84 results about "Problem transformation" patented technology

The invention relates to an automatic generating method for a day-ahead plan power flow in a power grid and belongs to the technical field of electric power system scheduling automatization. According to the invention, problems generated by the day-ahead plan power flow are resolved into active subproblems and reactive subproblems which are solved step by step; the subproblems are converted into constraint non-linear optimization problems which are solved by an interior point algorithm; the active subproblems are adopted to preliminarily eliminate mismatching quantity between a system electric generating set power generation plan and bus load projection data and coordinate the inconsistency between the electric generating set power generation plan and a communication section power transmission plan; and the reactive subproblems and active readjustment procedures are adopted to further eliminate the mismatching quantity between a system generating power and a load power, and an active power transmitted by the communication section is controlled according to the plan, and as a result, a reasonable and majorizing day-ahead reactive voltage plan is obtained. According to the invention, a credible day-ahead plan AC power flow can be effectively provided for a day-ahead plan safe checking module. Therefore, the method has favorable reliability and practicability.

The invention discloses a dispute focus automatic identification method based on a hierarchical attention neural network model. The method comprises the steps of firstly, extracting a dispute focus statement of a court for case induction from a document containing a dispute focus induced by the court, and constructing a dispute focus system by utilizing a hierarchical clustering method; and marking a plurality of different category labels for each document by utilizing the dispute focus system, constructing a data set, and converting a dispute focus identification problem into a multi-label multi-classification problem; then, training a hierarchical attention neural network model, paying more attention to important words, sentences and paragraphs containing more information, and forming adispute focus recognizer; and finally, inputting the text of which the dispute focus needs to be identified into a dispute focus identifier to obtain the dispute focus of the input text. The method ishigh in prediction accuracy, can accurately identify and judge the dispute focus of the document, and has good expandability.

The invention discloses a moving object destination prediction algorithm. The method comprises the steps that firstly, an improved track segmentation method with the minimum description length (IMDL)is provided, weight parameters are introduced, meanwhile, a Dijkstra algorithm is used for converting the track segmentation problem into the shortest path problem of an undirected graph, the optimalsegmentation of a track is obtained, and therefore the track is simplified; then, a deep learning prediction model EP-LSTM based on embedded processing and long-short-term memory fusion is provided, and a high-dimensional input vector is converted into an embedded vector to serve as input of the model. The method has the advantages that the problems of data sparseness and long-term dependence in moving object destination prediction are effectively solved, and an excellent prediction effect is achieved by performing a large number of experimental verifications on a real data set.

A method of realizing coordination control of a power-distribution-network static var compensator and a static synchronous compensator belongs to the electric power system technology field. A current method used for carrying out coordination control on a DSVC and DSTATCOM interaction influence problem is complex. By using the method of the invention, the above problem is solved. The method is characterized in that a coordination problem between the controllers is converted into a multi-target optimization problem; a novel multi-target heredity algorithm with a great influence is introduced, wherein the algorithm is a rapid non-dominated sorting genetic algorithm with an elitist strategy; parameters of the power-distribution-network static var compensator and the static synchronous compensator are coordinated. Compared to a traditional multi-target algorithm, by using the method of the invention, an obtained Pareto optimal solution is distributed uniformly; convergence and robustness are good. The method can be used for solving a coordination control problem of the power-distribution-network static var compensator and the static synchronous compensator.

The invention relates to a method for extracting characteristics of a three-dimensional model based on a section ray, which aims at the problems of a high calculation complexity, a larger distortion, not having a common research method and being difficult to obtain higher research precision in the existing method for extracting the characteristics of a mode based on contents. The method takes distance sets from specific points on a principal axis section of the model to the model surface as a characteristic description of the model, thus a problem of the model matching is transformed into the problem of comparing a group of distance sets. The method is divided into two sections of characteristic extraction and characteristic matching; when extracting the characteristics, the method can simply and rapidly obtain the model characteristics which have small reconfiguration distortion and can correctly describe a stereochemical structure and structural characteristics of the three-dimensional model. The method has the advantages of simplification, wide applicability, good research quality and high efficiency, therefore, the method has higher practical applicability and wider application range.

