Project full life cycle driven dynamic payment prediction method, device and terminal

Abstract

The invention relates to the technical field of artificial intelligence and financial science and technology crossing, in particular to a project full-life-cycle-driven dynamic payment prediction method and device and a terminal. The method comprises the following steps: acquiring project data, and processing the project data to obtain target project data and a payment rule boundary; generating a business rule curve of a project payment full life cycle based on an electric power project management specification, an actual business process, target project data and a payment rule boundary; taking the target project data as a training set, training a CART decision tree as a weak learner, integrating the weak learner to obtain a strong learner, and dynamically adjusting parameters of the strong learner in the training process according to project pre-arrangement data, historical payment data and a business rule curve of a to-be-tested year; and outputting a project payment prediction result according to the strong learner and the business rule curve. The problems that in the prior art, model parameters are solidified, and static budget lacks flexibility can be solved.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and financial technology, and in particular to a dynamic payment prediction method, device and terminal driven by the entire project life cycle. Background Technology

[0002] With the increasing diversification of enterprise projects, payment behaviors for various projects, such as infrastructure construction, digital integration, and service procurement, vary significantly due to factors such as project type and construction cycle, highlighting the volatility and cyclical nature of payments. Against this backdrop, accurately grasping payment patterns across projects and optimizing cash management have become key requirements for enterprises to improve the quality of their financial strategic decision-making.

[0003] Currently, project payment forecasting primarily employs static statistical models such as ARIMA and traditional regression, combined with static budgeting methods. These techniques are based on historical payment data and utilize fixed parameters to build predictive models, mainly used to assist companies in capital allocation and preliminary design of financing plans.

[0004] Existing technologies have obvious limitations: fixed model parameters cannot dynamically respond to sudden disturbances such as policy adjustments and extreme weather, resulting in a rapid decline in prediction accuracy as the business environment changes; static budgets lack flexibility and are difficult to deal with the mismatch between payment rhythm and capital income cycle, which can easily cause short-term liquidity pressure on enterprises and restrict project progress efficiency. Summary of the Invention

[0005] This invention provides a dynamic payment prediction method, device, and terminal driven by the entire project lifecycle, to solve the problems of fixed model parameters and lack of flexibility in static budgets in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a dynamic payment prediction method driven by the entire project lifecycle, comprising: Acquire project data and process the project data to obtain target project data and payment rule boundaries; Based on power project management standards, actual business processes, target project data, and payment rule boundaries, a business pattern curve for the entire lifecycle of project payment is generated. The target project data is used as the training set, and a Classification and Regression Tree (CART) decision tree is trained as a weak learner. The weak learners are integrated to obtain a strong learner. During the training process, the parameters of the strong learner are dynamically adjusted according to the pre-arranged data of the project in the year to be tested, historical payment data, and the business pattern curve. Based on the strong learner and the business pattern curve, the predicted result of project payment is output.

[0007] In one possible implementation, the project data includes: structured project data and unstructured project data; The project data is processed to obtain target project data and payment rule boundaries, including: Outliers in the project's structured data are processed to obtain the target project's structured data; The unstructured data of the project is processed into a payment rule matrix and a clause-stage-amount mapping table; based on the payment rule matrix and the mapping table, the payment rule boundaries for each lifecycle stage of the project are obtained.

[0008] In one possible implementation, the unstructured data of the project is processed to obtain a payment rule matrix and a terms-stage-amount mapping table, including: Extract payment-related keywords from the unstructured data of the project; Analyze the keywords to determine their structural features; Based on the structured features, a payment rule matrix and a terms-stage-amount mapping table are constructed.

[0009] In one possible implementation, generating a business pattern curve for the entire lifecycle of project payment based on power project management standards, actual business processes, target project data, and payment rule boundaries includes: Based on power project management standards, actual business processes, and the boundaries of the payment rules, project payment business is divided into four standard stages: project initiation, execution, acceptance, and quality assurance. The payment-specific characteristics of the four standard stages are extracted, and a dual-drive prediction architecture based on rules and data is constructed. The payment statistics of the target project's structured data are calculated using a sliding window method, and payment inertia is captured based on the payment statistics. The historical payment curve is decomposed into trend terms, periodic terms, and residual terms. The mapping table is transformed into computable constraints. Based on the payment inertia, trend terms, periodic terms, residual terms, constraints, and payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed. Based on the fusion results of the two channels in the dual-channel architecture, a payment curve is fitted, and labels for each stage of the project's entire life cycle and key payment node annotations are embedded in the payment curve to obtain a business pattern curve for the entire life cycle of project payment with confidence intervals.

[0010] In one possible implementation, after converting the mapping table into computable constraints, the following steps are also included: A boundary verification function is established based on the constraints. The boundary verification function is used to filter the trend item and the period item by rules, and automatically triggers rule correction for the trend item and the period item that do not conform to the rules.

[0011] In one possible implementation, after calculating the payment statistics of the target item's structured data using a sliding window approach and capturing payment inertia based on the payment statistics, the method further includes: Extract payment features based on project and time dimensions; The features with positive impact among the payment features are selected, and the features with the least impact among the positive impact features are removed to obtain the first feature set; The features in the first feature set are optimized to obtain the second feature set; Based on the payment inertia, the trend term, the periodic term, the residual term, the constraints, and the payment rule matrix, a rule-guided, data-driven dual-channel architecture is constructed, including: Based on the second feature set, the trend term, the period term, the residual term, the constraint conditions, and the payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed.

