34 results about "Behavior space" patented technology

The invention provides a method, a mobile terminal and a server for carrying out the intelligent search at a mobile terminal. The method comprises the following steps: acquiring the lexical space of network resources, and carrying out the conceptual clustering on the information to be searched in a search request, to form a conceptual space; retrieving lexes identical with the content semantics of the conceptual space from the content of the lexical space, to form a semantic space; analyzing user behavior to generate a user behavior space; generating an associated business library according to the mapping relation between the feasible business recorded in a business space and provided by operators and the content of the user behavior space; establishing the association relation between the content in the associated business library and the lexes in the semantic space; and selecting the search results from the network resources. By introducing the analysis on the user behavior, the invention improves the accuracy of search, so the search can be more convenient, direct and effective, and the information exchange via the network can be more user-friendly; and by analyzing the user behavior, the favorite advertising information and the most acceptable advertising information of the user can be recommended at the mobile terminal.

The invention discloses a humanoid robot stable control method of an RBF-Q learning frame. The method comprises the following steps: the RBF-Q learning frame which solves the problems of state space serialization and behavior space serialization in a Q learning process is brought forward; an online motion adjusting stable control algorithm of the RBF-Q learning is brought forward, loci of the hip joint, the knee joint and the ankle joint of a support leg are generated, and a humanoid robot is controlled to walk stably through calculation of angles of other joints; and finally, the feasibility and validity of an RBF-Q learning frame method are verified on the Vitruvian Man humanoid robot platform designed by the laboratory. The method provided by the invention can generate a stable walking gait of the humanoid robot in an online learning process.

The invention relates to a human behavior identification method based on multi-feature fusion, which comprises the following steps of: acquiring a human body behavior video by using a camera, extracting a foreground image of each frame of image, and carrying out cavity filling and interference filtering to obtain a human body silhouette image sequence; Calculating the similarity between adjacent frames in the image sequence, and obtaining the weight of each frame of image representing the behavior posture; According to each frame of image in the human body silhouette image sequence and the corresponding weight thereof, obtaining an action energy diagram representing a behavior process through weighted average; And extracting a Zernike moment, a gray histogram and a texture feature of the action energy diagram to form a multi-dimensional feature fusion vector containing behavior space-time characteristics; Constructing feature vector template libraries of different standard behaviors; And in the behavior identification process, extracting feature vectors of to-be-identified behaviors according to the to-be-identified videos, matching the feature vectors of the to-be-identified behaviors with the feature vectors of the standard behavior template library one by one, determining behavior types according to matching results, and realizing accurate identification of human body behaviors. According to the method, the time change and spatial attitude characteristics of human body behaviors are represented by constructing the action energy diagram, the behavior recognition accuracyand real-time performance can be improved, and certain practical value is achieved.

The invention relates to a method for realizing optimization combined connection admission and routing control in a wireless mesh network, which comprises the steps of determining state space in the wireless mesh network and current state of a system, determining corresponding action space, determining and selecting corresponding action in the action space according to maximized network operating average reward criteria, executing the determined and selected action, updating the current state of the system, and returning and repeatedly executing the steps when a next event occurs in the wireless mesh network. The adoption of the method for realizing optimization combined connection admission and routing control in the wireless mesh network can optimally control admissible access or inadmissible access as well as admissible new connection or admissible switch connection access, ensures that the switch connection has highest priority in all connection requests of the same level, adapts to a plurality of service types and service conditions of the wireless mesh network, effectively improves the network service quality and the operating income, reduces the computational complexity, and has stable and reliable working performance, and wider application range.

The invention provides a driver behavior recognition method based on a deep hybrid encoding and decoding neural network. The method comprises the steps: constructing a driver behavior recognition dataset; constructing a coding and decoding space-time convolution network; constructing a convolutional long short-term memory network; constructing a classification network; training three networks inthe driver behavior recognition model; adopting the trained driver behavior recognition model to recognize videos in the data set; and sending the video sample into a trained encoding and decoding space-time convolutional network to obtain a short-term behavior space-time feature representation, sending the short-term behavior space-time feature representation into a trained convolutional long-term and short-term memory network to obtain a long-term behavior space-time feature representation, and outputting a final driver behavior classification result by a trained classification network. According to the method, the implicit motion information can be effectively extracted from the short-term video clip, the driver behavior characteristic coding in the long video is realized through space-time fusion, the identification precision is high, and the driver behavior identification in the monitoring video can be realized.

The invention discloses a multi-level semantic feature extraction method for behavior data of intelligent wearable equipment, and the method comprises the following steps: S101, constructing a single-level behavior space, and finding a group of bases in the behavior space; S102, constructing a multi-level behavior space, and analyzing the behavior data from different granularities; S103, extracting the multi-level semantic features of behavior data in the multi-level behavior space. The method does not need to manually mark the data, and can be suitable for any behavior while greatly reducing the manual cost. Meanwhile, the extracted semantic features can be used for analyzing the behaviors from different granularities, thereby guaranteeing the high recognition precision. Compared with a conventional method based on a pre-defined semantic feature space, the method greatly improves the accuracy. Compared with a conventional method based on a supervised depth neural network, the method can provide higher recognition precision.

