91 results about "Space optimization" patented technology

The invention discloses a territorial space planning optimization method and system based on three-dimensional modeling, and relates to the technical field of three-dimensional modeling, and the method comprises the steps: obtaining multi-source territorial space data, constructing a three-dimensional space data set, carrying out the three-dimensional geometric modeling, and generating a territorial space three-dimensional model; carrying out multi-dimensional space analysis on the land space three-dimensional model, obtaining a space conflict feature set, carrying out planning constraint, and formulating a space optimization suggestion; executing the space optimization suggestion to perform three-dimensional dynamic modeling, generating a multi-time sequence planning simulation result to perform space planning evaluation, generating a land space planning score, performing planning compensation based on the backtracking space optimization suggestion, updating the space optimization suggestion, and obtaining a space planning optimization scheme. The technical problem that in the prior art, land space planning depends on two-dimensional data, the space conflict and the dynamic evolution process are difficult to comprehensively recognize, and the planning scheme is insufficient in scientificity is solved, and the technical effect of improving the space conflict recognition precision and the planning decision scientificity is achieved.

The invention discloses a land space optimization configuration method under terrain gradient constraint. The method comprises the following steps: S1, collecting multi-source geographic data to construct a spatial database; s2, factors such as gradient and elevation are extracted, and terrain gradient partitions are divided; s3, constructing a three-dimensional conflict matrix based on suitability evaluation and landscape pattern analysis; s4, constructing a driving index system, and analyzing a driving path of the spatial conflict; s5, setting a multi-objective optimization function, and constructing game rules and spatial conversion constraint conditions of the three types of subjects; s6, integrating the C-FLUS model, the MAOOAM model and the R2D3 model, and constructing a coupling system to execute dynamic optimization; and S7, when the convergence condition is met, outputting an optimal configuration scheme and generating a compensation and control strategy. According to the method, efficient optimization and intelligent decision support of multi-target land configuration under the terrain differentiation condition are realized.

The invention discloses a vegetation carbon sink space optimization method and system based on interpretable machine learning. The method comprises the steps that firstly, multi-source environment data such as administrative division, environment factors and target variables of a target area are acquired, and a space table data set is formed; training an XGBoost regression model based on the data set, predicting the vegetation carbon sink amount and evaluating the performance of the model; quantifying the contribution of each environmental factor to the carbon sink by combining an SHAP interpreter, generating a feature importance and interpretability map, fitting through a LOWESS algorithm to obtain an environmental factor and carbon sink response relation curve, and extracting an optimal threshold interval; further predicting a carbon sink gain according to an optimization threshold value, delimiting optimization partitions, and forming a space optimization scheme; and finally outputting a vegetation carbon sink optimization suggestion report. According to the method, scientific quantification of carbon sink space optimization can be realized, and the interpretability and decision support value of a carbon sink improvement strategy are improved.

The invention discloses a method for improving, regulating and controlling functions of a farmland ecological barrier in a plain area, relates to a related algorithm based on a multi-dimensional coupling model, and aims to solve the technical problems of system element splitting, single ecological function, poor space adaptability and the like in the conventional farmland ecological barrier construction. Different land utilization type areas are taken as decision variables, and underground water footprint reduction, farmland carbon reserve increase, economic benefit improvement, crop water resource utilization efficiency improvement and pollutant emission reduction are taken as objective functions; the method comprises the following steps of: constructing a water volume, water quality, economy, efficiency, carbon, grain and ecology multi-dimensional coupled optimal configuration model by taking available water volume and land utilization type occupied area as constraint conditions, carrying out gridding land utilization distribution through a space optimization algorithm, and applying an optimal configuration scheme to a space to finish multi-dimensional element coupled water and soil resource optimal configuration. The method can be used in agricultural planning field.

The invention discloses a knowledge graph-driven line storage space joint optimization scheduling model construction method. The method comprises the following steps: determining an optimal scheduling target and an overall target function of row storage space joint flood control scheduling, and setting constraint conditions; under a knowledge graph theoretical framework, constructing a response relationship between single-class line storage space scheduling and a river flood discharge state; selecting an excess peak water level exceeding a river embankment control water level as a coordination variable, decomposing the overall objective function into system association constraint equations of a river embankment, a reservoir, a flood storage and detention area and a flood drainage pump station, and reducing a multi-class line storage space joint scheduling problem into a problem of optimization scheduling of each single-class line storage space scheduling; and respectively solving by adopting an improved particle swarm algorithm to obtain an optimal scheduling process of each single-class line storage space. The method not only solves the problem that global optimum is difficult to ensure by directly applying a traditional optimization algorithm due to many row storage space optimization targets and a competitive relationship, but also reduces the calculation time of the model.

