165 results about "Algorithm convergence" patented technology

The invention relates to a network topology dynamic optimization method and equipment for large-scale power supply and distribution equipment networking. The method comprises the following steps: step S101, real-time data acquisition and state sensing; step S102, carrying out network topology modeling and performance evaluation; step S103, dynamic risk assessment and optimization target generation; s104, carrying out topological optimization decision making based on a feasibility maintenance type genetic algorithm; step S105, carrying out optimal strategy verification and seamless switching; selecting an optimal network topology reconstruction scheme from the Pareto optimal solution set according to a preset decision strategy; and after the reconstruction scheme is verified on a control level, generating an equipment cascade and open circuit control instruction sequence, and guiding related nodes to complete undisturbed switching of the network topology in a preset time window through a distributed cooperative control mechanism. According to the method, generation of invalid solutions can be avoided, and the convergence efficiency of a large-scale network topology optimization algorithm is remarkably improved.

The invention discloses a machine room group control multi-objective optimization digital twin platform and an optimization method, and relates to the technical field of machine room group control. Constructing a multi-objective optimization model, wherein the multi-objective optimization model comprises a three-dimensional objective function and constraint conditions; the three-dimensional objective functions comprise an energy consumption minimization function, a temperature and humidity control precision maximization function and an equipment life loss minimization function. Compared with a traditional static twinborn model, the method has the advantages that the deviation is greatly reduced, and a precise virtual-real mapping basis is provided for optimization decision making. According to the whale optimization algorithm, the global search capability and the NSGA-IIPareto solution screening capability are fused, the convergence speed of the algorithm is improved through fuzzy membership degree fitness calculation and crowding degree screening, when deviation exceeds a threshold value, optimization can be shortened to 15 seconds, compared with an existing algorithm, the response speed is greatly improved, and the real-time control requirement under the dynamic load of a machine room is met.

The invention proposes a robot path planning algorithm based on a particle swarm optimization algorithm and a dynamic window method, and relates to the field of control theories and electronic information, and the method comprises the steps: introducing an inertia weight updating strategy based on a state factor, setting adaptive parameters and crossover and mutation operators to improve the global search capability, increase the population diversity, and improve the robot path planning precision. On-line self-adaptive updating of inertia weight and learning factors is realized on line in combination with Q-learning, gene combination modes are enriched through crossover operators, convergence and exploratory performance of the algorithm are improved, an obstacle avoidance strategy of a traditional DWA algorithm is improved, weight parameters of a dynamic window evaluation function are dynamically adjusted according to real-time information of a target and an obstacle, and an obstacle avoidance algorithm is established. According to the method, the global planning is adopted, the path points generated through global planning are adopted as temporary targets, fusion of MOQLCOPSO and the improved DWA algorithm is achieved, local optimum is effectively avoided, the planning efficiency and path safety are improved, and the method is suitable for mobile robot navigation under the complex three-dimensional terrain.

The invention discloses a robot control optimization method based on deep reinforcement learning, and the method comprises the steps: firstly initializing a control system, constructing an experience playback buffer pool, and setting an Actor network strategy function, a Critic network weight, a hyper-parameter, and a target network parameter; then, the robot generates an action according to the Actor network, collects environment feedback after executing the action, and stores the state transition tuple into a buffer pool; next, a small batch of data is sampled, a time difference error is calculated, and a network parameter is updated accordingly to minimize a value estimation deviation, synchronously optimize the network parameter, maximize an expected value of a state-action value function, and improve control performance. According to the method, a multi-thread architecture and a deep reinforcement learning technology are fused, the real-time performance, stability and generalization ability of robot control are improved, the problem of algorithm convergence in a non-stationary environment is effectively solved, the cost of adjustment and calculation is reduced, and the robustness of long-term stable operation and deployment feasibility in a complex scene are enhanced.

