48results about How to "Strong global search ability" patented technology

The invention discloses a method for improving a BP (back propagation) neutral network and based on a genetic algorithm. The method includes coding the BP network to determine structure of the neutral network, wherein the structure includes the number of hidden layers and the number of units of each layer; adopting real-number coding to code by taking weight and threshold as genes, wherein each neutral network corresponds to a chromosome after coding; using the genetic algorithm to perform selection optimization on the network, wherein selection optimization includes the steps of selection, crossing and variation; training the BP network to acquire a final result; decoding an optimal individual selected by the genetic algorithm to generate a new neutral network, and training the new network by applying a BP training algorithm to acquire a final result. The method combines the genetic algorithm with the BP network, thereby being capable of fully utilizing advantages of the both, the problem that initial weight and threshold of the BP network are difficult to determine can be solved, searching range can be narrowed, training speed of the BP network can be increased, and the problem of local minimum can be improved.

The invention discloses a water unmanned ship local hierarchical path planning method based on the navigation error constraint. The method mainly includes the steps of local static path planning based on the navigation error constraint and combined with a genetic algorithm; local dynamic path planning fused with the maritime affair rule. The hierarchical thinking is adopted, local path planning is divided into two levels of static path planning and dynamic path planning, the static barrier avoiding problem is solved, and the dynamic barrier avoiding problem is also solved. According to static path planning based on the navigation error constraint and combined with the genetic algorithm, adverse influences of navigation errors on path selection are reduced, and the safety of path planning is improved; according to dynamic path planning based on the marine affair rule, the dynamics constraint of a water unmanned ship is considered, and the feasibility of path planning is improved.

The invention relates to a prediction method of transformer winding hot-spot temperature based on a neural network. After the neural network is optimized through the adoption of a genetic algorithm, the transformer winding hot-spot temperature value is obtained through the computation; the method comprises the following steps: (1) constructing the neural network; (2) normalizing input data and output data of the neural network; (3) optimizing a weight value and a threshold value of the neural network through the genetic algorithm; (4) training the neural network optimized through the genetic algorithm; (5) obtaining the output data through the utilization of real-time measured input data and trained neural network, and computing the hot-spot temperature value through computation. Compared with the prior art, the method disclosed by the invention has the advantage of being high in measurement precision.

A customized public transit network optimization method based on intelligent search comprises the following steps: using network terminals to obtain passenger travel demand data, and building a passenger demand database; building an evaluate index mathematics model satisfying customization public transit and constrained conditions; using N stations ranking in the front in terms of passenger travel population as basis, and building the start point of a backup route set of the customization public transit and initialization routes of N backup route sets; using the initialization routes of N backup route sets as the basis, combing with public transit GIS and station passenger demands, and building a backup route set; using a passenger nonstop rate as a fitness function on the basis of the backup route set, and using a heredity algorithm to search the customized public transit network satisfying the customization public transit evaluate index mathematics model. The method can optimize the passenger nonstop rate so as to build the customized public transit network, thus alleviating partial transport problems, and providing conveniences for passengers to travel.

The invention discloses a short-term impulse load forecasting model building method based on an signal decomposition and intelligent optimization algorithm, comprising the following steps: S1, according to the non-stationarity of load data, adopting complementary set empirical mode decomposition (CEEMD) to decompose the time series of the original load into several intrinsic mode functions (IMFs);complementary empirical mode decomposition (CEEMD) adding positive and negative white noise to the original time series, which not only guarantees the same decomposition effect as empirical mode decomposition (EEMD), but also reduces the reconstruction error caused by adding white noise. The invention adopts the decomposition technology to decompose the sequence into a plurality of modal components, optimizes the parameters of the prediction model combined with the optimization algorithm, and finally superimposes the prediction results of each component as the final prediction value. Comparedwith other models, the combined model can obtain higher prediction accuracy in the short-term impact load prediction.