The invention relates to the technical field of space planning, in particular to an intelligent optimization method for territorial space planning fusing multi-objective evolution. The method comprises the following steps: carrying out multi-modal territorial space characterization analysis on territorial space heterogeneous data to generate multi-modal territorial grid characterization space data; performing territorial multi-modal feature analysis processing on the multi-modal territorial grid representation spatial data to generate territorial planning unit multi-modal feature data; carrying out evolution trend analysis on the territorial planning unit multi-modal characteristic data to generate territorial planning unit evolution trend data; performing territorial space planning multi-target optimization strategy analysis based on the territorial planning unit evolution trend data to generate territorial space planning multi-target optimization strategy data; and carrying out multi-objective evolution territorial space planning intelligent optimization processing through the territorial space planning multi-objective optimization strategy data to generate territorial space planning intelligent optimization data. According to the invention, intelligent land and land space optimization planning is realized.

The invention relates to the field of industrial automation, and particularly provides an intelligent scheduling industrial robot logistics warehousing system. Based on an algorithm of reinforcement learning and real-time environment perception, the robot is endowed with the capability of dynamically planning the carrying path, the carrying efficiency is improved, and the collision risk is reduced. Based on an algorithm of space optimization and an intelligent storage strategy, storage space management is redefined, the space utilization rate is increased, and the storage retrieval process is optimized. The algorithms push the system to be efficient, accurate and intelligent from multiple dimensions, and a revolution is brought to the logistics storage industry.

The invention relates to the technical field of building intelligent management, and discloses a BIM-based indoor optimization design method and system, and the method comprises the following steps: a) building a building information model through BIM software, and collecting the space size, material parameters, structure parameters, environment data and user demand data of a building; and b) an indoor space optimization target is set according to design specifications and user requirements, design constraint conditions are formulated, and the design constraint conditions comprise building specifications, fire protection requirements, human flow lines, functional requirements and ergonomic parameters. According to the invention, through multi-source data fusion, high-precision and low-delay data acquisition is realized; through hybrid modeling, the balance between the accuracy of energy consumption prediction and the calculation efficiency is realized; through reinforcement learning and fuzzy control, adaptive optimization of building environment adjustment is realized; through the modular architecture and the block chain, high expansibility and data security of the system are realized.

The invention discloses a space optimization-based low-efficiency camellia oleifera forest transformation method and system. The method comprises the steps of collecting standard camellia oleifera forest data to construct a standard production database; the method comprises the following steps: acquiring stand layout data and single plant data of a camellia oleifera low-efficiency forest to be transformed, and constructing a twin digital forest of the camellia oleifera low-efficiency forest; based on data in the camellia oleifera standard production database, constructing a camellia oleifera low-efficiency forest reconstruction prediction model with maximization of forest stand expected total yield and minimization of reconstruction implementation complexity as targets, inputting data in the twin digital forest into the model, and performing optimization calculation to obtain a low-efficiency camellia oleifera low-efficiency forest reconstruction prediction model. Generating a production optimization scheme containing an optimal plant layout and a single plant transformation suggestion; based on the optimization scheme, generating a visual construction drawing or an AR guidance instruction to guide reconstruction construction; after transformation, output data are collected to verify the effect. According to the method, through digital modeling and multi-target space optimization, refinement, intelligence and scientific decision-making of camellia oleifera low-efficiency forest transformation are realized, and the overall output benefit is effectively improved.

The invention discloses an intelligent boxing method and system based on space optimization in an automatic loading and unloading scene, and relates to the technical field of logistics automation, and the method comprises the steps: obtaining the basic information of a to-be-loaded box body, measuring the space size of a loading container, and initializing the constraint condition of boxing; calculating an initial boxing plan based on a three-dimensional space optimization algorithm, and establishing a spatial index mechanism; and guiding the loading equipment to place the current box body, detecting the actual placement position of the box body, updating the position information of the loaded box body, recalculating the boxing plan of the remaining box bodies, and repeating the steps until the loading of all the box bodies is completed. The method is specially designed for an automatic loading and unloading scene, can be seamlessly integrated into intelligent equipment such as a loading and unloading robot, and provides key algorithm support for logistics automation; the logistics transportation cost is effectively reduced by improving the loading efficiency and the space utilization rate; through a real-time detection and dynamic adjustment mechanism, the system can automatically deal with abnormal conditions in the loading process.