The invention discloses a satellite edge computing task scheduling method and system based on multi-agent deep reinforcement learning, and the method comprises the steps: S1, constructing a system model which comprises a user equipment model and a satellite edge computing server model; s2, constructing a target function, wherein the target is to minimize the total delay and total energy consumption of task processing of the user equipment and the satellite edge computing server; s3, modeling an original optimization problem into a decentralized partial observation Markov decision process, proposing an efficient cross-plot Transform multi-agent near-end strategy optimization algorithm, and training a task unloading and resource allocation model until the algorithm converges; and S4, the user equipment makes an appropriate task unloading and resource allocation decision based on the trained task unloading and resource allocation model according to the current satellite edge computing environment state and task requirements. According to the method, satellite movement and calculation and communication resource constraints of equipment in a satellite edge calculation environment are fully considered, joint optimization of task unloading and resource allocation strategies is realized through a multi-agent deep reinforcement learning algorithm, the total delay and total energy consumption of the system are effectively reduced, and meanwhile, the task completion rate is improved.

The invention discloses a workshop layout optimization method for a discrete manufacturing system, and aims to construct a multi-row facility layout mathematical model considering a channel loading and unloading point mechanism aiming at the process storage behavior and cross-row logistics path problems in the manufacturing process. According to the method, on the basis of analysis of workshop parameters and logistics paths, reasonable assumed conditions are set, decision variables such as facility space positions and arrangement relations are defined, a constraint system meeting uniqueness, non-overlapping performance and boundary limitation of facilities is established, and material handling cost minimization is taken as a target. And a Manhattan distance and a Floyd algorithm are adopted to measure the same-row transportation distance and the inter-row transportation distance respectively. In order to solve the model, an improved genetic algorithm fusing bidding selection and a random immigrant mechanism is designed, and the algorithm convergence performance and the global search capability are improved. According to the method, the inter-bank logistics cost can be remarkably reduced, the compactness of the facility layout and the path continuity are optimized, and the method is suitable for manufacturing scenes with process storage characteristics such as welding, assembling and warehousing transfer.

The invention discloses an unmanned aerial vehicle area coverage flight path planning method and system based on graph segmentation, and relates to the technical field of unmanned aerial vehicle path planning. According to the method, a coverage point pool with high quality, comprehensive coverage and robustness is constructed for subsequent greedy selection through a strategy of adaptively generating candidate points; then, a greedy selection mechanism based on effective scores is adopted, the current optimal coverage point is accurately selected in each round of iteration, the new coverage area can be increased to the maximum extent, the overlapping area can be effectively controlled, and finally the whole target area is completely covered with the minimum number of circles. In the aspect of a path planning algorithm, a hybrid initialization mode of a greedy algorithm and a random generation strategy is innovatively combined, and the quality of an initial solution and population diversity are ingeniously balanced; meanwhile, in each generation of evolution of the genetic algorithm, a 2-opt local search strategy is embedded into a new non-elite individual, so that the convergence speed of the algorithm is increased, and the path optimization efficiency is remarkably improved. In conclusion, more efficient and more reliable planning of the unmanned aerial vehicle area coverage flight path can be realized.

The invention discloses a radar interference resource allocation method based on a grey wolf algorithm. The method mainly solves the problems that in the prior art, interference resource allocation efficiency is low, combination explosion occurs, and engineering implementation is not facilitated. The scheme comprises the following steps: 1) constructing an interference task scene, and constraining a solution space through a coding strategy optimization method; 2) constructing an interference resource allocation objective function, and initializing an allocation scheme according to a good point set theory; 3) on the basis of the grey wolf algorithm, a nonlinear convergence factor, a dynamic weight and a reverse learning method are introduced to complete updating of a grey wolf position, namely updating of a distribution scheme is completed; and 4) repeating the iterative updating process in the step 3) until the algorithm converges, and taking the wolf pack position corresponding to the currently obtained interference benefit as a final interference resource allocation result. The method can effectively improve the optimization speed in the resource allocation process, avoids falling into a local optimal solution, and can be used for improving the resource allocation task efficiency in a radar cooperative interference scene.