The invention provides an algorithm for identifying and positioning crowd abnormal behaviors by utilizing accumulated light streams. The algorithm comprises the following operation processes: step 1,performing space-time structure analysis on crowd behaviors; step 2, establishing a crowd behavior space-time structure and an identification rule; step 3, carrying out crowd behavior identification and positioning through the monitoring video; and step 4, comparing experimental analysis and result evaluation, and giving an alarm for an abnormal event. According to the invention, research is carried out on crowd behavior analysis and identification highly related to emergencies; and real-time detection of emergencies is realized, and the algorithm is applied to real-time monitoring and post-event analysis of high-crowd-density public places, has great practical significance and economic value for construction of safe cities and guarantee of public safety, and has wide application prospectsin deployment and implementation of video monitoring systems in urban public places.

The invention discloses a reinforcement learning method based on a mixed behavior space, and relates to the field of reinforcement learning, and the invention consists of a plurality of parallel Actornetworks which jointly act to output structured behaviors, and a Critic network which guides the training of the Actor networks. The Actor network comprises a state coding network, a discrete Actor network and a continuous parameter Actor network, the state coding network codes the state and inputs the state into the discrete Actor network and the continuous parameter Actor network, the discreteActor network is used for generating discrete actions, and the continuous parameter Actor network is used for generating continuous parameters corresponding to the discrete actions. According to the invention, the mixed behavior space with continuous actions and discrete actions can be processed, and the method can be expanded to all behavior spaces with hierarchical structures. According to the invention, a reinforcement learning result better than that of a previous mixed behavior space processing method can be obtained, the accuracy of behaviors is not lost, and the problem of over-parameterization is avoided through mask operation.

The invention relates to a data packet transmission intelligent decision-making method based on deep reinforcement learning. The method comprises the following steps: constructing a deep neural network model; designing and initializing a state space and a behavior space; acquiring current state information and historical state information of data transmission, and inputting the current state information and the historical state information into a state space; storing historical state information by adopting an experience playback mechanism; iteratively executing the step (3) and the step (4) for T times, and ending the round; updating the target value neural network parameter [theta]', and endowing the target value neural network with the latest parameter theta of the original value neuralnetwork; and (3) iteratively executing the steps (2) to (5) until the number of iterations reaches a preset round upper limit N or the deep neural network is converged, and terminating and automatically obtaining a data transmission strategy which meets the multi-constraint condition limitation and is lower in energy consumption. According to the invention, the user service quality is improved, the data transmission energy consumption is reduced, and the intelligent decision-making capability of data transmission of a communication network in a highly complex dynamic environment is effectively improved.

The invention discloses a reinforcement learning cluster swarming control method based on a graph neural network. The method comprises the following steps: establishing a cluster swarming control model; determining a topological structure feature representation method of the cluster; determining an observation information feature representation method of the intelligent agent; designing a state space, a behavior space and a return function; designing a strategy network and evaluation network model in a deep reinforcement learning algorithm; designing an algorithm framework and a network parameter updating method; and designing a training process of a cluster swarming control algorithm. The cluster swarming control algorithm is realized by means of the deep reinforcement learning technology, topological structure features and observation information features of the cluster are extracted by using the graph neural network, the convergence speed of the cluster swarming control algorithm and the adaptive capacity to the dynamic environment are effectively improved, and meanwhile, the stability of the algorithm under the interference of control noise and the like can be ensured.

The invention discloses a power grid pricing method based on reinforcement learning, and relates to a power grid pricing method. Existing power grid pricing is not comprehensive. The method comprisesthe following steps: initializing a Q value table and an E value table, and setting an initial electricity price; obtaining a state space and a behavior space of the reinforcement learning model, andadjusting the initial electricity price; feeding back the electricity price adjustment behavior to the power grid electricity selling market environment, and waiting for a time interval T to obtain environment information; calculating a return function of the current time by using the environmental feedback information and the power price information proposed in the previous state to obtain an instant return R, and updating an E value table and a Q value table; selecting a new behavior by using the updated Q value table, and adjusting the electricity price;, after each time T, circulating thesteps S4-S7 until the Q value table converges to a set range, and obtaining the optimal power grid pricing. According to the technical scheme, real-time pricing of the power grid is realized accordingto factors of supply and demand balance, power grid reliability and real-time pricing.

The invention discloses a behavior recognition method based on dual-channel depth separable convolution of skeleton data, and belongs to the technical field of human body posture behavior recognition.The method comprises the following steps: 1, acquiring human body behavior posture joint skeleton point data; 2, processing the skeleton point data to extract behavior space features; 3, constructinga D2SE dual-channel depth separable convolution layer, and extracting behavior time features in a time dimension; 4, superposing the space information on the graph convolution and the time information on a D2SE network layer to extract the space-time information of the attitude behavior; and 5, using a ReLu function to obtain bone movement classification. The GCN network layer and the D2SE network layer are used, spatial image convolution is used for human body posture behavior skeleton data to extract spatial information. Based on double channels, extra complexity is not introduced when theperformance of a convolution framework based on deep separation is improved, and parameters of a convolution layer can be obviously reduced.

