44results about How to "Avoid getting stuck in a local optimum" patented technology

The invention discloses a micro-grid small signal stability analyzing and parameter coordinated setting method, and belongs to the technical field of power system micro-grid operation and control. The micro-grid small signal stability analyzing and parameter coordinated setting method comprises the steps of establishing micro-grid mathematical models which comprise a network and load small signal model and an inverter small signal model and are in need of parameterized design, determining each parameter of the micro-grid according to the root-locus method and using the parameters as initial values of the particle swarm algorithm, conducting sensitivity analyzing through a characteristic value, determining a leading parameter, determining the value range of all the parameters or the mathematical relationships among all the parameters, solving the constraint conditions in the process by the utilization of the particle swarm algorithm, determining the initial values, the parameters in need of optimization and the constraint conditions, and conducting parameter optimization and coordinated setting by the utilization of the particle swarm algorithm. By means of the micro-grid small signal stability analyzing and parameter coordinated setting method, determining of each parameter can be realized quickly and accurately in a coordinated mode, the complexity of setting the parameters one by one is avoided, and the system is made to be more stable on the basis of small signal stability; target functions and boundary conditions are flexibly adjusted, the working time is greatly shortened and the efficiency is improved.

The invention relates to a microgrid multi-energy economic method based on a positive and negative feedback particle swarm optimization algorithm. The method comprises: step one, an optimization objective function for optimizing the system generator output and minimum power generation cost is established under the circumstance of satisfying the system operation constraint condition; step two, setting a constraint condition of the optimization objective function from the step one; and step three, carrying out optimization calculation by using a positive and negative feedback particle swarm algorithm, setting various parameters of the algorithm, and then starting iterative calculation to obtain an optimal solution of the optimization objective function from the step one. According to the invention, with a dynamic double-population particle swarm structure and linear decreasing inertia weight calculation, multi-energy scheduling of the micro grid is optimized effectively.

The invention discloses a complex system designing method based on the resampling particle swarm optimization algorithm. The method comprises the following steps: (I) establishing a fitness function of specific optimization problems; (II) initializing the particle swarm; (III) performing resampling operation of the particle swarm; (IV) updating the particle position and speed; (V) updating the historical optimal position and swarm optimal position of each particle; and (VI) if the precision requirement is not met and the maximum number of iterations is not reached, adding one to the number of iterations and returning to the step (III); and otherwise, recording and outputting the result. In the invention, practical problems in the optimal design of a complex system can be effectively dealt with; and by introducing the resampling technology, the convergence speed is increased, the efficiency is improved, the global search ability is enhanced, the optimization precision is improved, and an aim of efficiently and accurately solving the problems in the optimal design of the complex system is finally achieved.

The invention provides a crowd evacuation simulation system based on a composite potential energy field. An improved potential energy field model is adopted for the system, a traditional potential energy field under Dirichlet boundary conditions is linearly combined with a potential energy field under Neumann boundary conditions, a local potential energy field for solving collision prevention problems between people is added into the combined potential energy field, and thus the composite potential energy field is obtained; by combining an update strategy of pedestrians and a pedestrian speed control method, the simulation system of personnel evacuation in emergency on different scales of scenes can be established according to the layout of the actual scene. In the crowd simulation process, influences of path plans of moving individuals on the evacuation are fully considered, and the potential energy field method can play a certain role in removing the influence factor. The system has extensive application prospects in research on simulation of safe and fast evacuation of a large number of personnel on difference scales of scenes, the design defects on the scenes can be found, and the crowd evacuation simulation system can assist in making execution schemes in emergency and is economically feasible.

The invention discloses a wheel hub motor optimization method based on the genetic annealing algorithm. The method includes the following technical steps that the target and constraint conditions of wheel hub motor optimization; the genetic annealing algorithm is adopted for inverse solution and solution of a wheel hub motor, and hub motor design variable parameter values are obtained; according to the obtained variable parameter values, the sustainedly stable torque of the wheel hub motor is generated. The genetic annealing algorithm is improved on the basis of the genetic algorithm, the genetic algorithm and the simulated annealing algorithm are organically combined together, the efficiency of the algorithm can be improved, the global control capability of the algorithm can be enhanced, the precision is high, the convergence speed is high and the efficiency is high. It is feasible to apply the method to the optimization design of the wheel hub motor of an electric wheel, and the method has wide engineering application value.