The invention discloses a particle swarm optimization method for terrestrial magnetism auxiliary navigation track programming. The method comprises the following steps: 1) taking regard of underwater environment constrained condition, establishing an underwater navigation device path programming evaluation model; 2) according to the time, function relation, constraint, target condition and variate in the model, converting the path programming problem to a track optimization problem; 3) employing a mixed algorithm of Dijkstra algorithm and particle swarm optimization algorithm for track programming, taking roughly selected track as an initial estimation of a optimal solution for input, coding the particle swarm optimization algorithm, and taking time information as particle in a search space, and improving the convergence performance of the algorithm. The optimization method can increase the accuracy and reliability of the terrestrial magnetism navigation track programming.

The invention discloses a wind electricity power probability density predicting method based on a genetic algorithm and support vector quantile regression. The method is characterized by comprising the following steps: 1, acquiring data of output power of a wind electric field and carrying out data cleaning; a, carrying out normalization processing on sample data, and selecting data of training sets and testing sets; 3, establishing a support vector quantile regression model; 4, optimizing support vector quantile regression parameters by using the genetic algorithm; and 5, establishing a wind electricity power probability density predicting model to obtain a final wind electricity power predicting result. By the genetic algorithm, global searching can be implemented for optimization, wind electricity power predicting precision is improved, nondeterminacy of wind electricity power can be quantified, and basis is provided for safe and stable running of wind electricity connection.

The invention discloses a heterogeneous multi-core embedded system energy consumption optimization scheduling method based on a reinforcement learning algorithm. In the hardware aspect, a DVFS regulator is loaded on each processor, and the hardware platform matching the software characteristics is dynamically constructed by adjusting the working voltage of each processor and changing the hardwarecharacteristics of each processor. In the aspect of software, aiming at the shortcomings of traditional heuristic algorithm (genetic algorithm, annealing algorithm, etc.), such as insufficient local searching ability or weak global searching ability, this paper makes an exploratory application of Q-Learning algorithm to find the optimal scheduling solution of energy consumption. The Q-Learning algorithm can give consideration to the performance of global search and local search by trial-and-error and interactive feedback with the environment, so as to achieve better search results than the traditional heuristic algorithm. Thousands of experiments show that compared with the traditional GA algorithm, the energy consumption reduction rate of the Q-learning algorithm can reach 6%-32%.

The invention discloses a shaft system thermal error modeling method based on an SLSTM neural network. The method comprises the following steps: 1) inputting thermal error data of a shaft system changing with time; 2) decomposing the thermal error data into N intrinsic mode components and a residual component by using an EMD algorithm, and respectively converting the component data into a three-dimensional input matrix; 3) encoding the initial time window size, the batch processing size and the unit number of each piece of component data to obtain an original generation bat population; 4) initializing the original generation bat population by adopting a BA algorithm to obtain SLSTM neural networks with different time window sizes, different batch processing sizes and different unit numbers; 5) training the SLSTM neural network by using the thermal error data of the shaft system to determine hyper-parameters; and constructing an EMD-BA-SLSTM network model by using the optimal hyper-parameter, and then reconstructing a prediction component to obtain the output of a prediction result, i.e., the invention also discloses a shaft system thermal error compensation system based on the SLSTM neural network.

The invention belongs to the technical field of remote sensing image processing, and particularly relates to a hyperspectral image waveband selecting method based on mimicry physics optimization algorithm. The method comprises the following steps of: (1) dividing subspaces; (2) generating initial population and initial speed; (3) obtaining an optimal fitness value and a worst fitness value; (4) obtaining the mass of individuals of the population, virtual acting force of other individuals on the individual and the resultant force of the acting force; (5) updating individual movement speed and position; (6) updating the optimal fitness value and the worst fitness value; and (7) judging whether the iterations t meet the condition that t is more than tmax; if so, stopping circulation; and if not, adding 1 to the iterations and continuing to execute steps (4)-(7).

