31results about How to "Solving the Curse of Dimensionality" patented technology

The invention discloses a nasopharyngeal-carcinoma (NPC) lesion automatic-segmentation method and nasopharyngeal-carcinoma lesion automatic-segmentation systems based on deep learning. The method comprises: carrying out registration on a PET (Positron Emission Tomography) image and a CT (Computed Tomography) image of nasopharyngeal carcinoma to obtain a PET image and a CT image after registration;and inputting the PET image and the CT image after registration into a convolutional neural network to carry out feature representation and scores map reconstruction to obtain a nasopharyngeal-carcinoma lesion segmentation result graph. The method carries out registration on the PET image and the CT image of the nasopharyngeal carcinoma, obtains a nasopharyngeal-carcinoma lesion by automatic segmentation through the convolutional neural network, and is more objective and accurate as compared with manual segmentation manners of doctors; and the convolutional neural network in deep learning isadopted, consistency is better, feature learning ability is higher, the problems of dimension disasters, easy falling into a local optimum and the like are solved, lesion segmentation can be carried out on multi-modal images of the PET-CT images, and an application range is wider. The method can be widely applied to the field of medical image processing.

The invention discloses a method for realizing the five-freedom-degree alternating current active magnetic bearing displacement self detection by utilizing a mixed kernel function support vector machine displacement predicating model. According to the method, magnetic bearing control current is used as an input sample, radial and axial displacement is used as an output sample, the sample data is collected, a mixed kernel function is selected, the performance parameters of the support vector machine are optimized through a particle swarm algorithm, the training sample and the performance parameters are utilized for training the least square support vector machine, and a non-linear predicting model is built. The predicting model is connected with a linear closed loop controller before being connected to a five-freedom-degree alternating current active magnetic bearing in series, the magnetic bearing displacement closed loop control is formed with a first and second expansion current hysteresis three-phase power inverter and a switch power amplifier, the self detection of a five-freedom-degree alternating current active magnetic bearing displacement-free sensor is realized, the cost of a magnetic bearing system is reduced, and the dynamic property of the system is improved.

The embodiment of the invention provides a multi-logistics-robot movement control method and device. The method comprises the steps of constructing an action space according to actions capable of being executed by robots; receiving an end point position in the task information; planning the moving path of the robots into one or more sub-target positions to form a sub-target position space; obtaining the final numerical values of the network parameters by training the numerical values of the network parameters of a hierarchical reinforcement learning network; and controlling the robots to complete the moving task through the hierarchical reinforcement learning network based on the final numerical values of the network parameters. The problem of dimension disasters of complex problems is solved while the mutual cooperative moving control of multiple logistics robots is realized, and the training efficiency is improved.

The invention discloses a constructing method for a decoupling controller of a bearingless permanent magnet synchronous motor with five degrees of freedom. Three Clark inverse transformers are respectively connected in front of three corresponding current tracking inverters in series, and the three current tracking inverters and one linear power amplifier are respectively connected in front of the synchronous motor and a load model thereof in series to form a complex controlled object; and a support vector machine inverter with 6 input nodes and 7 output nodes is constructed by a support vector machine with 17 input nodes and 7 output nodes and 11 integrators, a pseudo-linear system is formed, a corresponding pseudo-linear subsystem controller is designed, a linear closed-loop controller is constructed, and the decoupling controller of the synchronous motor is constructed by the linear closed-loop controller, the support vector machine inverter, the three Clark inverse transformers, the three current tracking inverters and one linear power amplifier. Nonlinear dynamic decoupling control on the rotor displacement and the rotating speed of the bearingless permanent magnet synchronous motor with five degrees of freedom can be realized.

The invention relates to a wireless self-backhaul small base station access control and resource allocation joint optimization method, which belongs to the field of wireless communication. The methodcomprises the steps of: establishing a multi-objective optimization model jointly maximizing spectrum efficiency and energy efficiency for a wireless resource allocation method of a wireless self-backhaul small base station by using a Markov decision-making process on the premise of an average delay constraint of each user and a transmission power constraint of each base station; and formulating the optimal access and resource allocation strategy for any dynamically-achieved user demand by adopting an approximate dynamic programming method in a random dynamic arrival environment of user data packets, so that a system obtains higher spectrum efficiency and energy efficiency in relatively long-term resource allocation. The wireless self-backhaul small base station access control and resourceallocation joint optimization method provided by the invention can maximize the long-term average spectral efficiency and energy efficiency while ensuring the average delay constraint and the transmission power constraint.