The invention discloses an image classification method based on an RGB-D fusion feature and sparse coding. The method comprises following steps of (1) extracting dense SIFT features and PHOG features of a color image and a depth image; (2) carrying out feature fusion on the extracted features of the images by use of a linear serial connection form so as to obtain four kinds of different fusion features finally; (3) using the K-means++ clustering method to carry out clustering processing on the different fusion features so as to obtain four kinds of different vision dictionaries; (4) carrying out local restriction linear coding on each vision dictionary to obtain different image expressing sets; and (5) using the linear SVM to classify the different image expressing sets and using a vote decision method to decide final classification conditions of the obtained classification results. According to the invention, the method is high in classification precision.

The invention provides a substation dynamic reactive power optimization method and a system based on an improved bat algorithm. A substation dynamic reactive power optimization mathematical model is built, the substation dynamic reactive power optimization mathematical model is solved by employing the improved bat algorithm, and the optical action strategy of a substation tap and a capacitor set is obtained. In the whole process, the bat algorithm is applied to the field of dynamic reactive power optimization, the bat algorithm is improved by employing a heuristic strategy aiming to security constraint and control equipment action constraint, the condition of local optimum of the algorithm is effectively avoided, optimal control of reactive power of the substation voltage can be realized, and normal and high-efficiency operation of the substation is guaranteed.

A method for optimizing public traffic network features that the simulated annealing algorithm is used as a frame, and the whole-day total operation cost of operation company is minimized as an objective to obtain initial line network under the frame. The initial line network is scattered to form the line network unit, which is used as the input network, and the genetic algorithm is embedded to optimize it. The public transportation line network optimization model is constructed to minimize the total travel time of all travelers, and the simplified new line network is formed. The change of operation cost is compared to determine whether the convergence condition is reached. The invention combines the simulated annealing algorithm with the genetic algorithm, which ensures the global searching ability of the optimization process and avoids the algorithm from falling into the local optimal solution, thereby improving the solution quality. At the same time, the design concept of 'element'is proposed to promote the combination of multi-objective optimization process, and the convergence condition of sub-heuristic algorithm is improved by two-temperature cooperative control iteration, thus overcoming the common shortcomings of sub-heuristic algorithm that the convergence condition is difficult to define.

The invention relates to a vehicle routing optimization method with a fuzzy random time window. In order to achieve vehicle routing optimization with the fuzzy random time window in the engineering transportation, a bilevel vehicle routing optimization method with the fuzzy random time window is provided. The method comprises the following steps of establishing a lower level model of a bilevel vehicle routing optimization problem model with the time window, and establishing a corresponding upper layer model according to the lower level model; obtaining a vehicle routing optimization problem overall model with the fuzzy random time window through the bilevel planning technology according to the upper layer model and the lower layer model, wherein the overall model is formed by organically combining the upper layer model and the lower layer model; solving the overall model by utilizing the particle swarm algorithm. The improved particle swarm algorithm technology is applied to the bilevel vehicle routing optimization method with the fuzzy random time window, and an optimal solution of a bilevel vehicle routing with the fuzzy random time window can be quickly and effectively obtained. The bilevel vehicle routing optimization method with the fuzzy random time window is applied to the field of engineering management.

The invention discloses a spark cluster optimal configuration parameter determination method, device and apparatus, and a computer readable storage medium. The method comprises the steps of creating spark cluster test environments configured by different hosts by utilizing a cloud platform; obtaining an optimal configuration parameter sample corresponding to each spark cluster test environment through a preset sampling method and a machine learning algorithm, and combining the optimal configuration parameter samples into a sample set; training the sample set to obtain an optimization model; and inputting the configuration data of a target spark cluster into a tuning model to obtain an optimal configuration parameter corresponding to the target spark cluster. Through the disclosed technicalscheme, an optimization model is obtained through spark cluster test environment creation, a sample set composed of optimal configuration parameter samples acquired and the sample set is trained, andoptimal configuration parameters corresponding to a target spark cluster are obtained through the optimization model, so that the efficiency and accuracy of optimal configuration parameter determination are improved.