The invention discloses a new energy consumption space optimization method and system, relates to the technical field of power system planning and operation, and solves the problem of large calculation and evaluation deviation of a new energy consumption space. On the basis of power grid hardware constraint, the environment temperature and the starting number of the thermal power generating units are introduced to generate the dynamic limit matrix, and the method can adapt to changes of the power adjusting range and the safe operation boundary of the direct-current power transmission system caused by starting and stopping of the thermal power generating units; and then a constraint equation set capable of representing AC / DC coupling characteristics is constructed, and a mixed integer programming model is established and solved by taking maximization of a new energy consumption space of the power grid as an objective function, so that errors caused by calculation after an AC / DC system of the power grid is decoupled in a traditional method can be reduced. Through dynamic constraint and refined modeling, the problem of evaluation deviation caused by static hypothesis in a traditional method is solved, and a high-credibility new energy consumption decision basis can be provided for power grid dispatching.

The invention discloses a hysteretic parameter identification method of a BWBN model based on improved particle swarm optimization, and belongs to the field of hysteretic parameter identification, and the method comprises the steps: determining an equivalent yield point of an RC pier under quasi-static reciprocating loading according to a skeleton curve of an actually measured hysteretic curve; giving a boundary constraint of a hysteretic parameter to be identified; obtaining an optimal value of the hysteresis parameter to be identified by using an improved particle swarm algorithm; and in each step of iteration of the improved particle swarm algorithm, optimizing the global historical optimal position by using a Levy flight strategy and a pattern search algorithm, and taking the global historical optimal position output in the last step as the value of each hysteresis parameter to be identified, thereby completing the hysteresis parameter identification of the BWBN model. The method solves the problems that when an existing method is used for processing a complex high-dimensional space optimization problem, the existing method is prone to falling into a local optimal solution in the later period, premature convergence is achieved, the convergence speed and fitting precision depend on parameter selection, evolution lacks enough exploration capacity, and the diversity of particle populations is difficult to guarantee.

The invention discloses a channel structure analysis method and system based on multi-modal data, and the method comprises the steps: obtaining the multi-source data of a target region, building a three-dimensional geological structure model of the target region based on the multi-source data, generating a point distribution scheme through a space optimization algorithm, generating a monitoring network based on the point distribution scheme, and carrying out the analysis of the three-dimensional geological structure model. Different types of tracers are injected into the injection points by adopting a graded injection strategy, monitoring data are collected, an underground structure model is generated based on the monitoring data, hydraulic parameters are calculated, the underground structure model is dynamically corrected based on the hydraulic parameters, and an underground pipeline three-dimensional model is obtained; and grading channel structures in the underground pipeline three-dimensional model by using a preset channel grading standard, and generating a channel three-dimensional map based on a grading result. According to the method, the three-dimensional model of the underground pipeline is generated through comprehensive analysis of the multi-modal data, and the structural analysis precision of the underground water pollution channel is improved.

The invention relates to the technical field of space optimization, and discloses a container loading space optimization method based on robot collaboration, and the method comprises the steps: building a multi-dimensional physical model of a package through obtaining the point cloud data and weight distribution of the package; generating a loading scheme containing a target pose, a pressure tolerance threshold and a loading sequence based on a reinforcement learning algorithm; the mechanical arm generates a motor current-position composite control instruction according to a pre-trained grabbing dynamical model, and accurate grabbing is achieved; in the loading process, the pose of the parcel is monitored in real time through a Kalman filtering algorithm, when deviation exceeds a threshold value, a local space KD tree is constructed, a collision probability gradient field is calculated, and a compensation scheme meeting stability constraints is generated; and finally, updating the quad-tree index and feeding back to the reinforcement learning algorithm for online optimization. The problem that in the robot loading process, the grabbing force control is not accurate, and the efficiency is reduced due to execution deviation accumulation is effectively solved, and efficient and accurate automatic loading is achieved.

