40results about How to "Solve the "curse of dimensionality" problem" patented technology

The invention provides a robot adaptive grabbing method based on deep reinforcement learning. The method comprises the following steps: when distanced a certain distance away from an object to be grabbed, a robot obtaining a picture of a target through a pick-up head in the front, then according to the picture, calculating position information of the target by use of a binocular distance measurement method, and applying the calculated position information to robot navigation; when the target goes into the grabbing scope of a manipulator, taking a picture of the target through the pick-up head in the front again, and by use of a DDPG-based deep reinforcement learning network trained in advance, performing data dimension reduction feature extraction on the picture; and according to a feature extraction result, obtaining a control strategy of the robot, and the robot controlling a movement path and the posture of the manipulator by use of a control strategy so as to realize adaptive grabbing of the target. The grabbing method can realize adaptive grabbing of objects which are in different sizes and shapes and are not fixedly positioned and has quite good market application prospect.

The invention discloses a reinforcement learning-based AUV behavior planning and motion control method, which belongs to the technical field of underwater robots and aims at solving the problems of limited training experience and difficult application in a real environment as complex task planning by the AUV depends much on manual experience and a control method designed based on an intelligent algorithm needs an accurate environment mode. AUV tunnel detection is defined as a general task; behaviors corresponding to task completion comprise target trending, wall tracking and obstacle avoidance; a control instruction generated by the robot under water to complete the planned behavior is defined as an action; and when the AUV executes a tunnel detection task, a deep reinforcement learning DQN algorithm is used for real-time behavior planning, a corresponding deep learning behavior network is constructed, and planning of the tunnel detection task is completed. The AUV action network is trained by the DDPG method, the AUV is regarded as an environment model, and force-to-state mapping is obtained, thereby realizing the action control of the AUV.

The invention discloses a privacy-protection index generation method for mass image retrieval, relates to the privacy protection problem in mass image retrieval and involves with taking privacy protection into image retrieval. The method is used for establishing an image index with privacy protection, and therefore, the safety of the privacy information of a user can be protected while the retrieval performance is guaranteed. The method comprises the steps of firstly, extracting and optimizing SIFT (Scale Invariant Feature Transform) and HSV (Hue, Saturation and Value) color histogram, performing feature dimension reduction by use of a use of a manifold dimension reduction method of locality preserving projections, and constructing a vocabulary tree by using the dimension-reduced feature data. The vocabulary tree is used for constructing an inverted index structure; the method is capable of reducing the number of features, increasing the speed of plaintext domain image retrieval and also optimizing the performance of image retrieval. The method is characterized in that privacy protection is added on the basis of a plaintext domain retrieval framework and the inverted index is double encrypted by use of binary random codes and random projections, and therefore, the image index with privacy protection is realized.

The invention provides a hexapod-robot real-time gait planning method based on deep reinforcement learning. The method comprises the following steps of using a hexapod robot to acquire environment road condition information and making an integral movement track; through a camera, acquiring an environment photograph, according to the photograph, using a binocular range finding method to calculate road condition information of a target track, and using the calculated road condition information of the track in robot center of mass movement track navigation; in a foot end swinging space range of robot legs, taking photographs of a road condition environment, and through a trained deep reinforcement learning network based on a depth determinacy policy gradient (DDPG), carrying out data dimension reduction and feature extraction on the photograph; and according to a feature extraction result, acquiring a control policy of a hexapod robot, wherein the hexapod robot controls foot laying of a robot according to the control policy so that real-time walking of the hexapod robot is realized. By using the gait planning method, a complex and non-structural environment of a road condition can be planned in real time. The method has an important meaning for increasing an environmental adaptive capacity of the hexapod robot.

The invention discloses an instrument pointer jittering recognition method based on a support vector machine during interment detection. The instrument pointer jittering recognition method based on the support vector machine during interment detection solves the problems that due to the facts that the accuracy degree of an instrument is decreased and the instrument jitters, errors are caused for the industry, the exertion time of a control variable is long, the magnitude is not accurate, and an existing mode is not accurate or not enough in recognition accuracy. The method comprises the steps of (1) distinguishing subclasses of the instrument, (2) obtaining a training sample, (3) preprocessing the training sample, (4) obtaining a feature space, (5) generating a training model, (6) obtaining the optimal SVM training model, (7) recognizing and segmenting an image to be tested, (8) determining whether an instrument pointer jitters or not, and the like. The instrument pointer jittering recognition method is applied to the field of jittering of the instrument pointer.

