160 results about "Hierarchical modeling" patented technology

The invention provides an intelligent scheduling and control method and device for an integrated energy system. According to the method, power, gas and heat resource operation data are acquired, multi-scale layered modeling is performed according to a time scale and a space scale, and a power resource state space model, a gas resource flow continuity model, a heat resource heat balance model and a multi-energy coupling characteristic constraint model are established; carrying out feature extraction and dimension reduction representation by adopting a deep auto-encoder network; cooperative training of multiple groups of cognitive models is carried out through a split hierarchical federal learning framework, and a global intelligent model is obtained; constructing a neural architecture search network with a hybrid bionic learning rule, setting a hierarchical scheduling target, and generating a hierarchical intelligent scheduling strategy; and a fault-tolerant control mechanism is constructed, and error detection and correction of operation deviation are realized. According to the invention, multi-time scale collaboration, collaborative learning under multi-device group privacy protection and high-reliability fault-tolerant control are realized, and the operation efficiency and reliability of the integrated energy system are remarkably improved.

Hierarchical modelling and advanced feature engineering discover abnormalities in time series data with irregular trends. Data is collected in real time to ensure temporal integrity in the invention. Extraction filters and isolates useful data. Data cleansing removes noise and extraneous data after preliminary analysis identifies patterns and abnormalities. Feature engineering organizes cleansed data for machine learning algorithms. Primary storage stores this data for fast retrieval and extensive trend analysis. Holidays and weekends provide unique patterns in trend analysis. These trends are used to cluster data and create hierarchical predictive models, starting with a first-order model for general trends and increasing in order to refine residuals. Serializing these models improves storage and retrieval. Trend clusters are created from new data points, and algorithms detect pattern deviations. Statistical tests and machine learning classifiers identify anomalies and create alerts and remedial measures. The system monitors and analyzes incoming data to detect anomalies.

The invention relates to a battery internal short circuit early warning method and system based on a physical information neural network, and belongs to the technical field of battery fault detection, and the method comprises the steps: constructing a first-stage PINN model through a linearized Butler-Volmer equation, and carrying out the training through a multi-source signal of a battery; taking the output of the first-stage PINN model as prior input, and constructing a second-stage PINN model by using a nonlinear electrochemical-thermal-mechanical coupling equation, and training the second-stage PINN model; using the trained first-stage PINN model and the trained second-stage PINN model to output prediction results of the voltage and the expansive force of the target battery; comparing the difference between the prediction result of the model and the measured value of the sensor to generate a voltage residual error and an expansive force residual error; and realizing identification and alarm of the short circuit risk in the target battery based on the voltage residual error and the expansive force residual error. According to the method, a multi-fidelity PINN hierarchical modeling mechanism is introduced, so that high-precision characterization and prediction of the internal state of the battery can be realized by using experimental data and mechanism model information at the same time under different precision and complexity hierarchies.

The invention relates to a few-sample image classification method based on hyperbolic space image-text local feature alignment, and belongs to the technical field of image recognition and artificial intelligence, and the method comprises the steps: generating word-level attribute description for a support set image through employing a multi-mode large language model; encoding the image and the text by adopting a vision-language model; constructing a hyperbolic local feature alignment module in a hyperbolic space, screening most relevant image local features for text local features by calculating hyperbolic cosine similarity, and fusing by using hyperbolic weighted average; designing a hyperbolic cross attention module, and aggregating key information from the multi-modal local features of the support set to construct a category prototype by taking query image aggregation features as guidance; and finally performing classification based on the hyperbolic geodesic distance. According to the method, the hierarchical modeling capability of the hyperbolic space and the semantic priori knowledge of the large language model are fully utilized, fine-grained multi-modal feature alignment is realized, and the small sample image classification performance is improved.

The invention belongs to the technical field of power system resource scheduling, and provides a building flexible resource load optimal scheduling method and system based on a game mechanism, and the method comprises the steps: obtaining a distributed architecture of a building flexible resource load; performing department-level and building-level layered modeling on the acquired distributed architecture to obtain a department-level model and a building-level model; carrying out distributed optimization solution on the constructed department-level model and building-level model by adopting an alternating direction multiplier method to obtain a distributed collaborative model; considering an aggregator, and constructing a single-leader-multi-follower game model which takes the transaction center as a dominant and takes the aggregator and a power generator as followers; and solving the constructed game model, and completing optimal scheduling of the building flexible resource load based on the game mechanism.