[0012] In one possible implementation, the target project data is used as a training set, a CART decision tree is trained as a weak learner, and the weak learner is integrated to obtain a strong learner, including: The target project data and the second feature set are used as the training set. The sample weights are optimized using the mapping table as business features. Combined with residual calculation, the CART decision tree is trained as a weak learner. After multiple rounds of iterative training, a weighted voting method is used to integrate the weak learners obtained from multiple rounds of training into a strong learner; Design a streaming data fusion framework to integrate pre-planned project data and historical payment data for the year under test in real time, and dynamically adjust the parameters of the strong learner by combining the business pattern curve and the features in the second feature set.

[0013] In one possible implementation, after outputting the predicted project payment result based on the strong learner and the business pattern curve, the method further includes: A full-process analysis interface for projects is built, dynamically displaying the stage characteristics of project initiation, contract execution, and payment through a "lifecycle sandbox".

[0014] Secondly, embodiments of the present invention provide a dynamic payment prediction device driven by the entire project lifecycle, comprising: The data processing module is used to acquire project data and process the project data to obtain target project data and payment rule boundaries. The regularity curve generation module is used to generate a business regularity curve for the entire life cycle of project payment based on power project management specifications, actual business processes, target project data, and payment rule boundaries. The model training module is used to use the target project data as a training set, train a CART decision tree as a weak learner, integrate the weak learner to obtain a strong learner, and dynamically adjust the parameters of the strong learner during the training process based on the pre-arranged data of the project in the year to be tested, historical payment data, and the business pattern curve. The model application module is used to output the prediction results of project payment based on the strong learner and the business pattern curve.

[0015] Thirdly, embodiments of the present invention provide a terminal, 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 steps of the project lifecycle-driven dynamic payment prediction method as described in the first aspect or any possible implementation thereof.

[0016] This invention provides a dynamic payment prediction method, device, and terminal driven by the entire project lifecycle. By acquiring and processing project data, target project data and payment rule boundaries are obtained. Then, based on power project management standards, actual business processes, target project data, and payment rule boundaries, a business pattern curve for the entire project payment lifecycle is generated. The target project data is used as a training set to train a CART decision tree as a weak learner. The weak learners are integrated to obtain a strong learner. During training, the parameters of the strong learner are dynamically adjusted based on the project pre-arrangement data for the year to be tested, historical payment data, and the business pattern curve. This solves the problem of the inability to dynamically respond to sudden disturbances such as policy adjustments and extreme weather caused by fixed model parameters in existing technologies, thus achieving stable prediction accuracy. Based on the strong learner and the business pattern curve, the predicted project payment result is output. This invention achieves accurate modeling and dynamic prediction of payment rhythm by constructing a business pattern curve for the entire life cycle of project payments. Compared with traditional power project payment prediction, which mostly uses static statistical models, this application not only deeply integrates rules and data into the payment prediction model logic, but also establishes a dynamic prediction framework for project payments by mining the patterns of different project life cycles. Finally, it constructs an adaptive dynamic prediction model, thereby improving the flexibility of the payment prediction model, matching the payment rhythm with the capital income cycle, reducing the short-term liquidity pressure of enterprises, and improving project progress efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the implementation of the dynamic payment prediction method driven by the entire project lifecycle provided in this embodiment of the invention. Figure 2 This is a flowchart illustrating the implementation of the method for generating a business pattern curve for the entire lifecycle of project payments, as provided in this embodiment of the invention. Figure 3 This is a schematic diagram of payment feature filtering provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a strong learner provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the project-based operation architecture provided in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the panoramic analysis of the project dimension prediction model provided in this embodiment of the invention; Figure 7 This is a schematic diagram of the dynamic payment prediction device driven by the entire project lifecycle provided in this embodiment of the invention; Figure 8 This is a schematic diagram of the terminal provided in an embodiment of the present invention. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0021] In current technologies, the mismatch between payment schedules and funding risks is a key issue in financial management. Large-scale projects such as infrastructure and technological upgrades often experience a mismatch between peak payment periods and revenue cycles, leading to short-term liquidity pressures. For example, early payments for construction work are concentrated, while electricity revenue is cyclical and cannot fully cover payment needs, exacerbating the funding gap. Furthermore, sudden policy changes or extreme weather events can further strain funding. Traditional budgeting methods, lacking flexibility, struggle to dynamically respond to these risks and cannot adjust funding arrangements in a timely manner to cope with uncertainty. This problem not only affects the stability of a company's cash flow but may also limit project progress efficiency.

[0022] In the scenario of capital forecasting in the power industry, the contradiction between the static nature of algorithms and the dynamic nature of business is manifested in the fact that the fixed parameters of the model cannot adapt to the rapidly changing business environment, resulting in the decay of prediction accuracy and decision lag. In application, this is specifically manifested as prediction lag failure. Static models, such as ARIMA and traditional regression, have fixed parameters after training and cannot respond to dynamic disturbances such as policy changes and emergencies. The prediction error rate increases exponentially with changes in the business environment. The main reason is that static algorithms lack an update mechanism, which leads to the untimely iteration of dynamic business rules after one-time embedding. This highlights the limitations of static algorithms in offline training paradigms and the rigid shortcomings at the technology stack level.

[0023] To address the aforementioned problems, this invention provides a dynamic payment prediction method driven by the entire project lifecycle. Figure 1 A flowchart illustrating the implementation of a dynamic payment prediction method driven by the entire project lifecycle.