The invention discloses an improved moss growth algorithm-based assembly process parameter optimization method. The method comprises the following steps of (1) selecting a plurality of assembly precision indexes; (2) analyzing and evaluating the influence degree of each influence factor on the assembly precision index, selecting the influence factor with relatively large influence, inputting the influence factor into the improved support vector machine precision prediction model, and outputting an assembly precision prediction value; formulating a process optimization objective function by using the assembly precision predicted value; (3) selecting assembly process parameters needing to be optimized, formulating assembly process parameter optimization constraint conditions, and constructing a process parameter optimization model in combination with the process optimization objective function; (4) solving optimal assembly process parameters by the process parameter optimization model by using an improved moss growth algorithm; and (5) outputting the optimal assembly process parameters to form an assembly process adjustment scheme. According to the method, the convergence speed and optimization capacity of an existing algorithm are improved, the assembly precision and quality are improved, and the repair cost caused by trial and error and correction is reduced.

The invention provides a whole vehicle manufacturing coating resource scheduling method based on multi-agent deep reinforcement learning, and belongs to the technical field of artificial intelligence and intelligent manufacturing. The method is characterized by comprising the following steps: firstly, deeply analyzing whole vehicle manufacturing coating production characteristics and dynamic resource attributes in a cloud environment, and constructing a three-stage coating resource scheduling model considering manufacturing energy consumption and completion time; on the basis, through a multi-agent dynamic interaction mechanism in the cloud platform, a multi-agent near-end strategy optimization (MAPPO) algorithm is adopted to solve the whole vehicle manufacturing and coating resource scheduling problem in three stages; meanwhile, in order to enhance the interpretability of a scheduling strategy and ensure the convergence of the algorithm, KAN (Kolmogorov-Arnold Networks) is adopted as a neural network structure of an intelligent agent. The method is widely applied to finished vehicle manufacturing and coating production enterprises, the provided model and method can obtain a scheduling scheme meeting the requirements of energy consumption and completion time, and dynamic events such as resource maintenance and vehicle body return can be efficiently and autonomously processed.

The invention relates to an integrated energy system optimization scheduling method based on knowledge-guided reinforcement learning, and is suitable for the field of operation regulation and control of an integrated energy system. The method comprises the following steps: setting a state variable, an action variable and a reward function of reinforcement learning; acquiring a system operation state vector, and determining a control action vector corresponding to the state vector based on the strategy network; performing border crossing judgment on the control action vector; if the control action vector is judged to be border crossing, the control action influencing the operation cost is corrected, and a corrected control action vector meeting border crossing judgment conditions is obtained; based on the corrected control action vector, obtaining the reward result of the current round and the running state vector of the next round; storing a tetrad consisting of the running state of the current round, the control action, the reward result and the running state vector of the next round into an experience playback pool; and updating the policy network, the value network and the old policy network based on the data in the experience playback pool, and repeating the steps until the algorithm converges.

The invention relates to an adaptive Layered Threshold Min-Sum (ALTMS) decoding method for an LDPC (Low Density Parity Check Code) code, which is characterized in that the adaptive Layered Threshold Min-Sum (Amplitude Layered Threshold Min-Sum) decoding method is provided; according to the method, the advantages of a threshold normalized minimum sum decoding algorithm and a threshold offset minimum sum decoding algorithm are combined, correction factors under different iterations are dynamically adjusted, and check node information updating in each iteration is adapted, so that the algorithm convergence speed and the decoding performance are effectively improved. Simulation results show that compared with three threshold minimum sum decoding algorithms under hierarchical scheduling, the ALTMS decoding algorithm provided by the invention realizes certain performance gain, and compared with a fixed parameter minimum sum decoding algorithm, the ALTMS decoding algorithm provided by the invention has better error correction performance.

