A strategy generation method, system, platform and storage medium based on dynamic search of decision parameters

Abstract

The application discloses a strategy generation method, system and platform based on decision parameter dynamic search and a storage medium; the application scheme constructs a prediction model with counterfactual prediction capability, encodes strategy logic into an optimizable mathematical parameter, and adopts a gradient-guided evolutionary search algorithm to perform efficient optimization in a mixed parameter space, while being supplemented with an abnormality detection and disturbance simulation dual verification mechanism to screen a robust strategy, finally forming an automated closed loop of prediction, search, verification, execution, feedback and re-prediction, realizing intelligent generation of strategy parameters, accurate pre-judgment before execution, and continuous self-adaptive dynamic adjustment of the system to environmental changes, thereby improving decision quality while reducing artificial parameter adjustment cost and trial-and-error risk, and ensuring long-term stability and generalization ability of the strategy in a complex dynamic environment.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a strategy generation method, system, platform, and storage medium based on dynamic search of decision parameters. Background Technology

[0002] In many technical fields that require decision-making, allocation, and parameter configuration for large-scale objects, strategy formulation and optimization are key factors affecting overall system efficiency. Taking resource scheduling, task allocation, and service routing as examples, decision-making systems typically need to allocate appropriate execution paths, execution entities, and specific execution parameter configurations to each target object based on its attribute characteristics.

[0003] Currently, existing strategies are typically based on fixed rules (such as categorization by attribute or region), and these rules remain unchanged for a long period once set. Because the external environment (such as changes in the capabilities of the implementing entity or shifts in object distribution) is dynamically evolving, static rules cannot adapt to these changes, leading to a continuous decline in the timeliness and accuracy of the strategies. Furthermore, the evaluation of strategy effectiveness highly depends on complete feedback of the execution results, which typically takes a long time. The system may need to wait months or even longer to obtain complete data for evaluating the strategy's merits, causing strategy adjustments to lag significantly behind changes in the environment and business needs.

[0004] Furthermore, existing solutions lack the ability to counterfactually infer the effects of different strategies. Decision-makers cannot predict the merits of a strategy before implementation and must rely on costly trial and error or A / B testing to explore its potential, which is unacceptable in large-scale systems. Moreover, the parameter space is enormous, making manual tuning difficult. The strategy parameters typically involve multi-dimensional decision-making logic (such as path selection thresholds, weight allocation, and parameter ratios), resulting in a vast parameter space containing a mixture of discrete and continuous variables. Manual parameter tuning is not only inefficient but also prone to getting trapped in local optima, failing to guarantee a globally optimal solution.

[0005] Therefore, in order to address the aforementioned technical deficiencies, there is an urgent need to design and develop a strategy generation method, system, platform, and storage medium based on dynamic search of decision parameters. Summary of the Invention

[0006] To overcome the shortcomings and difficulties of the existing technology, the present invention aims to provide a strategy generation method, system, platform and storage medium based on dynamic search of decision parameters, which can accurately predict the effect of unexecuted strategies; can automatically and efficiently search for optimal strategy parameters in a large parameter space based on the prediction results; and can continuously collect feedback data during execution to achieve adaptive dynamic adjustment of the strategy.

[0007] The first objective of this invention is to provide a strategy generation method based on dynamic search of decision parameters; the second objective of this invention is to provide a strategy generation system based on dynamic search of decision parameters; the third objective of this invention is to provide a strategy generation platform based on dynamic search of decision parameters; and the fourth objective of this invention is to provide a computer-readable storage medium.

[0008] The first objective of this invention is achieved by the method comprising: First data corresponding to the target object is generated and acquired. The first data is then combined with multiple candidate decision codes. Combined with a trained prediction model, second data corresponding to the target object and under each candidate decision code is constructed and acquired. The first data is feature data, and the second data is predicted state data. Based on the second data, target parameters that satisfy the preset optimization objective are dynamically searched in a predefined parameter space; wherein, the parameters in the parameter space are used to generate the decision code; the parameters in the parameter space are used to define a decision function, which outputs the object allocation path, execution subject identifier, and execution parameter configuration in sequence according to the feature data; the parameters include the weight parameters and bias parameters of the decision function; The stability verification process handles the target parameters, generates third data corresponding to the target parameters, and generates and outputs corresponding fourth data based on the third data; wherein, the third data is the execution parameters and is the target parameters that have passed the verification; the fourth data is the control instructions generated based on the execution parameters.

[0009] Further, the step of creating, generating, and acquiring first data corresponding to the target object, combining the first data with multiple candidate decision codes respectively, and combining with a trained prediction model to construct, generate, and acquire second data corresponding to the target object under each candidate decision code, further includes: A fifth set of data corresponding to the target object is generated and extracted; wherein, the fifth set of data is initial attribute data; The continuous data in the fifth data is normalized, and the discrete data in the fifth data is mapped to quantization. Based on the processed continuous data and the discrete data, corresponding first data is created.

