A financial full-channel business dynamic self-adaptive scheduling optimization method and system

Abstract

The application discloses a kind of financial full-channel business dynamic self-adapting scheduling optimization method and system, it is related to computer processing technical field, the method includes the following steps: S01, based on the data acquisition chain established real-time acquisition service request data and processing resource state data, and the data obtained are handled, obtain calibrated data set;S02, the calibrated data set is input into the multi-model fusion time series prediction model constructed in advance, to obtain each channel service request volume in future preset period;Every interval preset period compares predicted value with actual business data, calculates deviation rate and triggers dynamic correction, uses filtering method to smooth the prediction result, output corrected service volume prediction result.The application breaks the channel island through unified data interface and cross-channel collaborative scheduling, significantly improves resource utilization and business response efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer processing technology, and in particular to a dynamic adaptive scheduling optimization method and system for omnichannel financial business. Background Technology

[0002] With the development of fintech, financial institutions (such as banks, securities companies, and insurance companies) are continuously expanding their business channels, now encompassing multiple business lines including centralized operations centers, remote video banking, offline branches, mobile sales, online trading platforms, and customer service centers. These channels exhibit characteristics such as cyclical fluctuations in business requests, frequent sudden peaks, highly differentiated business types, and stringent timeliness requirements under service level agreements. Simultaneously, financial systems face complex challenges due to diverse roles, significant differences in capabilities between primary and secondary staff, limited equipment resources, dynamic load changes, and cumulative fatigue.

[0003] In existing technologies, each business channel typically uses an independent scheduling system, forming "data silos" that cannot achieve a unified view and collaborative scheduling of resources across all channels, resulting in both low resource utilization and business backlog. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention employs a dynamic adaptive scheduling optimization method for omnichannel financial services, comprising the following steps:

[0005] S01. Based on the established data acquisition chain, collect business request data and processing resource status data in real time, and process the acquired data to obtain a calibrated dataset;

[0006] S02. Input the calibrated dataset into a pre-built multi-model fusion time-series prediction model to obtain the business request volume of each channel within a future preset time period;

[0007] At each preset interval, the predicted value is compared with the actual business data, the deviation rate is calculated and dynamic correction is triggered, the prediction result is smoothed and corrected by filtering method, and the corrected business volume prediction result is output.

[0008] S03. Construct a resource attribute library that includes static attributes and dynamic states. Based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and device status, construct a multi-dimensional resource scoring formula. Update the comprehensive score and ranking of each resource at preset intervals and output a resource priority queue.

[0009] S04. Construct a task priority weight formula based on business urgency, service level agreement timeliness, expected processing time, business value, task complexity, and channel characteristics. Calculate the comprehensive task priority by combining the business volume prediction results. Update the task priority queue at preset intervals.

[0010] S05. Construct a matching cost matrix based on the task priority queue and the resource priority queue, and process it according to the optimal algorithm based on the real-time business scenario to obtain the scheduling and allocation instruction;

[0011] S06. Optimize and solve the parameters of the entire process by backtracking historical data and using intelligent optimization algorithms, and feed the optimization results back to each stage to update the parameters of the entire process.

[0012] S07. Based on the rolling planning concept, steps S02-S06 are re-executed at each preset interval to generate a new scheduling scheme.

[0013] When the actual business volume deviates from the predicted value by more than a preset threshold, an emergency rescheduling is immediately triggered and the prediction model parameters are updated.

[0014] Preferably, in step S02, the process of comparing the predicted value with the actual business data at preset intervals, calculating the deviation rate and triggering dynamic correction, and using a filtering method to smooth and correct the prediction results, outputting the corrected business volume prediction results, includes:

[0015] S21. For business data with periodicity, a first time series model is used for fitting; for business data with seasonal effects, a second time series model is used for nonlinear fitting; and for business data with complex nonlinear relationships, a third time series model is used to capture long-term dependence and abrupt change characteristics.

[0016] S22. A weighted average integration strategy is adopted to assign dynamic weights to each model based on historical prediction accuracy and output the initial business volume prediction value.

[0017] S23. At each preset interval, the predicted value is compared with the real-time actual data. When the error rate exceeds the preset threshold, the integrated weights of each model are dynamically adjusted, and Kalman filtering is used to smooth and correct the prediction results.

[0018] Preferably, the multidimensional resource scoring formula in step S03 is as follows:

[0019] S=w1*R+w2*H+w3*C+w4*M+w5*Fe+w6*D;

[0020] Where R is the role matching degree, H is the historical performance score, C is the current load score, M is the multi-tasking capability score, F is the fatigue reverse score, D is the device status score, and w1-w6 are configurable weights.

[0021] Preferably, the task priority weight formula in step S04 is as follows:

[0022] T=α*U+β*De+γ*Ex+δ*Bu+ε*Co+ζ*Ch

[0023] Where α, β, γ, δ, ε, and ζ are configurable weights, U is the task urgency, De is the proximity to the SLA deadline, Ex is the expected processing time, Bu is the business value score, and Co is the task complexity score.

[0024] Based on the business volume forecast, when it is predicted that the business will enter a peak period in the future, the weight coefficient of the urgent task will be increased, and the idle order of high processing efficiency resources will be reserved in advance.

