42results about How to "Good initial value" patented technology

The invention discloses a clustering and deep belief network-based leakage initial locating method. The method comprises the steps of firstly performing calculation to obtain a pipe network monitoringpoint sensitivity coefficient matrix; secondly dividing a pipe network into multiple leakage regions based on K-means clustering, and generating a leakage sample by utilizing hydraulic simulation software; thirdly building and training a deep belief network-based leakage region identification model; and finally identifying the leakage regions according to actually measured pressure data. The method solves the problem of leakage sample scarcity during modeling, realizes quick locating of the leakage regions in an initial leakage stage, and has relatively high identification precision and relatively strong operability.

The present invention relates to a method and a system for predicting the load of a power distribution network. The method comprises the steps of obtaining a unsupervised training sample set, a supervised training sample set and a test sample set according to the time information and on the basis of the historical load influence factor of the power distribution network and the historical load value of a to-be-predicted area; according to the unsupervised training sample set, subjecting a DBN model layer in a pre-established load prediction model to unsupervised training layer by layer, wherein the load prediction model includes a DBN model layer and a linear neural network layer; adopting network parameters obtained through during the unsupervised training step as the network parameter initial values of the load prediction model; according to the supervised training sample set, subjecting the load prediction model to supervised training, and obtaining an optimal load prediction model; testing the test sample set by using the optimal load prediction model so as to obtain a load prediction value of the to-be-predicted area. The present invention can realize the high-precision loadprediction of the intelligent power grid environment under the influence of various factors.

A moving object detection method based on coarse-to-fine image segmentation comprises the following steps: n frame detection images are acquired; feature points are extracted from the first frame image, and the feature points are extracted from the next n-1 frame image to generate motion cues, and a classification of the feature points in the n-th frame image is determined by performing motion segmentation on the motion cues; superpixel segmentation is performed on the n-th frame image to reduce the dimension of the n-th frame image; according to the classification of the mark points in the n-th frame image, the superpixels are marked, and the superpixels are clustered by the similarity measure method, and the rough segmentation of the image is completed on the basis of clustering. The coarse segmentation image is constructed into a quaternary image, and then the quaternary image is fine segmented, so as to realize the accurate detection of the moving object. Through the segmentation from the coarse to the fine, the invention not only has high accuracy rate to the detection of the moving object in the image, but also has faster detection speed, and is more suitable for the detection of the fast moving object.

The invention discloses a double-filter fusion positioning method and system for park automatic driving, and aims to design the double-filter fusion positioning method and system for park automatic driving, so that by fusing a vehicle kinematics model and IMU integration through an extended Kalman filtering technology, a predicted pose of a vehicle is obtained. A particle filtering technology is used to fuse a vehicle prediction pose and a laser odometer so as to improve positioning precision and robustness of an automatic driving vehicle in a park environment.

The invention discloses an online identification method for pneumatic parameters of a vertical take-off and landing vehicle. According to the method, on the basis of wind tunnel data, the pneumatic parameters under different flight conditions are identified by using a maximum likelihood method, and then the relationship between the flight conditions of the vehicle and the pneumatic parameters is established by training a neural network so as to adapt to the condition that the pneumatic parameters of the vehicle change along with the flight conditions. Pneumatic parameter identification is performed based on the wind tunnel data and the maximum likelihood method, and the pneumatic parameters obtained by the wind tunnel data pneumatic interpolation table can provide good initial values for parameter identification by the maximum likelihood method; the trained neural network can be used for pneumatic parameter on-line identification, is good in real-time performance, and has good engineering practicability. The method solves the problems that a group of pneumatic parameters correspond to various flight conditions in offline identification of the pneumatic parameters of the vertical take-off and landing vehicle and are not consistent with reality, and the online identification real-time performance of the pneumatic parameters of the vehicle is poor, and can be used for online pneumatic parameter identification.

The invention discloses a grid map positioning method fusing NDT and ICP, and the method comprises the following steps: S1, obtaining the initial pose of a laser radar through employing an NDT algorithm for a last frame of laser point cloud and a current laser frame of point cloud; and S2, matching the point cloud of the current laser frame with the laser point cloud of the previous frame based on the initial pose, and obtaining the current pose of the laser radar. The repositioning problem of the mobile robot after abnormal conditions such as'binding 'or restarting and the like occurs is solved, and accurate positioning is realized only through a laser sensor.

