43results about How to "Avoid the "curse of dimensionality" problem" patented technology

The invention relates to a quick DP control method applied to PHEV, which adopts a method based on energy to define an on-vehicle storage cell SOC, and derived equations of an SOC state transition equation, a relation equation of the optimal specific fuel consumption and the power of an engine, a quadratic performance index function, a quadratic optimal performance index of the specific fuel consumption, a multiple-message merged equation for calculating the total power required for vehicle running, and the like, wherein the on-vehicle storage cell SOC is defined by the method based on energy; the SOC state transition equation has a linear form; the optimal specific fuel consumption and the power of the engine have a quadratic function form; the performance index function is in a 1*1 dimensional quadratic function form; and the optimal performance index of the specific fuel consumption is in the iteration form of the quadratic analytical function of the SOC. The quick DP control method can quickly calculate the optimal power distribution ratio of an engine / motor and the optimal speed ratio of the speed threshold of a speed changer, realize the economical global optimal control of fuel, and ensure that the on-vehicle storage cell SOC is maintained in the expected operation interval.

The invention provides a method for identifying the failure state of the rotor vibration signal of an aircraft engine. The method mainly includes the following steps of: firstly extracting 11 time domains and frequency domains and characteristics of time and frequency domains of the rotor vibration signal; selecting the proper characteristics by clustering; training a classifier by using the clustering result; at last introducing new data into the trained classifier to generate classification result. According to the method, the problem that, when in existing aircraft engine rotor failure state identification, the lots of extracted diagnosis characteristics have many uncorrelated or uncorrelated due to objective and subjective reasons is overcome. The method of classifying rotor vibration signal faults of the aircraft engine by using the combination of the clustering method and the classifier is adopted to provides a frame of the method for identifying the failure state of the rotor vibration signal of the aircraft engine, the key characteristic sensitive to the fault characteristic can be automatically identified, the correlated characteristics insensitive to the fault characteristic can be removed, and the performance of identifying the failure state of the rotor vibration signal of the aircraft engine can be effectively promoted.

The invention discloses a method for detecting weak small targets for a marine navigation radar, and specifically comprises a training process and a detecting process, wherein the training process includes a sea clutter data phase space reconfiguration and a gray neural network training, and specifically the detecting process is a detection for sea clutter targets. The method is based on the differences between inherent properties of echoes of the sea clutters and inherent properties of the targets, sea clutter data without the targets is used for training the gray neural network, and then under the condition of pure sea clutter, an overall error root mean square value or a compensation error tends to be zero; when the sea clutter data contain targets, an overall error root mean square value and a compensation error are large, thereby the detection for the weak small targets can be performed. Compared with a traditional constant false alarm rate target detection method, the method for detecting weak small targets for the marine navigation radar is capable of detecting the weak small targets under the background of strong sea clutters; compared with a method for detection by a radial basis function (RBF) neural network, the method for detecting weak small targets for the marine navigation radar has a faster training speed, less required sample data information, and has an excellent performance for detecting the weak small targets under the background of the sea clutters.

The invention discloses an industrial production process target data prediction method of a multi-feature fusion deep neural network. Collecting other variable time sequence data related to the key variable in the industrial equipment through equal-interval sampling by utilizing a sensor, and performing predictive analysis on the time sequence data of the key variable in the process industry; inputting into a pre-designed and pre-constructed deep convolutional neural network for training; segmenting historical data of the key variables according to time steps and then inputting the segmented historical data into a deep gated recurrent neural network for learning; a multi-feature fusion method is utilized to fuse output features obtained by two networks and then input the output features into a full connection layer, network parameters are optimized through back propagation, and prediction precision is improved. According to the method, reliable and effective target variable parameter prediction is provided for process monitoring in industrial production, and the hysteresis quality of measurement of key variables such as the silicon content of molten iron in industrial production isrelieved.