The invention discloses an underwater multi-target routing method based on an improved non-dominated sorting genetic algorithm. The method comprises the following steps: firstly, arranging sensor nodes in a target water area, networking, and encoding a feasible path from a source node to a target node by adopting a segmented structure; secondly, constructing a routing multi-target fitness function, and introducing a dynamic weighting function to form a comprehensive fitness index; and finally, grading the population by using an improved non-dominated sorting genetic algorithm, and exploring a better path through crossover and mutation operations. And designing a reference point elite selection strategy, improving the coverage of the solution set in the target space, and outputting an optimal routing path until the algorithm converges. According to the method, the dynamic balance among different optimization targets can be realized, and the link quality is considered while the energy consumption and the time delay are reduced, so that the network routing performance is improved.

The present invention relates to the technical field of wireless communications, and in particular to an intelligent reflecting surface-assisted group selection beamforming method. First, a raccoon position is initialized, and an RIS phase shift matrix is selected. Then, users are grouped on the basis of spatial correlation, and a beamforming matrix and an overall system rate are computed. The RIS phase shift matrix is updated on the basis of an obtained result by means of a raccoon optimization algorithm. The proposed algorithm is iterated multiple times to obtain the maximum overall system rate and a corresponding beamforming matrix. The intelligent reflecting surface-assisted group selection beamforming method provided by the present invention achieves joint optimization of the RIS phase shift matrix and the beamforming matrix by introducing the raccoon optimization algorithm, and solves the problems of slow convergence and susceptibility to local optima of other algorithms. By grouping the users when computing the beamforming matrix, the computational complexity of the system is reduced while inter-user interference is reduced, thereby improving the overall system rate.

The invention relates to the technical field of industrial engineering. The tourism highway facility ant colony intelligent layout method and system based on multi-source data collaboration are provided, and the method comprises the steps that collaborative fusion processing is carried out on multi-source heterogeneous data, and a dynamic demand field model is generated; performing topology constraint network construction processing based on the facility coordination rule to generate a topology constraint network; carrying out facility layout candidate solution set generation processing under the dynamic demand field model and the topology constraint network through an improved ant colony algorithm, and generating a facility layout candidate solution set; and carrying out multi-objective optimization and man-machine collaborative decision processing on the facility layout candidate solution set, and outputting a facility layout scheme so as to achieve the technical effects of improving the collaborative linkage capability among facilities, shortening the algorithm convergence time, reducing the local optimal trap occurrence rate and enhancing the multi-objective collaborative optimization efficiency.

The invention provides a vehicle-mounted radar antenna array design method and device, equipment and a storage medium, and relates to the technical field of antenna array design. The method comprises the steps that an initial arrangement matrix is obtained, a first optimization process is executed, the process comprises multiple target processes, and the target processes comprise the steps that based on the initial arrangement matrix, an improved genetic algorithm is adopted for optimization iteration, and local optimization arrangement is obtained; the improved genetic algorithm adopts a dynamic optimal solution retaining strategy; combining the local optimization arrays obtained in each target process to obtain an elite population matrix; aiming at the elite population matrix, executing a second optimization process, and performing iterative optimization in the process by adopting an improved genetic algorithm to obtain a plurality of candidate optimization arrays; based on a fitness function evaluation system, an optimal array in a plurality of candidate optimization arrays is determined as a design result of the antenna array, and through the dual iteration cooperation mechanism, the contradiction between the algorithm convergence speed and the optimization quality is effectively solved.

The invention discloses a task allocation method and system based on an improved whale optimization algorithm framework, relates to the technical field, and is used for optimizing order type adaptive cross-domain traffic control network task allocation and improving the key task response capability of a time-sensitive traffic system. ICWOA initializes a population through Chebyshev mapping, and introduces Levy flight disturbance to enhance the optimization ability; dynamically balancing local and global search by means of adaptive parameters; relieving population diversity attenuation through randomness retention, diversity maintenance and boundary constraint; dimension type pinhole imaging reverse learning is fused to reduce high-dimensional optimization dimension interference. According to the algorithm, the convergence speed and the solving precision are better, sub-second calculation time is kept under different task scales, the task distribution efficiency is improved, and the time-sensitive scene task success rate is remarkably improved. The method solves the problems that an existing algorithm is insufficient in adaptive order type'order application-order sending 'structure and prone to falling into local optimum, population diversity attenuation and calculation speed.