The invention discloses a dynamic frequency estimation and measurement method after disturbance of a power system. The method mainly comprises the following steps: A, training: A1, carrying out dynamic time domain simulation to obtain 10 initial input vectors; a2, calculating 11 secondary input vectors; a3, splicing the vectors in the steps A1 and A2 into an input eigenvalue vector; A4, carrying out dynamic time domain simulation to obtain a dynamic frequency; a5, completing the training of the prediction model according to the data in the steps A3 and A4; b, testing: B1, monitoring 10 initialtest input vectors on line; b2, if the zero element exists in the electromagnetic power vector of the generator, judging that disturbance occurs, and turning to the step B3; b3, calculating 11 secondary test input vectors; b4, connecting the vectors in the steps B1 and B3 in series to form a test input characteristic value vector; and B5, inputting the test input eigenvalue vector into a prediction model, and outputting the disturbed estimated dynamic frequency by the model. The method is rapid in estimation and measurement, high in precision and small in error.

The invention discloses a multi-task neural network architecture searching method based on evolutionary computation, which comprises the following steps: firstly, initializing a population; evaluating the multi-task generalization abilities of individuals in the population; then randomly obtaining two chromosomes through a binary tournament selection algorithm; comparing the multi-task generalization performance of the two chromosomes; selecting the chromosome with better performance as a parent; then carrying out crossover and mutation operations on two parents to generating two children; evaluating the multi-task generalization performance of the children; then combining the children and the parents; carrying out environment selection according to an evaluation result; generating a new population; carrying out a new round of evolution until a predetermined termination condition is reached; and outputting the individual with the best multi-task generalization ability. According to the method, a genetic algorithm is used for optimizing the multi-task network model system structure, the neural network model suitable for multi-task learning can be automatically searched out without manual participation, and the cross-task information fusion capability of the multi-task network is improved.

The invention relates to an ant colony algorithm applied to robot path planning. Aiming at the problems of slow convergence speed caused by early-state pheromone insufficiency, weight parameters Alpha (information inspiring factor) and Beta (expected inspiring factor) are improved by the ant colony algorithm, and the two parameters are dynamically adjusted; and moreover, a local optimal direction guidance mechanism is added to built new path selection probability, and the convergence speed of the ant colony algorithm is faster.

The invention provides a method for constructing a prediction model based on a colibacillus algorithm. The method comprises the following steps: acquiring sample data and normalizing the acquired sample data; optimizing a penalty factor C and a kernel width gamma of a support vector machine by utilizing a colibacillus-based algorithm; and on the basis of the obtained penalty factor C and kernel width gamma, constructing a prediction model by utilizing the normalized data, and classifying and predicting a to-be-classified sample on the basis of the constructed prediction model. By implementingthe method, the penalty factor and the kernel width of the SVM are optimized on the basis of the colibacillus algorithm, the convergence rate and the convergence precision of the algorithm and the capacity of the algorithm for escaping from the local optimal solution can be fully utilized, and a better global approximate optimal solution is found to obtain the SVM model with higher classificationprecision.

The invention discloses an energy storage scheduling method based on new energy consumption and a computer medium, and the method comprises the steps: extracting the operation data of a power grid, and building a power grid energy storage scheduling model based on the operation data of the power grid by taking the maximum combined output of wind and light power generation and the minimum operation cost of a thermal power generating unit as a target function; setting constraint conditions of the power grid energy storage scheduling model, wherein the constraint conditions comprise an energy storage device constraint condition, a power grid stable operation constraint condition and a generator set constraint condition; using a multi-objective genetic algorithm to set weights for the objective functions, converting the objective functions into single objective functions, and solving the single objective functions to obtain an optimal charging and discharging curve of the energy storage device; and determining a scheduling scheme of the energy storage device according to the optimal charging and discharging curve. According to the method, the multi-objective genetic algorithm based on the Pareto front is adopted, and the weight is introduced to convert the multi-objective function into the single-objective function, so that the situation of falling into a local optimal solution in the optimization process is effectively avoided, and the calculation speed is improved.

