207 results about "Augmented learning" patented technology

The invention relates to the field of converter station equipment parameter analysis, and provides a converter station equipment parameter analysis method and system based on multi-parameter fusion monitoring, and the method comprises the following steps: collecting multi-parameter characteristic data of converter station equipment; performing streaming processing and abnormal mode identification on the multi-parameter characteristic data to obtain equipment abnormity early warning information of the converter station equipment; constructing a time-varying graph neural network model based on the multi-parameter feature data, obtaining a space-time dependency relationship among parameters through a causal reasoning method, and dynamically adjusting the feature weight of the time-varying graph neural network model in combination with a reinforcement learning algorithm to obtain a running state evaluation model; and inputting the real-time state feature sequence into the operation state evaluation model to obtain an equipment operation state evaluation result, and visually displaying the equipment operation state evaluation result and the abnormal early warning information to obtain a comprehensive operation state evaluation report of the converter station equipment. According to the invention, real-time monitoring and evaluation of the operation state of the converter station equipment are realized.

The embodiment of the invention provides an intelligent learning path recommendation method and system based on a dynamic state. The method is applied to the technical field of intelligent learning recommendation, and comprises the following steps: calculating a skill adaptation weight, a load weight and an emergency degree weight in real time based on a dynamic state of a target employee, including a current skill improvement condition, a workload and a task emergency degree; combining the skill matching degree, the learning strength and the task association degree of each learning unit in the candidate learning path, and performing comprehensive evaluation by using a multi-weight scoring function; and sorting the paths according to the comprehensive score, generating a personalized recommended learning path set and sending the personalized recommended learning path set to an employee terminal, thereby realizing intelligent learning path recommendation with dynamic adaptation and accurate matching. According to the scheme, personalized learning path pushing aiming at actual post requirements and working states of the employees can be realized, correlation, urgency and acceptability of learning contents are improved, learning efficiency and task adaptability are remarkably enhanced, and the employees are helped to quickly compete with post targets.

The invention discloses a weak supervision video anomaly detection method based on prompt learning knowledge enhancement, and belongs to the technical field of video intelligent analysis. A video side gives a section of abnormal scene video, video sequence features and audio sequence features are obtained through a feature extraction network, then a trained and complete feature aggregation network is input to carry out multi-modal feature aggregation, an abnormal score is obtained through a score prediction network, and text representation is carried out based on prompt learning. A prompt template is constructed for abnormal video tags through a knowledge graph, semantic expansion is performed on normal tags through a plurality of learnable parameters, cross-modal alignment is performed on the normal tags and a video side, so that features of the video side are close to different normal semantics, knowledge enhancement is performed by introducing external information, positive abnormal boundaries of the video are learned, and the detection performance is improved. And finally, multi-task joint optimization is carried out through different loss functions, and abnormal video clip positioning is carried out.

The invention relates to the technical field of reinforcement learning intelligent scheduling, in particular to an intelligent scheduling management method for prefabricated part production resources based on reinforcement learning. The specific implementation process comprises the steps of collecting order demands, material distribution and production pedestals in real time, and mapping the order demands, the material distribution and the production pedestals into production mold load tensors; when a scheduling request is triggered, inputting the production modulo tensor into a resource arrangement network, and performing search and reasoning by using a scheduling strategy based on a multi-head attention mechanism to generate a resource scheduling matching instruction; calculating a state difference tensor, and outputting an efficiency reward signal in combination with a delivery constraint and a cost constraint; and packaging the production modulo tensor, the resource scheduling matching instruction and the reward signal into semantic interaction experience, storing the semantic interaction experience into a scheduling experience playback pool for gradient modulation, and iteratively optimizing a scheduling strategy. According to the method, learning can be carried out from a large amount of historical data by utilizing reinforcement learning, iterative optimization of the scheduling strategy is realized, the calculation time consumption of scheduling instruction generation is reduced, and the flexibility of production scheduling is improved.