[0024] The study on the forecasting of payment business across the entire scope revealed two core characteristics of payment behavior at the project level: First, the volatility of payments is affected by project type and project construction cycle. The distribution of payment amounts differs significantly among different project types, such as long-term milestone payments for infrastructure projects, phased acceptance payments for digital system integration projects, and linearly allocated payments for service procurement projects. Second, the periodicity of payments for the same project type is constrained by contract payment paradigms and terms, involving key elements such as advance payment ratios, progress payment milestones, and retention of quality assurance deposits.

[0025] The theoretical value of this dual-characteristic research lies in constructing a project-level payment behavior feature library, providing differentiated parameter inputs for predictive models. By introducing a dynamic response mechanism, real-time monitoring and correction of payment plan deviations can be achieved. In practice, the research findings can effectively support financial strategic decision-making: in the capital allocation stage, financing scheme design can be optimized, such as achieving intertemporal matching of payment cash flows through structured financing products.

[0026] In the dynamic payment forecasting method driven by the entire project lifecycle, the project-based forecasting business scope includes three business types: service procurement, material procurement, and employee reimbursement. It involves project types such as power grid infrastructure, production overhaul, production technology transformation, and marketing under the categories of comprehensive planning and special costs. By using big data analysis technology to identify the annual cycle patterns of payments for each project type, fund payment forecasting data and business characteristics are constructed to support the construction of annual forecasting models at the project level.

[0027] Details are as follows: Step 101: Obtain project data and process the project data to obtain target project data and payment rule boundaries.

[0028] Project data can include historical project data, pre-planned project data, and contract terms. This data constitutes the raw data pool, providing the foundation for training the subsequent predictive model. Pre-planned project data mainly refers to the preliminary arrangements related to projects planned for the target year, which is the year for which payment forecasts need to be made. Historical project data includes project types and payment records, while pre-planned project data includes a list of projects for the target year. Contract terms can be stored in a contract terms repository.

[0029] In one embodiment, the project data includes: structured project data and unstructured project data; The project data is processed to obtain the target project data and payment rule boundaries, including: Outliers in the project's structured data are processed to obtain the target project's structured data; The unstructured data of the project is processed into a payment rule matrix and a clause-stage-amount mapping table. Based on the payment rule matrix and mapping table, the payment rule boundaries for each stage of the project lifecycle are obtained.

[0030] The project's structured data includes: basic fields, transaction attributes, and time-related information. The basic fields include project number, project type, project start and end dates, service confirmation / delivery confirmation time, contract number, order number, and order type. Transaction attributes include payment amount, payee bank information (bank name and interbank code), and structured tags for payment summary. Payee bank information includes bank name and interbank code. The time dimension includes payment initiation time and actual payment time, with the payment initiation time accurate to the second.

[0031] The unstructured data for the project includes payment contract templates, policy documents, and project execution progress logs for each project type; the payment contract templates include payment terms and acceptance conditions, and the policy documents include subsidy policies and financial policies.

[0032] Optionally, for outliers, first analyze the rationality of the outlier's source based on the business context. If the outlier's cause does indeed have a special business background and meaning, then process it separately and conduct a specialized analysis, establishing a supplementary data model corresponding to this type of business logic. Based on this, delete erroneous outlier data; for missing data, use the historical mean method, or employ linear regression and ARIMA models to predict the expected value of the data, and supplement it with the predicted value; for cases where the numerical value is abnormal but the trend is correct: use the historical average value within adjacent years ± The data was replaced; and the deviation analysis based on the prediction results can further highlight the data anomaly problem, so as to address it in a more targeted manner. For example, the prediction error of the two cost types under the special project, production operation and maintenance project and power marketing project, is too high. By analyzing the data, it was found that the data anomalies in individual years of the training set data affected the model training. For example, the data of production operation and maintenance project in 2019 and 2020 were abnormally high, and the data of power marketing project in 2020 and 2021 were abnormally high. The mean method was used to process the outliers, and the processed data conformed to the general trend of increasing year by year.

[0033] In one embodiment, the unstructured data of the project is processed into a payment rule matrix and a term-stage-amount mapping table, which may include: Extract payment-related keywords from the project's unstructured data; Analyze keywords and determine their structural characteristics; Based on the structured characteristics, a payment rule matrix and a terms-stage-amount mapping table are constructed.

[0034] Optionally, unstructured data such as contract documents can be read, and BERT models can be used to extract key terms, such as "prepayment ratio" and "penalty clause". NLP parsing can be used to locate the structured features of the key terms, such as "70% payment upon acceptance" → payment ratio = 0.7, trigger condition = acceptance completion. Based on this, a payment rule matrix can be established, and a "term-stage-amount" mapping table can be constructed to determine the payment rule boundaries of each lifecycle stage. For example, "70% payment upon acceptance" corresponds to the payment ratio constraint of the "acceptance stage", and "30% payment upon equipment delivery" corresponds to the trigger condition of the "execution stage".

[0035] Step 102: Based on power project management standards, actual business processes, target project data, and payment rule boundaries, generate a business pattern curve for the entire lifecycle of project payments.

[0036] In one embodiment, such as Figure 2 As shown, based on power project management standards, actual business processes, target project data, and payment rule boundaries, a business pattern curve for the entire lifecycle of project payments is generated, which may include: Step 201: Based on the power project management standards, actual business processes, and payment rule boundaries, the project payment business is divided into four standard stages: project initiation, execution, acceptance, and quality assurance. The payment-specific characteristics of the four standard stages are extracted, and a dual-drive prediction architecture based on rules and centered on data is constructed.