The SVM-based wireless terminal security access method is characterized by comprising the following steps: 1, establishing a wireless network security problem model based on a communication scene of awireless terminal and a wireless base station, and establishing a channel feature vector based on the wireless network security problem model and a frame message received by the wireless base station; 2, checking the channel feature vector of the frame message by using a pre-trained SVM authentication model, and judging whether the frame message is a legal signal or not; and 3, determining that the wireless terminal sending the at least one frame message is a legal terminal based on judgment on whether the at least one frame message is a legal signal. Based on the method provided by the invention, the frame message sent by the wireless terminal accessed to the wireless base station can be checked by improving the SVM parameter combination optimization method of differential evolution andthe differential evolution method of control parameter self-adaptive improvement, so that the legal terminal can be effectively judged.

According to the invention, a routing protocol is fully understood and researched, the invention provides a method for improving routing transmission quality of a wireless distributed sensor network.The invention provides a routing transmission quality evaluation summary function which is used for calculating the adaptive value of each path and evaluating the quality of the selected path throughthe adaptive value of the path, the design of an improved routing transmission algorithm is recorded in detail, and the implementation method of the improved routing transmission algorithm is recordedin the algorithm implementation process. According to the routing transmission algorithm, the energy factor of each sensor node is introduced, energy balance of sensor nodes in the whole network canbe kept, premature loss of some key nodes due to energy exhaustion is avoided, the working time of the whole network is prolonged, various route transmission quality parameters are comprehensively considered, an improved bird flock foraging algorithm is applied to the process of finding the optimal route by the route, and the method has good feasibility and reliability.

The invention discloses a multi-microgrid coordinated optimization scheduling method considering load characteristics and demand response. The method comprises steps: firstly, establishing different load electricity price characteristic models and demand side response cost function models for different types of micro-grids; and then solving the optimization model by adopting an improved artificialfish algorithm by taking the minimum total cost of multi-microgrid coordinated optimization scheduling as a target; setting a satisfaction threshold of the demand side response by considering the irrational behaviors of the users, participating in the demand side response according to the threshold when the satisfaction of the microgrid users is lower than the threshold, and performing iterativesolution again after the problem is optimized after updating until the number of iterations is met or the satisfaction of each microgrid user is higher than the threshold; otherwise, each microgrid user takes the satisfaction value as a threshold value to participate in demand side response. Therefore, the multi-microgrid coordinated scheduling result obtained by the method has more practical significance.

The invention discloses a base station clustering method based on density and minimum distance in an ultra-dense network, comprising the following steps: firstly, calculating the distribution densityand clustering density threshold value of each micro-cell base station in the ultra-dense network, so that the micro-cell base stations of which the distribution density is greater than the clusteringdensity threshold value form an initial clustering center pool; calculating the minimum value of the distance between each micro-cell base station in the initial cluster center pool and the micro-cell base station with the distribution density higher than that of the micro-cell base station, defining the product of the distribution density of the micro-cell base stations and the minimum distanceas the weighted distribution density, and obtaining a to-be-selected cluster center pool according to the weighted distribution density; calculating a cluster center isolation distance, and sequentially removing the cluster center with a smaller weighted distribution density value in two cluster centers of which the distance between every two cluster centers is greater than the cluster center isolation distance in the to-be-selected cluster center pool from the to-be-selected cluster center pool; and finally, taking the number of cluster centers in the to-be-selected cluster center pool and the geographic position of the cluster center base station as parameters of traditional K-means algorithm, and executing K-means algorithm to obtain a clustering result. According to the method, the problem of non-uniform clustering is solved.