The invention discloses a ligand molecule massive characteristic screening method in drug design. In the drug molecule virtual screening based on the ligand, dimensionality (every dimensional characteristic represents a substructure) of the ligand molecule fingerprint feature generated by using the most popular ECFP method at present is massive and even ten thousands of dimensions due to the large number of ligand molecule, so the method can meet the problem of 'dimensionality disaster' in actual task. The method screens the massive ECFP molecule fingerprint features by using the LASSO method based on EDPP rule, and acquires related characteristics of the ligand molecule by using a robustness selecting method. The activity of the ligand molecule is often related to a few number of substructure; the method can rapidly and largely remove uncorrelated characteristics, select related characteristics of robustness, solve the problem of the 'dimensionality disaster', acquire the substructure related to the ligand activity and push the wider application of the ECFP method in the drug design.

The invention discloses a video target tracking method based on a manifold particle filter algorithm. According to the method, projective transformation of a video image is constructed to a matrix lie group, the projective transformation parameter of a target is used as state variable, and a state transition model is established on the lie group; a target area in the video image is described by means of covariance description; a state sample is extracted along a manifold geodesic by means of the particle filter algorithm on the lie group; an internal implication mean value is solved by means of the internal implication Gaussian Newton arithmetic, state estimation of a system is obtained, and target tracking is achieved. According to the video target tracking method based on the manifold particle filter algorithm, effect of the noise statistical property of the Euclidean space on weight variance is reduced, solving of the particle degenerating problem is benefited, and tracking accuracy and robustness of the algorithm are improved.

The invention relates to a feature selection method and system, a computer readable storage medium and electronic equipment. The method comprises the steps that a first feature subset is acquired, anda first evaluation index corresponding to the first feature subset is determined; Obtaining a second feature subset and a third feature subset; Determining a second evaluation index corresponding tothe third feature subset; When the second evaluation index is superior to the first evaluation index, executing an updating operation; And when it is determined that the first updating cut-off condition is met, determining the current target feature subset as the target input feature of the target model, otherwise, determining the third feature subset as a new first feature subset, determining thesecond evaluation index as a new first evaluation index, and returning to the step of obtaining the second feature subset and the third feature subset. Therefore, the target input features of the better target model can be quickly and effectively screened out, the dimension of the feature set is greatly reduced, the problem of 'dimension disaster' is solved to a great extent, and the calculationefficiency is improved.

The invention discloses a 'source-network-load 'two-stage multi-target control method and system for a power distribution network, and aims to provide a control method and system capable of realizingthe stability of safe operation of a power distribution system. The method comprises the following steps: establishing a two-stage multi-objective mathematical optimization model; establishing a first-stage centralized optimization control objective function and a second-stage distributed optimization control objective function; solving a first-stage centralized optimization control objective function of the current day by adopting a multi-objective harmony search algorithm; setting weight preferences to obtain a network structure of the power distribution network and the total number of partition feeders; adopting a harmony search algorithm to solve a second-stage distributed optimization control objective function of the partition feeder line; if the current time point is equal to 24 hours, generating distributed power supply and flexible load power sequence data of the 24 time points; and if the current date is equal to the preset value, ending the whole calculation process. According to the method, the frequency of global information transmission processing is reduced, and the security is improved.

The invention relates to a word computing method and a word computing device based on a broad general type-2 fuzzy set. The method comprises the steps of building a broad general type-2 fuzzy set word model; building a univariate rule module, measuring proximity between an actual input assessment criteria linguistic value and a corresponding univariate rule module rule first component to serve as a weight and a rule consequent, and aggregating to acquire an intermediate output word linguistic value; operating to acquire a univariate rule module weight via a constrained multi-objective optimization algorithm, and aggregating the univariate rule module weight with the intermediate output word linguistic value to acquire a final output word linguistic value; determining a plurality of object relations to be evaluated through sorting via a fractional function method, or computing proximity between the final output word linguistic value and the known output word linguistic values and selecting the value with the maximum proximity to serve as the actual output word linguistic value of the object to be evaluated. According to the method and the device provided by the invention, the problem of curse of dimensionality caused by rapid increase of fuzzy rules is solved, word computing efficiency and accuracy in artificial intelligence equipment are effectively improved, and wider application prospects are provided.