[0037] Based on power project management standards and aligning with actual business processes such as project initiation, bidding, construction, and acceptance, the standard stages for project payment execution can be preliminarily determined: project initiation, execution, acceptance, and quality assurance. By decoupling complex payment behaviors, stage-specific patterns are extracted, and a dual-drive predictive architecture based on rules and centered on data is constructed. Taking infrastructure projects as an example, the payment rules and data performance logic for the four standard stages are shown in Table 1 below.

[0038] Table 1

[0039] Step 202: The payment statistics of the target project's structured data are calculated using a sliding window method, and payment inertia is captured based on the payment statistics.

[0040] Considering the large volume and long time span of the structured data for the target project, a sliding window approach is adopted to continuously calculate the payment statistics for the past three periods. These payment statistics can include the payment mean, variance, and coefficient of variation. Payment inertia is captured based on these statistics.

[0041] Optionally, after calculating the payment statistics of the target item's structured data using a sliding window approach and capturing payment inertia based on the payment statistics, the method may further include: Extract payment features based on project and time dimensions; The first feature set is obtained by filtering out the features with positive impact among the payment features and removing the features with the least impact among the positive features; The features in the first feature set are optimized to obtain the second feature set.

[0042] When extracting payment features based on project and time dimensions, the project dimension can be categorized by project type, such as infrastructure, technological upgrading, and major repairs, and specific features for each project type can be extracted, such as project type, initial payment time / amount, material payment time / amount, service payment time / amount, total amount of prepayment / delivery payment / operation payment / warranty payment / progress payment, maximum payment amount, minimum payment amount, total number of payments within the cycle, number of material / service payments, and number of prepayment / delivery payment / operation payment / warranty payment / progress payment payments, etc. The time dimension can be set to an annual dimension according to actual needs. The annual dimension can distinguish between ongoing projects (with complete historical data) and new projects (relying on pre-arranged data) based on the project initiation node, and set differentiated feature weights, such as the payment ratio / time in the year of project initiation, the payment ratio / time in the following year, and the payment ratio / time in the year of historical project initiation, etc.

[0043] A baseline model is built using all payment features. This baseline model is then used for feature evaluation and selection. To evaluate model performance, one feature is removed sequentially, and the model is retrained. After each feature removal, the model's performance is re-evaluated to quantify the positive or negative impact of that feature. Features that positively influence the baseline model's performance are selected to form a new feature set. A recursive feature elimination method is then used to iteratively train the model and remove unimportant features to reduce model complexity and computational cost while improving generalization. This results in the final feature set, the first feature set. See also... Figure 3 The diagram shown illustrates the payment feature filtering process. The final effective features obtained after filtering include important features such as progress payment cycle, number of prepayments / payments upon receipt, monthly payment execution, project completion date, estimated project amount, and project profit center.

[0044] After feature selection, the features are further optimized so that the optimized features can more accurately train the prediction model.

[0045] When optimizing the features in the first feature set to obtain the second feature set, the following steps can be taken: The features in the first feature set are used for algorithm training and result testing. It is found that the initial feature set did not include employee expense reimbursement features. After updating, it is found that adding employee expense reimbursement features increases the prediction error. Feature analysis of the added part reveals that the employee expense reimbursement feature itself does not exhibit strong regularity. Therefore, reducing employee expense reimbursement-related features is equivalent to reducing the weight of employee expense reimbursement-related features. After optimization, the model's prediction error is further reduced. Simultaneously, it is found that applying scaling strategies such as standardization and normalization to the features does not reduce some evaluation metrics but rather increases them. This is because the XGBoost algorithm itself is not sensitive to feature scale, and the tree model does not pre-assume the data distribution but directly learns the relationship between features and labels. Therefore, feature scaling does not have a significant improvement effect, so the original features are used directly.

[0046] Step 203: Decompose the historical payment curve into trend terms, periodic terms, and residual terms, and transform the mapping table into computable constraints. Based on payment inertia, trend terms, periodic terms, residual terms, constraints, and payment rule matrix, construct a rule-guided and data-driven dual-channel architecture.

[0047] The trend term corresponds to the project's progress, the periodic term corresponds to quarterly management characteristics, and the residual term corresponds to unexpected disturbances. Specifically, linear regression is used to fit the trend term, Fourier transform is used to extract the periodic term, and the ± values ​​of payment deviations caused by unexpected disturbances are used to form the residual term. Fluctuation boundary library.

[0048] In one embodiment, after converting the mapping table into computable constraints, the process may further include: A boundary verification function is established based on constraints. The boundary verification function is used to filter trend items and periodic items according to rules, and automatically triggers rule correction for trend items and periodic items that do not conform to the rules.

[0049] In one embodiment, the dual-channel architecture includes a data-driven channel and a rule-guided channel. The data-driven channel takes into account trend terms, periodic terms, residual terms, and a second feature set, and captures temporal dependencies through a Long Short-Term Memory (LSTM) network. The rule-guided channel takes into account constraints and a payment rule matrix, and generates a compliance score through a symbolic inference engine. The reinforcement learning agent dynamically adjusts the weights of the dual channels with the reward function of minimizing prediction error and maximizing compliance.

[0050] The dual-channel architecture is forced to meet business constraints through a matrix of constraints and payment rules, such as "acceptance phase payment accounts for 10%-15%" and "quality assurance deposit release period is 1-3 years".

[0051] The reward function for a reinforcement learning agent is: ,in, As a reward value, and These are the weighting coefficients. For prediction error, For compliance scoring.