[0025] Preferably, step S05, which involves constructing a matching cost matrix based on the task priority queue and the resource priority queue, and processing it using the optimal algorithm according to the real-time business scenario to obtain scheduling and allocation instructions, includes:

[0026] S51. Construct a two-dimensional cost matrix, the elements of which represent the matching cost of allocating tasks to resources. The matching cost integrates role matching degree, task urgency and resource efficiency, current resource load, skill matching degree and geographical distance factor.

[0027] S52. When the real-time requirement is higher than the global optimal requirement, a greedy algorithm is used to match the resource with the lowest cost for the highest priority task in turn.

[0028] S53. When the requirement for global optimality is higher than the requirement for real-time performance, the Hungarian algorithm is used to find the minimum cost global optimal match.

[0029] Preferably, step S06, which involves optimizing the entire process parameters using historical data backtracking and intelligent optimization algorithms, and feeding the optimization results back to each stage to update the entire process parameters, includes:

[0030] S61. Construct a key performance indicator model for scheduling evaluation, and monitor the key performance indicator model for scheduling evaluation in real time using statistical methods. Trigger an abnormal alarm when the indicator deviates from the preset range.

[0031] S62. After the daily business is completed, a full replay simulation is performed using historical business data and scheduling decisions. Intelligent optimization algorithms are used to optimize and solve the scheduling-related weight parameters to generate the optimal parameter combination.

[0032] S63. Use the optimal parameter combination as the initial parameters for the next scheduling cycle.

[0033] Preferably, step S07, which involves immediately triggering emergency rescheduling and updating the prediction model parameters when the deviation between the actual business volume and the predicted value exceeds a preset threshold, includes:

[0034] S71. Set a fixed scheduling cycle. Before the start of each cycle, re-execute the entire process of business volume forecasting, resource scoring, task sorting, and matching decision-making to generate the optimal scheduling plan for that cycle.

[0035] S72. When the real-time business volume is detected to exceed the preset proportion of the predicted value, the emergency dispatch mechanism is triggered, the matching restrictions of high-load resources are relaxed, and idle secondary resources are included in the matching pool, and an emergency allocation plan is regenerated.

[0036] S73. Real-time monitoring of triggering conditions, including active release, timeout failure, handling anomalies, personnel offline, and task escalation, and execution of re-matching or cross-pool matching rescheduling processing.

[0037] A dynamic adaptive scheduling and optimization system for omnichannel financial services, used to implement the dynamic adaptive scheduling and optimization method for omnichannel financial services described in the above scheme, includes:

[0038] The data acquisition module is used to collect business request data and processing resource status data in real time based on the established data acquisition chain, and to process the acquired data to obtain a calibrated dataset.

[0039] The prediction and correction module is used to input the calibrated dataset into a pre-built multi-model fusion time-series prediction model to obtain the business request volume of each channel within a future preset time period; and to compare the predicted value with the actual business data at preset intervals, calculate the deviation rate and trigger dynamic correction, use filtering methods to smooth and correct the prediction results, and output the corrected business volume prediction results.

[0040] The data processing module is used to build a resource attribute library that includes static attributes and dynamic states. It constructs a multi-dimensional resource scoring formula based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and equipment status. It updates the comprehensive score and ranking of each resource at preset intervals and outputs a resource priority queue.

[0041] The data calculation module is used to construct a task priority weight formula based on business urgency, service level agreement timeliness, expected processing time, business value, task complexity and channel characteristics, calculate the comprehensive task priority in combination with the business volume prediction results, and update the task priority queue at preset intervals.

[0042] The matching and scheduling module is used to construct a matching cost matrix based on the task priority queue and the resource priority queue, and to process it using the optimal algorithm according to the real-time business scenario to obtain scheduling and allocation instructions.

[0043] The parameter optimization feedback module is used to optimize and solve the parameters of the entire process through historical data backtracking and intelligent optimization algorithms, and to feed back the optimization results to each module to update the parameters of the entire process.

[0044] The rolling planning and emergency scheduling module is used to re-trigger the prediction and correction module, the resource priority calculation module, the task priority calculation module, and the matching scheduling module at preset intervals based on the rolling planning concept, so as to generate a new scheduling scheme; and to immediately trigger emergency rescheduling and update the prediction model parameters when the deviation between the actual business volume and the predicted value exceeds a preset threshold.

[0045] The present invention has at least the following beneficial effects:

[0046] 1. This invention can simultaneously adapt to the multiple complex characteristics of financial operations, such as periodicity, seasonality, and sudden fluctuations. By combining half-hour-level dynamic deviation correction with Kalman filter smoothing correction mechanism, it significantly reduces prediction errors, providing a highly reliable decision-making basis for subsequent task sequencing and resource matching, and fundamentally avoiding scheduling mishaps caused by inaccurate predictions.

[0047] 2. A multi-dimensional resource scoring model encompassing static attributes and dynamic states was constructed, and combined with a task priority weight formula, enabling scheduling decisions to respond in real-time to changes in resource fatigue, current load, and task urgency. Through adaptive switching between greedy and Hungarian algorithms, the real-time performance of peak-hour scheduling and the global optimality of off-peak-hour scheduling are balanced. By continuously iterating scheduling parameters through daily review and intelligent optimization algorithms, the scheduling system possesses the ability to continuously self-evolve, maintaining optimal operational efficiency in the long term.