[0010] Furthermore, the prediction model is trained in the following manner: Obtain a historical record set; wherein the historical record set contains feature data of historical objects, historical decision codes, and corresponding historical state results; The feature data of the historical object and the historical decision encoding are input into the prediction model to be trained to generate the prediction state result; The parameters of the prediction model to be trained are adjusted based on the difference between the predicted state and the historical state, as well as the propensity score of the historical decision encoding.

[0011] Furthermore, the step of dynamically searching for target parameters that satisfy a preset optimization objective in a predefined parameter space based on the second data also includes: Create an initial population containing individuals with multiple parameters, and repeat the following steps until the iteration stopping condition is met: The fitness value corresponding to each parameter individual in the population is calculated based on the second data; wherein, the fitness value is used to characterize the difference between the expected return of the strategy determined by the corresponding parameter individual and the load balancing cost; the expected return is determined based on the efficiency index, timeliness index and success probability index output by the prediction model; Based on the gradient information of the prediction model, the continuous parameter components in the individual parameters are mutated. Cross-recombination processing is applied to the discrete parameter components in the individual parameters. Based on the fitness value, individuals are selected from the current population to enter the next generation population; The iteration stopping condition includes any one of the following: the number of iterations reaches a preset threshold; the change in the fitness value of the best individual in the population within a consecutive preset number of iterations is less than a change threshold.

[0012] Furthermore, the stability verification process, which involves processing the target parameter, creating third data corresponding to the target parameter, and generating and outputting corresponding fourth data based on the third data, further includes: Construct and obtain a decision distribution corresponding to the target parameters, and calculate the deviation between the generated decision distribution and the reference distribution; The first data is perturbed, and the fluctuation range of the effect of generating the target parameter under the perturbed feature data is evaluated. The third data is selected from multiple target parameters based on the deviation and the fluctuation range of the effect.

[0013] Furthermore, the method also includes: The system generates and acquires sixth data after the control command is executed, and updates the model parameters of the prediction model based on the sixth data; wherein the sixth data is actual feedback data. When a preset trigger event is detected, the step of dynamically searching for target parameters in the parameter space is re-executed based on the updated prediction model.

[0014] The second objective of this invention is achieved by the system used to implement the strategy generation method based on dynamic search of decision parameters, the system comprising: A generation unit is created to generate and acquire first data corresponding to a target object, and to combine the first data with multiple candidate decision codes respectively. Combined with a trained prediction model, second data corresponding to the target object and under each candidate decision code is constructed and acquired. The first data is feature data, and the second data is predicted state data. A dynamic search unit is used to dynamically search for target parameters that satisfy a preset optimization objective in a predefined parameter space based on the second data; wherein, the parameters in the parameter space are used to generate the decision code; the parameters in the parameter space are used to define a decision function, the decision function outputting an object allocation path, an execution entity identifier, and an execution parameter configuration in sequence according to the feature data; the parameters include the weight parameters and bias parameters of the decision function; The verification processing unit is used to perform stability verification processing on the target parameters, create and generate third data corresponding to the target parameters, and generate and output corresponding fourth data based on the third data; wherein, the third data is an execution parameter and is a target parameter that has passed verification; the fourth data is a control instruction generated based on the execution parameters.

[0015] Furthermore, the creation and generation unit further includes: A generation module is established to generate and extract fifth data corresponding to the target object; wherein, the fifth data is initial attribute data; The normalization processing module is used to normalize the continuous data in the fifth data and to process the discrete data in the fifth data by quantization mapping. The first generation module is used to create corresponding first data based on the processed continuous data and the discrete data; The prediction model is trained using the following modules: An acquisition module is used to acquire a historical record set; wherein, the historical record set contains feature data of historical objects, historical decision codes, and corresponding historical state results; The second generation module is used to input the feature data of the historical object and the historical decision code into the prediction model to be trained, and generate the prediction state result. The adjustment processing module is used to adjust the parameters of the prediction model to be trained based on the difference between the predicted state result and the historical state result, as well as the propensity score of the historical decision code. And / or, the dynamic search unit further includes: The third generation module is used to create an initial population containing individuals with multiple parameters, and repeats the following module until the iteration stopping condition is met: The fourth generation module is used to calculate and generate a fitness value corresponding to each parameter individual in the population based on the second data; wherein, the fitness value is used to characterize the difference between the expected return of the strategy determined by the corresponding parameter individual and the load balancing cost; the expected return is determined based on the efficiency index, timeliness index and success probability index output by the prediction model; The mutation processing module is used to mutate the continuous parameter components in the individual parameters based on the gradient information of the prediction model. The recombination processing module is used for cross-recombination processing of discrete parameter components in the individual parameters; The first screening module is used to select individuals from the current population to enter the next generation population based on the fitness value; The iteration stopping condition includes any one of the following: the number of iterations reaches a preset threshold; the change in the fitness value of the best individual in the population within a consecutive preset number of iterations is less than a change threshold. And / or, the verification processing unit further includes: The fifth generation module is used to construct, generate, and obtain a decision distribution corresponding to the target parameters, and calculate the deviation between the generated decision distribution and the reference distribution; A perturbation processing module is used to perturb the first data and evaluate the effect fluctuation of the target parameter generated under the perturbation-processed feature data; The second filtering module is used to filter out the third data from multiple target parameters based on the deviation and the effect fluctuation range. And / or, the system further includes: An update processing unit is used to create, generate, and acquire sixth data after the execution of the control command, and update the model parameters of the prediction model according to the sixth data; wherein, the sixth data is actual feedback data; The detection and determination unit is used to re-execute the step of dynamically searching for target parameters in the parameter space based on the updated prediction model when a preset trigger event is detected.