The invention provides a large model parameter optimization and self-adaptive adjustment method based on reinforcement learning, and relates to the technical field of optimization and self-adaptive adjustment of large model parameters based on reinforcement learning in machine learning, and the method comprises an overall architecture fusing a large model main body, a reinforcement learning module, an environment perception module and a reward feedback mechanism. The method is used for realizing dynamic optimization and adaptive adjustment of large model parameters, and can rapidly increase the learning rate according to environment feedback in the initial stage of large model training, so that the model parameters rapidly approach to the direction of an optimal solution, and the early-stage exploration time of training is greatly shortened. In the later stage of training, the learning rate can be accurately reduced, model oscillation is avoided, and it is ensured that the model is stably converged to a globally optimal solution. According to the dynamic adjustment mechanism, the number of iterations of training is effectively reduced, the training efficiency is greatly improved, a large number of computing resources and time cost can be saved, and for example, in large-scale image classification model training, the training time can be shortened by more than 30%.

The invention relates to the technical field of large model optimization, and discloses a large model parameter optimization and adaptive adjustment method and device based on reinforcement learning, and the method comprises the steps: obtaining a current operation state of a target large model, and carrying out the comprehensive evaluation of the operation state of the target large model through combining with data features, and obtaining an operation state vector; constructing an intelligent agent, and inputting the operation state vector into the intelligent agent to obtain a parameter optimization strategy; optimizing parameters of the target large model based on a parameter optimization strategy to obtain an optimized large model; calculating a plurality of preset indexes for optimizing the large model by utilizing a reward function to serve as reward results; and adaptively adjusting parameters of the intelligent agent based on the reward result. In training and practical application of the large model, large model parameters are dynamically adjusted by means of an intelligent agent, it is ensured that rapid convergence can be achieved in the initial stage of training, the global optimal solution can be accurately approached in the later stage of training, the training efficiency and quality are effectively improved, and by means of a reward mechanism, the intelligent agent is made to adjust the parameters in a self-adaptive mode, and the generalization ability of the large model is improved.

The invention relates to the technical field of virtual simulation teaching. The e-commerce virtual simulation teaching resource construction method and system based on the ability map are provided, and the method comprises the following steps: on the basis of e-commerce industry technical post requirements, disassembling post capabilities, and generating a structured ability map comprising a core ability unit, a skill point layer and a knowledge point layer; performing classification design processing on the virtual simulation scene resources according to a skill point layer in the structured ability map, and generating a virtual resource library comprising production type scene resources, interactive type scene resources and creative type scene resources; based on the mapping relation between the nodes of the structured ability graph and the virtual resource library, performing label labeling processing on the virtual simulation task to generate a task-knowledge point association network, the technical effects of improving the matching precision of resources and post requirements, optimizing the classification systematicness of virtual simulation resources and enhancing the dynamic relevance of learning paths and knowledge points are achieved.

A system and method for stimulating the activation of neural mirroring mechanisms through personalized avatars to enhance learning and skill development. The system creates hyper-realistic or stylized digital representations of users and animates these avatars to present educational content or demonstrate skills, stimulating mirror neuron activation when users observe themselves demonstrating mastery. Applications include academic learning, language acquisition, professional training, personal development, athletic performance enhancement, and therapeutic interventions including for Autism Spectrum Disorder.

The invention discloses a laboratory detection equipment intelligent management system based on edge calculation and reinforcement learning, and relates to the technical field of equipment intelligent management, the system comprises a multi-dimensional data acquisition module, a model training and feature library construction module, a deviation degree research and judgment module, a model optimization module and an interaction and execution module; according to the invention, through integration of the multi-dimensional data acquisition module and the model training and feature library construction module, all-directional precision monitoring of the operation state of laboratory detection equipment is realized, and various feature data in the operation process of the equipment can be captured in real time through the multi-dimensional sensor array; the model training and feature library construction module uses a deep learning model of a CNN-LSTM mixed structure, combines historical operation data and fault cases of equipment, calculates each feature weight, and dynamically constructs and updates a health state feature vector library of the equipment, thereby improving the accuracy and timeliness of equipment state evaluation, and improving the reliability of equipment state evaluation. Therefore, managers can find potential faults in advance and take preventive maintenance measures.

The invention relates to the technical field of electromagnetic compatibility and signal processing, and discloses a multi-dimensional feature fusion-based electromagnetic leakage signal classification and identification method, which comprises the following steps of: preprocessing an original electromagnetic signal to obtain standardized data, extracting time domain, frequency domain and space domain features to construct a nine-dimensional feature vector, and extracting a three-dimensional feature vector; constructing an initial feature library and completing the initial feature library through a self-supervision verification mechanism, training the initial feature library through a CNN-LSTM fusion model with an attention mechanism to obtain a classification model, classifying signals to be identified, starting a self-supervision reinforcement learning mechanism optimization model according to an F1 score, and dynamically optimizing the feature library each month; the method solves the defects of the traditional technology, improves the recognition accuracy and the system adaptability, and can be used for cable information safety leakage protection.