The invention belongs to the technical field of identification, provides a behavior identification method of a deep network model based on space-time context association, and aims to solve the problemthat the spatial feature learning range in a CNN model is limited by the size of a sensing domain, reduce the loss of behavior feature representation by the model, and improve the accuracy of behavior recognition. According to the main scheme, user behavior data is imported for convolution mapping operation; then a behavior space-time characteristic graph TSF is obtained by utilizing a grid LSTMneural network, the behavior space-time characteristic graph TSF is imported into an attention gate module to carry out different time characteristic weight learning, a behavior characteristic graph is obtained, and the behavior characteristic graph is input into a softmax classifier to calculate probability distribution D of a behavior category; cross entropy loss function operation is carried out on the probability distribution D of the behavior category and the training set behavior label Y to obtain a loss Loss0, and an l2 loss function is introduced as a final total loss function L; and according to the total loss function L, the numerical value of the model imaginary-seating parameter is modified by using a back propagation operation to obtain the deep network model M.

The invention discloses a traffic organization scheme optimization method based on multi-signal lamp reinforcement learning, and belongs to the field of traffic signal lamp control. Firstly, an Actor network containing a state space set and a behavior space set is constructed, then an observed value is introduced, high-latitude information is compressed into low-latitude information through processing of a Subnet network, the behavior deflection probability is calculated, then initial state information, updated state information and the behavior deflection probability are introduced into a Critic network for centralized learning, finally, track reconstruction is carried out. In a multi-intersection traffic environment, multiple intelligent agents improve the road network unblocked rate by means of an Actor-Critic algorithm framework. Meanwhile, a method of centralized learning and distributed execution between intelligent agents is used, and the advantages of centralized learning and distributed execution are combined, so that the convergence speed of the algorithm is greatly improved.

The invention discloses a Deep Q-Learning-based cluster area coverage method. The method comprises the following steps of establishing a dynamic model of a cluster system; determining a neighbor set of agents in the cluster; establishing a motion control model of the cluster system; constructing an information map, and encoding the information map; defining a state space, a behavior space and a return function required by reinforcement learning according to the information map; a network model required by a Deep Q-Learning algorithm is designed; a Deep Q-Learning area coverage algorithm under the free area is designed; and adjusting the obtained points as required to obtain a Deep Q-Learning area coverage algorithm under the obstacle area. According to the method, training and learning of a cluster area coverage control algorithm are realized by means of a Deep Q-Learning technology, cluster area coverage in a free area and an obstacle area is realized, the cluster area coverage efficiency is effectively improved, and meanwhile, the stability of the algorithm in a weak communication environment can be ensured.

The invention discloses a cross axle double-flow-line flat bed house type which mainly structurally comprises walls, a floor and middle partition plates in a defined mode, and a three-room two-hall expansion structure is formed. The cross axle double-flow-line flat bed house type is characterized in that the interior of a house is divided into master and service cross axle double flow lines. I is the living flow line of a master and includes a living room, a dining room, a study, a master bedroom and a subaltern bedroom. II is the service working flow line perpendicular to I and includes a working balcony, a storage room, two hallways, a small bedroom, a kitchen, the dining room and the living room. The study is a flexible space and is changeable in pattern; the double-hallway system has the decoration and storage functions; the axial symmetric ceremony sensing dinning room is comfortable in angle of view. Both the living behavior space and the living article space are emphasized in the design, and the functions of safety, convenience, harmony and pleasantness are achieved on living people.

The invention discloses an analysis method for targets entering a substation environment. The method comprises the following steps: an infrared sensing network is established around the substation, and the infrared sensing network surrounds the edge of a set area around the substation; constructing an infrared human behavior database, wherein a plurality of infrared human behavior templates are stored in the infrared human behavior database; The infrared sensing network performs infrared detection on the target entering the set area in real time, tracks the behavior of the target, and generates an infrared video image; and the infrared sensing network performs infrared detection on the target entering the set area in real time. Gaussian mixture model and background subtraction algorithm are adopted to extract the behavior spatio-temporal contour of the object in the infrared video image, and the behavior spatio-temporal contour of the object is enhanced in contrast. The behavior space-time contour of the target is compared with a plurality of infrared human behavior templates stored in the infrared human behavior database one by one, and similarity analysis is performed on the behavior space-time contour of the target and the infrared human behavior templates.

The embodiment of the invention provides a network selection method and device based on a Markov decision process, and the method comprises the steps: building a Markov decision process model, and determining the state space, behavior space and transition probability of a terminal selection network; obtaining a network performance normalization weight vector of each network and a preset service quality performance index weight vector; obtaining transfer income according to the network performance normalization weight vector, the service quality performance index weight vector, the state spaceand the behavior space; and according to the transfer income and the transfer probability, obtaining a corresponding system state when the long-term income is maximum, and obtaining a network selection result of the terminal. According to the network selection method and device based on the Markov decision process provided by the embodiment of the invention, the performance demand difference of the terminal task under different service qualities is fully considered to obtain the transfer income, the long-term income of the system is considered, the system state of the system at the maximum income is calculated, and the terminal network selection scheme is obtained.