[0052] For different project types such as infrastructure, technological upgrading, and major repair, we train exclusive curve parameters. For infrastructure projects, we strengthen the slope of the trend term and amplify the amplitude of the period term. For technological upgrading projects, we extend the period of the period term and narrow the fluctuation of the residual term. For major repair projects, we stretch the tail of the quality assurance stage curve and reduce the weight of the period term.

[0053] Based on the payment cycle characteristics of typical projects such as infrastructure and production technology upgrades, patterns are extracted from historical payment data. Payment rhythms are categorized according to the characteristics of project lifecycles, and a predictive model research framework suitable for various project types is constructed accordingly. For example, the steep rise and gradual decline type, taking infrastructure project monetization as an example, shows rapid growth in payment amount after project initiation, reaching a peak in the 8th month, with an annual payment ratio of 26.74%, reaching 70.49% by the second year, and 90% by the third year. This type has a long payment cycle, with payments declining steadily after reaching the peak. It is mainly driven by construction progress and equipment delivery time, and is sensitive to external disturbances (such as supply chain delays). The steady increase and decrease type, taking other projects in the special project as examples, shows steady growth in payment amount after project initiation, reaching a peak in the 9th month, with an annual payment ratio of 37.49%, reaching 92.13% by the second year, and 99.35% by the third year. This type has a short payment cycle, steadily rising to a peak and then steadily declining. It is influenced by the special project's annual payment principle, with clear milestone nodes. Key performance indicators are summarized in Table 2.

[0054] Table 2

[0055] Step 204: Based on the fusion results of the two channels in the dual-channel architecture, fit the main body of the payment curve, and embed the labels of each stage of the project's entire life cycle and the annotations of key payment nodes into the main body of the payment curve to obtain the business pattern curve of the entire life cycle of project payment with confidence interval.

[0056] Optionally, based on the dual-channel fusion results, an improved Logistic model is used to fit the main body of the payment curve, with the vertical axis representing the cumulative percentage of payment amount and the horizontal axis representing the percentage of project lifecycle, and the residual term ± The fluctuation range serves as the boundary, generating a strip curve that includes a baseline curve and upper and lower risk boundaries. Labels for each stage of the project's entire lifecycle and key payment nodes are embedded into the payment curve, and curve parameters are updated continuously via real-time data streams. The generated curve undergoes rule verification, data verification, and robustness testing. Rule verification involves calling the "term-stage-amount" mapping table to verify whether the payment percentage for each stage complies with contractual rules. Data verification compares the curve's deviation from historical payment curves of similar projects. Robustness testing simulates extreme residual events to verify the curve's confidence interval coverage, resulting in a business pattern curve for the entire project payment lifecycle with confidence intervals.

[0057] Step 103: Use the target project data as the training set, train the CART decision tree as a weak learner, integrate the weak learners to obtain a strong learner, and dynamically adjust the parameters of the strong learner according to the pre-arranged data of the project in the year to be tested, historical payment data, and business pattern curves during the training process.

[0058] In one embodiment, using the target project data as a training set, training a CART decision tree as a weak learner, and integrating the weak learners to obtain a strong learner may include: The target project data and the second feature set are used as the training set. The sample weights are optimized by using a mapping table as the business feature. Combined with residual calculation, the CART decision tree is trained as a weak learner. After multiple rounds of iterative training, a weighted voting method is used to integrate the weak learners obtained from multiple rounds of training into a strong learner; Design a streaming data fusion framework to integrate pre-planned data of projects in the year to be tested with historical payment data in real time, and dynamically adjust the parameters of the strong learner by combining business pattern curves and features in the second feature set.

[0059] The first CART decision tree (a weak learner) is trained using the training set, outputting the predicted payment amount. The predicted payment amount is compared with the actual payment amount to obtain the sample residuals, which reflect patterns the model has not yet learned. An exponential loss function is used to increase the weights of samples with larger residuals, allowing the model to focus more on harder-to-learn samples, thus improving its ability to handle complex data and its robustness to payment biases. A new CART decision tree is then trained using the training set of samples with updated weights, repeating the residual calculation-weight update-training process.

[0060] like Figure 4 The diagram shown illustrates the generation of a strong learner. After multiple iterations, all weak learners are integrated through weighted voting to form a stronger strong learner with greater predictive power, thus achieving more accurate predictions. For example, an infrastructure project might integrate 20 weak learners, covering multiple dimensions such as "construction period, quarterly period, and disturbances."

[0061] For new projects, by pre-arranging data to match project type and stage characteristics, we can associate them with the full life cycle business pattern curves of similar projects. For example, for new infrastructure projects, we can match the curves of steep rise in the execution stage and gradual decline in the quality assurance stage, thus dynamically injecting prior patterns into the strong learner.

[0062] For ongoing projects, based on the latest payment data, such as monthly payment records, we update trend items / period items / residual items in real time and adjust the parameters of the strong learner, such as correcting the prediction time of quarterly payment peaks, to ensure that the model is synchronized with actual business.

[0063] Based on the construction logic of the strong learner described above, by iteratively integrating the weak learner, optimizing sample weights, and introducing regularization, a highly efficient ability to capture complex data features is formed. For project-level payment amount prediction, multi-dimensional data such as historical expenditure amount, project type, payment time, and project attribute indicators can be used as input features. Relying on the advantages of strong learners in mining spatiotemporal patterns and business association patterns, the hidden patterns in expenditure data are analyzed. The model has both data-driven accuracy (fitting historical payment patterns) and rule-guided compliance (meeting contract and management requirements). At the same time, it can dynamically adapt to the payment prediction needs of different scenarios such as new / continued projects and regular / disruptive projects.