The invention discloses a target detection method and device based on a multi-gating hybrid expert model. The method comprises the steps of obtaining a target feature map and a potential target frame of an area where a potential target is located in an image; processing the target feature map by using an expert model, and outputting a target classification subtask result corresponding to the target feature map and a frame regression parameter determination subtask result; processing the target feature map by using a gating network, and outputting an adaptive weight value of each expert model corresponding to the target classification sub-task and an adaptive weight value of each expert model corresponding to the frame regression parameter determination sub-task; and according to the adaptive weight value, the target classification subtask result and the frame regression parameter subtask result, determining the category and the frame of the target through full-connection neural network processing. And target classification and regression learning are performed through the multi-gating hybrid expert model, so that the efficiency of classification and regression task joint learning is improved, and the accuracy of target detection is improved.

The invention relates to a design method of a small thrust phase modulation maneuver on an elliptic orbit, and belongs to the technical field of spacecraft orbital maneuvers. The method comprises the steps of proposing two feasible phase modulation policies based on a rule of an influence of the thrust direction on a phase change of a spacecraft, simplifying the complex phase modulation maneuver to an optimization problem only containing three parameters by using techniques such as constraint relaxation and orbit equalization, and conducting simplification processing on a design model according to the characteristic of a small-thrust elliptic orbit phase modulation task by adopting the orbit equalization technique and neglecting the secondary constraint, so as to efficiently and rapidly obtain the initial value of an elliptic orbit phase modulation maneuver parameter. The method has the advantages that the method is simple in algorithm, high in robustness and high in computational efficiency. The method is applicable to the initial design of the small thrust phase modulation maneuver on the same elliptic orbit, and the initial design of the phase modulation maneuver between different elliptic orbits with smaller orbit deviation.

The invention discloses an airship propeller reliability estimation method based on a chaos initialization SSA-BP neural network. The method comprises the following steps: determining main factors influencing blade strain of a propeller under a design parking working condition; constructing a training / testing input data set of the chaotic initialization SSA-BP neural network; solving the strain value of the maximum strain position of the propeller under the working condition of taking the input data set as the working condition; establishing a chaotic initialization SSA-BP neural network model; normal distribution discretization is carried out on the designed flight height and rotating speed of the propeller under the mission profile according to the 3 sigma principle, a new input data set is obtained, and normal distribution discretization is carried out on the allowable strain value of the propeller according to the variable coefficient of the allowable strain value according to the 3 sigma principle; and solving the failure rate of the airship propeller and the MTBF (mean time of failure). According to the method, the situation of falling into a local optimal solution is effectively avoided, the prediction precision and the prediction efficiency are improved, the reliability of the propeller can be quickly estimated, and the method has great engineering value.

The invention discloses a method for tracking a maximum power point of a solar street lamp based on cloud evolution. The method comprises the following steps: setting a set iteration number needing to carry out the cloud evolution and a cloud variation algebraic threshold in the cloud evolution on a controller; randomly generating a population of solar cells; obtaining a predetermined number of individuals in the population according to the power of the individuals from large to small, and carrying out cloud evolution calculation to obtain an updated population; carrying out cloud evolution update on the individuals in the updated population; when the evolution algebra of the cloud evolution update is less than the cloud variation algebraic threshold, updating the cloud evolution update individuals again by adopting the cloud evolution; when the evolution algebra of the cloud evolution update is less than the set iteration number and reaches or exceeds the cloud variation algebraic threshold, carrying out a cloud variation operation on the cloud evolution update individuals, and adopting the cloud evolution to update the obtained individuals of a variant population; and when the evolution algebra of the cloud evolution update reaches or exceeds the set iteration number, obtaining the maximum output power point of the solar cells. By adopting the method disclosed by the invention, the maximum power point of the solar street lamp can be accurately tracked.