The invention relates to the field of quality detection of mechanical parts, in particular to an intelligent recognition system for surface defects of mechanical parts based on deep learning. Comprising a data acquisition module, a differential geometric feature extraction module, a small sample learning module, an automatic labeling module, a data reinforcement learning module, a neural network training module, a defect identification module and a defect quantitative evaluation module. Extracting geometric features including curvature features and differential invariant features; small sample learning is adopted to solve the problem of sample scarcity; the efficiency is improved by utilizing automatic labeling; simultaneously capturing microcosmic details and macroscopic morphology through a multi-scale differential analysis framework; organic fusion of surface image information and depth geometric information is realized; according to the method, the problems of insufficient samples, difficulty in labeling, inaccurate defect evaluation and the like in industrial production are effectively solved.

The invention provides a contract generation method and system based on artificial intelligence. Belongs to the technical field of artificial intelligence and comprises the steps that a user inputs basic information and personalized requirements of a contract through an interactive interface, and a management system analyzes the input information, extracts key features and constructs a requirement vector; constructing an intelligent decision model based on a reinforcement learning algorithm; a database containing various contract templates is constructed, and the management system selects the most matched template from the template database as a starting point of contract generation according to the output of the reinforcement learning model. Through an artificial intelligence technology, especially natural language processing and a reinforcement learning algorithm, user demands can be rapidly analyzed and contract texts meeting the requirements can be generated, so that the time of traditional manual contract writing is greatly shortened.

The invention relates to the technical field of e-commerce, and discloses a cross-border e-commerce abnormal order processing method and system based on reinforcement learning, and the method comprises the steps: collecting historical transaction records and real-time behavior records of a user, and obtaining an original data set; performing format unification according to the original data set to obtain a structured data set, and performing multi-dimensional feature extraction to obtain multi-dimensional behavior features; according to the multi-dimensional behavior characteristics, abnormal sequence analysis and risk level classification are carried out, and risk control adjustment parameters are matched; according to the risk control adjustment parameters and the structured data set, environment grouping, risk order identification and abnormal feature extraction are carried out to obtain abnormal distribution features; according to the abnormal distribution characteristics, comprehensive risk analysis is carried out through a pre-constructed neural network model, and a comprehensive risk score is obtained; and performing abnormal order judgment and risk control processing according to the comprehensive risk score and the historical transaction record, and optimizing a neural network model. The method improves the accuracy of abnormal order detection.

The application relates to the technical field of data processing, in particular to a kiln data monitoring and early warning method based on reinforcement learning, which comprises the following steps: acquiring intelligent gas meter readings at each collection time in each cycle during the operation of a rotary kiln and temperatures at each collection time at each monitoring point; calculating heat space diffusivity at each monitoring point at each collection time in each cycle, calculating heat time dissipation and heat propagation index at each monitoring point at each collection time in each cycle; constructing heat abnormality degree at each monitoring point; calculating energy consumption abnormality degree of the rotary kiln; calculating heat preservation degree at each collection time in each cycle; constructing comprehensive abnormality degree of the rotary kiln; constructing a prediction autoregressive term number according to the comprehensive abnormality degree of the rotary kiln, acquiring a predicted energy consumption of the rotary kiln, and monitoring abnormal conditions of rotary kiln operation energy consumption in combination with a preset energy consumption. The application aims to improve the reliability of rotary kiln operation energy consumption abnormality monitoring.

The invention provides an aluminum substrate drilling tool wear prediction method based on reinforcement learning, and the method comprises the steps: collecting working condition parameters, such as main shaft rotation speed, feeding speed, cutting depth, material type and historical vibration energy, through the deployment of an industrial sensor network, and achieving data standardization through feature extraction and normalization processing; a dynamic cognitive map with process meaning is constructed by adopting semantic analysis and a map generation strategy, key semantic nodes are identified by utilizing a map attention network, reinforcement learning is driven through a structured reward function, and self-adaptive optimization of a wear prediction strategy is realized; according to the method, abnormal detection and working condition abrupt change increment updating are supported, map consistency optimization is achieved by fusing expert experience, the tool wear prediction precision and the environment adaptability can be improved, and the intelligent sensing and dynamic response capacity to the complex manufacturing process can be enhanced.