The invention provides a general visual unsupervised defect detection method based on hierarchical reconstruction, and the method comprises the steps: firstly processing an input image through a self-attention mechanism and a hierarchical jump-connection reconstruction module, carrying out the hierarchical modeling and reconstruction of a foreground, a background and a mask of the image through the reconstruction module, and generating the multi-dimensional reconstruction representation of an original image; by means of the design, structural decoupling can be achieved, the model is made to learn accurately, and complex background textures are reconstructed; then, a defect judgment module is used for judging the feature distribution of the original image and the reconstructed image so as to determine whether the input sample has defects or not; according to the method, global semantic information and multi-level structure features are fused, the robustness, accuracy and real-time performance of defect detection are effectively improved, and the method is suitable for complex and changeable industrial visual detection scenes and has good application prospects.

The invention relates to the technical field of new energy power systems, in particular to a new energy consumption prediction and early warning method, system, equipment and medium, and the method comprises the steps: collecting and preprocessing multi-source data in real time, and constructing a feature vector to support hierarchical collaborative prediction: firstly, training a new energy power generation prediction model based on historical weather and power generation data; secondly, combining load characteristics and influence factors to establish a load prediction model, and finally combining real-time parameters of a power grid to construct a consumption capability prediction model; early warning is triggered by dynamically comparing the generated power with the consumption capability predicted value, and a source-grid-load-storage cooperative control strategy is generated to execute regulation and control; the data coupling relation is deeply mined through the hierarchical modeling architecture, accurate quantification of the consumption potential and risk prospective early warning are achieved, the prediction timeliness and the power grid toughness are remarkably improved, power abandoning is effectively restrained, and the operation economical efficiency is optimized.

The invention relates to the technical field of posture recognition, in particular to a motion posture recognition method and system based on deep learning, and the method comprises the following steps: based on a human body motion image sequence, analyzing a skeleton center space trajectory, adjusting an acquisition visual angle, recognizing staged skeleton nodes, and generating a node time sequence trigger sequence; and optimizing joint point angle feature grouping and expression to obtain a hierarchical feature expression structure. In the invention, through detection for interference dynamic change characteristics, visual angle adjustment and node staged response in a time sequence are combined, compensation reasoning of a space trajectory is enhanced, and the hierarchical expression capability of motion capture is improved; the input features have the advantages of actively screening background disturbance, dynamically adapting the motion direction, separating action core nodes and perfecting node response association and grouping angle time sequence structures, the integrity and identification degree of action feature structure expression and hierarchical modeling are guaranteed, and higher identification accuracy and complex environment adaptability are brought.

The invention discloses a system engineering meta-model generation method based on a scene and a DM2 theory, and relates to the technical field of system engineering, and the method comprises the steps: obtaining scene description information related to a target business scene; performing multi-level analysis on the scene description information to obtain a corresponding service model; performing hierarchical modeling and conversion on the business model based on the hierarchical architecture of the DM2 framework to generate structured model data; and according to the expansion requirement of the system engineering field, performing field expansion and integrated processing on the model entity and the semantic relationship in the structured model data to generate a system engineering meta-model. Through the method, the technical problems of incomplete sorting of the meta-model, separation from actual business and poor ductility in a traditional mode are effectively solved.

The invention discloses a hierarchical OvO-SVM nondestructive testing method for the attack degree of potato blackheart disease, which comprises the following steps: preparing a detection sample, and collecting a VIS / NIR original transmission spectrum; preprocessing the collected original spectrum; adopting a random forest algorithm to screen sensitive wavebands; a two-stage hierarchical classification framework is constructed, health / disease coarse classification-disease degree subdivision is carried out, cross validation type Platt calibration is carried out on decision scores of each group of SVMs, and corresponding probabilities are output; according to the method, a classification threshold value is introduced, the final grade of the morbidity degree is judged through the maximum value of a pseudo probability / threshold value, and through the technical path of spectrum preprocessing, sensitive band screening, two-stage hierarchical modeling, probability calibration and classification threshold value shaping, hierarchical modeling and OvO local boundary learning are utilized to judge the morbidity degree. According to the method, the boundary confusion problem of'health-light 'and'moderate-heavy' is remarkably relieved, the spectrum dimension is reduced by screening sensitive wavebands, and accurate grading of the attack degree of the blackheart disease is achieved.

