows the model to call upon it in real time, realizing knowledge accumulation and reuse.

[0075] Furthermore, after each anomaly intervention (e.g., within 1 hour for mild anomalies, within 24 hours for severe anomalies), four-dimensional data including scenario chain characteristics, intervention plan, execution effect, and resource scheduling data is automatically collected, and model parameters and rule base are updated in real time based on this data. When encountering similar scenario chains in the future, the fixed optimal solution is automatically applied, significantly improving processing efficiency.

[0076] For example, a grid encountered an anomaly due to a scenario chain of "competitor's low-price package → resource shortage." An intervention plan of "reducing the target by 10%, dispatching one person across regions, and increasing channel subsidies" reduced the deviation rate from 25% to 7%. The plan then immediately solidified the "increased channel subsidy weight by 15% and set the dispatch priority to the highest" within that scenario chain into the model. Subsequently, the processing cycle for similar scenarios was shortened from 4 hours to 1.5 hours.

[0077] Through the aforementioned self-learning optimization mechanism, we can continuously learn from historical interventions and continuously improve the intelligence level and response efficiency of budget management.

[0078] In summary, this invention constructs a full-link intelligent budget management system that integrates historical data, business data, and scenario data. This system enables accurate prediction of budget targets, optimal global resource allocation, efficient closed-loop handling of anomalies, and real-time model evolution. It is perfectly adapted to the five-level organizational structure and multi-role requirements of telecom operators, and solves the core pain points of existing technologies, such as poor dynamic adaptation, weak global collaboration, low processing efficiency, and one-sided attribution.

[0079] Example 2 like Figure 2 As shown, this embodiment provides an intelligent budget management system 200, including: The data acquisition module 201 is used to acquire historical data and real-time data for the current budget period; wherein, the historical data includes historical data of budget indicators and historical business data; the real-time data includes external environment scenario data, business progress data and resource status data of each budget unit; Feature processing module 202 is used to perform feature engineering processing based on the historical data and the real-time data to generate feature vector data; The prediction module 203 is used to input the feature vector data into a pre-trained prediction model to generate scene feature weights and budget target prediction values, wherein the budget target prediction values ​​are budget indicator values ​​for the next budget period. The decomposition module 204 is used to calculate the initial decomposition coefficients of each budget unit based on the feature vector data and the scene feature weights through a pre-trained decomposition model. The initial decomposition coefficients of each budget unit represent the proportion of the budget target prediction value allocated to the budget unit. The allocation module 205 is used to adjust the allocation of budget indicators for each budget unit based on the predicted budget target value, the initial decomposition coefficient of each budget unit, and the resource status data of each budget unit, and to generate a budget indicator allocation scheme for each budget unit.

[0080] Optionally, in an embodiment, the feature processing module 202 includes: The preprocessing unit is used to perform data cleaning and standardization on the historical data and the real-time data; The tagging unit is used to tag the processed external environment scene data and extract scene features. The feature processing unit is used to perform feature extraction processing based on the processed historical data and the processed real-time data to obtain multi-dimensional features. The multi-dimensional features include at least the resource carrying capacity index and resource idle rate of each budget unit. The splicing unit is used to perform feature splicing processing based on the scene features and the multidimensional features to generate feature vector data.

[0081] Optionally, in this embodiment, the pre-trained prediction model includes an LSTM network, a scene-weight mapping sub-model, and an XGBoost model; the prediction module 203 includes: The hidden feature extraction unit is used to extract temporal hidden features based on the historical data using the LSTM network; The mapping unit is used to perform mapping processing based on the feature vector data using the scene-weight mapping sub-model to generate scene feature weights; The prediction unit is used to perform prediction processing using the XGBoost model based on the temporal hidden features extracted by the LSTM network, the real-time data, and the scene feature weights, to obtain the predicted value of the budget target.

[0082] Optionally, in an embodiment, the decomposition model includes a clustering module, a carrying capacity assessment sub-model, and a random forest regressor; the decomposition module 204 includes: Clustering units are used to divide each budget unit into different potential categories using the clustering module; The scoring unit is used to assess the carrying capacity of each budget unit based on the resource carrying capacity index of each budget unit in the feature vector data, and to obtain the carrying capacity score of each budget unit. The decomposition unit is used to calculate the initial decomposition coefficients of each budget unit based on its potential category and carrying capacity score using the random forest regressor.

[0083] Optionally, in an embodiment, after the decomposition unit, the following is further included: An adjustment unit is used to compare the bearing capacity score of each budget unit with the bearing capacity assessment threshold. When the bearing capacity score of the budget unit is lower than the bearing capacity assessment threshold, the initial decomposition coefficient of the budget unit is reduced. The bearing capacity assessment threshold is dynamically adjusted based on the scene feature weights.