The invention belongs to the technical field of computer vision and model generation, and particularly relates to a diffusion generation method for hierarchical condition control and dynamic jump sampling, and the method is realized through a hierarchical control architecture: macroscopic time scheduling adopts a bidirectional LSTM dynamic combination de-noising step, and the equivalence is ensured through differential transition constraint; in microscopic space optimization, a key area is positioned through an uncertainty thermodynamic diagram, only local attention is executed on the key area, and an implicit network is used for background generation; according to the multi-modal fusion, dynamic routing gating is adopted to calculate the weight and adversarial training is combined, according to the diffusion generation method of hierarchical condition control and dynamic jump sampling, the sampling step number is compressed from 1000 + to 50-80, 95% of non-critical area calculation is reduced, the multi-modal guide quality loss is reduced by 37%, the peak video memory occupation is reduced by 60%, the efficiency is improved while the quality is guaranteed, and the method is suitable for large-scale popularization and application. And mobile terminal deployment is supported.

The invention provides a balance calibration modeling method and system based on sample errors and space optimization. The method comprises the steps of obtaining a calibration data set and dividing the calibration data set into a sample set and a verification set; calculating a load space weight and a load complexity weight of the sample set, and obtaining a comprehensive weight based on the load space weight and the load complexity weight; and constructing a polynomial model of the sample set, and solving a coefficient matrix in the polynomial model based on the comprehensive weight to obtain a final balance calibration model. The method further comprises the following steps: verifying and evaluating the balance calibration model by using the sample set and the verification set, and optimizing the balance calibration model according to an evaluation result. According to the balance calibration modeling method cooperating with the sample space distance and the load complexity weight provided by the invention, nonlinear system errors introduced by calibration load complexity are fully considered, so that the influence of error distribution of calibration samples under various load combinations on modeling is more uniform and reasonable; and meanwhile, the problem of sample space occupation ratio is also solved.

This invention discloses a method and system for spatial optimization of vegetation carbon sinks based on interpretable machine learning. The method first acquires multi-source environmental data, including administrative divisions, environmental factors, and target variables for the target area, forming a spatial tabular dataset. Based on this dataset, an XGBoost regression model is trained to predict vegetation carbon sink capacity and evaluate model performance. A SHAP interpreter is then used to quantify the contribution of each environmental factor to carbon sink, generating feature importance and interpretability maps. The Lowes algorithm is used to fit the relationship curve between environmental factors and carbon sink response, extracting the optimal threshold interval. Further, based on the optimization threshold, carbon sink gain is predicted, optimization zones are delineated, and a spatial optimization scheme is formed. Finally, a vegetation carbon sink optimization recommendation report is output. This method enables the scientific quantification of carbon sink spatial optimization, improving the interpretability and decision support value of carbon sink enhancement strategies.

The present invention discloses a method for evaluating and optimizing the resilience of territorial space, which relates to the technical field of spatial evaluation and optimization. The method includes collecting multi-modal data of territorial space and performing preprocessing; constructing a spatio-temporal coupling model based on the multi-modal data to calculate global coupling characteristics, and constructing a resilience evaluation model to analyze the resilience of territorial space; optimizing the territorial space based on the GAN model and genetic algorithm, and evaluating the optimized territorial space plan; forming optimization suggestions for the optimized territorial space plan for display and storing them in a database. By collecting territorial space data to generate a topological network and constructing a spatio-temporal coupling matrix to extract features of the topological network, and obtaining the resilience of territorial space through constructing a resilience evaluation model, the present invention effectively improves the accuracy and flexibility of territorial space evaluation, realizes efficient and rapid territorial space evaluation. At the same time, an optimized territorial space plan is generated and evaluated according to the GAN model and genetic algorithm, improving the effectiveness of the optimized territorial space plan.

The method for constructing urban ventilation corridors based on local climate zone classification and remote sensing inversion belongs to the technical field of urban climate regulation and space optimization. The present application constructs a multi-scale urban thermal environment analysis framework, from macro, meso to micro level, to identify the coupling mechanism of urban form and thermal effect, and to realize the fine modeling and regulation path identification of thermal environment. At the same time, combined with machine learning and explainable analysis method, the scientificity and transparency of thermal environment driving factor identification are enhanced. Through remote sensing inversion, form factor construction and three-dimensional simulation simulation, a complete technical path of "cold source identification-resistance modeling-path optimization-microclimate response evaluation" is put forward, which effectively improves the ventilation and cooling potential identification ability of blue-green infrastructure, and is suitable for ventilation corridor planning and climate adaptive updating design of various urban spaces, and has wide application prospect.