The invention discloses a software defect prediction method based on feature set division and ensemble learning, and the method comprises the steps: dividing an original data set into a training dataset and a test data set, and dividing the training data set into a plurality of feature subsets; selecting K base classifiers for ensemble learning, and synthesizing an ensemble classifier of each feature subset according to the base classifiers and the corresponding weights; selecting a feature subset most similar to the input instance, performing defect prediction on the input instance by usingan integrated classifier of the feature subset, and establishing a software defect prediction model; dividing the test data set and searching a feature subset most similar to the input instance; searching optimal values of the centroid set and the weight set, and optimizing the software defect prediction model by combining the most similar feature subset of the test data set. The method has the advantages that redundant features in the defect prediction data set can be removed, the search space of the algorithm is reduced, and the problem of high feature dimension of historical data of software defects can be effectively relieved.

The invention provides a QSFLA-SVM-based perceptive intrusion detection method. The method comprises the steps of setting related parameters; initializing the position of a frog population; transmitting position information of each frog individual to an SVM abnormal sequence detection model, taking a calculated correct rate of test set classification as a fitness function value of each frog individual, performing descending order arrangement on the frog population and performing sub-population division on the arranged population; updating worst individuals of each frog sub-population by utilizing a quantum particle swarm update mechanism, until a local maximum iterative frequency is reached; and performing global information exchange, and if a global maximum iterative frequency is reached,returning a global optimal frog individual, and outputting an optimal test set classification result, wherein at the moment, position information of the global optimal frog individual is an optimal parameter value when the SVM abnormal sequence detection model obtains the maximum correct rate of the test set classification. According to the method, intrusion detection is performed in combinationwith a quantum particle swarm search mechanism-based QSFLA and an SVM.

The invention relates to an optimal torque distribution method based on a distributed electric drive vehicle. The torques of four drive wheels are reasonably distributed, and meanwhile the drive system efficiency and driving safety of the distributed electric drive vehicle are improved. The torque distribution method comprises the following steps of (1) adopting a response surface analysis methodfor conducting regression analysis on test data of a hub motor to obtain a drive motor efficiency function; (2) based on a demand torque value of a distributed electric drive system, establishing objective functions which characterize the efficiency optimization of the drive system and the driving safety optimization of the vehicle respectively; adopting a linear weighting method of a self-adaptive weight coefficient for converting solutions of the two objective functions into a multi-objective optimization problem under constraint conditions; (3) integrating the respective advantages of a genetic algorithm and a taboo search algorithm to put forward a hybrid genetic taboo search algorithm (HGTSA) for solving the multi-objective optimization problem, and obtaining the optimal torque distribution of the distributed electric drive system accordingly.

The invention relates to a hybrid self-adaptive hydropower station group intelligent optimization scheduling method and system. The method comprises the following steps: determining a scheduling objective function according to a scheduling task of a hydropower station group; determining a scheduling constraint condition, and processing the scheduling constraint condition according to types; carrying out population initialization by adopting improved Tent chaotic mapping; calculating particle fitness, an individual optimal solution and a global optimal solution based on a particle swarm algorithm; calculating particle energy and a threshold thereof, and particle similarity and a threshold thereof; introducing a search strategy, searching a particle neighborhood, and updating an original solution; and updating the positions and speeds of the particles until a termination condition is reached. According to the method, Tent chaotic mapping is adopted to generate an initial population; particle energy and a threshold value thereof are introduced, particle similarity and a threshold value thereof are introduced to improve population evolution quality, continuous adaptive adjustment can be carried out along with iteration, good local refinement capability is achieved in the later period, premature is inhibited, and the defects that previous premature convergence occurs, and a solved solution is a local optimal solution instead of a global optimal solution are overcome.

The invention discloses a fault line selection analysis method of a large-data distribution network based on an improved particle swarm algorithm, This method improves the traditional particle swarm optimization algorithm, which not only optimizes the parameters of SVM model more quickly but also is not easy to fall into the local optimum, and combines the fifth harmonic method with wavelet packettransform method to realize the efficient fault line selection of resonant system. The invention not only has high realization rate and good accuracy, but also is not influenced by factors such as grounding resistance, fault distance and the like.