[0048] 3. When business volume forecasts deviate or sudden peaks occur, emergency scheduling can be automatically triggered, dynamically relaxing resource constraints and activating all available resources to quickly clear backlogged tasks. Simultaneously, real-time monitoring of various rescheduling trigger conditions and execution of automated processing strategies greatly improves response speed and system robustness to emergencies, effectively reduces SLA violation rates, and ensures high stability and continuity of financial operations. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0050] Figure 1 A flowchart of a dynamic adaptive scheduling optimization method and system for omnichannel financial services provided in Embodiment 1 of the present invention;

[0051] Figure 2 This is a flowchart of S02 provided in Embodiment 1 of the present invention;

[0052] Figure 3 This is a flowchart of S05 provided in Embodiment 1 of the present invention;

[0053] Figure 4 A flowchart of S06 provided in Embodiment 1 of the present invention;

[0054] Figure 5 The flowchart for S07 provided in Embodiment 1 of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0057] Example 1

[0058] This embodiment provides a dynamic adaptive scheduling optimization method and system for omnichannel financial services. The method includes the following steps: Figure 1 As shown:

[0059] S01. Based on the established data acquisition chain, collect business request data and processing resource status data in real time, and process the acquired data to obtain a calibrated dataset;

[0060] Specifically, the data acquisition chain in the above embodiment refers to the method of establishing a data serial port to acquire data from a predetermined website. The method of establishing the data serial port is to input the corresponding website link, user account and password into the system, and then set the execution command to actively acquire data according to a predetermined period.

[0061] Furthermore, the acquired data is processed, specifically including: after completing the real-time collection of business request data and processing resource status data, the system performs unified standardization processing on the acquired multi-source heterogeneous data to generate a high-quality calibrated dataset. First, for missing values ​​that may appear during data collection, mean interpolation or median interpolation methods are used to fill in the gaps. For example, for window business data with reporting delays, the median of data from the same period in the previous time period is used to fill in the gaps. Simultaneously, abnormal spikes or drops in data are identified and removed based on the interquartile range algorithm or Z-score method to avoid outliers interfering with subsequent predictions and scheduling. Second, the cleaned data is normalized or standardized. Min-Max normalization maps indicators such as business volume and resource load of different dimensions to a unified numerical range, ensuring that multi-source data are comparable in scale. Finally, feature construction is performed, which involves extracting derived features that are strongly related to scheduling decisions from the raw data. These features include, but are not limited to, task urgency features (based on customer complaint markings or timeout warning levels), target response time features (calculating the remaining response window based on SLA timeliness and the current time), peak business period features (marking fixed peak periods and dynamically identifying peak intervals), channel load balancing features (statistically calculating the ratio of current queue length to processing capacity for each channel), and task pool health features (comprehensive backlog, average waiting time, and other indicators).

[0062] S02. Input the calibrated dataset into the pre-built multi-model fusion time series prediction model to obtain the business request volume of each channel within the future preset time period;

[0063] At preset intervals, the predicted values ​​are compared with actual business data, the deviation rate is calculated, and dynamic correction is triggered. A filtering method is used to smooth and correct the prediction results, and the corrected business volume prediction results are output, including (e.g.) Figure 2 (as shown)

[0064] S21. For business data with periodicity, a first time series model is used for fitting; for business data with seasonal effects, a second time series model is used for nonlinear fitting; and for business data with complex nonlinear relationships, a third time series model is used to capture long-term dependence and abrupt change characteristics.

[0065] S22. A weighted average integration strategy is adopted to assign dynamic weights to each model based on historical prediction accuracy and output the initial business volume prediction value.

[0066] S23. At each preset interval, the predicted value is compared with the real-time actual data. When the error rate exceeds the preset threshold, the integrated weights of each model are dynamically adjusted, and Kalman filtering is used to smooth and correct the prediction results.

[0067] Specifically, through multi-model fusion and dynamic feedback correction mechanisms, accurate prediction of future business request volumes across all financial channels is achieved. First, the system inputs the calibrated, standardized time-series dataset into a pre-constructed multi-model fusion time-series prediction model. This model is not a single algorithm but integrates three types of time-series prediction models tailored to the different characteristics of financial business data: For data with strong periodicity and stability (such as corporate transfers in centralized operations centers, whose volume typically exhibits stable fluctuations on a weekly or monthly basis), a first time-series model (e.g., ARIMA model or autoregressive integral moving average model) is used for fitting, utilizing its autoregressive and moving average components to capture the inherent periodicity and trends of the data; for business data containing holiday effects and seasonal fluctuations (such as retail business showing significant peaks on bank closing dates, "Double Eleven," etc.), a second time-series model (e.g., ARIMA model or autoregressive integral moving average model) is used. For example, the Prophet model, an open-source decomposable time series model from Facebook, decomposes the time series into trend, seasonal, and holiday terms by introducing external regression variables such as weekends, public holidays, and specific promotional days, thereby flexibly fitting nonlinear changes. For business data with high noise levels, affected by sudden policies or market events, and complex nonlinear relationships (such as a surge in cloud cabinet requests or customer service complaints), a third time series model (such as the LSTM model, Long Short-Term Memory network) is used. It utilizes the memory units and gating mechanisms in its recurrent neural network structure to learn long-term dependencies in the sequence and respond quickly to sudden changes.