[0016] The third objective of this invention is achieved as follows: it includes a processor, a memory, and a strategy generation platform control program based on dynamic search of decision parameters; wherein the strategy generation platform control program based on dynamic search of decision parameters is executed on the processor, the strategy generation platform control program based on dynamic search of decision parameters is stored in the memory, and the strategy generation platform control program based on dynamic search of decision parameters implements the strategy generation method based on dynamic search of decision parameters.

[0017] The fourth objective of this invention is achieved as follows: the computer-readable storage medium stores a strategy generation platform control program based on dynamic search of decision parameters, and the strategy generation platform control program based on dynamic search of decision parameters implements the strategy generation method based on dynamic search of decision parameters.

[0018] This invention constructs a predictive model with counterfactual prediction capabilities, encodes policy logic into optimizable mathematical parameters, and employs a gradient-guided evolutionary search algorithm for efficient optimization in a hybrid parameter space. Simultaneously, it utilizes a dual verification mechanism of anomaly detection and perturbation simulation to screen robust policies, ultimately forming an automated closed loop of prediction, search, verification, execution, feedback, and re-prediction. This achieves intelligent generation of policy parameters, accurate pre-execution prediction, and continuous adaptive dynamic adjustment of the system to environmental changes. Thus, while improving decision-making quality, it reduces the cost of manual parameter tuning and trial-and-error risks, ensuring the long-term stability and generalization ability of the policy in complex dynamic environments. Attached Figure Description

[0019] 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.

[0020] Figure 1 This is a schematic diagram of the process steps of a strategy generation method based on dynamic search of decision parameters according to the present invention. Figure 2 This is a schematic diagram of a strategy generation system architecture based on dynamic search of decision parameters according to the present invention. Figure 3 This is a schematic diagram of a strategy generation platform architecture based on dynamic search of decision parameters according to the present invention. Figure 4 This is a schematic diagram of a computer-readable storage medium architecture in one embodiment of the present invention. Detailed Implementation

[0021] To facilitate a clearer understanding of the objectives, technical solutions, and advantages of this invention, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art can easily understand other advantages and effects of this invention from the content disclosed in this specification.

[0022] This invention can also be implemented or applied through other different specific examples, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of this invention.

[0023] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0024] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Secondly, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0025] Preferably, the strategy generation method based on dynamic search of decision parameters of the present invention is applied in one or more terminals or servers. The terminal is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0026] The terminal can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal can interact with the customer via a keyboard, mouse, remote control, touchpad, or voice control device.

[0027] This invention provides a strategy generation method, system, platform, and storage medium based on dynamic search of decision parameters. For example... Figure 1 The diagram shown is a flowchart of a strategy generation method based on dynamic search of decision parameters provided in an embodiment of the present invention.

[0028] In this embodiment, the strategy generation method based on dynamic search of decision parameters can be applied to terminals or fixed terminals with display functions. The terminals are not limited to personal computers, smartphones, tablets, desktop computers or all-in-one computers with cameras, etc.

[0029] The policy generation method based on dynamic search of decision parameters can also be applied to a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes, but is not limited to, a wide area network (WAN), a metropolitan area network (MAN), or a local area network (LAN). The policy generation method based on dynamic search of decision parameters in this embodiment can be executed by the server, by the terminal, or by both the server and the terminal.

[0030] For example, for a terminal that needs to perform strategy generation based on dynamic search of decision parameters, the strategy generation function based on dynamic search of decision parameters provided by the method of this invention can be directly integrated into the terminal, or a client for implementing the method of this invention can be installed. Alternatively, the method provided by this invention can also run on servers or other devices in the form of a Software Development Kit (SDK), providing an interface for the strategy generation function based on dynamic search of decision parameters in the form of an SDK. Terminals or other devices can then implement the strategy generation function based on dynamic search of decision parameters through the provided interface. The invention will be further described below with reference to the accompanying drawings.

[0031] like Figure 1 As shown, this invention provides a strategy generation method based on dynamic search of decision parameters, the method comprising the following steps: S01. Create and acquire first data corresponding to the target object, and combine the first data with multiple candidate decision codes respectively. Combine the first data with the trained prediction model to construct and acquire second data corresponding to the target object under each candidate decision code; wherein, the first data is feature data; the second data is prediction state data. S02. Based on the second data, dynamically search for target parameters that satisfy the preset optimization objective in a predefined parameter space; wherein, the parameters in the parameter space are used to generate the decision code; the parameters in the parameter space are used to define a decision function, the decision function outputting the object allocation path, the execution subject identifier, and the execution parameter configuration in sequence according to the feature data; the parameters include the weight parameters and bias parameters of the decision function; S03. The stability verification process is performed on the target parameters, and third data corresponding to the target parameters is created and generated. Based on the third data, a corresponding fourth data is generated and output. The third data is the execution parameters and is the target parameters that have passed the verification. The fourth data is the control instructions generated based on the execution parameters.