The invention provides a cross-visual-angle gait identification method based on tensor simultaneous discriminant analysis. The cross-visual-angle gait identification method comprises steps of constructing a gait characteristic expressed on the basis of Gabor, online model training and an offline test. The cross-visual-angle gait identification method not only adopts a coupling measurement study principle to weak a data heterogeneous problem under the cross-visual-angle, combines with the tensor gait characteristic which is expressed on the basis of the Gabor and the tensor discriminant analysis principle, improves classification performance of the gait and avoids a problem of a small sample because of lack of samples.

The present invention discloses a multi-Agent short-term optimization dispatching method for a hydroelectric station group. The solution speed and operation efficiency of optimization dispatching of the hydroelectric station group are emphatically improved and the problem that an existing technology cannot satisfy short-term optimization dispatching of large-scale hydroelectric station groups is solved. According to the method, the hydroelectric benefit of the whole hydroelectric station group can reach the maximum. The method is of great significance in promoting the development of optimization dispatching of cascade hydroelectric stations and improving the economic operation level.

The invention provides an electrical power system real-time probabilistic load flow online computing method, belongs to the technical field of power grid safety running under the large-scale wind, light and intermittent power supply connection backgrounds and aims at solving the problem that an existing probabilistic load flow online computing method is non-accurate in computing result and causes 'dimensionality curse' of a large number of network nodes of an actual electrical power system, complicated computing process, overlarge hardware and software investment cost and the like. The electrical power system real-time probabilistic load flow online computing method comprises the specific steps of establishing joint probability distribution models among load-side running mode characteristic quantities, obtaining joint probability distribution among the characteristic quantities and sampling the joint probability distribution to obtain a sample set; establishing the joint probability distribution models among intermittent power-supply-side running mode characteristic quantities, obtaining joint probability distribution and sampling the joint probability distribution to obtain a sample set; utilizing load flow high-dimensional models of key nodes and key lines to compute probabilistic load flow distribution and main distribution characteristics of a power grid in a next time period and obtaining a power grid running state. The electrical power system real-time probabilistic load flow online computing method is applied to an electrical power system.

The invention provides a modeling method for power system transient state stability assessment. The method includes the following steps that power generator track information and a transient state calculation result are collected; a feature value of a power generator is calculated; an initial mathematic model X and a transient state stability assessment mathematic model F are formed; dimensionality reduction is conducted on the transient state stability assessment mathematic model F to obtain a two-dimensional standard mathematic model D. Multidimensional counting is conducted on power generator track curves, a statistic curve can reflect main mean values, change ranges and change rate features of each track curve, statistic features are irrelevant to the scale of a system, the dimensionality curse problem caused by increasing of an analysis system in the data modeling process can be effectively avoided, and meanwhile adverse of power generator information loss on calculation can be reduced to the largest extent.

The invention discloses a two-stage stochastic optimal operation method of a hydropower station reservoir based on the residual period benefit function approximation, which comprises the following steps of dividing the hydropower station reservoir operation period into a current stage and a residual period, and constructing a two-stage decision model framework; considering the influence of residual water situation and residual storage capacity on residual benefit, constructing the approximate function of residual benefit; based on stochastic dynamic programming (SDP), proposing a stepwise iterative method to obtain the approximate residual benefit function; according to the actual runoff forecasting level, establishing a two-stage stochastic optimal scheduling model. The method of the invention avoids the dimension disaster of the SDP, takes the artificial neural network as the approximator of residual period benefit function, avoids the artificial assumption of residual period benefitfunction, and can obtain the continuous curved surface of residual period benefit. The two-stage stochastic optimal dispatching model transforms the multi-stage reservoir dispatching decision-makingproblem into a two-stage optimal decision-making problem, which can be directly used to guide the rolling updating of medium and long-term generation plans.