The invention discloses a monocular depth estimation adversarial sample generation method based on a diffusion model. The monocular depth estimation adversarial sample generation method is specifically implemented according to the following steps: step 1, selecting and preprocessing an existing data set; step 2, loading a monocular depth estimation model and evaluating baseline performance; step 3, construction of a diffusion model and initialization of a back diffusion process; 4, performing disturbance-resisting submerged space optimization and constraint generation; and 5, decoding the adversarial sample, enhancing the physical realizability and verifying the attack effect. According to the method, on the premise of keeping the imperceptibility of confrontation disturbance, the prediction precision of the monocular depth estimation model in real and physical environments is remarkably reduced, so that the robustness and safety of the monocular depth estimation model are evaluated and improved.

The invention relates to the technical field of ocean space planning, in particular to an ocean function space configuration method based on landscape ecology. The method comprises the following steps: acquiring multi-source geographic space data of a target sea area, and integrating the data to generate a reference marine landscape map; calculating a group of predefined marine landscape pattern indexes according to the reference digital marine landscape map, and quantitatively evaluating the space structure of the map through the marine landscape pattern indexes; an iterative space optimization algorithm is executed, and the spatial layout of the functional area is reconstructed on the premise that constraint conditions are met by taking optimization of an ecological structure as a target; and finally, outputting a marine function space configuration diagram after ecological optimization. According to the method, the limitation of a traditional qualitative zoning method is overcome, a scientific, quantitative and repeatable process is provided for ocean space planning, and therefore the ecological integrity of ocean management planning is effectively improved.

The invention relates to an optimal configuration method and system for a blue-green space. The optimal configuration method of the blue-green space comprises the following steps: acquiring acquisition data of a to-be-evaluated area; performing landscape pattern quantitative analysis on the to-be-evaluated area based on the collected data; carrying out fusion degree analysis on the to-be-evaluated area based on a landscape pattern quantitative analysis result; performing landscape optimization evaluation on the to-be-evaluated area based on a landscape pattern quantitative analysis result and a fusion degree analysis result to obtain an optimal fusion degree; and optimizing the to-be-evaluated region based on the optimal fusion degree. According to the method, accurate evaluation of the blue-green space can be realized, and space optimization configuration can be effectively carried out, so that the space distribution of the blue-green space is reasonable, and the problems of fragmentation, poor connectivity and the like do not exist.

The application provides a sea wind farm layout optimization method based on tail flow constraint random reconstruction, aiming at the problems that the tail flow effect and electrical fault control are separated in the prior art, and the aerodynamic interference is ignored in fault reconstruction. The application first establishes an initial layout model with tail flow awareness, maximizes wind energy capture efficiency through nonlinear space optimization, then constructs a probability scenario of fault and wind speed, and quantifies the influence of cable random fault on tail flow propagation. Specifically, the application designs a dynamic reconstruction optimization layer, embeds aerodynamic power constraints and tail flow sequence into electrical topology decision, has aerodynamic and electrical regulation and control capability in fault state, realizes investment planning and reconstruction optimization through progressive solution, and introduces a tail flow feasibility cutting mechanism to improve robustness. The application effectively solves the limitations of traditional layout design, improves power generation efficiency, reduces tail flow loss, shortens fault recovery period, and provides a high-precision scheme for sea wind farm layout optimization.

The invention discloses an intelligent anchor position detection algorithm based on multilevel spatial constraints. The intelligent anchor position detection algorithm comprises the following steps: establishing a ship safety distance model based on a Monte Carlo stochastic simulation technology; establishing an anchoring safe water depth detection model based on KD-Tree; and screening anchor location points which simultaneously meet safety spacing and water depth constraints. According to the method, the advantages of global fast search and local fine detection are effectively integrated by constructing a layered progressive space optimization framework. The method comprises the following steps: firstly, establishing a multi-level spatial constraint model for ship anchor position detection on a theoretical level, and defining a collaborative optimization mechanism between different levels; and secondly, organic fusion of spatial index and dynamic filtering is realized in algorithm design, an anchoring position point meeting the safe distance between ships is obtained through an anchoring safety detection model based on plane distance constraint, an anchoring position point meeting the water depth constraint is obtained through water depth detection based on KD-Tree, and compared with an existing method, the calculation efficiency is remarkably improved.

Techniques are described for encoding symbols using a new algorithm that provides a flexible Huffman tree approximation and can be used for low-latency encoding. For example, the new algorithm can perform encoding using one or more of the following stages: Shannon-based binning, code space optimization, tree completion, and code allocation.