[0032] The step of creating, generating, and acquiring first data corresponding to the target object, combining the first data with multiple candidate decision codes, and combining this with a trained prediction model to construct, generate, and acquire second data corresponding to the target object under each candidate decision code, further includes: S011. Establish, generate, and extract fifth data corresponding to the target object; wherein, the fifth data is initial attribute data; S012. Normalize the continuous data in the fifth data and process the discrete data in the fifth data by quantization mapping. S013. Based on the processed continuous data and the discrete data, create corresponding first data.

[0033] The prediction model is trained in the following way: S014. Obtain the historical record set; wherein, the historical record set includes the feature data of the historical object, the historical decision code, and the corresponding historical state result; S015. Input the feature data of the historical object and the historical decision code into the prediction model to be trained to generate the prediction state result; S016. Adjust the parameters of the prediction model to be trained based on the difference between the predicted state result and the historical state result, and the propensity score of the historical decision code.

[0034] The step of dynamically searching for target parameters that satisfy a preset optimization objective in a predefined parameter space based on the second data further includes: S021. Create an initial population containing individuals with multiple parameters, and repeat the following steps until the iteration stopping condition is met: S0211. Calculate and generate a fitness value corresponding to each parameter individual in the population based on the second data; wherein, the fitness value is used to characterize the difference between the expected return of the strategy determined by the corresponding parameter individual and the load balancing cost; the expected return is determined based on the efficiency index, timeliness index and success probability index output by the prediction model; S0212. Based on the gradient information of the prediction model, perform mutation processing on the continuous parameter components in the individual parameters; S0213. Cross-recombination process is performed on the discrete parameter components in the individual parameters. S0214. Based on the fitness value, select individuals from the current population to enter the next generation population; The iteration stopping condition includes any one of the following: the number of iterations reaches a preset threshold; the change in the fitness value of the best individual in the population within a consecutive preset number of iterations is less than a change threshold.

[0035] The stability verification process, which involves processing the target parameter, creating third data corresponding to the target parameter, and generating and outputting corresponding fourth data based on the third data, further includes: S031. Construct, generate, and obtain a decision distribution corresponding to the target parameters, and calculate the deviation between the generated decision distribution and the reference distribution; S032. Perturb the first data and evaluate the effect fluctuation of the target parameter generated under the perturbated feature data; S033. Based on the deviation and the effect fluctuation range, select the third data from multiple target parameters.

[0036] The method further includes: S041. Create, generate, and acquire the sixth data after the execution of the control command, and update the model parameters of the prediction model according to the sixth data; wherein, the sixth data is actual feedback data; S042. When a preset trigger event is detected, the step of dynamically searching for target parameters in the parameter space is re-executed based on the updated prediction model.

[0037] Specifically, in this embodiment of the invention, a strategy generation method based on dynamic search of decision parameters is proposed, comprising the following steps: Step S1: Obtain the feature data of the target object; its output serves as the input for steps S2 and S3.

[0038] Step S2: Construct a prediction model and learn the mapping relationship of "target object features × decision code → state data" based on historical data; its output is used as the input for step S4.

[0039] Step S3: Define a policy parameterization representation framework and encode the policy into an optimizable parameter vector; its output serves as the input for step S4.

[0040] Step S4: Based on the prediction model output from step S2 and the policy parameterization representation output from step S3, construct an optimization objective function and find the optimal parameters by dynamically searching the parameter space; its output serves as the input for step S5.

[0041] Step S5: Perform stability verification on the candidate optimal parameters output in step S4; the output is used as the input for step S6.

[0042] Step S6: Execute the optimal parameters confirmed in step S5 to generate control commands and output them, and continuously collect actual feedback data to update the prediction model parameters in step S2, forming a closed-loop iteration.

[0043] Step S1: Acquisition of target object feature data: Definition of feature dimensions, for the first... For each target object, extract features in the following dimensions to construct a feature vector: (1) in, Total number of feature dimensions: object attribute features; environmental features; timeliness features; Feature preprocessing and embedding: The original features are preprocessed as follows, with continuous features normalized using quantiles: (2) in, Features Sort order among all objects The total number of objects, with a range of 1. .

[0044] Categorical features are represented using learned embedding vectors: (3) in, The preset embedding dimension; The final object feature vector is the concatenation of the normalized values ​​of continuous features and the embedding vectors of categorical features: (4) in, For continuous characteristic numbers, For the number of category features, This represents the final feature dimension.

[0045] The output of step S1 is the set of feature vectors of all objects to be processed. This serves as the input for steps S2 and S3.

[0046] Step S2, Prediction Model Construction: Model Architecture, constructing a conditional prediction model. Its input is the object feature vector. and decision coding The output is the predicted multidimensional state data. : (5) in: The decision encoding vector contains the subject identifier encoding, priority parameter, and time limit parameter, with dimensions of [dimension number missing]. The data source is the parameterized representation of the strategy in step S3; The model can learn a set of parameters; The predicted state data vector contains the following three components: Predictive efficiency index, with a value range of ; : Normalized value of the forecast timeliness indicator, with a value range of ; : Probability of successful prediction, with a value range of .