[0068] After obtaining the prediction results of the three types of models, the system uses a weighted average ensemble strategy to fuse them. Specifically, the ensemble weights are dynamically allocated based on the historical prediction accuracy of each model over a recent period (such as the average absolute percentage error over the past 7 days), meaning that the model with higher accuracy has a larger weight.

[0069] To overcome the problem that a single integrated weight cannot adapt to real-time business fluctuations, the system introduces a dynamic deviation correction mechanism. At preset intervals (e.g., half an hour), the system compares the actual business volume of each channel with the initial forecast value within that period, calculating the deviation rate as (actual value - forecast value) / forecast value. When the deviation rate exceeds a preset threshold, the prediction is deemed inaccurate. The system immediately and dynamically adjusts the integrated weights of the three models, predicts the current business volume based on the correction value from the previous moment, and then uses the current actual observation value (or the weighted forecast value) to optimally correct the prediction result, thereby outputting a smoother and more stable final business volume prediction result.

[0070] S03. Construct a resource attribute library that includes static attributes and dynamic states. Based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and device status, construct a multi-dimensional resource scoring formula. Update the comprehensive score and ranking of each resource at preset intervals and output a resource priority queue.

[0071] The multidimensional resource scoring formula mentioned above is as follows:

[0072] S=w1*R+w2*H+w3*C+w4*M+w5*Fe+w6*D;

[0073] Where R is the role matching degree, H is the historical performance score, C is the current load score, M is the multi-tasking capability score, F is the fatigue reverse score, D is the device status score, and w1-w6 are configurable weights.

[0074] Specifically, during runtime, an attribute library needs to be established for each processing resource, encompassing both static attributes and dynamic status. Static attributes, once configured, are relatively stable and include the resource's primary job type (management role), a list of secondary job types (basic roles), historical performance (processing success rate, average processing time, customer satisfaction rate), qualification certificates, service scope, and hourly cost. Dynamic status is continuously updated through real-time data streams, including the current number of queued tasks, the amount of tasks processed, real-time busy / idle status (idle / busy / standby), fatigue score (linearly or exponentially increasing based on continuous working hours), real-time geographical location (suitable for on-site service scenarios), and device concurrency status (such as the number of video streams processed simultaneously by a cloud cabinet terminal).

[0075] The aforementioned role matching score indicates the degree of matching between the primary / secondary role of the resource and the type of task to be assigned. The score is set to 1.0 when the primary role matches, 0.5 when the secondary role matches, and 0 when there is no match. The historical performance score is a normalized weighted value of indicators such as processing success rate and average processing time. The current load score is inversely proportional to the current number of queued tasks or task slack of the resource. The multi-tasking capability score is calculated based on the number of task types or concurrency capability that the resource can process in parallel. The fatigue score decreases as the continuous working time increases to prevent excessive fatigue from causing a decline in quality. The equipment status score is dynamically assigned based on online status, concurrency margin, etc., and resets to zero when offline or fully loaded.

[0076] Furthermore, before each scheduling cycle (e.g., every 30 minutes) or each triggered rescheduling, the system iterates through all available resources, obtains their dynamic status in real time, and calculates a comprehensive score for each resource using the scoring formula mentioned above. Subsequently, resources are sorted from highest to lowest score, generating a resource priority queue. Resources with high scores indicate the best role match, lightest load, lowest fatigue, and best equipment condition, thus receiving priority for task allocation. Primary role resources, naturally superior to secondary role resources in role matching, are usually placed at the front of the queue, prioritizing primary roles. Secondary role resources, even when the primary task pool is full or waiting times exceed limits, can still enter the tail of the queue and be activated for elastic matching based on their role matching score (0.5 for secondary roles) in the scoring formula. Equipment resources (such as cloud cabinets) have their availability dynamically adjusted based on their equipment status score.

[0077] S04. Construct a task priority weight formula based on business urgency, service level agreement validity period, expected processing time, business value, task complexity, and channel characteristics. Calculate the overall task priority by combining the business volume forecast results. Update the task priority queue at preset intervals.

[0078] The formula for the task priority weighting mentioned above is as follows:

[0079] T=α*U+β*De+γ*Ex+δ*Bu+ε*Co+ζ*Ch

[0080] Where α, β, γ, δ, ε, and ζ are configurable weights, U is the task urgency, De is the proximity to the SLA deadline, Ex is the expected processing time, Bu is the business value score, and Co is the task complexity score.

[0081] Based on the business volume forecast, when it is predicted that the business will enter a peak period in the future, the weight coefficient of the urgent task will be increased, and the idle order of high processing efficiency resources will be reserved in advance.

[0082] Specifically, by calculating the overall priority of each task to be processed in real time and using a priority queue data structure to achieve dynamic sorting and updating, tasks with high urgency, high value and those close to the Service Level Agreement (SLA) deadline can be scheduled with priority.