The invention belongs to the technical field of carbon emission prediction, and in particular relates to a carbon dioxide emission prediction method comprising the steps of collecting data including the historical CO2 emission, population, GDP per capita, urbanization rate, secondary industry added value proportion, energy consumption structure, energy strength, overall coal consumption, carbon emission strength and total export-import volume; performing non-dimensionalization on the data, computing a gray association degree between each piece of data and the CO2 emission, and screening CO2 emission influence factor indexes input by the model according to the gray association degrees to achieve feature dimension reduction; using a gray prediction model GM(1,1) to predict the screened CO2 emission influence factor index; and using predicted values of the CO2 emission influence factors to serve as model inputs, and then using an improved shuffled frog leaping algorithm to optimize a least square support vector machine model for predicting the CO2 emission. The method provide by the invention has efficient computing performance and excellent global searching ability.

The invention discloses a forming workshop energy-saving dispatching method based on a genetic simulated annealing algorithm. The forming workshop energy-saving dispatching method comprises the following steps that 1, an initial production scheme is generated; 2, an analysis period is set; 3, a production model for the total processing energy consumption and the processing time of the forming workshop is constructed; 4, a corresponding constraint condition is set to form a multi-objective optimization model; 5, relevant information is acquired, and the genetic simulated annealing algorithm isused for solving the multi-objective optimization model according to the collected information to obtain a production scheme of the analysis period; 6, it is judged that whether the production schemeof the analysis period is superior to the production scheme of previous analysis period, if yes, production is arranged according to the production scheme of the analysis period, and if not, production is continued according to the production scheme of the previous analysis period; and 7, a next production period begins and the step 5 is executed. The method can obtain an optimal production schemewith lowest energy consumption and shortest time in real time and perform production and processing, so that productivity is improved, and production cost and energy consumption are reduced.

The invention provides a torque ripple suppression method for a permanent magnet synchronous motor injected with harmonic current. The method comprises the steps of constructing a single electric period sequence model of harmonic flux linkage and harmonic current, further deriving a torque sequence model containing k-th harmonic torque, establishing a target function with a minimum torque peak-to-peak value as a target, optimizing the target function by adopting a genetic algorithm, solving the optimal harmonic current, taking the optimal solution as a reference value of the harmonic current,and respectively controlling the fundamental current and harmonic current tracking reference values, thereby suppressing the torque ripple. According to the method, the single electric period sequencemodel of the harmonic flux linkage and the harmonic current is constructed, the torque sequence model containing the k-th harmonic torque is derived, the influence of the flux linkage harmonic amplitude and phase is comprehensively considered, the actual waveform of the torque in one electric period is fitted, and the result is real and reliable; and by introducing a genetic algorithm, the robustness is high, the global optimal solution of the harmonic current under the current working condition can be quickly solved, and the convergence is very good.

A steam pipe network friction resistance coefficient identification system based on the genetic algorithm belongs to the technical field of parameter identification and calculation of a steam pipe network and comprises a relation data base server, a real-time data base server, an application server, an engineer station and an application module. The relation data base server is connected with the engineer station and the application server, and the application server is connected with the relation data base server, the real-time data base server and the engineer station and keeps data exchange with the same. The application module comprises a relation data base, a data acquiring module, a data result display module, a waterpower and heating power coupled calculating module and a pipe network friction resistance coefficient identification module. The steam pipe network friction resistance coefficient identification system based on the genetic algorithm has the advantages that a target function formed by the quadratic sum of pressure of nodes of the pipe network, real-time pipe flow measuring valve and a calculated value is used as a criterion function, calculation of the steam pipe network friction resistance coefficient identification and calculation are realized quickly and accurately, so that the pipe network model calculation can be more accurate, and analysis and maintenance of the pipe network are facilitated.