[0047] For model training, based on historical datasets Training the model, where This refers to the actual observed state data.

[0048] Training loss function: (6) in: Huber loss function; Binary cross-entropy loss function; : Preset multi-task loss weights; : Preset regularization coefficient.

[0049] Regarding counterfactual prediction capability, the core capability of the model lies in making counterfactual predictions about the state data of the same object under different decisions. This can be achieved by changing the decision code in the input. Obtain the prediction results under different strategies: (7) in, Let be the number of all possible decision combinations.

[0050] To enhance the reliability of counterfactual predictions, the model employs a training strategy based on propensity score weighting: (8) in, The historical decision-making tendency score is estimated by the auxiliary classification model. for The lower limit is truncated for the pre-defined tendency score.

[0051] The weighted training loss is: (9) The output of step S2 is the trained prediction model. This serves as the input for step S4.

[0052] Step S3, Policy Parameterization Representation Framework: Policy Structure Definition, defining the policy as a parameterized decision function. Its input is the object feature vector (provided by step S1), and its output is the decision code. ; This represents the policy parameter vector. The decision function employs a hierarchical structure: First layer, routing decision: (10) in, : object The identifier of the assigned execution path; Total number of execution paths; : Routing decision weight matrix; : Routing decision bias vector; The second layer involves the decision-making process of the implementing entity. (11) in, : object The specific identifier of the assigned implementing entity; :path The allocation decision weight matrix within; path The context embedding vector, with dimension . ; The third layer executes parameter configuration decisions: (12) in, Configure the execution parameter vector, with the value range of each component as follows: , This is the Sigmoid activation function.

[0053] The final decision code is: (13) The policy parameter vector, the complete policy parameter vector is: (14) The total dimension of the parameters is .

[0054] The output of step S3 is the policy parameterized representation framework and the initial parameter vector. This serves as the input for step S4.

[0055] Step S4, Optimize the objective function and search algorithm: Optimize the objective function and define the strategy. In the current collection of objects The objective function for the expected effect is: (15) in, , , In strategy Next case Decision coding The predicted values ​​of the state data obtained after inputting into the prediction model are calculated by the model in step S2; : object The quantified value of the scale; , , : Preset multi-objective weighting coefficients; : Preset load balancing regularization coefficient; Load balancing penalty items: (16) in, : for strategy Sub-assignment to path main body The number of objects; :path The total number of entities within; : as the main body , Standard capacity; Gradient-guided evolutionary search algorithms are used to optimize objective functions that are not fully differentiable due to the presence of discrete decisions in the decision function. Initialization, generating the initial population ,in Based on the initial parameters set in step S3, the remaining individuals are generated by random perturbation within their neighborhood. ， The preset population size.

[0056] Iterative optimization (the first) Generation: Fitness assessment, for each individual in the population. Calculate the objective function value Gradient-guided mutation, for continuous parameter parts ( ), using gradient information from the prediction model to guide the direction of mutation: (17) in, : This is the preset gradient guidance step size; : Preset random exploration noise intensity; Standard normal random noise.

[0057] Cross operation, for discrete parameter part ( A uniform crossover strategy is adopted, with a preset crossover probability. Exchange parameter blocks between two parent individuals; The selection process uses a tournament selection option, with the tournament size set to preset parameters. ; Elite retention, retaining the top performers with the highest adaptability. Each individual directly enters the next generation; Termination condition, when consecutive The improvement in optimal fitness is lower than the preset convergence threshold. The iteration terminates at the specified time. The output of step S4 is the top-fittest individuals in the population. A set of candidate policies consisting of policy parameter vectors This serves as the input for step S5.

[0058] Step S5, Robustness Verification: For each policy in the candidate policy set output in Step S4 Perform the following two verifications: Deviation identification based on anomaly detection algorithms, and candidate strategies The generated decision distribution is compared with the decision distribution of historical reference strategies to identify abnormal deviations: strategy The decision set generated on the current object set { Calculate its statistical eigenvector : (18) Anomaly detection model trained using the Isolation Forest algorithm Calculate the outlier scores for candidate strategies: (19) when Exceeding the preset abnormal threshold When this happens, the strategy is marked as significantly deviating.

[0059] Robustness evaluation of candidate strategies based on perturbation simulation By adding random perturbations to object characteristics to simulate data uncertainty, the stability of the policy effect is evaluated. (20) in, : Characteristic perturbation noise; : is the preset standard deviation of disturbance intensity; : This represents the preset number of simulations; Robustness metrics for computational strategies: (twenty one) The final strategy selection is based on the combined results of the two validations, and the final strategy score is calculated. (twenty two) in, This is the preset penalty coefficient for abnormal deviations; This represents the maximum value of the robustness index among the candidate strategies; The default robustness preference index is used; select... The highest strategy This is the optimal strategy that has been ultimately confirmed.

[0060] The output of step S5 is the finally confirmed optimal policy parameters. This serves as the input for step S6.