The invention discloses a multi-satellite in-orbit collaborative scheduling method based on deep reinforcement learning and heuristic rule fusion, and relates to the field of multi-satellite in-orbit intelligent scheduling. The method comprises the following steps: establishing a collaborative decision model based on a multi-agent depth deterministic strategy gradient algorithm; a prior experience playback mechanism is introduced to enhance the learning efficiency; designing a heuristic rule for guiding agent decision making; self-adaptive fusion of a reinforcement learning strategy and a heuristic rule is realized through a dynamic mixing coefficient; the agent learning convergence is accelerated by adopting a reward shaping technology; and dynamically adjusting a scheduling scheme according to satellite resource constraints and task priorities. According to the method, intelligent collaborative scheduling of multiple satellite tasks can be realized under complex constraint conditions, the task completion rate and the resource utilization efficiency are improved, and the method has important significance in improving the autonomous decision-making capability of a satellite system and the robustness of dealing with emergencies.

The invention discloses a colorectal early-stage tumor data analysis early-warning method based on reinforcement learning, and the method comprises the steps: achieving the semantic alignment and weighted fusion through the unified expression of different modal features, and employing a differentiable attention unit; dynamic strategy optimization is carried out in combination with a reinforcement learning model, and the generalization ability and early warning accuracy of heterogeneous data are improved; a dynamic weight adjustment and online learning mechanism is introduced, continuous self-adaptive updating of the model and compensation of a data missing scene are realized, the accuracy and stability of colorectal tumor risk early warning are improved, and the method has high clinical application and popularization value.

The invention discloses a graph reinforcement learning reactive voltage control method considering domain knowledge, and the method comprises the steps: (I) enabling a reactive voltage control problem to be built into a partially observable Markov decision process, and enabling an intelligent agent corresponding to each inverter to only observe the local state quantity of a limited number of nodes around; (2) constructing a space-time diagram in which domain knowledge is embedded; (3) utilizing a graph attention neural network to carry out domain knowledge-based feature reinforcement learning of the multi-agent; and (IV) performing multi-inverter action decision making based on a GAMARL method and the like. According to the invention, by integrating domain knowledge and an advanced graph reinforcement learning technology, the voltage regulation capability of the novel power distribution system under the condition of high-proportion distributed photovoltaic access is remarkably improved, and the intelligent and sustainable development of a power distribution network is promoted.

The invention provides an online course MOOC learning prediction method based on heterogeneous feature fusion, and belongs to the field of computer-aided intelligent education. The method comprises the following steps: acquiring an MOOC data set, and preprocessing the data set to obtain a test set; constructing an IHFNet network comprising a multi-agent adaptive static feature selector module, a hierarchical time sequence feature extractor module and a heterogeneous feature fusion module; a multi-agent adaptive static feature selector module screens key static features; the hierarchical time sequence feature extractor module extracts behavior time sequence features with high discrimination ability; the heterogeneous feature fusion module carries out adaptive fusion on the key static features and the behavior time sequence features and carries out classification prediction; an IHFNet network is trained; and collecting MOOC data of a to-be-predicted learner, and inputting the MOOC data to the trained IHFNet network for learning risk prediction. According to the invention, modeling is carried out by fusing the static features of the learner and the behavior time sequence features, the feature representation ability of the learner is enhanced, and the learning risk prediction effect is improved.

The invention discloses an intelligent food storage tank internal environment adaptive control system and control method, and belongs to the technical field of artificial intelligence and Internet of Things control. The system specifically comprises a multi-mode sensing module, a feature extraction and preprocessing module, a reinforcement learning module, an instruction adaptive control module and a digital simulation module. The multi-modal sensing module deploys a sensor and an environment adjusting device in an array mode; the feature extraction and preprocessing module carries out filtering, standardization and feature extraction on the data; the reinforcement learning module generates an optimization control strategy by using an algorithm; the instruction self-adaptive control module converts the strategy into an instruction and accurately adjusts parameters of the environment adjusting device; the digital simulation module constructs a digital model for simulation learning and evolution of a strategy agent. Compared with a traditional monitoring system, the method has the technical advantage of generating an optimization strategy, solves the problem of insufficient adaptive control capability caused by a fixed strategy of the traditional monitoring system, and provides a more efficient monitoring service.