[0083] In actual operation, it is assumed that the value is set to (0.3, 0.25, 0.1, 0.15, 0.1, 0.1) during normal periods, while α can be dynamically increased to 0.5 during peak periods to enhance emergency response. The calculation method for each factor is as follows: Task urgency is assigned based on business tags (e.g., customer complaints and regulatory reports are assigned the highest level of 1.0, while regular business is assigned 0.2); SLA deadline proximity is 1 - (remaining processing time / total allowed time). When the remaining time approaches zero, the factor approaches 1, triggering priority processing; Expected processing time is normalized using the average processing time of similar tasks in history, and then the reciprocal or inverse is taken to give higher priority to tasks with shorter processing times; Business value is quantified by combining customer level and expected revenue (e.g., VIP customer tasks are assigned 0.9, while ordinary customer tasks are assigned 0.3); Task complexity is assessed based on the number of skills required, the required qualification level, etc. (level 5 complexity is assigned 0.8, level 1 is assigned 0.2) to prevent complex tasks from accumulating in the long term; Channel characteristic coefficient is dynamically adjusted based on the channel's real-time load and resource scarcity.

[0084] S05. Construct a matching cost matrix based on the task priority queue and resource priority queue, and process it using the optimal algorithm according to the real-time business scenario to obtain scheduling and allocation instructions, including (such as...) Figure 3 (as shown)

[0085] S51. Construct a two-dimensional cost matrix, whose elements represent the matching cost of allocating tasks to resources. The matching cost takes into account the role matching degree, task urgency and resource efficiency, current resource load, skill matching degree and geographical distance factor.

[0086] S52. When the real-time requirement is higher than the global optimal requirement, a greedy algorithm is used to match the resource with the lowest cost for the highest priority task in turn.

[0087] S53. When the requirement for global optimality is higher than the requirement for real-time performance, the Hungarian algorithm is used to find the minimum cost global optimal match.

[0088] Specifically, the system constructs a two-dimensional matching cost matrix based on the generated task priority queue and resource priority queue. The rows of the matrix correspond to tasks to be scheduled, and the columns correspond to available processing resources. Each matrix element c(i,j) represents the comprehensive matching cost of assigning task i to resource j. This cost is not a single-dimensional value, but rather a weighted result of multiple key factors: role matching degree (the degree to which the resource's primary or secondary role matches the task type), task urgency and resource efficiency (higher task priority and higher resource processing efficiency result in lower costs), current resource load (the more tasks currently queued and the heavier the load, the higher the cost), skill matching degree (whether the resource possesses the qualifications or certificates required for the task), and geographical distance factor (for scenarios such as on-site services, the distance between the resource's real-time location and the task location). By weighting and summing these factors, the system can quantify complex scheduling constraints into a unified cost metric.

[0089] After the cost matrix is ​​constructed, the system automatically selects the optimal matching algorithm based on the real-time business scenario. When the business is at its peak or the system detects a surge in task backlog, the real-time requirement takes precedence over the global optimal requirement. In order of task priority from high to low, the system selects the lowest-cost and available resource in the current cost matrix for each task and immediately generates an allocation instruction.

[0090] S06. Optimize and solve the parameters of the entire process through historical data backtracking and intelligent optimization algorithms, and feed the optimization results back to each stage to update the parameters of the entire process, including (e.g.) Figure 4 (as shown)

[0091] S61. Construct a key performance indicator model for scheduling evaluation, and monitor the key performance indicator model for scheduling evaluation in real time using statistical methods. Trigger an abnormal alarm when the indicator deviates from the preset range.

[0092] S62. After the daily business is completed, a full replay simulation is performed using historical business data and scheduling decisions. Intelligent optimization algorithms are used to optimize and solve the scheduling-related weight parameters to generate the optimal parameter combination.

[0093] S63. Use the optimal parameter combination as the initial parameters for the next scheduling cycle.

[0094] Specifically, statistical methods such as moving averages and standard deviations are used to monitor various indicators in real time, setting reasonable fluctuation ranges for each indicator. When any indicator deviates from the preset range, the system immediately triggers an anomaly alarm and pushes it to operations personnel through multiple channels, indicating the potential failure risk of the current scheduling strategy. Simultaneously, the system records all daily business requests, scheduling decisions, resource status, and execution results, forming a complete scheduling log. After the daily business concludes, the system uses this historical data for full-scale replay simulation: based on the actual arrival sequence of tasks and changes in resource status, the system recalculates the matching scheme using the saved scheduling strategy, and compares the simulation results with the actual execution results to quantitatively evaluate the impact of each parameter on the scheduling effect. Based on this, intelligent optimization algorithms (such as genetic algorithms or particle swarm optimization) are used to optimize the weight parameters throughout the process. A set of weight parameters is encoded as individuals, with the optimization objectives of minimizing the average task waiting time and maximizing resource utilization. Through selection, crossover, and mutation operations, the system iteratively searches within the parameter space, converging to the optimal parameter combination after approximately 100 generations of evolution.

[0095] S07. Based on the rolling planning concept, steps S02-S06 are re-executed at each preset interval to generate a new scheduling scheme.