The application relates to a dynamic particle swarm optimization-based X structure Steiner minimum tree construction method. The method mainly comprises the following three effective strategies: (1) a dynamic subgroup and information exchange strategy enables subgroups to exchange information with other subgroups while maintaining independence, and increases subgroup diversity; (2) an improved particle learning strategy can combine the advantages of local topological structure in particle diversity and optimization precision with the advantages of global topological structure in algorithm convergence speed; and (3) a transition from multi-group local learning to single-group global learning strategy enables particles to obtain a better line length optimization rate. The application takes optimization of line length as the target, and finally optimizes the important target of line length.

The invention provides an underground cable position detection method and system based on a ground penetrating radar, relates to the technical field of ground penetrating radar imaging, and provides a fast search method of an optimal migration speed combined with F-FK, which does not need to traverse all speed values in a scanning interval, and only needs to determine a small number of key search points based on a specific sequence. The search interval is dynamically reduced by comparing the information entropy values of the search points, the number of times of calculation of the imaging entropy values is greatly reduced, and the cable data processing efficiency is greatly improved; secondly, the algorithm does not need a fixed scanning interval, the search point determined through the golden section proportion can adaptively focus the optimal speed area, even if the cable penetrates through the concrete-wet soil mixed medium section, the speed sudden change point can be accurately captured, the problem that the real speed is missed at the fixed interval is avoided, and finally, the algorithm is high in convergence speed and high in accuracy. Compared with a traditional method, the processing time of GPR data of a long measuring line and a high-density sampled cable is greatly shortened, and the method can be applied to emergency troubleshooting.

The invention discloses a federated learning method and system with a communication efficient mechanism in a heterogeneous environment, and the method comprises the steps: carrying out each round of training by a client through employing a gradient tracking algorithm with a momentum mechanism, obtaining the variable quantity of a trained local model matrix parameter, compressing the variable quantity through employing an error feedback mechanism, and uploading the compressed variable quantity to a central server; the central server calculates the average value of the compressed variable quantity uploaded by the participating clients and sends the average value to all the clients, and the clients and the central server calculate a new round of global model based on the average value; and repeating until the global model converges, and completing the training of federal learning. According to the invention, the calculation amount of the client in federal learning can be reduced, and the communication traffic between the client and the central server is reduced, so that the algorithm convergence time is reduced, and the federal learning efficiency is improved.

A gas temperature field reconstruction method using thermocouple measurement correction is disclosed. First, an Inventor 3D model and an ANSYS temperature field model are established for the gas temperature field. Then, a mapping relationship is established between the control parameters of the gas temperature field and the boundary conditions of the ANSYS temperature field model. Based on the principle of maximizing the influence of fuzzy boundary conditions on the temperature field reconstruction results, experiments are designed for precise boundary conditions and experimental data are obtained. Next, given the boundary conditions of the ANSYS temperature field model, the model is run to obtain the temperature field reconstruction results. A model quality assessment algorithm is run to calculate the fitting degree between the measured curve and the simulation curve, obtaining the algorithm fitting parameters. The algorithm fitting parameters are selected to form the loss function of the gradient descent algorithm. The learning rate, termination condition, and initial iteration parameters are determined. The gradient descent algorithm is run until convergence. Finally, the final iteration parameters are taken as the correction result of the fuzzy boundary conditions of the ANSYS temperature field model. This invention improves the spatiotemporal resolution of the temperature field reconstruction results.