The invention discloses a crowd evacuation simulation method and system based on deep reinforcement learning. The method comprises the steps of initializing a constructed evacuation scene simulation model according to scene information and crowd parameter information; grouping crowds, and dividing leaders and followers of each group; adopting a hierarchical path planning method to obtain evacuation paths of crowds, wherein a leader in an upper-layer group performs global path planning through an EMADDPG algorithm to obtain an optimal evacuation path, and followers in a lower-layer group avoidobstacles and follow the leader to evacuate along the optimal evacuation path. A learning curve and a high-priority experience playback strategy are introduced on the basis of a traditional MADDPG algorithm, an EMADDPG algorithm is formed, the learning efficiency of the algorithm is improved, a hierarchical path planning method is provided on the basis of the EMADDPG algorithm and used for planning evacuation paths of crowds, the path planning time is effectively shortened, and the crowd evacuation efficiency is improved. People can be better guided to evacuate, and the crowd evacuation efficiency is improved.

The invention provides a Chinese microblog topic detection method and system based on semantics, time and social relations, which is used for solving the problem that in topic detection, microblog data is poor in topic detection effect due to the defects of short text, spoken language, polysemy and the like. The method comprises the steps of collecting the microblog data of related topics at a certain time interval, performing pre-training on acquired microblog data by using a pre-training language model BERT (Binary Encoder Transformers), and performing pre-training on the acquired microblogdata by using the pre-training language model BERT to obtain pre-trained microblog data; conducting vectorization representation on the microblog text through a pre-trained BERT model, and acquiring microblog semantic representation based on context semantics; proposing a text clustering algorithm comprehensively considering a time factor and a forwarding relationship between microblogs so that the problem that the traditional microblog topic detection only considers text semantic similarity is solved. The invention is mainly used for microblog search tasks, and the topic detection results ofrelated microblogs are used for improving the microblog search hit rate.

The invention discloses a quality detection method of Dendrobium catenatum Lindley. The quality detection method comprises: 1) sequencing by using ITS-26SE and ITS-17SE as primers to identify the variety of a to-be-detected Dendrobium herb; 2) carrying out chromatographic detection on n samples to obtain detection data using a chemical small molecule component schaftoside and/or naringenin as a reference component; 3) respectively carrying out fingerprint detection on the samples to obtain the fingerprint peak area value of the full chemical components of Dendrobium catenatum Lindley; and 4) establishing an analysis model by using the content value of the schaftoside and/or naringenin in the chromatographic data as the response variable and using the peak area value of other components inthe fingerprint as the independent variable, screening the variable through a Lasso method, and establishing the fingerprint model related to the chemical small molecule component. According to the present invention, through the first-generation sequencing and the characteristic fingerprinting, the quality of Dendrobium catenatum Lindley is precisely identified and controlled.

The invention discloses a central air-conditioning control method based on multi-agent deep reinforcement learning, and the method comprises the steps: carrying out the model-free optimization control of the starting and stopping states and working parameters of a cooler, a cooling water pump and a cooling water tower fan in a central air-conditioning system according to the current indoor demand cooling load and outdoor wet bulb temperature, including the operation sequence control of the cooler; according to the control method, an accurate central air-conditioning system model does not need to be established in the actual deployment process, the working frequency of the cooling water pump and the working frequency of the cooling water tower fan can be respectively controlled only by using a single agent, and the working frequency of the cooling water pump and the working frequency of the cooling water tower fan can be controlled by relying on a small amount of historical data. An efficient and accurate control strategy is trained in a short time, the unnecessary refrigerating capacity is reduced, the workload of a refrigerator, a cooling water pump and a cooling water tower fan is reduced, the service life is prolonged, the failure rate is reduced, and the energy consumption of the whole central air-conditioning system and even the total energy consumption of a building are greatly reduced.

The invention discloses a simplification order reduction method for a Volterra series behavior model. The method comprises steps that, a general Volterra series behavior model in a discrete mode is established, unknown Volterra nucleuses are simplified according to physical characteristics of a to-be-described circuit, and Volterra series interception is further carried out; the Volterra nucleuses at each order are trimmed by employing a preset trimming method; input signals are recombined according to the trimmed Volterra nucleuses, and an input signal matrix is established; a simplified Volterra series behavior model is acquired through identification by utilizing a preset identification method. Through the method, a problem of curse of dimensionality existing in a general Volterra series behavior model is solved, so the general Volterra series behavior model can be applied to strong non-linear and memory length nonlinear circuit modeling, and application to system-grade electromagnetic compatibility simulation of a nonlinear circuit is convenient.