[0096] When the deviation between the actual business volume and the predicted value exceeds a preset threshold, an emergency rescheduling is immediately triggered and the prediction model parameters are updated, including (such as...). Figure 5 (as shown)

[0097] S71. Set a fixed scheduling cycle. Before the start of each cycle, re-execute the entire process of business volume forecasting, resource scoring, task sorting, and matching decision-making to generate the optimal scheduling plan for that cycle.

[0098] S72. When the real-time business volume is detected to exceed the preset proportion of the predicted value, the emergency dispatch mechanism is triggered, the matching restrictions of high-load resources are relaxed, and idle secondary resources are included in the matching pool, and an emergency allocation plan is regenerated.

[0099] S73. Real-time monitoring of triggering conditions, including active release, timeout failure, handling anomalies, personnel offline, and task escalation, and execution of re-matching or cross-pool matching rescheduling processing.

[0100] Specifically, during runtime, the system sets a fixed-length time window as the basic scheduling cycle, typically half an hour or one hour. Before each cycle begins, the system automatically triggers a complete scheduling process: re-executing step S02 (business volume prediction, updating model parameters based on the latest historical data), step S03 (recalculating and sorting the comprehensive score based on the real-time load, fatigue, and equipment status of each resource), step S04 (task priority reordering, considering changes in the urgency of newly arrived tasks and unfinished tasks), and step S05 (matching decision), generating the optimal scheduling scheme for that cycle.

[0101] Secondly, the system has an independent monitoring thread that compares the actual volume of business received with the predicted value output in step S02 in real time. When the actual volume of business received exceeds the preset proportion of the predicted value or the absolute deviation exceeds the safety threshold, the regular periodic waiting is immediately interrupted, triggering the emergency rescheduling mechanism. The core logic of emergency scheduling is to dynamically relax the matching constraints on high-load resources, re-include resources that were originally restricted from matching due to excessive load into the candidate pool, forcibly activate all idle secondary resources (even if their primary task pool is not completely idle), and temporarily increase the weight of the "processing efficiency" factor and decrease the weight of the "load balancing" factor in the matching cost matrix, thereby quickly generating an emergency allocation plan within seconds to centrally digest sudden backlog of tasks.

[0102] Example 2

[0103] This embodiment is used to implement the method in Embodiment 1 above, and its system includes:

[0104] The data acquisition module is used to collect business request data and processing resource status data in real time based on the established data acquisition chain, and to process the acquired data to obtain a calibrated dataset.

[0105] The prediction and correction module is used to input the calibrated dataset into the pre-built multi-model fusion time series prediction model to obtain the business request volume of each channel within a preset time period in the future; and to compare the predicted value with the actual business data at preset intervals, calculate the deviation rate and trigger dynamic correction, use filtering methods to smooth and correct the prediction results, and output the corrected business volume prediction results.

[0106] The data processing module is used to build a resource attribute library that includes static attributes and dynamic states. It constructs a multi-dimensional resource scoring formula based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and equipment status. It updates the comprehensive score and ranking of each resource at preset intervals and outputs a resource priority queue.

[0107] The data calculation module is used to construct a task priority weight formula based on business urgency, service level agreement timeliness, expected processing time, business value, task complexity, and channel characteristics. It also calculates the comprehensive task priority by combining the business volume prediction results and updates the task priority queue at preset intervals.

[0108] The matching and scheduling module is used to construct a matching cost matrix based on the task priority queue and the resource priority queue, and to process it using the optimal algorithm according to the real-time business scenario to obtain the scheduling and allocation instructions.

[0109] The parameter optimization feedback module is used to optimize the parameters of the entire process by backtracking historical data and using intelligent optimization algorithms, and to feed the optimization results back to each module to update the parameters of the entire process.

[0110] The rolling planning and emergency scheduling module is used to re-trigger the prediction and correction module, resource priority calculation module, task priority calculation module and matching scheduling module at preset intervals based on the rolling planning concept, so as to generate a new scheduling scheme; and to immediately trigger emergency rescheduling and update the prediction model parameters when the deviation between the actual business volume and the predicted value exceeds a preset threshold.

[0111] Example 3

[0112] This invention provides a non-transitory computer-readable storage medium storing at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the following steps:

[0113] Based on the established data acquisition chain, business request data and processing resource status data are collected in real time, and the acquired data is processed to obtain a calibrated dataset;

[0114] The calibrated dataset is input into a pre-built multi-model fusion time series prediction model to obtain the business request volume of each channel within a future preset time period;

[0115] At each preset interval, the predicted value is compared with the actual business data, the deviation rate is calculated and dynamic correction is triggered, the prediction result is smoothed and corrected by filtering method, and the corrected business volume prediction result is output.

[0116] Construct a resource attribute library that includes static attributes and dynamic states. Build a multi-dimensional resource scoring formula based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and device status. Update the comprehensive score and ranking of each resource at preset intervals and output a resource priority queue.

[0117] A task priority weight formula is constructed based on business urgency, service level agreement validity period, expected processing time, business value, task complexity and channel characteristics. The comprehensive task priority is calculated in combination with business volume prediction results, and the task priority queue is updated at preset intervals.

[0118] A matching cost matrix is ​​constructed based on the task priority queue and the resource priority queue. The optimal algorithm is then used to process the data according to the real-time business scenario to obtain the scheduling and allocation instructions.