The invention discloses a breast nodule detection method and system based on ultrasonic medicine. The method comprises the steps of obtaining breast ultrasonic video data, carrying out the preprocessing, carrying out training on the preprocessed data through a transfer learning method, building a breast nodule detection model, inputting to-be-tested video data into the breast nodule detection model, and obtaining a result. The breast nodule detection method and system based on ultrasonic medicine aim to establish a precise method for judging benign and malignant breast tumors and carrying outpathological typing by taking ultrasonic breast video image examination as a main means, and solve the difficult clinical problem that dynamic identification and identification of breast tumors are difficult. The establishment of the highly accurate breast tumor qualitative system is of great clinical significance to the diagnosis and treatment of breast tumor patients.

The invention provides a three-phase asynchronous motor parameter off-line acquisition method and system. The method includes the steps: driving an inverter by a controller to apply excitation voltage to any two phases of a three-phase asynchronous motor at different time periods, and acquiring response current generated by the three-phase asynchronous motor according to the excitation voltage at different time periods; acquiring the leakage inductance of the three-phase asynchronous motor according to the excitation voltage, the response current and a first equivalent circuit, and acquiring a first time constant of the three-phase asynchronous motor according to the response current and a second equivalent circuit. According to the technical scheme, parameters of the motor can be statically recognized under the condition of finishing installation of the motor, good initial values are provided for an online identification algorithm, safe and reliable starting of the three-phase asynchronous motor is further ensured, and convergence of the online identification algorithm is accelerated.

The invention relates to a multipath channel estimation method based on an all-pole model, which comprises the following steps: converting an impulse response of a multipath channel from time domain to frequency domain to obtain a frequency domain response of the multipath channel; expressing the frequency domain response with an all-pole model to obtain an all-pole expression of the multipath channel model; recurring the all-pole model with the corresponding sending end data and receiving end data to obtain model parameters, and constructing an all-pole model with the model parameters; searching poles in the constructed all-pole model to obtain an initial estimated value of a multipath channel time-lag from the searched poles; and determining the impulse response of the multipath channel model according to the path number of the multipath channel and the initial estimated value of the multipath channel time-lag, and obtaining a multipath channel model. The method roughly estimates the multipath channel time-lag, which can prevent the multipath channel parameters trapping in local optimum in the subsequent refining, so as to reduce the complexity of the available multipath channel model estimation and improve the robustness, the anti-jamming property and accuracy.

The invention discloses a method for improving the expansibility of a deep neural network power flow model, which mainly comprises the following steps: (1) acquiring basic data of a power system; (2)determining a feature vector; (3) establishing an original DNN power flow model; (4) training the original DNN power flow model to obtain a trained original DNN power flow model; (5) expanding the original DNN power flow model to obtain an expanded DNN power flow model; and (6) calculating the probabilistic power flow of the expanded system to obtain a probabilistic power flow result. The method can be widely applied to the calculation of the probabilistic power flow of the power system, and is particularly suitable for the situation in which the DNN of the original system is not suitable forbeing used due to the expansion of the system.

The invention discloses a robust sound source positioning system and method. The method comprises the following steps: step 1, establishing a time difference equation set; 2, estimating a time difference tauij; 3, implementing a sound velocity processing mode; 4, solving an equation by a regularization Newton method; 5, setting an initial solution; and 6, carrying out weight initialization and self-adaptive updating. In combination with other structures or methods, the defects that in the prior art, errors can be caused by a time difference positioning method under sound source positioning, and large positioning errors can be generated under the environment that a microphone has large noise are effectively overcome.

The invention discloses a method for estimating the relative posture of a human face by combining key points of the human face with RGBD data. According to the method, face key points on an RGB imageare recognized, an initial posture angle difference is calculated, depth data is combined, RGBD point cloud is matched, and high-precision relative posture estimation of a face between different angles is obtained. Different from a previous recognition method, RGB-D data are used in the method, the requirement for the size of a data set is low, and high-precision relative attitude estimation can be achieved on data of small samples. Face key point information is also combined, so that the robustness of attitude estimation is enhanced.