The invention discloses a standard file automatic classification method based on artificial intelligence, and relates to the technical field of text classification, and the method comprises the steps: obtaining original data of a to-be-classified file, and carrying out the analysis and preprocessing of the original data of the to-be-classified file, and obtaining metadata, chapter structure information and plain text content; inputting the plain text content and the chapter structure information into a multi-granularity semantic pyramid model for analysis and fusion, and generating a document feature vector; constructing a standard classification system knowledge graph by using the standard classification system data to obtain a classification name mapping table, and performing reinforcement learning on the standard classification system knowledge graph by using a graph neural network technology to generate an enhanced feature vector; and calculating the semantic similarity between the document feature vector and the enhanced feature vector. According to the method, hierarchical reasoning is performed in the knowledge graph based on the similarity to obtain the target classification code, and the classification result is matched and output through the mapping table.

The invention discloses an import and export commodity intelligent classification method based on knowledge graph metadata topology, and relates to the technical field of reinforcement learning, and the method comprises the steps: inputting an initial data packet into a dynamic interaction model, carrying out explicit association mining through a semantic enhancement layer, optimizing a rule matching path through a rule evolution layer, and constructing a dynamic commodity knowledge graph; performing topological structure derivation on the dynamic commodity knowledge graph to generate a graph topological analysis report and a metadata list, and performing knowledge reasoning integration on the graph topological analysis report and the metadata list to generate an intelligent navigation engine; calling an intelligent navigation engine to execute multi-path semantic query and rule verification on the dynamic knowledge graph to generate a candidate classification scheme set; and performing multi-target collaborative optimization on the candidate classification scheme set to generate a sorting scheme sequence, performing traceability packaging on the sorting scheme sequence, and outputting an intelligent classification scheme. According to the invention, through the dynamic interaction model and multi-target collaborative optimization, the rule adaptation efficiency in a complex scene is improved.

The invention relates to the technical field of hydroelectric generating set intelligent optimization control, in particular to a hydroelectric generating set optimization system and method based on the reinforcement learning technology, and the system comprises a multi-source sensing fusion unit, a reinforcement learning decision unit and an instruction execution unit. A unified time sequence feature tensor is generated through dynamic time warping time sequence alignment, Kalman filtering noise reduction, gradient normalization and feature cascade fusion, a reinforcement learning decision unit extracts three types of features by using a multi-scale time convolution network, and the coupling strength is quantified through a multi-head attention mechanism. The double branches respectively generate a start-stop sequence carrying start-stop loss punishment and a load distribution proportion of a coupling flow power function, a triple optimization mechanism and staged course learning are combined, an optimal matching scheme is output, an instruction execution unit converts the scheme into a control instruction, and the control instruction is output to a unit PLC control system after verification and compliance.

The invention discloses a method for generating a controlled file template based on deep learning, and relates to the technical field of intelligent document processing, and the method comprises the steps: collecting document data, obtaining a multi-modal data set through preprocessing, and generating a domain knowledge graph; based on an optimization learning strategy, optimizing the template generation and content evaluation task through a reinforcement learning algorithm to obtain a file template; performing format specification evaluation and content rationality evaluation on the file template by adopting a self-supervised learning method, and adjusting a template layout structure and content terms and details to generate an optimized file template; according to the real-time feedback data, feedback learning and template adjustment are conducted on the optimized file template through an incremental learning method, and a controlled file template is generated. The template quality is improved, manual intervention is reduced, and a basis is provided for continuous optimization.

The invention discloses a fire-fighting emergency evacuation guiding system driven by reinforcement learning, relates to the technical field of emergency evacuation guiding, and provides the following scheme that the fire-fighting emergency evacuation guiding system comprises the steps of gridding a monitoring area and calculating a comprehensive danger value of each grid by collecting temperature, smoke concentration, personnel density and a standard path in a building in real time; after the high-risk grids are removed, the remaining grid risk values are input into a reinforcement learning decision model to generate candidate paths; a safety score is obtained by calculating the matching degree of a path and a standard path, and the score is dynamically adjusted in combination with the danger value of each grid and the personnel density; meanwhile, obtaining an efficiency score based on a path length ratio and performing corresponding optimization; and finally, determining an optimal evacuation path from the candidate paths by adopting a multi-objective optimization algorithm, and driving an intelligent indicating device to guide evacuation. According to the invention, the limitation of a fixed evacuation scheme in a dynamic fire scene is solved, intelligent emergency evacuation command is realized, and the safety and efficiency of the evacuation process are improved.