The invention provides a metal surface anti-rust performance test board and an anti-rust performance evaluation method. The test platform can realize an accelerated corrosion experiment on the surface of a metal sample, and the data acquisition platform can perform image data acquisition on the metal sample of the test platform. The method comprises the following steps: analyzing image data obtained by a test board, constructing a set of corrosion region segmentation method, segmenting a corrosion region of a metal surface from an image, calculating a corrosion ratio index, comparing the corrosion ratio index with a preset threshold value, and judging whether a metal sample meets an experiment termination condition or not at the moment; when the corrosion ratio index of the metal sample exceeds a preset threshold value, regarding that the rust resistance test of the sample is completed; and after the whole experiment is completed, counting the time from the first operation of the test platform to the last taking out of each metal sample, regarding the time difference is the quantitative index of the rust resistance test of the metal sample, wherein the longer the time difference is, the better the rust resistance of the metal sample is.

The invention provides a cross angle of view gait recognition method based on two-dimensional coupling margin Fisher analysis. According to the invention, the characteristics of a gait energy graph at different angles of view and the superiority of coupling metric learning in the aspect of cross-domain biometric feature recognition are combined; two-dimensional coupling margin Fisher analysis is provided; data difference of the cross angle of view gait energy graph in matrix space is weakened; local relation among samples is kept; the intra-class divergence is the largest; the intra-class divergence is the smallest; and the cross angle of view gait recognition performance is greatly improved.

The invention discloses an SAR image target recognition method based on kernel fuzzy Foley-Sammon transformation. The method includes the steps that first, an SAR image is stretched according to columns to form row vectors and then dimension reduction processing is performed through principal component analysis; second, a fuzzy K-nearest neighbor algorithm and a fuzzy C-means clustering algorithm are used for achieving fuzzing of data, then, the characteristic vector corresponding to the maximum characteristic value of kernel fuzzy linear discriminant analysis is calculated and serves as the first characteristic vector in an optimal identified vector set of the method, next, the optimal identified vector set of the method is calculated according to the mutual orthogonal rule of neighbor identified vectors, and finally nonlinear transformation of fuzzy Foley-Sammon transformation is achieved through a kernel function. The problem that the linear impartibility problem of the fuzzy Foley-Sammon transformation is difficult to solve is solved, the nonlinear identification information of an SAR image radar target can be extracted, and the radar target identification accuracy rate is high.

The invention discloses an electric vehicle energy storage aggregation modeling method based on a Markov process. According to the technical scheme, the method comprises the following steps: firstly, providing an electric vehicle charge state division method, discretizing continuously changing charge states in a charging process of an electric vehicle, and converting the discretized charge states into a double-layer interval nested discrete structure; then, considering the probability distribution of the battery capacity of the electric vehicle, solving a one-step transition probability of a small interval (an interval obtained by second-layer division) based on a Markov theory, and further obtaining an expected one-step transition probability between large intervals (an interval obtained by first-layer division); and finally, discussing the dynamic change process of the electric vehicle load in the interval according to conditions, deducing a transition probability matrix and a state space expression of an electric vehicle cluster, namely an electric vehicle aggregation model, and verifying the accuracy of the model by using a Monte Carlo simulation method. The method can be used for establishing a state space model of large-scale electric vehicle charge state discretization, thousands of electric vehicles are converted into linear state space expressions with few dimensions, and the dimensions of the expressions are irrelevant to the number of the electric vehicles. Therefore, the dimension reduction of the electric vehicle cluster control variable can be realized, so that the time and space pressure calculated by the control algorithm can be relieved to a great extent.

The invention discloses a dynamic industrial process fault diagnosis method based on a GRU depth neural network. The method divides original data into a plurality of sequence units as the input of theGRU, a GRU network is established through a batch normalization algorithm, extract dynamic characteristics can be effectively extracted from the sequence units, by adopting a softmax regression method, faults are classified according to the dynamic characteristics extracted by the GRU, and the probability interpretation of the classification is provided, so that the diagnosis result is further accurate, the problem of dimension disaster is avoided, and the accuracy of the fault diagnosis of the dynamic industrial process is improved.