The invention relates to a satellite task planning simulation analysis method and system based on a key path, and belongs to the technical field of spaceflight. The method comprises the steps: 1, calculating satellite observation window and yaw angle data by utilizing simulation software; 2, generating an initial task allocation scheme; 3, calculating individual fitness by utilizing a key path method; And 4, judging whether iteration is finished or not. The invention further relates to a satellite task planning simulation analysis system based on the key path, the satellite task planning simulation analysis system comprises an observation window calculation module, an optimization module and a result output module, and the three modules are executed in sequence when satellite task planningsimulation is carried out. A genetic algorithm and a key path method are adopted to carry out simulation analysis on satellite task planning, rapidity is guaranteed, and meanwhile it is guaranteed that the method has good global searching capacity.

The invention discloses a hybrid intelligent optimization method (PP algorithm) for a spatial structure. The method omprises the following steps: firstly, utilizing the global searching ability of particle swarm optimization algorithm (PSO) with high inertia weight to carry out preliminary screening on feasible region space, selecting a feasible solution which falls near the global optimum solution and taking it as initial growth point to provide initial value for subsequent simulated plant growth algorithm (PGSA); Then, the optimal solution obtained by PSO is used as the initial growth point,and the global optimal solution satisfying the requirements can be found quickly and accurately by using the growth space optimization method based on the strong local search ability of PGSA and thegood or bad value of objective function.

The invention discloses an identity recognition method based on PPG signal sparse decomposition. The method comprises following steps: step one, obtaining a PPG signal of a person to be identified, and preprocessing the PPG signal by filtering, moving average, and zero-mean methods; step two, detecting the time domain features of the preprocessed signal to extract the time domain feature value andextracting the optimal waveband of the preprocessed signal; step three, cutting the waveform of extracted optimal waveband to obtain a plurality of monocycle waveforms; step four, carrying out signalsparse decomposition on the monocycle waveforms to obtain optimal atomic feature parameter characteristics of the signal; step five, using the time domain feature value and the optimal atomic featureparameter characteristics to carry out feature fusion to obtain a training template and a test sample; and step six, utilizing a support vector machine to match the test sample and the training template to identify the identity of the person. The provided method solves the problem that in the prior art, a conventional identity recognition method is easily influenced by the external environment and the operation is complicated. The recognition rate of the method can reach 98% or more.

The invention discloses a method of establishing a user relation prediction model, comprising the steps of S1) obtaining a subnetwork of two user relations from the original social relation network through random sampling; S2) respectively extracting four topology features including the common friends number, common friends clustering coefficient, friends clustering coefficient and the shortest path distance of a user two-tuples connected to each edge in the subnetwork; S3) establishing the user relation prediction model, which has a feedforward neural network structure; S4) obtaining an optimal individual by means of genetic algorithm based on a training set and the established user relation prediction model, wherein the individual is a well-trained user relation prediction model. In addition, the invention provides a method of predicting the user dynamic relation, which can predict the dynamic change of user relation. The prediction method is advantageous in that the method is not restricted by the shortest path distance D when predicting the user relation; the accuracy of predicting the user relation and the analysis capability of weak relation can be improved.

The invention belongs to the technical field of energy prediction, in particular to an energy consumption prediction method. The method comprises the following steps: collecting sample data includinghistorical energy consumption, population, GDP, industrial structure, energy consumption structure, energy intensity, carbon emission intensity and total import and export amount; the sample data being subjected to dimensionless processing, and the grey relational degree of each sample data and energy consumption structure being calculated, and the input factors of the model being selected according to the order of grey relational degree; multiple IMF components being obtained by integrated empirical mode decomposition (EMD) based sequence denoising; the parameters of LS-SVM being optimized byusing the improved hybrid frog leapfrog algorithm, and the forecasting model being established to reconstruct the forecasting results, and the final energy consumption forecasting results being obtained. Experiments proved the EMD-ISFLA-LSSVM model predicts the energy consumption, and the predicting effect is remarkable.