[0119] By backtracking historical data and using intelligent optimization algorithms, the parameters of the entire process are optimized and solved, and the optimization results are fed back to each stage to update the parameters of the entire process.

[0120] Based on the concept of rolling planning, the above steps are repeated every preset period to generate a new scheduling scheme.

[0121] When the actual business volume deviates from the predicted value by more than a preset threshold, an emergency rescheduling is immediately triggered and the prediction model parameters are updated.

[0122] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0123] 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 used as 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.

[0124] Example 4

[0125] This invention provides an electronic device, including a processor and a memory, wherein the memory stores at least one instruction or at least one program segment, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement the following steps:

[0126] Based on the established data acquisition chain, business request data and processing resource status data are collected in real time, and the acquired data is processed to obtain a calibrated dataset;

[0127] The calibrated dataset is input into a pre-built multi-model fusion time series prediction model to obtain the business request volume of each channel within a future preset time period;

[0128] At each preset interval, the predicted value is compared with the actual business data, the deviation rate is calculated and dynamic correction is triggered, the prediction result is smoothed and corrected by filtering method, and the corrected business volume prediction result is output.

[0129] Construct a resource attribute library that includes static attributes and dynamic states. Build a multi-dimensional resource scoring formula based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and device status. Update the comprehensive score and ranking of each resource at preset intervals and output a resource priority queue.

[0130] A task priority weight formula is constructed based on business urgency, service level agreement validity period, expected processing time, business value, task complexity and channel characteristics. The comprehensive task priority is calculated in combination with business volume prediction results, and the task priority queue is updated at preset intervals.

[0131] A matching cost matrix is ​​constructed based on the task priority queue and the resource priority queue. The optimal algorithm is then used to process the data according to the real-time business scenario to obtain the scheduling and allocation instructions.

[0132] By backtracking historical data and using intelligent optimization algorithms, the parameters of the entire process are optimized and solved, and the optimization results are fed back to each stage to update the parameters of the entire process.

[0133] Based on the concept of rolling planning, the above steps are repeated every preset period to generate a new scheduling scheme.

[0134] When the actual business volume deviates from the predicted value by more than a preset threshold, an emergency rescheduling is immediately triggered and the prediction model parameters are updated.

[0135] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic adaptive scheduling and optimization method for omnichannel financial services, characterized in that, Includes the following steps: S01. Based on the established data acquisition chain, collect business request data and processing resource status data in real time, and process the acquired data to obtain a calibrated dataset; S02. Input the calibrated dataset into a pre-built multi-model fusion time-series prediction model to obtain the business request volume of each channel within a future preset time period; At each preset interval, the predicted value is compared with the actual business data, the deviation rate is calculated and dynamic correction is triggered, the prediction result is smoothed and corrected by filtering method, and the corrected business volume prediction result is output. S03. Construct a resource attribute library that includes static attributes and dynamic states. Based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and device status, construct a multi-dimensional resource scoring formula. Update the comprehensive score and ranking of each resource at preset intervals and output a resource priority queue. S04. Construct a task priority weight formula based on business urgency, service level agreement timeliness, expected processing time, business value, task complexity, and channel characteristics. Calculate the comprehensive task priority by combining the business volume prediction results. Update the task priority queue at preset intervals. S05. Construct a matching cost matrix based on the task priority queue and the resource priority queue, and process it according to the optimal algorithm based on the real-time business scenario to obtain the scheduling and allocation instruction; S06. Optimize and solve the parameters of the entire process by backtracking historical data and using intelligent optimization algorithms, and feed the optimization results back to each stage to update the parameters of the entire process. S07. Based on the rolling planning concept, steps S02-S06 are re-executed at each preset interval to generate a new scheduling scheme. When the actual business volume deviates from the predicted value by more than a preset threshold, an emergency rescheduling is immediately triggered and the prediction model parameters are updated.

2. The dynamic adaptive scheduling optimization method for omnichannel financial services according to claim 1, characterized in that, In step S02, the predicted value is compared with the actual business data at each preset interval, the deviation rate is calculated and dynamic correction is triggered, and a filtering method is used to smooth and correct the prediction result, outputting the corrected business volume prediction result, including: S21. For business data with periodicity, a first time series model is used for fitting; for business data with seasonal effects, a second time series model is used for nonlinear fitting; and for business data with complex nonlinear relationships, a third time series model is used to capture long-term dependence and abrupt change characteristics. S22. A weighted average integration strategy is adopted to assign dynamic weights to each model based on historical prediction accuracy and output the initial business volume prediction value. S23. At each preset interval, the predicted value is compared with the real-time actual data. When the error rate exceeds the preset threshold, the integrated weights of each model are dynamically adjusted, and Kalman filtering is used to smooth and correct the prediction results.

3. The dynamic adaptive scheduling optimization method for omnichannel financial services according to claim 1, characterized in that, The multidimensional resource scoring formula in step S03 is as follows: S=w1*R+w2*H+w3*C+w4*M+w5*Fe+w6*D; Where R is the role matching degree, H is the historical performance score, C is the current load score, M is the multi-tasking capability score, F is the fatigue reverse score, D is the device status score, and w1-w6 are configurable weights.