The invention discloses a pose estimation method based on obb large and small frames, and the method comprises the steps: constructing a global-local two-stage OBB detection architecture, respectively generating a global bounding box and a local bounding box for the overall contour of an object RGB image and a directivity feature sub-component region, carrying out the standardization processing, marking parameters according to a preset condition, building a fixed space constraint, and carrying out the positioning of the object RGB image. The problem of direction estimation deviation caused by traditional OBB parameter chaos is solved; the global detection result and the local detection result are fused, the main direction of the local bounding box serves as a reference, the rotation angle of the global bounding box is corrected through vector operation, and 180-degree rotation ambiguity is eliminated; the corrected rotation angle is fused to an instance segmentation result, and a Mask fine contour and an OBB accurate direction are combined; and finally, a depth value is extracted from the depth map to generate a point cloud, an ICP algorithm is called for registration after processing, a local optimal trap is avoided, a precise pose under a world coordinate system is output, and the final estimation precision of the 6D pose and the algorithm convergence efficiency are remarkably improved.

The invention relates to the field of intelligent deployment of farmland equipment, in particular to an intelligent deployment system of intelligent farmland Internet of Things equipment, and the system comprises a farmland environment and crop coupling module; the mapping module is used for establishing mapping relations between crops and environments; a monitoring module; the area and equipment preprocessing module is used for quantitatively evaluating the sensitivity degree of each area of the farmland to environment change; the searching module is used for introducing an optimization searching mechanism based on an RANSAC algorithm and searching an optimal equipment layout scheme; a deployment module; and converting the optimal equipment layout scheme into an executable equipment deployment task column. According to the method, sampling and iteration logic of the RANSAC algorithm is optimized, accurate quantification of perception coverage is performed by combining an equipment vector field, the convergence speed of the algorithm is increased, and it is ensured that the output layout scheme reaches the optimal balance between the coverage effect and resource configuration.

This invention relates to a 3D UAV path planning method based on an improved gray wolf optimization algorithm, comprising: acquiring a 3D environment model; setting the start and end positions of the UAV based on the 3D environment model; employing the improved gray wolf optimization algorithm, using the objective function as the fitness function; updating the positions of all gray wolves synchronously during the iterative process; and outputting the optimal path. This invention introduces a circle chaotic mapping to the gray wolf optimization algorithm to increase population diversity; introduces an adaptive exploration factor to increase the balance between global and local search; combines the global and local development formulas of the dung beetle optimization algorithm to compensate for the slow convergence speed of the traditional gray wolf optimization algorithm; subsequently introduces a sine and cosine strategy to help the algorithm search for potential optimal solution regions, thus better exploring the global optimal solution; and finally introduces a pooling mechanism to randomly recombine stored historical poor solutions to generate new solutions, accelerating the convergence speed and improving the robustness of the algorithm.

The invention belongs to the technical field of wireless communication, and particularly relates to a post-disaster wireless communication network resource allocation method. The method comprises the following steps: arranging a corresponding strategy network for each agent, training each strategy network by adopting a multi-agent reinforcement learning model comprising each strategy network to obtain an optimal network parameter of each strategy network, and deploying each strategy network configured with the optimal network parameter to the corresponding agent. Therefore, resource allocation in the network is guided; according to the method, algorithm convergence is accelerated, single-step decision delay is reduced, the network throughput is finally improved, and real-time and reliable service quality guarantee is provided for users.

The invention discloses a comprehensive energy scheduling method and system based on genetic variation optimization, and the method comprises the steps: firstly issuing a price strategy through an energy manager based on a Stackelberg game theory, and then responding to a price signal and optimizing the own energy consumption behavior or equipment scheduling through a user aggregator, an energy supplier and an energy storage operator, therefore, a double-layer game feedback cycle taking the price as the core is formed. According to the method, a price initialization strategy inspired by improved load characteristics is designed to perform population initialization, and a more representative initial solution set is constructed, so that the convergence efficiency of the algorithm is expected to be improved from the source; meanwhile, a self-adaptive variation mechanism based on the load change rate is introduced, variation intensity is dynamically adjusted to balance exploration and utilization, and therefore the solving quality and the convergence efficiency of the algorithm are effectively improved, and the method adapts to the complex dynamic characteristics of the multi-main-body energy system.