[0061] Step S6, Policy Execution and Closed-Loop Iterative Update: Policy execution, applying the optimal policy output in step S5. Collect actual feedback data from processed objects to construct an incremental training dataset: (twenty three) Generate specific case assignment instructions (including but not limited to target disposal channels, target disposal agencies / personnel, disposal time limit parameters, repayment target parameters, and phased assessment nodes).

[0062] Collection of payment results and model updates are performed within a pre-defined observation and feedback cycle. Collect actual repayment results from already assigned cases and construct an incremental training dataset: (twenty four) The prediction model parameters in step S2 are updated using an online learning method: (25) in, This is the preset online learning rate.

[0063] Strategy iteration is triggered when any of the following conditions are met: After the prediction model is updated, the prediction performance of the currently executed strategy decreases by more than a preset strategy failure threshold. : (26) The cumulative deviation between the actual feedback data and the predicted data exceeds the preset prediction drift threshold. : (27) The maximum validity period of the strategy has exceeded the preset time since the last strategy optimization. .

[0064] To achieve the above objectives, the present invention also provides a strategy generation system based on dynamic search of decision parameters, such as... Figure 2 As shown, the system is applied to the strategy generation method based on dynamic search of decision parameters, and the system includes: A generation unit is created to generate and acquire first data corresponding to a target object, and to combine the first data with multiple candidate decision codes respectively. Combined with a trained prediction model, second data corresponding to the target object and under each candidate decision code is constructed and acquired. The first data is feature data, and the second data is predicted state data. A dynamic search unit is used to dynamically search for target parameters that satisfy a preset optimization objective in a predefined parameter space based on the second data; wherein, the parameters in the parameter space are used to generate the decision code; the parameters in the parameter space are used to define a decision function, the decision function outputting an object allocation path, an execution entity identifier, and an execution parameter configuration in sequence according to the feature data; the parameters include the weight parameters and bias parameters of the decision function; The verification processing unit is used to perform stability verification processing on the target parameters, create and generate third data corresponding to the target parameters, and generate and output corresponding fourth data based on the third data; wherein, the third data is an execution parameter and is a target parameter that has passed verification; the fourth data is a control instruction generated based on the execution parameters.

[0065] The creation and generation unit further includes: A generation module is established to generate and extract fifth data corresponding to the target object; wherein, the fifth data is initial attribute data; The normalization processing module is used to normalize the continuous data in the fifth data and to process the discrete data in the fifth data by quantization mapping. The first generation module is used to create corresponding first data based on the processed continuous data and the discrete data; The prediction model is trained using the following modules: An acquisition module is used to acquire a historical record set; wherein, the historical record set contains feature data of historical objects, historical decision codes, and corresponding historical state results; The second generation module is used to input the feature data of the historical object and the historical decision code into the prediction model to be trained, and generate the prediction state result. The adjustment processing module is used to adjust the parameters of the prediction model to be trained based on the difference between the predicted state result and the historical state result, as well as the propensity score of the historical decision code. And / or, the dynamic search unit further includes: The third generation module is used to create an initial population containing individuals with multiple parameters, and repeats the following module until the iteration stopping condition is met: The fourth generation module is used to calculate and generate a fitness value corresponding to each parameter individual in the population based on the second data; wherein, the fitness value is used to characterize the difference between the expected return of the strategy determined by the corresponding parameter individual and the load balancing cost; the expected return is determined based on the efficiency index, timeliness index and success probability index output by the prediction model; The mutation processing module is used to mutate the continuous parameter components in the individual parameters based on the gradient information of the prediction model. The recombination processing module is used for cross-recombination processing of discrete parameter components in the individual parameters; The first screening module is used to select individuals from the current population to enter the next generation population based on the fitness value; The iteration stopping condition includes any one of the following: the number of iterations reaches a preset threshold; the change in the fitness value of the best individual in the population within a consecutive preset number of iterations is less than a change threshold. And / or, the verification processing unit further includes: The fifth generation module is used to construct, generate, and obtain a decision distribution corresponding to the target parameters, and calculate the deviation between the generated decision distribution and the reference distribution; A perturbation processing module is used to perturb the first data and evaluate the effect fluctuation of the target parameter generated under the perturbation-processed feature data; The second filtering module is used to filter out the third data from multiple target parameters based on the deviation and the effect fluctuation range. And / or, the system further includes: An update processing unit is used to create, generate, and acquire sixth data after the execution of the control command, and update the model parameters of the prediction model according to the sixth data; wherein, the sixth data is actual feedback data; The detection and determination unit is used to re-execute the step of dynamically searching for target parameters in the parameter space based on the updated prediction model when a preset trigger event is detected.

[0066] In the system solution embodiment of the present invention, the specific details of the method steps involved in the generation of a strategy based on dynamic search of decision parameters have been described above. That is to say, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, and will not be repeated here.

[0067] To achieve the above objectives, the present invention also provides a strategy generation platform based on dynamic search of decision parameters, such as... Figure 3 As shown, the system includes a processor, a memory, and a strategy generation platform control program based on dynamic search of decision parameters. The processor executes the strategy generation platform control program based on dynamic search of decision parameters, which is stored in the memory. This control program implements the steps of the strategy generation method based on dynamic search of decision parameters. The specific details of these steps have been described above and will not be repeated here.