The invention relates to the technical field of complex equipment data management, in particular to a model-driven complex equipment data relation automatic construction method and system. The method comprises the following steps: constructing a business entity meta-model standard framework comprising a base layer and a domain extension layer through a meta-model technology, and establishing a dual-check system; a hierarchical modeling method is adopted to construct a data model system, and model management is performed through a multi-source heterogeneous data reference mechanism and semantic version control; constructing a data association relationship by adopting a multi-modal fusion method and integrating structural analysis and semantic reasoning; and finally, four-dimensional quality evaluation of integrity, consistency, timeliness and accuracy is carried out on the data association relationship. According to the method, the problems of non-uniform data standards, low association efficiency and incomplete change transmission in complex equipment development are solved, and an automatic closed loop from data association construction to quality control is realized.

The invention relates to the technical field of intelligent traffic simulation and heavy vehicle cooperative control, and discloses an intelligent cooperative control simulation verification system for a tractor or a trailer, which comprises three modules and four modules: a three-dimensional environment modeling module based on multi-source data fusion and hierarchical modeling and combined with a timing and event double-trigger mechanism to update dynamic obstacles; the cooperative motion control module integrates Bezier trajectory planning, a geodesic theoretical dynamics coupling model and triangular collision detection, and realizes vehicle-towing cooperation and risk early warning. The visual simulation interaction module constructs a multi-window view based on OpenGL / Unity 3D, and simulates light and shadow and friction characteristics in combination with material attributes; the result verification module outputs a conclusion based on the quantitative index and the qualitative report. According to the system, the goodness of fit between a simulation environment and a real scene is remarkably improved, the collision early warning response is less than or equal to 100ms, more than 80% of roll-on-roll-off ship scenes can be covered, the entity test cost is reduced by 60%, and a high-fidelity verification tool is provided for vehicle-tow intelligent cooperation.

The invention discloses an AI business process automatic generation method based on AI model combinatorial optimization, and the method comprises the following steps: constructing a multi-level AI capability decoupling and reconstruction module, and carrying out the bottom-up hierarchical modeling and top-down modular decoupling, dynamic mapping of an AI atomic power layer, an AI modular production capacity layer, an AI general capability layer and an AI application business layer is realized; an elastic AI capability combinatorial optimization module is constructed, AI capability combinatorial optimization oriented to three dimensions of data, features and models is carried out based on service quality requirements, and multiplexing, combination and arrangement of AI capabilities are realized through evolutionary computation optimization driven by an agent model; and an AI business process automatic generation module is constructed, an AI service containerization deployment scheme is generated and optimized based on data-driven process mining and a hyper-heuristic algorithm, and AI business process automatic generation is realized. According to the invention, the adaptability and execution efficiency of the AI technology in a complex scene can be improved.

The invention relates to the technical field of power distribution network analysis, and provides a power distribution network distributed photovoltaic bearing capacity evaluation method, system and device, and a medium, and the method comprises the steps: obtaining an energy storage charging and discharging power sequence of a power distribution network according to an energy storage peak load shifting optimization model; an energy storage power initial state is simulated according to a power flow obtained by the energy storage charging and discharging power sequence; global time sequence power flow simulation and layered time sequence power flow simulation are carried out on the power distribution network to obtain power distribution network target access photovoltaic bearing capacity including power grid target feeder line photovoltaic bearing capacity, power grid target distribution transformer photovoltaic bearing capacity and voltage level target photovoltaic bearing capacity of each voltage level; and carrying out bearing capacity evaluation on the to-be-evaluated photovoltaic access scheme according to the target access photovoltaic bearing capacity of the power distribution network to obtain a bearing capacity evaluation result. Based on layered modeling and time sequence power flow analysis, the accuracy of photovoltaic bearing capacity evaluation of the power distribution network can be effectively improved, and the stability and economical efficiency of the power grid after photovoltaic access are ensured.

The invention discloses a fusion method and device based on multi-source water conservancy data, equipment and a medium, and relates to the technical field of water conservancy data processing, and the technical scheme is characterized in that the method comprises the steps: collecting water conservancy original data, carrying out the classified storage of the water conservancy original data according to a water conservancy data specification, and obtaining a water conservancy classification data set; preprocessing the water conservancy classification data set to obtain a preprocessed water conservancy classification data set, combining similar data of multiple sources, and carrying out missing compensation processing on the similar data to obtain a fused similar data set; and carrying out hierarchical modeling on the fused data sets of the same kind by utilizing distributed processing and a multi-bin hierarchical architecture design to obtain a standardized unified water conservancy data mart. The system has the advantages of high-efficiency water conservancy data acquisition, high-quality similar data fusion, high-standard heterogeneous data fusion, high-availability water conservancy data service and the like.