[0064] Call the terms-stage-amount mapping table to verify whether the forecast complies with business rules, such as whether the percentage of accepted payments is within the 10%-15% range. If it violates the rules, the regularization constraints of the rule guidance layer will be triggered to force the forecast to be corrected.

[0065] Step 104: Based on the strong learner and the business pattern curve, output the predicted results of project payment.

[0066] The project prediction model includes a strong learner and business pattern curves.

[0067] The project prediction model uses pre-arranged data as its business foundation, combines the full lifecycle patterns of various project types, selects effective features, and completes model construction. Comprehensive testing and evaluation of the model's prediction performance are conducted, targeted optimization measures are developed, and the model prediction data is dynamically updated in conjunction with various factors such as company plan deployment, policy adjustments, and actual execution. This effectively improves the model's accuracy and adaptability. The project-based operational architecture is as follows: Figure 5 As shown.

[0068] Optionally, based on the strong learner and business pattern curves, output project payment heatmaps, prediction curves, risk warning information, and fund management dashboard data.

[0069] After outputting the project payment prediction results based on the strong learner and business pattern curve, it also includes: building a full-process analysis interface for the project dimension, and dynamically displaying the stage characteristics of project initiation, contract execution and payment through the "life cycle sand table".

[0070] In one embodiment, the prediction process of the project prediction model can be visualized. By constructing a full-process analysis interface across project dimensions, a visual tracking system from data tracing to prediction generation can be achieved. A "lifecycle sandbox" is used to dynamically display the characteristics of project initiation, contract execution, and payment stages. A "term-stage-amount" mapping table is called to verify the consistency between the prediction results and contract rules to validate the rationality of the prediction, thus achieving a visual tracking system from data tracing to prediction generation. Figure 6The diagram illustrates a comprehensive overview of the project dimension prediction model. This design transforms the black box of machine learning into a visualization tool, preserving the intelligence of algorithmic predictions while helping finance personnel understand feature association rules and verify the rationality of results through a transparent modeling process. Ultimately, this enhances the confidence and business guidance value of the predicted amounts in fund management.

[0071] This invention provides a dynamic payment prediction method driven by the entire project lifecycle. By acquiring and processing project data, target project data and payment rule boundaries are obtained. Then, based on power project management standards, actual business processes, target project data, and payment rule boundaries, a business pattern curve for the entire project payment lifecycle is generated. The target project data is used as a training set to train a CART decision tree as a weak learner. The weak learners are integrated to obtain a strong learner. During training, the parameters of the strong learner are dynamically adjusted based on the project pre-arrangement data for the year to be tested, historical payment data, and the business pattern curve. This solves the problem of the inability to dynamically respond to sudden disturbances such as policy adjustments and extreme weather caused by fixed model parameters in existing technologies, thus achieving stable prediction accuracy. Based on the strong learner and the business pattern curve, the predicted project payment result is output. This invention achieves accurate modeling and dynamic prediction of payment rhythm by constructing a business pattern curve for the entire life cycle of project payments. Compared with traditional power project payment prediction, which mostly uses static statistical models, this application not only deeply integrates rules and data into the payment prediction model logic, but also establishes a dynamic prediction framework for project payments by mining the patterns of different project life cycles. Finally, it constructs an adaptive dynamic prediction model, thereby improving the flexibility of the payment prediction model, matching the payment rhythm with the capital income cycle, reducing the short-term liquidity pressure of enterprises, and improving project progress efficiency. It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0072] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0073] Figure 7 The diagram shows a schematic of a dynamic payment prediction device driven by the entire project lifecycle, as provided in an embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 7 As shown, the dynamic payment prediction device 7 driven by the entire project lifecycle includes: a data processing module 71, a regular curve generation module 72, a model training module 73, and a model application module 74.

[0074] The data processing module 71 is used to acquire project data and process the project data to obtain target project data and payment rule boundaries. The regularity curve generation module 72 is used to generate a business regularity curve for the entire life cycle of project payment based on power project management specifications, actual business processes, target project data, and payment rule boundaries. The model training module 73 is used to use the target project data as a training set to train the CART decision tree as a weak learner, integrate the weak learners to obtain a strong learner, and dynamically adjust the parameters of the strong learner according to the pre-arranged data of the project in the year to be tested, historical payment data, and business pattern curves during the training process. Model application module 74 is used to output the prediction results of project payment based on the strong learner and business pattern curve.

[0075] In one possible implementation, project data includes: structured project data and unstructured project data; Data processing module 71 processes project data to obtain target project data and payment rule boundaries, and then uses it for: Outliers in the project's structured data are processed to obtain the target project's structured data; The unstructured data of the project is processed into a payment rule matrix and a clause-stage-amount mapping table. Based on the payment rule matrix and mapping table, the payment rule boundaries for each stage of the project lifecycle are obtained.

[0076] In one possible implementation, when the data processing module 71 performs rule-based processing on the unstructured project data to obtain the payment rule matrix and the terms-stage-amount mapping table, it is used for: Extract payment-related keywords from the project's unstructured data; Analyze keywords and determine their structural characteristics; Based on the structured characteristics, a payment rule matrix and a terms-stage-amount mapping table are constructed.