The invention belongs to the field of knowledge graph automatic question answering, and particularly relates to a knowledge graph common sense question answering method and system based on staged query, and the method comprises the steps: obtaining a question statement, preprocessing the question statement, and converting the question statement into a question sequence; inputting the question sequence into a trained improved question and answer model to obtain a question and answer result, wherein the improved question and answer model comprises an entity recognition model, a constraint language recognition model and a question structure classification model. According to the method, complex questions or simple questions are converted into the question semantic structure tree based on the statement structure, the answers to the questions are searched step by step by updating the undetermined nodes, and the searching process is simplified from complexity to simplicity.

The invention discloses a multi-objective reactive power optimization method for solving a Pareto optimal solution set. The method of the invention converts the problem into a problem of solving reactive power optimization containing two objective functions, boundary points of the two objective functions are respectively solved directly in a tangent method, an inequality constraint method is then adopted to solve Pareto frontier formed by the two objective functions, during the calculation process, algorithm for solving a single-objective reactive power optimization model adopts a primal dual interior point method, and a point for an initial value adopts a result of power flow calculation. The multi-objective reactive power optimization method for solving a Pareto optimal solution set has the advantages that optimization of multiple objectives is comprehensively considered during the reactive power optimization process, subjectivity generated when a single objective reactive power optimization model is built by other algorithms can be effectively avoided, the distribution condition of each objective function can be intuitively seen from the Pareto frontier surface, and a reasonable and reliable decision which can meet needs of system operation can be made conveniently by a decision maker.

The invention provides a counter-fact prediction method about a set type decision effect, and belongs to the technical field of machine learning. According to the method, a decision variable and confusion variable decorrelation problem is converted into a lower-dimension decision variable implicit representation and confusion variable decorrelation problem, a probability density ratio estimation method based on a deep neural network, is adopted to take the probability density ratio of the joint distribution of the decision variable implicit representation and the confusion variable corresponding to the observation data sample to the joint distribution of the decision variable implicit representation and the confusion variable unassociated as the weight of a data point formed by the decision variable implicit representation and the confusion variable. A variational sample reweighting method is adopted to synthesize the weight of a data point formed by decision variable implicit representation and confusion variables into the weight of a sample in observation data, and the weighted observation data sample is utilized to train a counter-fact prediction model to perform counter-fact prediction on the effect of an individual under the influence of a specific decision. According to the invention, the accuracy of counter-fact prediction is improved, and the method has a very high application value.

The invention provides a neutral network extreme value control method based on chaotic annealing and parameter perturbation, and a system thereof. The control method transforms a control problem of an extreme value search system to solve an extreme value point problem with zero rate of slope in an output function of a controlled system. According to the extreme value point solving problem, a pair of dual problems with constraint conditions are constructed. A neutral network solving dual problem based on chaotic annealing and parameter perturbation is established, which comprises a chaotic annealing initial search phase, a parameter perturbation middle search phase and a final search phase of a natural network. The global optimum search variable can be obtained by the solving of the neutral network extreme value control method. According to the obtained global optimum search variable, the output value of the extreme value search system is driven to converge to a global extreme value point of the output function, thus realizing the control purpose of the extreme value search system. The control system provided by the invention is divided into a module simulation mode and a real time control mode and realizes the application of the control method in the extreme value search system respectively from two aspects of off-line module simulation and real time system control.

A genetic algorithm-based routing method under the constraints of designated points is used to solve the technical problems of long search time and heavy path weight in routing selection in communication. The implementation steps are: 1) Read the original network graph information to obtain the adjacency matrix of the network graph and the set of specified points; 2) Preprocess the preprocessing of the network graph to find the starting point, end point and the relationship between the specified points 3) Encoding according to the starting point and specified point information of the network graph, thus transforming the problem into a TSP-like problem; 4) Solving the TSP-like problem by using the improved genetic algorithm to obtain a starting point, an end point and a specified point 5) judging the legitimacy of the path, performing correction processing, and finally obtaining a complete routing path. The invention can effectively reduce the solution space of routing path search, and obtain a better routing path in limited time and resources.