The invention discloses an ancient building parameterized three-dimensional modeling method and system based on point cloud and texture features, and relates to the technical field of computer vision and three-dimensional reconstruction. According to the invention, high-precision cross-modal registration is carried out on the point cloud and texture of the ancient building; intelligent component segmentation is realized by fusing geometric features and texture features of the ancient building and adopting an improved density sensitive DBSCAN algorithm; identifying the category of the component by using a PointNet-CNN hybrid network, and generating an optimal component parameter through parameter template matching and LM optimization solution; a BIM component is generated through layered LOD modeling, and a damaged area is repaired by adopting a Pix2Pix-HD network; the components and BIM metadata are integrated to construct a complete model, and closed-loop feedback correction is achieved through geometry / texture double-precision verification and error type diagnosis; the problems of precision-efficiency-cost imbalance, texture-geometric splitting and component intelligence are solved, and technical support and data basis are provided for high-precision, high-efficiency and low-cost digital archiving, research, monitoring and repairing of ancient buildings.

The invention discloses a text-based human body action content generation method and system, and belongs to the field of computer graphics and artificial intelligence. According to the technical scheme, the method comprises the following steps: constructing a MambaTrans mixed backbone network, and fusing the capability of a Mama model for processing long-sequence data and the capability of a Transform model for capturing a global dependency relationship; a layered adaptive feature enhancement mechanism is introduced, wherein an adaptive frame weighting module dynamically allocates frame weights and a multi-scale feature fusion module fuses different time scale features; the motion is decomposed into layered tokens through a residual vector quantization auto-encoder, a basic token is generated by using a mask Transform, and then a residual token is generated layer by layer by using MambaTrans. According to the method, higher generation quality, higher layered modeling capability and higher reasoning speed are realized, and the method is suitable for virtual reality, augmented reality, movie animation production and game role control.

The invention relates to the technical field of data processing, and discloses a data online modeling and data management method for dimension modeling. The method comprises the following steps: performing dimension characteristic baseline acquisition on a multi-source heterogeneous data stream to generate a dimension attribute characteristic set; performing dimension hierarchical modeling based on the feature set, and outputting a result containing an atomic layer, an aggregation layer and an application layer; calculating a dimension correlation degree according to the topological relation of the service entities, and generating a correlation constraint parameter; detecting dynamic modeling conflicts based on the parameters, and outputting a conflict dimension identification set; distributing a query direction optimization weight according to the identifier set to obtain a dimension query optimization strategy; incremental data loading is controlled based on a strategy, and an incremental loading sequence divided according to time windows is generated; performing closed-loop consistency verification on the sequence, and outputting a verification result set containing data integrity and association consistency indexes; recording a modeling state snapshot according to the result set, and generating a snapshot with a version mark; and analyzing the abnormal backtracking path based on the snapshot, and outputting a dimension data recovery path instruction.

The application discloses a kind of friction coefficient fluctuation and mean value index layered modeling prediction method, belong to mechanical engineering and intelligent monitoring technical field, including the following steps: step S1, multi-source information acquisition and pre-processing;Step S2, multi-source information feature extraction and correlation analysis;Step S3, based on step S2 carries out friction coefficient fluctuation index prediction;Step S4, based on step S3 carries out friction coefficient mean value index prediction.The application uses the above-mentioned friction coefficient fluctuation and mean value index layered modeling prediction method, by introducing layered modeling mechanism and mixed training strategy, realize the modeling of the physical rationality stronger, higher prediction accuracy of each type of index of friction coefficient, with good physical explainability, applicable popularization and engineering practical value.

The application provides a video summarization method based on multi-dimensional features and fine-grained hierarchical modeling, and relates to the technical field of video processing. In practical application, the video summarization technology can facilitate large-scale video retrieval and browsing. The method comprises the following steps: firstly, frame extraction is performed on an input video to obtain a frame sequence, and a multi-dimensional feature extraction network composed of a 2D network and a 3D network is used to extract multi-dimensional features; then, hierarchical temporal modeling is performed to complete the modeling process of the temporal dependence of the entire video sequence; finally, a regression network is used to obtain the importance score of each frame and generate a video summary. The application further explores the influence of the spatiotemporal features extracted by 3D feature extractors with different spatiotemporal complexities on the video summary result. The application shows excellent performance on the video summary datasets SumMe and TVSum. Whether from the application scene or the performance index, the application has strong practical value.