The invention discloses a high-dimensional complex system uncertainty analysis method based on statistical machine learning, and the method comprises the steps: selecting uncertainty factors affectinga high-dimensional complex system, and obtaining a high-dimensional random variable input sample matrix; inputting the high-dimensional random variable into a sample matrix, and converting the samplematrix into a low-dimensional random variable sample matrix; inputting the high-dimensional random variables into a sample matrix for one-by-one calculation to obtain an output response quantity matrix; performing accurate modeling on the random response surface agent model to obtain a random response surface model highly approximate to the researched high-dimensional complex system; obtaining amean value and a variance of an output response quantity of the random response surface model by a formula derivation method; and analyzing the uncertainty factors according to the mean value and thevariance to obtain an uncertainty analysis result. The method has the advantages that the calculation result has high accuracy, the calculation amount is reduced and the calculation efficiency is improved on the basis of ensuring the calculation precision, the problem of dimensionality disasters is avoided, the flexibility degree is high and the like.

The embodiment of the invention discloses a traffic signal control method and device, electronic equipment and a readable storage medium, and the method comprises the steps: obtaining the state information of a current stage of a current intersection; receiving a control action of the previous stage of the adjacent intersection of the current intersection; determining an average value of the codesof the control actions of the previous stage close to the intersection; and determining the control action of the current stage of the current intersection based on the state information of the current stage of the current intersection and the average value of the codes of the control action of the previous stage of the adjacent intersection, thereby taking the interaction between all intelligentagents and the traffic flow environment as a random game problem. The dimension of an action space is reduced through average field approximation, the interaction problem of the intelligent agent andother intelligent agents is converted into the interaction problem of a certain average effect of the intelligent agent and an adjacent intelligent agent, and the problem of dimension disasters is avoided.

The invention discloses a multi-kernel hash learning-based large-scale medical image retrieval method. The method concretely comprises the steps of: constructing a kernel matrix through fusion of a plurality of different kernel functions; completely converting an image into hash codes by use of the learned hash function, and carrying out compression on the hash codes; solving distances of medical images through hamming distance measurement, sorting according to an ascending order, selecting m images with minimum distances to be returned to a user; optimizing the sorting of the retrieved images again by the user by use of a relevance feedback algorithm until the sorting satisfies the requirement of the user. The method is high in calculation efficiency, rapid in retrieval speed, small in memory space, high in retrieval precision, clear in steps and strong in pertinence, contributes to medical diagnosis of a doctor, reduces the workload of the doctor and improves the working efficiency.

Provided is a distribution terminal location selection and setting method based on multi-stage coordination. The method comprises the following steps of A, dividing a distribution network into multiple zones, determining annual average power outage times of each zone, and determining a total outage loss cost in a whole life cycle of the distribution network with an average annual outage time as anindependent variable; B, constraining through a power supply reliability index and a staged investment cost index, using a whole life cycle total outage loss cost as an optimization target, and establishing a power distribution terminal dynamic programming model; C, based on the method of a discrete binary particle swarm algorithm, solving a dynamic programming model of a distribution terminal, and obtaining a location selection and setting scheme of the distribution terminal. The method not only improves the economics of the distribution network operation, but also improves the reliability of the distribution network operation.

The invention provides a hybrid recommendation method based on user commodity portrait and potential factor feature extraction. The hybrid recommendation method comprises the following steps of S100,extracting explicit feature representation of a user and a commodity through information of the user portrait and the commodity; s200, mapping the user and the commodity to a potential space to obtainpotential factor feature representation of the user and the commodity; and S300, performing feature extraction on the explicit features and the potential factor features by using a stack type noise reduction auto-encoder to obtain low-dimensional feature representation with higher robustness. According to the invention, the explicit feature space and the potential factor feature space of the userand the commodity are considered at the same time; two feature spaces are comprehensively considered, the defect of a single recommendation model is overcome, the problem of cold start of articles issolved, meanwhile, SDAE is adopted for extracting high-dimensional features, the problem of dimensionality disasters is effectively avoided, and due to the fact that random noise is added in the training process, the robustness of the algorithm is greatly improved.