The invention discloses a method for evaluating a link quality assessment model by a multi-attribute group decision making theory. The method comprises the following steps: constructing an evaluation index system of the link quality assessment model by taking a packet reception rate as a stability standard of the assessment model and taking a signal-to-noise ratio as an agility standard of the assessment model; evaluating the assessment model in different link states, and calculating the influence degrees of the link states on the performance of the assessment model; assessing the influence degrees of evaluation indexes on the performance of the link quality assessment model, and calculating weights of the indexes; and transforming an optimum seeking problem of the link quality assessment model into a multi-attribute group decision making problem, fusing the plurality of evaluation indexes by a positive-negative ideal solution, and quantifying and optimizing the performance of the link quality assessment model. Through adoption of the method, the evaluation index system of the link quality assessment model is constructed; the performance of the link quality assessment model is quantified by the multi-attribute group decision making theory; and sequencing optimization is performed on the assessment model according to a quantification result.

The invention discloses a business data processing model training method and a business data processing method and device, relates to the computer data processing field. The model training method at least comprises the steps of obtaining user information samples of a plurality of overdue repayment samples in a first time period and overdue urging days in a second time period, and extracting a target input variable based on the user information samples; classifying the users according to the overdue urging days to obtain a user classification result; and training by taking the target input variable as input and the user classification result as output to obtain a business data processing model. According to the business data processing model training method, the obtained sample data is comprehensive, and the accuracy of the model obtained through training is high; complex and diversified problems are converted into dichotomy problems in supervised learning, the problems are simplified,and the model training difficulty is reduced; according to the business data processing method based on the model, user risk level classification can be carried out for overdue customers under different conditions to provide data support for decision makers.

The invention relates to a design method of motor optimization based on Game Theory, which comprises the following steps: (1) determining variables to be optimized in motor design and establishing a partial objective functional equation and a constraint equation; (2) determining a noninferior solution set which is then taken as a strategy set in Game Theory; (3) taking an optimization objective as a player and taking a partial objective function and taking definite purpose as utilities of each player, disintegrating design variables into strategies for each player according to the correlation between the utilities and each partial objective function, and changing the solving problem of the optimal motor design scheme into a Game problem according to the above steps, wherein each player can cooperate with each other so as to maximize the utility of each player; and (4) adopting a method of cooperative Game for integratedly optimizing each optimization objective from Nash equilibrium solution, thus obtaining the global optimum solution. The design method of motor optimization fully takes the relation among different optimization objectives into account so as to search the integrated optimum solution, thus shortening the motor design cycle, lowering development difficulty and laying foundation for the intelligentization of motor optimization design.

The invention belongs to the technical field of social network entity anchor link identification, and particularly relates to a heterogeneous social network user anchor link identification method based on meta-paths. In view of a scene of multi-entities, complex link relations and a small number of label entities in a heterogeneous social network, the problems of non-uniform data distribution, feature selection and the like are avoided based on unsupervised learning, and the link relationship related to a user is fully mined by utilizing the link relationship and attribute information of the user entity in the social network and adopting a meta-path technology. By means of definitions of a cost function, a matrix norm, an objective function and the like, a user anchor link identification problem is converted into a problem of optimizing an objective function.

The invention discloses a convex optimization and graph theory-combined generalized rank target sensing network spatio-temporal data positioning method and belongs to the technical field of wireless sensor networks. According to the method, an SDR and S-process convex optimization technology is adopted to perform modeling analysis; a nonlinear non-convex problem is modeled into an SDP convex optimization problem; convex optimization and a graph theory are combined to transform a target-measurement correlation unknown problem into a standard weighted complete bigraph (SWCB) matching problem, aconvex optimization positioning model is constructed; and a linear programming technique is adopted to perform positioning solving according to a transformation result. Experiments show that the performance of the method is significantly better than that of a conventional positioning algorithm under low-SNR and small-snapshot number scenarios; the algorithm has obvious advantages in tracking performance and can significantly improve the quality of multi-target tracking; the practical level of the spatio-temporal data mining target sensing method is improved; and theoretical support can be provided for solving the target sensing positioning problem of the spatio-temporal data of the sensor networks.