The invention discloses a Go question type recommendation system and method based on user ability evaluation, and relates to the technical field of smart learning. The system is composed of a data acquisition module, a data processing module, a learning recommendation module and a feedback evaluation module. The data acquisition module covers quantification of five-force dimensions, actual combat data analysis and training behavior records of users; and the data processing module generates an accurate three-dimensional feature vector through knowledge graph filtering, recent development area principle sorting and feature extraction. The learning recommendation module intelligently matches the most suitable question type library content according to the feature vector of the user, and ensures that the recommended question can not only fit the current capability level of the user, but also meet the specific learning target of the user; and the feedback evaluation module dynamically updates the capability value of the user by using an IRT model, optimizes recommendation algorithm parameters in combination with user feedback and strengthens error exercise to form a closed-loop feedback mechanism. And according to accurate question type recommendation, the learning effect is further enhanced through continuous feedback and optimization.

The present invention relates to the field of radio navigation system state evaluation. Provided is a radio navigation system state evaluation method based on multi-dimensional data-augmented learning. The method comprises: acquiring function operation monitoring data of a radio navigation system, and selecting data associated with system health; performing sliding window processing on the selected data, so as to obtain sample data including time information; dividing the sample data into a training set, a first test set and a second test set, and constructing positive and negative samples; establishing and training a contrastive learning model; establishing and training a deep one-class support vector machine; and using output results of the contrastive learning model and deep one-class support vector machine for the test sets to complete evaluation. In the present invention, deep features of measured data are learned by means of contrastive learning, and the deep one-class support vector machine is used to construct a state evaluation model. The present invention has the advantages of high detection precision and interpretability, has important theoretical and engineering practice significance, and realizes precise quantitative evaluation of the health state of an airborne radio navigation system.

The invention provides a fault diagnosis method and device, equipment, a medium and a product, and belongs to the technical field of communication network fault detection, the method comprises the following steps: receiving multi-source real-time information of each functional layer in a communication network, the multi-source real-time information comprising a dynamic network topology relation and fault information of at least one functional layer; fusing the multi-source real-time information into a multi-modal feature vector; the multi-modal feature vectors are input into a large language model, a fault diagnosis report output by the large language model is obtained, and the fault diagnosis report comprises fault root cause nodes; wherein the large language model is a model obtained by performing reinforcement learning according to corpora in the communication field. Multi-source information in the communication network is fused, cross-layer fault reasoning is carried out by means of the large language model obtained based on corpus training in the communication field, fault root cause points and fault diagnosis reports with practical guiding significance are obtained, and efficient, high-precision and real-time fault diagnosis services are provided for the large-scale communication network.

The invention relates to the technical field of industrial data automatic processing, and discloses an industrial protocol identification method and system, and the method comprises the steps: obtaining industrial data received based on a network interface in real time; performing key field identification on the industrial data to obtain a feature vector; inputting the feature vector into a pre-trained protocol type recognition model to obtain an initial protocol type recognition result; and based on the feature vector of the industrial data and the corresponding initial protocol type identification result, performing reinforcement learning by using a pre-trained reinforcement learning model to obtain a final protocol type corresponding to the industrial data. According to the method, the initial protocol type identification result is determined according to the feature vector of the industrial data, and the final protocol type is obtained by performing reinforcement learning by using the reinforcement learning model on the basis, so that the data acquisition efficiency is improved, and meanwhile, the expansibility, flexibility and economy of an industrial data acquisition system are improved; and the industrial system is promoted to develop towards a more efficient and intelligent direction.

The invention relates to the technical field of artificial intelligence and reinforcement learning, and discloses a cross-border trade intelligent decision optimization method based on reinforcement learning. The method comprises the following steps: constructing a composite state representation fusing tax, logistics, exchange rate, demand and competitive behavior; high-dimensional state compression is realized through an auto-encoder; designing a hierarchical action architecture to decouple a macroscopic strategy and a microscopic operation; a Nash equilibrium guided reward shaping function is introduced, and an equilibrium income deviation is estimated in combination with anti-factual reasoning; carrying out stable training by adopting a double-delay depth deterministic strategy gradient algorithm with state transition consistency constraint; an online fine tuning mechanism is deployed to adapt to a real business environment. According to the method, the strategy convergence speed, the annual profit rate and the responsiveness to policy mutation are remarkably improved.