The invention discloses a hierarchical dimensionality reduction dynamic planning method for load optimization distribution of a hydropower station, which is based on a dynamic planning method and overcomes the problem of curse of dimensionality of the dynamic planning method by using a hierarchical planning method. A table storage to-be-looked-up method is adopted. Firstly, an optimal operation general table of the whole plant is calculated off line, and an optimal operation scheme is obtained by looking up the table through a computer in the operation process. On the basis of unit design characteristic data of the power station, dynamic characteristics of each unit and each unit are calculated. On the basis, the optimal load distribution problem under the ideal condition that all the units in the plant can work normally (namely, all the units are good in state, free of faults and capable of working according to designed dynamic characteristics) is researched, a scheme of optimal load distribution of the hydropower station units is given, and a general conclusion of load distribution of the units is obtained. The method has the advantage that the operation efficiency of optimal distribution of the hydropower station unit load is improved. The calculation workload is reduced, the calculation time is shortened, and the real-time operation requirement is met.

The invention relates to a wheelchair control system based on detection of brain near-infrared function information. The system comprises a near-infrared collection unit, a control unit, a computer processing unit, a wireless transmission unit and a motor driving unit, wherein the output end of the near-infrared collection unit is connected with the input end of the control unit, the output end of the control unit is connected with the input end of the computer processing unit, the output end of the computer processing unit is connected with the input end of the wireless transmission unit, and the output end of the wireless transmission unit is connected with the input end of the motor driving unit. The system has the advantages that the brain movement function areas and four movement emotions, such as imagination of the left hand, imagination of the right hand, imagination of the left foot or imagination of the right foot, are recognized, and the different contents of aerobic hemoglobin and deoxygenated hemoglobin are detected, so the recognition purpose is realized, and a control instruction is sent; the cost is low, the resolution is high, and the operability is high.

The present invention discloses a variable-stage gradual optimization based joint optimization scheduling method for cascaded hydropower stations. The method comprises: step 1, acquiring an initial scheduling process of each hydropower station, and performing water level regulation in a constraint search manner; step 2, fixing water levels of an ith node and an (i+VP)th node, and performing optimization calculation for problems in VP stages by adopting a differential evolution algorithm; step 3, selecting a population size NP, wherein initial individuals are adjustable points of the cascaded hydropower stations; step 4, obtaining a water level process line of each cascaded hydropower station; step 5, by using the water level process line of each hydropower station obtained in the step 4 as an initial trajectory, repeating the steps 2 to 5 until a variable-stage convergent condition is satisfied; step 6, setting VP to be equal to the sum of VP and 1; repeating the steps 2 to 5 until a final convergent condition is satisfied; and step 7, ending the calculation, and obtaining an optimal water level process line of each hydropower station. The method disclosed by the present invention has very good global search capability for solving the problem in joint optimization scheduling of the cascaded hydropower stations, and effectively solves the local convergence problem of a conventional algorithm.

The invention discloses a multi-unmanned aerial vehicle cooperative search method under communication constraint, which belongs to the field of unmanned aerial vehicle regulation and control, and comprises the following steps of: firstly, determining the coverage area of each unmanned aerial vehicle; secondly, carrying out two-dimensional gridding processing on a given search region, and calculating an area coverage rate corresponding to each unmanned aerial vehicle at each moment; thirdly, setting an initial position, a speed and a direction of each unmanned aerial vehicle randomly , and determining a communication distance d between the unmanned aerial vehicles according to the communication quality; fourthly, establishing two neural networks with identical structures and different parameters for each unmanned aerial vehicle in each flight control system, and initializing the neural networks; fifthly, respectively calculating a node position of each unmanned aerial vehicle at the next moment by using the initialized unmanned aerial vehicles and a neural network, updating corresponding neural network parameters, carrying out loop iteration, and connecting all nodes to obtain a corresponding search path; and finally, enabling each unmanned aerial vehicle to fly according to the respective search path to complete the search task. According to the multi-unmanned aerial vehicle cooperative search method, the maximum search range is achieved, and meanwhile the communication quality is guaranteed.