[0068] In this embodiment of the invention, the built-in processor of the strategy generation platform based on dynamic search of decision parameters can be composed of integrated circuits. For example, it can be composed of a single packaged integrated circuit, or it can be composed of multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor uses various interfaces and lines to connect to various components, and executes programs or units stored in memory, and calls data stored in memory, to perform various functions and process data based on the strategy generation of dynamic search of decision parameters. The memory is used to store program code and various data. It is installed in the strategy generation platform based on dynamic search of decision parameters and enables high-speed and automatic access to programs or data during operation. The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0069] To achieve the above objectives, the present invention also provides a computer-readable storage medium, such as... Figure 4 As shown, the computer-readable storage medium stores a strategy generation platform control program based on dynamic search of decision parameters. The strategy generation platform control program based on dynamic search of decision parameters implements the strategy generation method steps based on dynamic search of decision parameters; the specific details of the steps have been described above and will not be repeated here.

[0070] In the description of embodiments of the present invention, it should be noted that any process or method description in the flowcharts or otherwise described herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0071] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processing module, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, a “computer-readable medium” can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, the computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0072] This invention constructs a predictive model with counterfactual prediction capabilities, encodes policy logic into optimizable mathematical parameters, and employs a gradient-guided evolutionary search algorithm for efficient optimization in a hybrid parameter space. Simultaneously, it utilizes a dual verification mechanism of anomaly detection and perturbation simulation to screen robust policies, ultimately forming an automated closed loop of prediction, search, verification, execution, feedback, and re-prediction. This achieves intelligent generation of policy parameters, accurate pre-execution prediction, and continuous adaptive dynamic adjustment of the system to environmental changes. Thus, while improving decision-making quality, it reduces the cost of manual parameter tuning and trial-and-error risks, ensuring the long-term stability and generalization ability of the policy in complex dynamic environments.

[0073] In other words, this invention parameterizes the strategy and dynamically searches within a vast parameter space to automatically generate optimal strategy parameters, overcoming the limitations and inefficiencies of manual parameter tuning based on experience. By constructing a predictive model with counterfactual prediction capabilities, it can simulate the effects of different decisions before strategy execution, providing precise guidance for strategy search and avoiding the costs of blind trial and error. By introducing a dual verification mechanism of anomaly detection and perturbation simulation, it ensures that the final selected strategy not only performs well on historical data but also maintains stable performance output when future data distribution fluctuates. Furthermore, by establishing an automated closed loop of execution, feedback, update, and re-optimization, the strategy system can continuously perceive environmental changes and proactively adjust itself in a timely manner, thereby maintaining optimal performance in the long term.

[0074] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A strategy generation method based on dynamic search of decision parameters, characterized in that, The method includes: First data corresponding to the target object is generated and acquired. The first data is then combined with multiple candidate decision codes. Combined with a trained prediction model, second data corresponding to the target object and under each candidate decision code is constructed and acquired. The first data is feature data, and the second data is predicted state data. Based on the second data, target parameters that satisfy the preset optimization objective are dynamically searched in a predefined parameter space; wherein, the parameters in the parameter space are used to generate the decision code; The stability verification process handles the target parameters, generates third data corresponding to the target parameters, and generates and outputs corresponding fourth data based on the third data; wherein, the third data is the execution parameters and is the target parameters that have passed the verification; the fourth data is the control instructions generated based on the execution parameters.

2. The strategy generation method based on dynamic search of decision parameters according to claim 1, characterized in that, The step of creating, generating, and acquiring first data corresponding to the target object, combining the first data with multiple candidate decision codes, and combining this with a trained prediction model to construct, generate, and acquire second data corresponding to the target object under each candidate decision code, further includes: A fifth set of data corresponding to the target object is generated and extracted; wherein, the fifth set of data is initial attribute data; The continuous data in the fifth data is normalized, and the discrete data in the fifth data is mapped to quantization. Based on the processed continuous data and the discrete data, corresponding first data is created.

3. A strategy generation method based on dynamic search of decision parameters according to claim 1 or 2, characterized in that, The prediction model is trained in the following way: Obtain a historical record set; wherein the historical record set contains feature data of historical objects, historical decision codes, and corresponding historical state results; The feature data of the historical object and the historical decision encoding are input into the prediction model to be trained to generate the prediction state result; The parameters of the prediction model to be trained are adjusted based on the difference between the predicted state and the historical state, as well as the propensity score of the historical decision encoding.

4. The strategy generation method based on dynamic search of decision parameters according to claim 1, characterized in that, The step of dynamically searching for target parameters that satisfy a preset optimization objective in a predefined parameter space based on the second data further includes: Create an initial population containing individuals with multiple parameters, and repeat the following steps until the iteration stopping condition is met: Based on the second data, a fitness value corresponding to each parameter individual in the population is calculated and generated. Based on the gradient information of the prediction model, the continuous parameter components in the individual parameters are mutated. Cross-recombination processing is applied to the discrete parameter components in the individual parameters. Based on the fitness value, individuals are selected from the current population to enter the next generation population; The iteration stopping condition includes any one of the following: the number of iterations reaches a preset threshold; the change in the fitness value of the best individual in the population within a consecutive preset number of iterations is less than a change threshold.