The invention discloses a face recognition method integrating EM algorithm and probabilistic two-dimensional CCA. Firstly, a probabilistic two-dimensional CCA model is established, and logarithmic likelihood expectation of left probabilistic model and right probabilistic model is calculated respectively. Then EM algorithm is used to maximize the expectation of logarithm likelihood to estimate theparameters of the model and optimize the left-right probability model. Finally, the probabilistic two-dimensional CCA projection matrix is used to project the observed data into the hidden space to reduce the dimension of the high-dimensional data, which greatly reduces the computational complexity and improves the accuracy of face recognition. A large number of test and recognition of the AR facedatabase verifies the robustness and superiority of the method proposed by the invention in dealing with the face samples of the illumination change, the expression change and the attitude change, and solves the dimension disaster and the small sample problem of the existing face recognition method in dealing with the high-dimensional data.

The invention relates to the field of knowledge maps, in particular to a traditional Chinese medicine and ethnic medicine knowledge map construction method. The method comprises the following steps: inputting medical data, converting the medical data into word vectors, carrying out entity labeling and automatic annotation on the word vectors, inputting the labeled and annotated word vectors into a traditional Chinese medicine and ethnic medicine database, and extracting knowledge graph information to obtain a traditional Chinese medicine and ethnic medicine knowledge graph model; converting unstructured data into structured data by utilizing a natural language processing technology. Therefore, the data can be conveniently identified by a computer, the utilization rate and accuracy of the data are improved, and a new rule is conveniently found by statistical data, thereby integrating minority medical resources, evaluating the medical resource level of minority regions, and improving the medical resource level of the minority regions. The minority medicine resource mining is accurately realized.

The invention discloses an image recognition method of an evolutionary algorithm based on gradient guidance. The method mainly comprises the following steps: 1, acquiring image samples to construct a training sample data set; 2, a parent population is initialized, a gSBX operator is used for the parent population in the mating selection process to obtain a filial population, the parent population is added into the filial population, non-dominated sorting is conducted on the filial population, and a plurality of first individuals are selected from the sorted population to serve as an optimal individual population; 3, dominated solutions are deleted from the optimal individual population, a sparse stochastic gradient method SGD is used for conducting fine adjustment on weight variables of all the remaining individuals in the population, and an attribute set of one individual is selected in an inflection point area on the Pareto leading edge face of the population to serve as a variable of a final training model. According to the method, the image recognition model is optimized through the evolutionary algorithm, so that the accuracy of the model in image recognition can be improved, and the training cost and memory consumption of the neural network are reduced.

The invention discloses a reservoir group adaptive scheduling method and system, and belongs to the field of water resource utilization. An initial population is generated in a search space, the fitness value of each individual in the population is evaluated, the individual optimal position and the global optimal position in the population are updated, global exploration is performed by adopting aglobal exploration strategy in the coevolution algorithm so as to improve the population convergence rate, and local search is performed by adopting a local development strategy in the coevolution algorithm so as to improve the individual precision. Individual positions in the population are updated through iterative evolution, and an individual output corresponding to the globally optimal solution is taken as a final solution after a search stop condition is reached. Examples prove that the method has good convergence rate and search precision, can coordinate the relationship between exploration and development, effectively solves the problem of cascade reservoir group scheduling, and significantly improves the economic benefits of the whole power grid.

The embodiment of the invention provides a correlation degree analysis method for corn starch process parameters and starch milk DE values, and belongs to the field of process modeling of big data. Comprising the following steps: performing feature extraction processing and dimension reduction processing on initial input data by applying a principal component analysis method to obtain a feature vector of the initial input data and principal component input data of the initial input data; training the principal component input data by using a convolutional neural network model, and calculating a total weight matrix of the trained convolutional neural network model; taking a matrix formed by sequentially combining the feature vectors of the initial input data as a weight matrix of the initial input data; and taking a matrix obtained by multiplying the total weight matrix of the convolutional neural network model by the weight matrix of the initial input data as a correlation matrix of the corn starch process sites corresponding to the DE values of the starch milk products. According to the correlation condition of the corn starch process site and the DE value of the starch milk product, a process adjustment scheme can be quickly formed for different raw materials, and the DE value of the starch milk is increased.