4. The dynamic adaptive scheduling optimization method for omnichannel financial services according to claim 1, characterized in that, The task priority weighting formula in step S04 is as follows: T=α*U+β*De+γ*Ex+δ*Bu+ε*Co+ζ*Ch Where α, β, γ, δ, ε, and ζ are configurable weights, U is the task urgency, De is the proximity to the SLA deadline, Ex is the expected processing time, Bu is the business value score, and Co is the task complexity score. Based on the business volume forecast, when it is predicted that the business will enter a peak period in the future, the weight coefficient of the urgent task will be increased, and the idle order of high processing efficiency resources will be reserved in advance.

5. The dynamic adaptive scheduling optimization method for omnichannel financial services according to claim 1, characterized in that, Step S05, which involves constructing a matching cost matrix based on the task priority queue and the resource priority queue, and processing it using the optimal algorithm according to the real-time business scenario to obtain scheduling and allocation instructions, includes: S51. Construct a two-dimensional cost matrix, the elements of which represent the matching cost of allocating tasks to resources. The matching cost integrates role matching degree, task urgency and resource efficiency, current resource load, skill matching degree and geographical distance factor. S52. When the real-time requirement is higher than the global optimal requirement, a greedy algorithm is used to match the resource with the lowest cost for the highest priority task in turn. S53. When the requirement for global optimality is higher than the requirement for real-time performance, the Hungarian algorithm is used to find the minimum cost global optimal match.

6. The dynamic adaptive scheduling optimization method for omnichannel financial services according to claim 1, characterized in that, Step S06, which involves optimizing the entire process parameters using historical data backtracking and intelligent optimization algorithms, and feeding the optimization results back to each stage to update the entire process parameters, includes: S61. Construct a key performance indicator model for scheduling evaluation, and monitor the key performance indicator model for scheduling evaluation in real time using statistical methods. Trigger an abnormal alarm when the indicator deviates from the preset range. S62. After the daily business is completed, a full replay simulation is performed using historical business data and scheduling decisions. Intelligent optimization algorithms are used to optimize and solve the scheduling-related weight parameters to generate the optimal parameter combination. S63. Use the optimal parameter combination as the initial parameters for the next scheduling cycle.

7. The dynamic adaptive scheduling optimization method for omnichannel financial services according to claim 1, characterized in that, Step S07, which states that when the deviation between the actual business volume and the predicted value exceeds a preset threshold, immediately triggers emergency rescheduling and updates the prediction model parameters, includes: S71. Set a fixed scheduling cycle. Before the start of each cycle, re-execute the entire process of business volume forecasting, resource scoring, task sorting, and matching decision-making to generate the optimal scheduling plan for that cycle. S72. When the real-time business volume is detected to exceed the preset proportion of the predicted value, the emergency dispatch mechanism is triggered, the matching restrictions of high-load resources are relaxed, and idle secondary resources are included in the matching pool, and an emergency allocation plan is regenerated. S73. Real-time monitoring of triggering conditions, including active release, timeout failure, handling anomalies, personnel offline, and task escalation, and execution of re-matching or cross-pool matching rescheduling processing.

8. A dynamic adaptive scheduling and optimization system for omnichannel financial services, characterized in that, A method for implementing the dynamic adaptive scheduling optimization of omnichannel financial services as described in any one of claims 1-7 includes: The data acquisition module is used to collect business request data and processing resource status data in real time based on the established data acquisition chain, and to process the acquired data to obtain a calibrated dataset. The prediction and correction module is used to input the calibrated dataset into a pre-built multi-model fusion time-series prediction model to obtain the business request volume of each channel within a future preset time period; and to compare the predicted value with the actual business data at preset intervals, calculate the deviation rate and trigger dynamic correction, use filtering methods to smooth and correct the prediction results, and output the corrected business volume prediction results. The data processing module is used to build a resource attribute library that includes static attributes and dynamic states. It constructs a multi-dimensional resource scoring formula based on role matching degree, historical performance, current load, multi-tasking capability, fatigue level, and equipment status. It updates the comprehensive score and ranking of each resource at preset intervals and outputs a resource priority queue. The data calculation module is used to construct a task priority weight formula based on business urgency, service level agreement timeliness, expected processing time, business value, task complexity and channel characteristics, calculate the comprehensive task priority in combination with the business volume prediction results, and update the task priority queue at preset intervals. The matching and scheduling module is used to construct a matching cost matrix based on the task priority queue and the resource priority queue, and to process it using the optimal algorithm according to the real-time business scenario to obtain scheduling and allocation instructions. The parameter optimization feedback module is used to optimize and solve the parameters of the entire process through historical data backtracking and intelligent optimization algorithms, and to feed back the optimization results to each module to update the parameters of the entire process. The rolling planning and emergency scheduling module is used to re-trigger the prediction and correction module, the resource priority calculation module, the task priority calculation module, and the matching scheduling module at preset intervals based on the rolling planning concept, so as to generate a new scheduling scheme; and to immediately trigger emergency rescheduling and update the prediction model parameters when the deviation between the actual business volume and the predicted value exceeds a preset threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the financial omnichannel business dynamic adaptive scheduling optimization method and system as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the dynamic adaptive scheduling optimization method and system for omnichannel financial business as described in any one of claims 1 to 7.

Images