5. A strategy generation method based on dynamic search of decision parameters according to claim 1 or 4, characterized in that, The stability verification process, which involves processing the target parameter, creating third data corresponding to the target parameter, and generating and outputting corresponding fourth data based on the third data, further includes: Construct and obtain a decision distribution corresponding to the target parameters, and calculate the deviation between the generated decision distribution and the reference distribution; The first data is perturbed, and the fluctuation range of the effect of generating the target parameter under the perturbed feature data is evaluated. The third data is selected from multiple target parameters based on the deviation and the fluctuation range of the effect.

6. The strategy generation method based on dynamic search of decision parameters according to claim 1, characterized in that, The method further includes: The system generates and acquires sixth data after the execution of the control command, and updates the model parameters of the prediction model based on the sixth data; wherein the sixth data is actual feedback data. When a preset trigger event is detected, the step of dynamically searching for target parameters in the parameter space is re-executed based on the updated prediction model.

7. A strategy generation system based on dynamic search of decision parameters, characterized in that, The system is applied to the strategy generation method based on dynamic search of decision parameters as described in any one of claims 1 to 6, and the system comprises: A generation unit is created to generate and acquire first data corresponding to a target object, and to combine the first data with multiple candidate decision codes respectively. Combined with a trained prediction model, second data corresponding to the target object and under each candidate decision code is constructed and acquired. The first data is feature data, and the second data is predicted state data. A dynamic search unit is used to dynamically search for target parameters that satisfy a preset optimization objective in a predefined parameter space based on the second data; wherein the parameters in the parameter space are used to generate the decision code; The verification processing unit is used to perform stability verification processing on the target parameters, create and generate third data corresponding to the target parameters, and generate and output corresponding fourth data based on the third data; wherein, the third data is an execution parameter and is a target parameter that has passed verification; the fourth data is a control instruction generated based on the execution parameters.

8. The strategy generation system based on dynamic search of decision parameters according to claim 7, characterized in that, The creation and generation unit further includes: A generation module is established to generate and extract fifth data corresponding to the target object; wherein, the fifth data is initial attribute data; The normalization processing module is used to normalize the continuous data in the fifth data and to process the discrete data in the fifth data by quantization mapping. The first generation module is used to create corresponding first data based on the processed continuous data and the discrete data; The prediction model is trained using the following modules: An acquisition module is used to acquire a historical record set; wherein, the historical record set contains feature data of historical objects, historical decision codes, and corresponding historical state results; The second generation module is used to input the feature data of the historical object and the historical decision code into the prediction model to be trained, and generate the prediction state result. The adjustment processing module is used to adjust the parameters of the prediction model to be trained based on the difference between the predicted state result and the historical state result, as well as the propensity score of the historical decision code. And / or, the dynamic search unit further includes: The third generation module is used to create an initial population containing individuals with multiple parameters, and repeats the following module until the iteration stopping condition is met: The fourth generation module is used to calculate and generate a fitness value corresponding to each parameter individual in the population based on the second data; The mutation processing module is used to mutate the continuous parameter components in the individual parameters based on the gradient information of the prediction model. The recombination processing module is used for cross-recombination processing of discrete parameter components in the individual parameters; The first screening module is used to select individuals from the current population to enter the next generation population based on the fitness value; The iteration stopping condition includes any one of the following: the number of iterations reaches a preset threshold; the change in the fitness value of the best individual in the population within a consecutive preset number of iterations is less than a change threshold. And / or, the verification processing unit further includes: The fifth generation module is used to construct, generate, and obtain a decision distribution corresponding to the target parameters, and calculate the deviation between the generated decision distribution and the reference distribution; A perturbation processing module is used to perturb the first data and evaluate the effect fluctuation of the target parameter generated under the perturbation-processed feature data; The second filtering module is used to filter out the third data from multiple target parameters based on the deviation and the effect fluctuation range. And / or, the system further includes: An update processing unit is used to create, generate, and acquire sixth data after the execution of the control command, and update the model parameters of the prediction model according to the sixth data; wherein, the sixth data is actual feedback data; The detection and determination unit is used to re-execute the step of dynamically searching for target parameters in the parameter space based on the updated prediction model when a preset trigger event is detected.

9. A strategy generation platform based on dynamic search of decision parameters, characterized in that, The system includes a processor, a memory, and a strategy generation platform control program based on dynamic search of decision parameters; wherein the strategy generation platform control program based on dynamic search of decision parameters is executed on the processor, the strategy generation platform control program based on dynamic search of decision parameters is stored in the memory, and the strategy generation platform control program based on dynamic search of decision parameters implements the strategy generation method based on dynamic search of decision parameters as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a strategy generation platform control program based on dynamic search of decision parameters, which implements the strategy generation method based on dynamic search of decision parameters as described in any one of claims 1 to 6.

Images