The invention discloses a state estimation method for a high-accuracy nonlinear system. The method comprises the following steps in sequence: configuring a high-order sampling point which is consistent with nonlinear state transfer function probability distribution; calculating a state vector estimate; calculating a state vector error covariance estimate; configuring a high-order sampling point which is consistent with nonlinear state transfer observation function probability distribution; calculating the estimate of measurement; calculating a filtering gain; calculating a posteriori estimate of a state vector; calculating a state vector error covariance estimate. The method has the advantages that 1, the state estimation accuracy is high; 2, the converging speed is high; 3, resolution of a Jacobian matrix is avoided; 4, redundant weight coefficient debugging distribution is not needed; 5, the problem of dimension disaster is solved.

The invention provides a text vectorization-based handling reference method in a fault power failure first-aid repair event. The method comprises the following steps: 1, obtaining a handled fault power failure event and handling scheme data thereof; 2, preprocessing and word segmentation are carried out on the description text of the handled fault power failure event; 3, performing vectorization representation on the power failure event description text; step 4, power failure event description text semantic similarity calculation and disposal scheme pushing: adopting cosine similarity to calculate description text vector semantic similarity of the newly added to-be-disposed fault power failure event and the stored disposed fault power failure event, and when the similarity exceeds a set threshold, pushing the disposal scheme of which the stock has disposed the fault power failure event to an operator as a reference. According to the invention, the stock-processed fault outage event with high semantic similarity with the newly-added to-be-processed fault outage event description text can be identified, the processing scheme, the processing duration and other information of the stock-processed fault outage event are provided for operating personnel as reference, and the event processing efficiency can be improved.

The invention provides a cross-view gait recognition method based on tensor simultaneous discriminant analysis, which includes constructing Gabor representation-based gait features, online model training and offline testing. The method of the present invention not only uses the idea of ​​coupling metric learning to weaken the heterogeneous problem of data under cross-view angles, but also combines the idea of ​​tensor gait features and tensor discriminant analysis based on Gabor representation to improve the classification of gaits performance and avoid the small sample problem caused by too few samples.

The invention discloses an interpretable mixed type fuzzy system optimization method based on the multi-objective ant colony algorithm. The method comprises the following steps: constructing an interpretable mixed type fuzzy system; constructing an initial reference rule vector by adopting a fuzzy set online clustering updating algorithm; and optimizing system parameters of the mixed type fuzzy system by adopting an improved multi-target leading edge oriented continuous ant colony optimization algorithm. According to the method, the interpretable mixed type fuzzy system is constructed according to the constraint of the uncertain coverage domain of the fuzzy set, so that the generation of a redundant interval type-2 fuzzy set is effectively avoided; an initial reference rule vector is constructed by adopting a fuzzy set online clustering updating algorithm, so that the calculation is relatively simple, and the characteristics of traditional similarity measurement based on a set theory are reserved; and finally, an improved multi-target leading edge oriented continuous ant colony optimization algorithm is adopted to optimize the control performance and the interpretability at the same time, and better balance between the interpretability and the control performance of the fuzzy controller is realized.

The invention relates to a method and system for measuring and calculating the production cost of a power system, which is characterized in that it includes the following steps: 1) setting the iteration cycle; Constrained thermal power unit combination model or safety-constrained economic dispatch model for production simulation simulation calculation; 3) Judging whether the production simulation calculation of the safety-constrained thermal power unit combination model or safety-constrained economic dispatch model in all periods of the iteration cycle is completed, and if completed, output the iteration The production simulation calculation results of the power system to be measured in all periods of the cycle; if not completed, go to step 4); 4) use part of the production simulation simulation results of this iteration as the parameters of the power system to be measured, and enter step 2), The present invention can be widely used in the field of power system production simulation until the production simulation calculation results of the power system to be measured in all time periods in the iterative cycle are obtained.