This invention provides a semi-automatic method for constructing and iterating an ontology in the financial and tax domain, comprising the following steps: S1, system initialization and configuration; S2, data acquisition and preprocessing; S3, based on a few-shot learning model and an expert rule base, automatically extracting core concepts from structured financial and tax data and inferring relationships between concepts, performing hierarchical modeling according to financial and tax business logic, and generating a draft ontology model; S4, multi-dimensional automatic verification and manual optimization of the draft; S5, ontology output and version management; S6, dynamic iterative update of the ontology; S7, iterative automatic verification of the ontology in multiple dimensions. This invention achieves four core advantages: improved efficiency, quality assurance, scenario adaptation, and dynamic updates, solving key technical challenges in the construction and application of ontology in the financial and tax domain, providing core support for the efficient operation of financial and tax intelligent agents, and possessing broad application prospects and commercial value.

The application discloses a kind of 6G network multidimensional resource AI model dynamic deployment optimization methods based on category theory, belongs to intelligent collaborative optimization technical field;Method is: the cross-layer consistency dependency of end-to-end AI reasoning service is formalized by functor form;Establish the joint optimization model with long-term average end-to-end delay minimization as target, while being constrained by multidimensional resource and service quality;Convert long-term random optimization problem into time-slot online decision problem;Get AI model dynamic deployment and task scheduling result.The application realizes cross-layer consistency description to task scheduling and model deployment through category theory unified modeling and functor composite mechanism, reduces the inconsistency and redundant constraint caused by hierarchical modeling, improves the structured degree and explainability of joint decision;Under the constraint of multidimensional resources such as calculation, memory, storage and bandwidth, dynamic adaptive optimization is realized, node resource over-limit and load imbalance are effectively avoided, and congestion and queuing delay are reduced.

This invention discloses a multi-scale method for identifying and controlling the propagation paths of airworthiness risks for autonomous vessels. The method includes: hierarchically modeling multi-source airworthiness risk factors based on ontology to construct a structured knowledge graph; establishing logical relationships between risk factors using structural equation modeling and converting them into a Bayesian network; dynamically evaluating input vessel operation data and identifying multi-scale airworthiness risk propagation paths and key risk nodes using graph theory topology analysis; and finally, generating a hierarchical control strategy based on the identification results. This invention enables dynamic and probabilistic assessment of the initial and ongoing airworthiness of autonomous vessels, effectively identifying risk propagation mechanisms and key links, and providing support for the safe design, operation, and supervision of autonomous vessels.

The invention relates to the technical field of data processing, in particular to a cross-end personalized reading recommendation method and system based on big data, and the method comprises the steps: carrying out the hierarchical division of a multi-terminal user through a cross-end user hierarchical modeling method, collecting the data of each layer of user, and constructing a user content interaction database; determining a reading preference threshold value and a content matching threshold value of each level user based on the database to obtain content feature sensitivity; the method comprises the following steps: acquiring cross-end reading behavior data and reading experience data of a user in real time, generating a cross-end recommendation candidate set when the data exceeds a preference threshold value or is lower than a matching threshold value, analyzing correlation between a user reading state and content characteristics, generating a recommendation parameter set, performing sensitivity grouping on the user according to the sensitivity of the content characteristics, according to the method, the progressive recommendation rule is generated to optimize the recommendation parameter set, and the cross-end reading platform is controlled to perform personalized content recommendation based on the optimization result, so that accurate and dynamic recommendation in a cross-end scene is realized, and the consistency and satisfaction of user experience are improved.

The invention discloses a control method and system for deep peak regulation of a 1000MW secondary reheating unit, and relates to the technical field of thermal power unit peak regulation control. Aiming at the problems of thermal inertia lag, multivariable coupling and low-load stability during deep peak regulation of a unit, a technical system is constructed from three dimensions of boiler side optimization, steam turbine protection and coordinated control upgrading: a three-stage stable combustion system of plasma strengthening, powder preparation cooperation and smoke temperature regulation is used for assisting a boiler to maintain stable combustion under low load; a parameter early warning-dynamic correction-rate grading protection system is adopted to assist safe operation of the steam turbine; the coordination control system is upgraded based on'multivariable feedforward + internal model optimization 'to relieve multivariable coupling influence. Meanwhile, through the implementation path of layered modeling, logic integration, staged verification and all-working-condition optimization, safe and economical deep peak regulation of the unit within the 30%-100% load interval is facilitated. According to the method, the unit peak regulation adaptability and stability can be improved, and the energy consumption and equipment loss risk can be reduced.