[0084] Optionally, in an embodiment, the allocation module 205 includes: The first adjustment unit is configured to receive manual adjustment values ​​for the predicted budget target value, and generate a final budget target value based on the manual adjustment values; The template generation unit is used to generate a decomposition template based on the final budget target value and the initial decomposition coefficients of each budget unit, and to distribute the decomposition template to each budget unit. The coordination processing unit is used to perform multi-level collaborative decomposition and cross-level resource scheduling processing based on the decomposition template, so as to match the budget indicators of each budget unit with the resource status data of each budget unit and generate the initial budget indicator allocation scheme of each budget unit. The second adjustment unit is used to receive manual adjustment information for the initial budget indicator allocation scheme for each budget unit, and to generate a budget indicator allocation scheme for each budget unit based on the manual adjustment information.

[0085] Optionally, in an embodiment, the system further includes: The execution data acquisition module is used to collect the decomposed execution data of each budget unit in real time during the execution of the budget indicator allocation plan in each budget unit. The anomaly detection module is used to detect anomalies based on the decomposed execution data of each budget unit, using the isolated forest algorithm combined with scenario-based dynamic thresholds. The report generation module is used to identify consecutively occurring related scenarios through time series analysis when an anomaly is detected, to form a scenario chain, and to trace the cumulative impact of each related scenario on the anomaly based on the scenario chain, and generate a scenario chain impact report. The early warning module is used to generate and push early warning information based on the scenario chain impact report. The early warning information includes abnormal indicators, deviation values, scenario chain attribution conclusions, and resource scheduling suggestions.

[0086] Optionally, in an embodiment, the system further includes: The effect data acquisition module is used to acquire the execution effect data after abnormal intervention, and the execution effect data includes at least the magnitude of the decrease in the deviation rate. The evaluation module is used to evaluate in real time whether the decrease in the deviation rate has reached a preset threshold, and to periodically review and evaluate multiple performance dimensions of budget management to obtain the review and evaluation results. The first processing module is used to update the prediction model and the decomposition model when the deviation rate decreases to a preset threshold, and to store the intervention plan and scheduling path corresponding to this intervention in a preset case library. The second processing module is used to trigger model adjustments of the prediction model and the decomposition model when the review and evaluation results indicate that any performance dimension fails to meet the standard.

[0087] In some embodiments, the intelligent budget management system 200 of the present invention can be implemented in a combination of hardware and software. As an example, the intelligent budget management system 200 of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute an intelligent budget management method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0088] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0089] Example 3 like Figure 3 As shown, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements an intelligent budget management method as described in Embodiment 1.

[0090] In other words, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store a computer program; the processor is used to execute an intelligent budget management method shown in any embodiment of the present invention by calling the computer program.

[0091] In one alternative embodiment, an electronic device is provided. Figure 3 The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may further include a transceiver 304, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of the electronic device 300 does not constitute a limitation on the embodiments of the present invention.

[0092] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0093] Bus 302 may include a path for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus 302 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0094] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0095] The memory 303 is used to store application code (computer program) for executing the present invention, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0096] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0097] It should be noted that, Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0098] Example 4 This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute an intelligent budget management method as described in Embodiment 1.

[0099] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0100] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned intelligent budget management method.

[0101] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0102] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0103] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0104] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods shown in the above embodiments.

[0105] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0106] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0107] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0108] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent budget management method, characterized in that, Applied to a terminal device, the method includes: Acquire historical data and real-time data for the current budget period; wherein, the historical data includes historical data of budget indicators and historical business data; the real-time data includes external environment scenario data, business progress data, and resource status data of each budget unit; Feature engineering is performed on the historical data and the real-time data to generate feature vector data; The feature vector data is input into a pre-trained prediction model to generate scene feature weights and budget target prediction values, wherein the budget target prediction values ​​are budget indicator values ​​for the next budget period. Based on the feature vector data and the scene feature weights, the initial decomposition coefficients of each budget unit are calculated through a pre-trained decomposition model. The initial decomposition coefficients of each budget unit represent the proportion of the budget target prediction value allocated to the budget unit. Based on the predicted budget targets, the initial decomposition coefficients of each budget unit, and the resource status data of each budget unit, the budget indicator allocation for each budget unit is adjusted to generate a budget indicator allocation scheme for each budget unit.

2. The intelligent budget management method according to claim 1, characterized in that, The step of performing feature engineering processing based on the historical data and the real-time data to generate feature vector data includes: The historical data and the real-time data are cleaned and standardized. The processed external environment scene data is labeled with scene characteristics and scene features are extracted. Based on the processed historical data and processed real-time data, feature extraction is performed to obtain multidimensional features, which include at least the resource carrying capacity index and resource idle rate of each budget unit. Feature vector data is generated by concatenating the scene features and the multidimensional features.