[0077] In one possible implementation, when the regularity curve generation module 72 generates a business regularity curve for the entire lifecycle of project payment based on power project management specifications, actual business processes, target project data, and payment rule boundaries, it is used for: Based on power project management standards, actual business processes, and payment rule boundaries, project payment business is divided into four standard stages: project initiation, execution, acceptance, and quality assurance. The payment-specific characteristics of the four standard stages are extracted, and a dual-drive prediction architecture based on rules and data is constructed. The payment statistics of the target project's structured data are calculated using a sliding window method, and payment inertia is captured based on the payment statistics. The historical payment curve is decomposed into trend terms, periodic terms, and residual terms. The mapping table is transformed into computable constraints. Based on payment inertia, trend terms, periodic terms, residual terms, constraints, and payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed. Based on the fusion results of the two channels in the dual-channel architecture, the main body of the payment curve is fitted, and the labels of each stage of the project life cycle and the annotation of key payment nodes are embedded in the main body of the payment curve to obtain the business pattern curve of the project payment life cycle with confidence interval.

[0078] In one possible implementation, after the regular curve generation module 72 converts the mapping table into computable constraints, it is also used for: A boundary verification function is established based on constraints. The boundary verification function is used to filter trend items and periodic items according to rules, and automatically triggers rule correction for trend items and periodic items that do not conform to the rules.

[0079] In one possible implementation, after the regularity curve generation module 72 uses a sliding window to calculate the payment statistics of the target project's structured data and captures payment inertia based on the payment statistics, it is also used for: Extract payment features based on project and time dimensions; The first feature set is obtained by filtering out the features with positive impact among the payment features and removing the features with the least impact among the positive features; The features in the first feature set are optimized to obtain the second feature set; Based on payment inertia, trend terms, periodic terms, residual terms, constraints, and the payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed, including: Based on the second feature set, trend term, period term, residual term, constraint conditions, and payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed.

[0080] In one possible implementation, the model training module 73 uses the target project data as a training set, trains a CART decision tree as a weak learner, and integrates the weak learners to obtain a strong learner for: The target project data and the second feature set are used as the training set. The sample weights are optimized by using a mapping table as the business feature. Combined with residual calculation, the CART decision tree is trained as a weak learner. After multiple rounds of iterative training, a weighted voting method is used to integrate the weak learners obtained from multiple rounds of training into a strong learner; Design a streaming data fusion framework to integrate pre-planned data of projects in the year to be tested with historical payment data in real time, and dynamically adjust the parameters of the strong learner by combining business pattern curves and features in the second feature set.

[0081] In one possible implementation, after the model application module 74 outputs the predicted project payment result based on the strong learner and the business pattern curve, it is further used for: A full-process analysis interface for projects is built, dynamically displaying the stage characteristics of project initiation, contract execution, and payment through a "lifecycle sandbox".

[0082] The above embodiments provide a dynamic payment prediction device driven by the entire project lifecycle. A data processing module acquires and processes project data to obtain target project data and payment rule boundaries. Then, a pattern curve generation module generates a business pattern curve for the entire project payment lifecycle based on power project management specifications, actual business processes, target project data, and payment rule boundaries. A model training module uses the target project data as a training set to train a CART decision tree as a weak learner. The weak learners are then integrated to form a strong learner. During training, the parameters of the strong learner are dynamically adjusted based on the project pre-arrangement data for the year to be tested, historical payment data, and the business pattern curve. This solves the problem of the inability to dynamically respond to sudden disturbances such as policy adjustments and extreme weather caused by fixed model parameters in existing technologies, thus achieving stable prediction accuracy. The model application module outputs the project payment prediction results based on the strong learner and the business pattern curve. This invention achieves accurate modeling and dynamic prediction of payment rhythm by constructing a business pattern curve for the entire life cycle of project payments. Compared with traditional power project payment prediction, which mostly uses static statistical models, this application not only deeply integrates rules and data into the payment prediction model logic, but also establishes a dynamic prediction framework for project payments by mining the patterns of different project life cycles. Finally, it constructs an adaptive dynamic prediction model, thereby improving the flexibility of the payment prediction model, matching the payment rhythm with the capital income cycle, reducing the short-term liquidity pressure of enterprises, and improving project progress efficiency.

[0083] Figure 8 This is a schematic diagram of a terminal provided in an embodiment of the present invention. Figure 8 As shown, the terminal 8 in this embodiment includes: a processor 80, a memory 81, and a computer program 82 stored in the memory 81 and executable on the processor 80. When the processor 80 executes the computer program 82, it implements the steps in the above-described embodiments of the dynamic payment prediction method driven by the entire lifecycle of projects, for example... Figure 1 Steps 101 to 104 are shown. Alternatively, when the processor 80 executes the computer program 82, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 7 The functions of each module / unit are shown.

[0084] For example, the computer program 82 can be divided into one or more modules / units, which are stored in the memory 81 and executed by the processor 80 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 82 in the terminal 8. For example, the computer program 82 can be divided into... Figure 7 The modules / units shown are shown.

[0085] The terminal 8 may include, but is not limited to, a processor 80 and a memory 81. Those skilled in the art will understand that... Figure 8 This is merely an example of terminal 8 and does not constitute a limitation on terminal 8. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal may also include input / output devices, network access devices, buses, etc.