The application discloses a kind of personalized design intelligent interaction methods based on topology optimization, including S1, construct the user behavior monitoring system based on multi-modal data acquisition, generate high-dimensional data set containing time sequence characteristics;S2, based on adaptive deep neural network model, dynamically predict and hierarchical modeling to user personalized demand;S3, the optimal solution of design parameter is generated in real time by dynamic topology reconstruction algorithm;S4, through reinforcement learning mechanism, iteratively update design scheme generation rule;S5, construct neural network supported multi-objective design space exploration and optimization framework, quickly filter and optimize design parameter;S6, through multidimensional data mapping and parameterization control means, realize the dynamic update and local optimization of design scheme;S7, establish data-driven continuous learning and evolution mechanism, constantly optimize user demand prediction model and design optimization algorithm.The application has the advantages of strong dynamic adaptability, high intelligent level and high personalized satisfaction precision.

The invention discloses a power load prediction method and system based on a conditional diffusion model, and a storage medium. The power load prediction method based on the conditional diffusion model comprises the steps of recombining historical power load time sequence data into a two-dimensional image matrix; performing multi-scale decomposition on the two-dimensional image matrix through multi-level average pooling to obtain trend component sequences with multiple resolutions; constructing a conditional diffusion model comprising a conditional network, a denoising network and a fusion decoder; and training the conditional diffusion model by using a historical trend component sequence to establish a mapping relationship between the input trend component sequence and the predicted power load time sequence data as an obtained target prediction model. The method aims at achieving multi-resolution conjoint analysis, hierarchical modeling and variable length processing of the power load time sequence, and the prediction precision and robustness of the complex multi-scale power load sequence are improved.

This invention provides a zero-shot anomaly detection method and system based on bidirectional alignment enhancement. The method includes: achieving direct interaction between global and local features through the cross-attention module CAGL; extracting fine-grained local features through the region token alignment module RegionAlign; and projecting global visual semantics onto the text embedding space through the text embedding enhancement module TextAug to generate domain-aware cue words. This invention solves the technical problems of poor generalization ability, neglect of local spatial details, and difficulty in hierarchical modeling of multi-scale visual anomalies.

The invention provides a multivariable time sequence anomaly detection method based on dynamic graph hierarchical modeling and learning, which comprises the following steps: inputting multivariable time sequence data as an original data source; dynamic graph layered modeling is carried out, and local, regional and global dynamic graphs are constructed; respectively applying a graph neural network on the local dynamic graph, the regional dynamic graph and the global dynamic graph, and extracting multi-level features; a layered contrast learning mechanism is introduced, so that normal samples are aggregated in an embedding space, and abnormal samples are pushed away in each layer; a dynamic and steady state difference learning mechanism is introduced to ensure that the model keeps balance between a long-term trend and a short-term dynamic state; and inputting the optimized features into a normalized flow model, and outputting an anomaly score and an anomaly positioning result. According to the method, multi-level graph structure modeling of time series data and a comparative learning mechanism are combined, and the accuracy, robustness and sensitivity of anomaly detection can be effectively improved.

The invention discloses a simulation task load dynamic change-oriented end-cloud integrated cluster dynamic container capacity planning method, and relates to the technical field of computer services. Modeling a micro-service, a container, a user and a server; load sensing and triggering judgment, including calculation of micro-service total load and load variation, calculation of load intensity and triggering condition judgment; mDP differential capacity adjustment: constructing an MDP state space, formulating a container capacity expansion and shrinkage strategy, and calculating an adjustment correction coefficient; key index calculation including a response time model, a system total cost model and a fairness index model; and carrying out multi-objective solution and verification, namely constructing an objective function, setting constraint conditions and solving by adopting fusion MDP. Hierarchical modeling is carried out by adopting micro-services, containers, users and servers, MDP decision differentiation adjustment is triggered through load sensing, multi-target solving is carried out, a four-layer technical architecture of modeling-sensing-decision-solving is formed, and dynamic optimization of cluster capacity can be realized.