This invention discloses a method for testing jailbreak attacks on large language models based on latent space optimization. First, standard rejection vectors pointing to rejection concepts at multiple network layers of the target large language model are obtained. Then, the set of key layers performing security defenses in the target large language model is located, and adaptive weights are assigned to each layer in the set of key layers. A cognitive hijacking prefix targeting malicious query commands is generated and compressed. Using the positive guiding part of the cognitive hijacking prefix as the optimization target, an initial suffix is ​​generated through a first-stage adversarial suffix optimization. Finally, using the complete cognitive hijacking prefix as the optimization target, combined with the standard rejection vectors and the adaptive weights, a second-stage adversarial suffix optimization is performed on the initial suffix to generate the final adversarial suffix. This method transforms the attack target from shallow output probability manipulation to deep cognitive state alignment, effectively eliminating the "pseudo-jailbreak" phenomenon.

The invention provides a block chain data storage device and a space optimization method, and relates to the field of data storage, and the block chain data storage device comprises a cache module which is used for storing block data indexes of high-frequency access; the distributed storage module is used for dividing the block chain data into dynamic fragments and distributing the dynamic fragments to different nodes for storage; the encryption module is used for encrypting the unstructured data and storing the unstructured data in a local database; the load balancing module is used for dynamically monitoring node loads and optimizing storage resource allocation; and the consensus mechanism module is used for defining data classification, fragmentation rules and migration logic through an intelligent contract, ensuring data consistency by adopting a right proof or PBFT algorithm, and automatically executing a storage rule and data verification. The overall storage cost is reduced, and the storage space is optimized.

The invention discloses a glass composition optimization method and system based on a neural network and a goat optimization algorithm, and relates to the technical field of glass composition optimization. The method comprises the following steps: optimizing hyper-parameters of a neural network by using a goat optimization algorithm, and constructing a high-performance neural network prediction model; then, the optimized neural network prediction model is embedded into a fitness evaluation link of a goat optimization algorithm, in the glass component optimization process, performance indexes of different component combinations are rapidly predicted through a neural network, the performance indexes serve as fitness values of the goat optimization algorithm, and the algorithm is guided to efficiently search for the optimal component combination. According to the method, through the high-precision prediction capability of the neural network and the global search and adaptive capability of the goat optimization algorithm, the efficiency and accuracy of glass component design are remarkably improved, an innovative solution is provided for research and development of high-performance glass materials, the separation dilemma of'prediction-design 'of glass components is broken through, and the method is suitable for large-scale popularization and application. And the core challenge of high-dimensional constraint space optimization is solved. And glass research and development is pushed from trial and error experiments to calculation design.

The present invention relates to a land space optimization method that integrates cross-scale planning objectives and ecological resilience feedback. First, based on the big data platform GEE, the random forest algorithm is used to perform long-term land space function classification and analyze driving factors. Planning texts from multiple administrative levels are collected, planning objectives are analyzed, and the target parameter group is obtained using the equal spacing method. Combined with the "double evaluation" results, a model is built to simulate the future land space structure. On this basis, the vegetation structure is analyzed, and the long-term resilience indicators of the global and local grids are simulated based on biophysical processes. Finally, the changing trends of all resilience indicators are analyzed. If there is a trend shift, the parameters are fed back and iterated until all resilience indicators remain stable or show a positive trend, the cycle is ended, and the output is output. This method quantifies for the first time the feedback relationship between planning objectives at different levels and ecosystem resilience, successfully achieving the effective integration of sustainability and resilience, and improving the scientific nature of land space structure optimization and spatial decision-making.

This invention discloses an adaptive reinforcement learning-driven multi-objective optimization method for accelerators, belonging to the technical field of accelerator optimization design. First, it analyzes the mechanism of action of key design variables such as convolution loop unrolling factor, quantization accuracy, and cache partitioning, establishes a system-level state modeling method, and designs the DDPG framework suitable for continuous action space optimization as a learner for the configuration strategy. Simultaneously, a lightweight searcher is implemented by running Python scripts on the CPU to search and allocate the block parameters of Loop-3 and Loop-4 in real time, thereby reducing the computational burden on the reinforcement learning agent. Finally, an adaptive configuration method driven by reinforcement learning and a corresponding search process are proposed to achieve joint optimization of unrolling scale, quantization accuracy, and block parameters.