The invention relates to a production plan Pareto optimization method considering carbon emission, punishment and balance, and belongs to the technical field of computational intelligence and production plan optimization. The method comprises the following steps: constructing a mathematical model of a heterogeneous parallel machine production plan problem considering carbon emission, rolling punishment and load balancing; and a non-dominated sorting hybrid algorithm based on Gaussian learning is proposed to optimize three conflicting targets, i.e., minimization of carbon emission, minimization of task rolling penalty and minimization of load imbalance, in the model. According to the method, a non-dominated sorting hybrid algorithm based on Gaussian learning is extracted and designed from an actual production scene, meanwhile, a Gaussian mixture model training strategy based on a disturbance solution set is designed to be combined with traditional genetic manipulation to generate new offspring individuals, and improvement of algorithm convergence and exploration ability is facilitated. The invention provides a group of high-quality non-dominated solutions for balancing carbon emission, rolling punishment and load balancing, and provides important management enlightenment for making an actual production plan of an enterprise.

The invention relates to the technical field of cloud computing and distributed computing scheduling, in particular to a server-free MapReduce multi-target scheduling optimization method. Comprising the following steps: constructing a directed acyclic graph, coding a resource allocation decision into a solution from a source node to a target node, generating an initial population by adopting a hybrid initialization strategy, evaluating fitness, calculating a congestion distance of a solution in each non-dominated leading edge, and sequentially executing tournament selection, a self-adaptive crossover strategy and a topological repair variation strategy. And periodically executing a solution injection strategy guided by heuristic search, carrying out crowding distance selection, updating the global Pareto optimal solution set, outputting the current global Pareto optimal solution set when the time is up, and otherwise, returning to iteration. The method has the positive effects that the execution cost and the execution time of the MapReduce operation are balanced, the global Pareto optimal solution set quality of the multi-target scheduling scheme is improved, and the convergence speed and the robustness of the algorithm are enhanced.

The technical scheme of the invention discloses a special accelerator for real-time optical flow estimation based on multi-scale block division PatchMatch. The invention provides a high-speed and high-energy-efficiency optical flow estimation accelerator through collaborative optimization of an algorithm and hardware. In the aspect of algorithms, the invention provides a low-complexity optical flow estimation algorithm, algorithm process design is carried out for optical flow estimation application, algorithm convergence is accelerated and accuracy is improved under the condition that the tag estimation number of each pixel is not remarkably increased, and the problems that the optical flow estimation algorithm is large in calculation difficulty and difficult to converge are solved. In the aspect of hardware design, a special accelerator is designed based on the algorithm, architecture optimization methods such as parallel image partitioning, motion vector assistance and optical flow accelerator special access design are provided, and the optimization methods greatly improve the operation efficiency of the accelerator.

The invention discloses an improved self-adaptive single-loop method based on a hybrid dynamic mean value conjugation algorithm and application, according to a Karush-Kuhn-Tucker (KKT) optimality condition, a normalized sensitivity vector is calculated based on the hybrid dynamic mean value conjugation algorithm, and then a minimum function target point is deduced. The invention discloses an improved self-adaptive single-loop method, and relates to the field of reliability design optimization. The improved self-adaptive single-loop method comprises the following steps that basic parameters are set and initialized; deriving a normalized steepest descent direction; calculating a static conjugate scalar factor; calculating a dynamic factor value; calculating a dynamic conjugate gradient vector; deriving a normalized sensitivity vector; calculating random variables and random parameters; and solving the equivalent deterministic optimization model until the algorithm converges. Compared with a mainstream double-loop method, a single-loop method and a decoupling method, the improved self-adaptive single-loop method disclosed by the invention has the advantages of high efficiency, high accuracy and high convergence.