The invention relates to a semi-supervised object pose estimation method combining generated data and unlabeled data. The method comprises the following steps: 1) generating point cloud data with a pose label, namely the generated data; 2) obtaining a color image and a depth image of a target object without labels, inputting the color image into the trained instance segmentation network to obtain an instance segmentation result, and obtaining a point cloud of the target object from the depth image according to the segmentation result, namely real data without labels; 3) in each training period, performing supervised training on the pose estimation network model by using the generated data, and performing self-supervised training on the pose estimation network model by using real data without labels; and 4) after each training period is finished, calculating the accuracy of the pose estimation network model by adopting part of real data. Compared with the prior art, the method mainly solves the problem that the 6D pose label is difficult to obtain, and accurate object pose estimation can be realized only by using synthetic data and unlabeled real data.

The invention discloses a transformer substation optimization site selection method based on gravity center regression and a particle swarm hybrid algorithm. The method is suitable for optimizing andplanning a site selection and volume determination scheme of a transformer substation. The method specifically comprises the steps of firstly determining the number n of substations needing to be newly built in a power grid, then dividing load nodes in the power grid into n load districts by adopting a gravity center regression algorithm, obtaining position coordinates and supplied loads of the substations for supplying power to the load districts, and initializing the positions of particles by taking the position coordinates and the supplied loads as initial values of a particle swarm algorithm; and then taking the minimum global load moment as a target fitness value, optimizing substation positions and supplied loads by adopting a particle swarm algorithm to obtain optimized n substationpositions and supplied loads, and finally solving the substation capacity of each substation according to a constraint relationship between the substation capacity of the substation and the suppliedloads. The calculation result of the design has good accuracy, stability and optimization effect.

The invention discloses a method for optimizing logistics distribution center site selection by applying an improved hybrid immune algorithm. The method comprises the following steps: (1) establishinga logistics center site selection model; (2) performing immune algorithm calculation; and (3) carrying out improved hybrid immune algorithm calculation. The method has the beneficial effects that aiming at the problem that a conventional immune algorithm is easy to fall into local optimum, simulated annealing is used in an immune link to realize dynamic threshold selection so as to modify a function expected value in real time, and meanwhile, random single-point crossover operation, high-frequency variation and other operations are adopted in immunization to ensure the diversity of antibodies. The improved algorithm improves antibody diversity so as to avoid falling into a local optimal value and accelerate convergence speed. Dynamic threshold selection, random single-point crossover operation, high-frequency variation and other operations are carried out by using function propagation expectation of a simulated annealing correction immune clone algorithm to improve population diversity so as to avoid falling into local optimum, time complexity is reduced, and convergence speed is also accelerated.

The invention discloses a step-by-step linear aggregation rainfall data scale conversion method. The method comprises the following steps: giving a station control range l; grading the ith link according to the length Li of the ith microwave link in the microwave network; discretizing the link according to the grading result, equally dividing the link according to the link grade, and taking the equally divided link center point of each segment as a virtual rainfall station; determining an initial station position and an estimated value; determining the preliminary estimation value of the unestimated virtual station with the lowest level; obtaining a link of the preliminary estimation value to carry out iterative optimization; and entering the next level of link calculation, and repeating until the calculation of all levels of links is completed. According to the concept of grading and step-by-step optimization provided by the invention, the problem that all links fall into a local optimal solution can be effectively avoided; the step-by-step optimization can effectively reduce the influence of large errors in the long-chain route aggregation data on the conversion process, and improves the precision of the conversion result while guaranteeing the convergence rate of the algorithm.

The present invention relates to a vehicle route optimization method with a fuzzy random time window. In order to solve the problem of vehicle route optimization with a fuzzy random time window in engineering transportation, the present invention provides a two-layer vehicle route optimization method with a fuzzy random time window. The steps are as follows : establish the lower layer model of the two-layer vehicle route optimization problem model with time window, and establish the corresponding upper layer model according to the lower layer model; obtain the vehicle with fuzzy random time window under the second layer programming technology according to the upper layer model and the lower layer model The overall model of the path optimization problem, the overall model is an organic combination of the upper model and the lower model; the particle swarm algorithm is used to solve the overall model. The invention applies the improved particle swarm algorithm technology to the optimization method for solving the vehicle route with fuzzy random time window on the second floor, and can quickly and effectively obtain the optimal solution of the vehicle route on the second floor with fuzzy random time window. The invention is applicable to the field of engineering management.