The invention relates to a NOx emission prediction method and device, electronic equipment and a computer storage medium, and the method comprises the steps: obtaining sample data including a training set, a verification set and a test set, and outputting the sample data including NOx emission; inputting the training set into the initial SVR model for training, generating a first NOx emission prediction model, and obtaining parameters influencing the performance of the first NOx emission prediction model; inputting the verification set into a first NOx emission prediction model, optimizing parameters based on a PSO algorithm to obtain optimized parameters, and obtaining a second NOx emission prediction model based on the optimized parameters; and inputting the test set into a second NOx emission prediction model for prediction, and outputting a NOx emission prediction result. According to the method, an accurate prediction result is obtained, and whether the NOx sensor breaks down or not is diagnosed in time.

The invention belongs to the technical field related to reservoir scheduling, and particularly relates to a dynamic planning dimension reduction reservoir scheduling method based on a dynamic search and discrete mechanism. The method comprises the following steps: S1, determining an upper limit and a lower limit of an initial reservoir capacity of a reservoir to be processed; S2, dispersing the continuous curve of the storage capacity and the time into a plurality of moments, and updating the upper limit and the lower limit of the storage capacity corresponding to each moment; S3, dividing a part between the upper limit and the lower limit of the storage capacity at each moment into a plurality of equal parts to obtain the number of discrete points so as to obtain the number of the required discrete points; and S4, inputting the required discrete points and the upper limit and the lower limit of the reservoir capacity corresponding to each moment into a target function corresponding to the maximum peak clipping criterion of the reservoir so as to obtain the maximum discharge flow and the calculation time of the reservoir. By means of the method, the problems that in the dynamic planning calculation process, the number of discrete points is small, the discharged flow is negative, the dispersion degree is large, and the calculation time is long are solved.

The invention relates to the technical field of data monitoring, in particular to a nonlinear contour data monitoring method based on LLE-SVDD, shows that the nonlinear contour monitoring model basedon LLE-SVDD has a good monitoring effect by comparing the effect of the this method with the monitoring effect of a traditional contour monitoring method, and provides a new visual angle and a new method for further exploring the nonlinear contour quality real-time monitoring problem. The method comprises the steps of performing S1, local linear embedding dimension reduction; S2, constructing an LLE-SVDD nonlinear quality monitoring model; and S3, carrying out a model monitoring process.

The invention discloses an electric vehicle cluster dispatching method. The method comprises the steps of S1, establishing a mathematical model of a single electric vehicle according to an energy boundary concept; S2, performing cluster division on the electric vehicles according to the time when the single electric vehicle is accessed to the power system; and S3, based on an improved firefly algorithm, solving the electric vehicle load cooperative dispatching model. The mathematical model of the electric vehicle load is accurately established, the problem of dimensionality disasters generatedunder large-scale electric vehicle access is solved, the operation cost of a power grid is reduced, and the peak-valley difference of the power grid is reduced.

The invention discloses a quad-rotor unmanned aerial vehicle formation reconstruction control system based on an ADP and an obstacle avoidance mechanism, and the system comprises a finite time comprehensive disturbance observer based on a sliding mode, which carries out the estimation of the external comprehensive disturbance, and designs a cost function which reflects the uncertainty, tracking errors and controlled quantity at the same time according to a disturbance estimation value; an ADP-based formation reconstruction controller approximates an optimal cost function through a neural network and solves an optimal control law, and a potential energy function is adopted to complete collision avoidance item design of the unmanned aerial vehicle; and the finite time attitude tracking controller based on the sliding mode performs stable response tracking control on the yaw angle, the rolling angle and the pitch angle of the attitude of the unmanned aerial vehicle, and realizes formation keeping and formation transformation control of the multi-unmanned aerial vehicle formation in the interference environment. By adopting the quad-rotor unmanned aerial vehicle formation reconstruction control system, formation generation, maintenance and reconstruction control of multiple unmanned aerial vehicles under the influence of comprehensive disturbance is realized, and high-quality flight of the unmanned aerial vehicles is ensured.