The invention discloses a flight track matching method based on a genetic algorithm and the problems of low flight track matching accuracy and a large calculation amount in case of many targets, manydisturbances, and many noises are solved. The method comprises a step of inputting radar and surveillance system ADS-B flight tracks to obtain a set, a step of forming an initial population, a step ofcalculating the individual fitness of the population, a step of carrying out competition selection, a step of carrying out gene crossover, a step of carrying out gene mutation, a step of calculatingthe individual fitness of the population again and judging whether the fitness satisfies an ending condition or not, outputting an optimal result if so, otherwise carrying out a new round of selection, crossover, and mutation, and finally obtaining an optimal flight track matching event set. According to the method, the selection and inheritance mechanisms of the natural world are simulated, poormatching flight tracks are continuously removed, a good match is retained, and the final finding of the optimal result is ensured. The method has good global search ability, small calculation amount and linear controllability, an effect of finding an optimal matching result in a finite time is improved, and the method is used for flight track matching between radar and a monitoring system ADS-B.

The invention discloses a speedless item particle swarm optimization algorithm based on a circuit breaker. The speedless item particle swarm optimization algorithm introduces a stock market circuit breaker by targeting defects that a traditional particle swarm algorithm is slow in convergence and easy to produce local optimization; a particle swarm iteration evolution process is divided into 20 segments; if the particle swarm evolution process is in segments (1,2,5,6,9,10) of front 10 segments or in last 10 segments and random selection probability is smaller than 1 / 3, a circuit breaker is started to update positions of particles, if not, the circuit breaker is not started; and a speedless item global optimal position gbest guiding iteration equation is adopted to update positions of the particles. The speedless item particle swarm optimization algorithm based on the circuit breaker introduces the circuit breaker to change motion directions and pace lengths of the particles, jumps out of local optimization, and adopts the speedless item global optimal position gbest guiding iteration equation to accelerate convergence speed. As a result, the speedless item particle swarm optimization algorithm has an excellent global searching capability and has fast convergence speed.

The invention discloses a water demand prediction method for an Elman neural network based on optimization combination, and the method comprises the steps: carrying out the screening of input variables based on an MIV (mean impact value) algorithm, and optimizing the weight and threshold of the Elman neural network through an MEA (thought evolution algorithm). The MIV algorithm can effectively eliminate information overlapping of influence factors and screen out better indexes; the MEA algorithm has good global search capability, and the problem that a pure artificial neural network is limitedby random selection of an initial weight and a threshold can be solved. Therefore, the method is higher in accuracy and better in prediction effect, can effectively predict and forecast the water demand of the crops, guarantees timely and reasonable adjustment of an irrigation system, and has a certain application value in prediction of the water demand of the crops.

The invention discloses a target identification method of a remote sensing image of an artificial immune network based on a self-adaptive PSO(Particle Swarm Optimization), mainly overcoming the disadvantages of low target identification precision and low convergence speed in the traditional method. The identification method comprises the following steps of: firstly, extracting 7 invariant moment characteristics of an image target and carrying out normalization treatment on the characteristic data; secondly, setting running parameters, selecting a training sample and initializing an immune network and immune cells; thirdly, calculating the affinity degree of the immune cells and cloning the immune cells; fourthly, executing hyper-mutation operation based on the self-adaptive PSO; fifthly, selecting an immune cell with highest affinity degree and adding the immune cell into the immune network; sixthly, carrying out network inhibition operation; seventhly, judging a stop condition, turning to the eighth step eight if the condition is satisfied, and otherwise, and otherwise jumping to the third step; and eighthly, inputting characteristic values of the remote sensing images which are not used as training samples into the immune network, and judging a category attribute value of each image by the immune network. The method has the advantages of high target identification accuracy and stable target identification performance and can be used for solving the problem of target identification of a remote sensing image set.