The invention relates to the field of reservoir scheduling and discloses an orthogonal successive approximation algorithm based reservoir scheduling method. The reservoir scheduling method includes the steps: firstly, determining number of reservoirs, and adopting the orthogonal successive approximation algorithm for calculation if one reservoir is available; otherwise, setting scattering-state step length and scattering number and selecting a proper orthogonal table according to the number of the reservoirs and number of scattering states; determining initial reservoir scheduling trajectories involved in optimal calculation; resolving the original reservoir optimal scheduling problem into multiple subproblems by the successive approximation algorithm; calculating each state combination and performing statistic analysis by a penalty function method according to reservoir condition combination evenly distributed in a constraint space of an orthogonal table structure, and rapidly acquiring the optimal state combination of the subproblems; performing the optimal calculation on the subproblems in sequence, and successively approximating an optimal solution of the original problem. By the application of the reservoir scheduling method, calculation dimensions of the reservoir optimal scheduling problems can be lowered, calculation efficiency can be improved, and the reservoir scheduling method is applicable to a single-reservoir optimal scheduling and group-reservoir optimal scheduling as well as water resource optimal allocation.

The invention provides a neighborhood morphological space artificial immunodetector multi-source generation algorithm, and belongs to the technical field of artificial intelligence immune systems. Theneighborhood morphological space artificial immunodetector multi-source generation algorithm uses a neighborhood morphological space, improves a neighborhood negative selection algorithm, and introduces a chaos theory and a genetic algorithm to form the detector multi-source generation algorithm, thereby solving the problems of each algorithm of a detector in a real-value morphological space.

The invention discloses a dynamic industrial process fault diagnosis method based on the GRU deep neural network. The method divides the original data into several sequence units as the input of the GRU, and establishes the GRU network through the batch normalization algorithm, which can effectively obtain the sequence units from the sequence units. The dynamic features are extracted from the GRU, and the softmax regression method is used to classify the faults according to the dynamic features extracted by the GRU, and the probability explanation of the classification is given to make the diagnosis results more accurate, thereby avoiding the problem of "curse of dimensionality" and improving the accuracy of dynamic industrial process fault diagnosis. Accuracy.

The invention discloses a method and system for judging broadband oscillation stability of a closed-loop power system. The method comprises the following steps: determining a broadband oscillation mode of a first subsystem of a closed-loop power system; determining an interaction variable between the first subsystem and the second subsystem according to a physical structure and an electrical relationship of the closed-loop power system; determining a transfer function model of the first subsystem and a transfer function model of the second subsystem according to the interaction variable; determining a generalized Phillips-Heffron model used for broadband oscillation analysis of the closed-loop power system; determining a generalized torque of the second subsystem to the broadband oscillation circuit; determining the sensitivity of the broadband oscillation mode to the generalized damping torque coefficient and the sensitivity of the broadband oscillation mode to the generalized synchronous torque coefficient; calculating a change value of the broadband oscillation mode of the closed-loop power system; and judging the broadband oscillation stability of the closed-loop power system according to the broadband oscillation mode of the first subsystem and the change value of the broadband oscillation mode of the closed-loop power system.

The invention provides a modeling method for power system transient state stability assessment. The method includes the following steps that power generator track information and a transient state calculation result are collected; a feature value of a power generator is calculated; an initial mathematic model X and a transient state stability assessment mathematic model F are formed; dimensionality reduction is conducted on the transient state stability assessment mathematic model F to obtain a two-dimensional standard mathematic model D. Multidimensional counting is conducted on power generator track curves, a statistic curve can reflect main mean values, change ranges and change rate features of each track curve, statistic features are irrelevant to the scale of a system, the dimensionality curse problem caused by increasing of an analysis system in the data modeling process can be effectively avoided, and meanwhile adverse of power generator information loss on calculation can be reduced to the largest extent.