3. The intelligent budget management method according to claim 1, characterized in that, The pre-trained prediction model includes an LSTM network, a scene-weight mapping sub-model, and an XGBoost model; the step of inputting the feature vector data into the pre-trained prediction model to generate scene feature weights and budget target prediction values ​​includes: Based on the historical data, the LSTM network is used to extract temporal hidden features; Based on the feature vector data, the scene-weight mapping sub-model is used to perform mapping processing to generate scene feature weights. Based on the temporal hidden features extracted by the LSTM network, the real-time data, and the scene feature weights, the XGBoost model is used for prediction processing to obtain the budget target prediction value.

4. The intelligent budget management method according to claim 2, characterized in that, The decomposition model includes a clustering module, a carrying capacity assessment sub-model, and a random forest regressor; the calculation of initial decomposition coefficients for each budget unit based on the feature vector data and the scene feature weights using the pre-trained decomposition model includes: The clustering module is used to divide each budget unit into different potential categories; Based on the resource carrying capacity index of each budget unit in the feature vector data, the carrying capacity assessment sub-model is used to assess the carrying capacity of each budget unit and obtain the carrying capacity score of each budget unit. Based on the potential category and carrying capacity score of each budget unit, the initial decomposition coefficients of each budget unit are calculated using the random forest regressor.

5. The intelligent budget management method according to claim 4, characterized in that, After calculating the initial decomposition coefficients of each budget unit using the random forest regressor based on the potential category and carrying capacity score of each budget unit, the process further includes: For each budget unit, the bearing capacity score of the budget unit is compared with the bearing capacity assessment threshold. When the bearing capacity score of the budget unit is lower than the bearing capacity assessment threshold, the initial decomposition coefficient of the budget unit is reduced. The load-bearing capacity assessment threshold is dynamically adjusted based on the scene feature weights.

6. The intelligent budget management method according to claim 1, characterized in that, The step of adjusting the budget indicator allocation for each budget unit based on the predicted budget target, the initial decomposition coefficients of each budget unit, and the resource status data of each budget unit, and generating a budget indicator allocation scheme for each budget unit, includes: Receive manually adjusted values ​​for the predicted budget target values, and generate a final budget target value based on the manually adjusted values; Based on the final budget target value and the initial decomposition coefficients of each budget unit, a decomposition template is generated and distributed to each budget unit. Based on the decomposition template, multi-level collaborative decomposition and cross-level resource scheduling are performed to match the budget indicators of each budget unit with the resource status data of each budget unit, thereby generating an initial budget indicator allocation scheme for each budget unit. Receive manual adjustment information for the initial budget indicator allocation scheme for each budget unit, and generate a budget indicator allocation scheme for each budget unit based on the manual adjustment information.

7. The intelligent budget management method according to claim 1, characterized in that, Also includes: During the execution of the budget indicator allocation plan in each budget unit, the decomposed execution data of each budget unit is collected in real time. Based on the decomposed execution data of each budget unit, an isolated forest algorithm combined with scenario-based dynamic thresholds is used for anomaly detection. When an anomaly is detected, time-series analysis is used to identify consecutively occurring related scenarios to form a scenario chain. Based on the scenario chain, the cumulative impact of each related scenario on the anomaly is traced, and a scenario chain impact report is generated. Early warning information is generated and pushed based on the scenario chain impact report. The early warning information includes abnormal indicators, deviation values, scenario chain attribution conclusions, and resource scheduling suggestions.

8. The intelligent budget management method according to claim 7, characterized in that, Also includes: Obtain execution effect data after abnormal intervention, wherein the execution effect data includes at least the decrease in deviation rate; Real-time assessment of whether the decrease in the deviation rate has reached a preset threshold, and periodic review and evaluation of multiple performance dimensions of budget management to obtain review and evaluation results; When the deviation rate decreases to a preset threshold, the prediction model and the decomposition model are updated, and the intervention plan and scheduling path corresponding to this intervention are stored in a preset case library. When the post-mortem evaluation results indicate that any performance dimension fails to meet the standard, the model adjustment of the prediction model and the decomposition model is triggered.

9. An intelligent budget management system, characterized in that, include: The data acquisition module is used to acquire historical data and real-time data for the current budget period; wherein, the historical data includes historical data of budget indicators and historical business data; the real-time data includes external environment scenario data, business progress data, and resource status data of each budget unit; The feature processing module is used to perform feature engineering processing based on the historical data and the real-time data to generate feature vector data; The prediction module is used to input the feature vector data into a pre-trained prediction model to generate scene feature weights and budget target prediction values, wherein the budget target prediction values ​​are budget indicator values ​​for the next budget period. The decomposition module is used to calculate the initial decomposition coefficients of each budget unit based on the feature vector data and the scene feature weights through a pre-trained decomposition model. The initial decomposition coefficients of each budget unit represent the proportion of the budget target prediction value allocated to the budget unit. The allocation module is used to adjust the allocation of budget indicators for each budget unit based on the predicted budget target value, the initial decomposition coefficient of each budget unit, and the resource status data of each budget unit, and to generate a budget indicator allocation scheme for each budget unit.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement an intelligent budget management method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the intelligent budget management method according to any one of claims 1 to 8.

Images