[0086] The processor 80 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0087] The memory 81 can be an internal storage unit of the terminal 8, such as a hard disk or memory of the terminal 8. The memory 81 can also be an external storage device of the terminal 8, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal 8. Furthermore, the memory 81 can include both internal storage units and external storage devices of the terminal 8. The memory 81 is used to store the computer program and other programs and data required by the terminal. The memory 81 can also be used to temporarily store data that has been output or will be output.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0089] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0090] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0091] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0094] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above embodiments of the dynamic payment prediction method driven by the entire lifecycle of each project. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0095] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A dynamic payment prediction method driven by the entire project lifecycle, characterized in that, include: Acquire project data and process the project data to obtain target project data and payment rule boundaries; Based on power project management standards, actual business processes, target project data, and payment rule boundaries, a business pattern curve for the entire lifecycle of project payment is generated. The target project data is used as a training set to train a CART decision tree as a weak learner. The weak learner is then integrated to obtain a strong learner. During the training process, the parameters of the strong learner are dynamically adjusted based on the pre-arranged data of the project in the year to be tested, historical payment data, and the business pattern curve. Based on the strong learner and the business pattern curve, the predicted result of project payment is output.

2. The dynamic payment prediction method driven by the entire project lifecycle as described in claim 1, characterized in that, The project data includes: structured project data and unstructured project data; The project data is processed to obtain target project data and payment rule boundaries, including: Outliers in the project's structured data are processed to obtain the target project's structured data; The unstructured data of the project is processed into a payment rule matrix and a clause-stage-amount mapping table; based on the payment rule matrix and the mapping table, the payment rule boundaries for each stage of the project's lifecycle are obtained.

3. The dynamic payment prediction method driven by the entire project lifecycle as described in claim 2, characterized in that, The unstructured data of the project is processed into a rule-based system to obtain a payment rule matrix and a term-stage-amount mapping table, including: Extract payment-related keywords from the unstructured data of the project; Analyze the keywords to determine their structural features; Based on the structured features, a payment rule matrix and a terms-stage-amount mapping table are constructed.

4. The dynamic payment prediction method driven by the entire project lifecycle as described in claim 3, characterized in that, The process of generating a business pattern curve for the entire lifecycle of project payments, based on power project management standards, actual business processes, target project data, and payment rule boundaries, includes: Based on power project management standards, actual business processes, and the boundaries of the payment rules, project payment business is divided into four standard stages: project initiation, execution, acceptance, and quality assurance. The payment-specific characteristics of the four standard stages are extracted, and a dual-drive prediction architecture based on rules and data is constructed. The payment statistics of the target project's structured data are calculated using a sliding window method, and payment inertia is captured based on the payment statistics. The historical payment curve is decomposed into trend terms, periodic terms, and residual terms. The mapping table is transformed into computable constraints. Based on the payment inertia, trend terms, periodic terms, residual terms, constraints, and payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed. Based on the fusion results of the two channels in the dual-channel architecture, a payment curve is fitted, and labels for each stage of the project's entire life cycle and key payment node annotations are embedded in the payment curve to obtain a business pattern curve for the entire life cycle of project payment with confidence intervals.

5. The dynamic payment prediction method driven by the entire project lifecycle as described in claim 4, characterized in that, After converting the mapping table into computable constraints, the process also includes: A boundary verification function is established based on the constraints. The boundary verification function is used to filter the trend item and the period item by rules, and automatically triggers rule correction for the trend item and the period item that do not conform to the rules.

6. The dynamic payment prediction method driven by the entire project lifecycle as described in claim 4, characterized in that, After calculating the payment statistics of the target project's structured data using a sliding window method and capturing payment inertia based on the payment statistics, the method further includes: Extract payment features based on project and time dimensions; The features with positive impact among the payment features are selected, and the features with the least impact among the positive impact features are removed to obtain the first feature set; The features in the first feature set are optimized to obtain the second feature set; Based on the payment inertia, the trend term, the periodic term, the residual term, the constraints, and the payment rule matrix, a rule-guided, data-driven dual-channel architecture is constructed, including: Based on the second feature set, the trend term, the period term, the residual term, the constraint conditions, and the payment rule matrix, a rule-guided and data-driven dual-channel architecture is constructed.

7. The dynamic payment prediction method driven by the entire project lifecycle as described in claim 6, characterized in that, Using the target project data as a training set, a CART decision tree is trained as a weak learner. The weak learners are then integrated to obtain a strong learner, including: The target project data and the second feature set are used as the training set. The sample weights are optimized using the mapping table as business features. Combined with residual calculation, the CART decision tree is trained as a weak learner. After multiple rounds of iterative training, a weighted voting method is used to integrate the weak learners obtained from multiple rounds of training into a strong learner; Design a streaming data fusion framework to integrate pre-planned data of projects in the year to be tested with historical payment data in real time, and dynamically adjust the parameters of the strong learner by combining the business pattern curve and the features in the second feature set.

8. The dynamic payment prediction method driven by the entire project lifecycle according to any one of claims 1-7, characterized in that, After outputting the predicted project payment result based on the strong learner and the business pattern curve, the method further includes: Build a full-process analysis interface for projects, and dynamically display the stage characteristics of project initiation, contract execution, and payment through a "lifecycle sandbox".

9. A dynamic payment prediction device driven by the entire project lifecycle, characterized in that, include: The data processing module is used to acquire project data and process the project data to obtain target project data and payment rule boundaries. The regularity curve generation module is used to generate a business regularity curve for the entire life cycle of project payment based on power project management specifications, actual business processes, target project data, and payment rule boundaries. The model training module is used to use the target project data as a training set, train a CART decision tree as a weak learner, integrate the weak learner to obtain a strong learner, and dynamically adjust the parameters of the strong learner during the training process according to the pre-arranged data of the project in the year to be tested, historical payment data, and the business pattern curve. The model application module is used to output the prediction results of project payment based on the strong learner and the business pattern curve.

10. A terminal, comprising a memory and a processor, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the dynamic payment prediction method driven by the entire project lifecycle as described in any one of claims 1 to 8.

Images