The invention discloses a sensitive data detection method and device based on machine learning. When data is synchronized from a production environment to a development test environment, sensitive fields must be desensitized. According to a machine learning principle, a data table field sensitivity identification problem is converted into a text classification problem, and an NLP technology is applied to train a model to identify sensitive fields. The method is combined with a conventional detection means, the recognition effect is continuously improved through a self-learning algorithm, the risk that sensitive fields are omitted to the external environment can be reduced, manual intervention is reduced, and the working efficiency is improved.

The invention discloses a dynamic target traversal access sequence planning method and system. The method comprises the following steps: giving an initial position and a terminal position of each target and a moving path of each target in a planning time period; converting a moving target access sequence planning problem into an intermittent moving target access sequence planning problem through adiscrete time axis, and constructing a problem model; solving the intermittent moving target access sequence planning problem model by adopting a dynamic target sequence planning ant colony algorithmto obtain an optimal target access sequence of the intermittent moving target access sequence planning problem; fixing the target access sequence so that an original moving target access sequence planning problem is converted into a sequence fixed access time planning problem, creating and solving a problem model to acquire the optimal access time of each target and finally, acquiring the optimalsolution of the original moving target access sequence planning problem. Optimization efficiency and the optimization success rate of the moving target access sequence planning problem are effectively improved.

The invention provides a multi-branch power flow transfer search algorithm based on a virtual branch model and FTIL. The method includes the following steps: I, giving a virtual branch model of powerflow transfer under the condition of multi-branch breaking, and deriving a distribution factor calculation formula after multi-branch breaking based on a classic power flow foldback process; II, giving the definition of a generalized power flow transfer severe area based on FTIL, and converting the determination of a transmission section into the definition of a generalized power flow transfer severe area; III, giving a power flow transfer search algorithm based on an improved BFS, and introducing FTIL as a heuristic item to the termination condition of BFS to make the search automatically avoid areas with weak FTIL; and IV, carrying out multi-branch braking experiment on a new England 39 node system, obtaining the power flow increase of other branches in the network through PSASP simulation, and determining whether the error between the transferred power flow on the branches and the power flow increase obtained by PSASP software simulation is reasonable through analysis and calculation.

The invention discloses a method for calculating the zero-sequence impedance of a distribution transformer, and the method comprises the steps: obtaining an operation electrical quantity of the distribution transformer through a station zone collection device during three-phase load unbalance, building a voltage loop equation set of a zero-sequence impedance to-be-solved quantity, converting a solving problem of the voltage loop equation set into an optimization problem of a target function, and calculating the zero-sequence impedance of the distribution transformer through a genetic algorithm. Aiming at a condition that the zero-sequence impedance of the distribution transformer is usually not tested during a delivery test, the nonlinear equation set comprising the zero-sequence impedanceto-be-solved quantity is built on the basis of the existing operation data of the distribution transformer in a station zone. Considering that the nonlinear equation set is difficult to solve througha common numerical method, the method enables the solving problem of the nonlinear equation set to be converted into a problem of target optimization, and the genetic algorithm is used for solving the target optimization function. Finally, the zero-sequence impedance of the distribution transformer is solved under different zero-sequence currents.

The invention discloses an ICT infrastructure site selection method based on variable neighborhood search, and the method comprises the following steps: 1, obtaining different double-target set coverage problem modeling and problem abstraction according to different ICT infrastructure site selection problems, and obtaining a target function of a single-target optimization problem; 2, solving an objective function of a single-objective optimization problem by utilizing a variable neighborhood random search method; according to the invention, a complex double-target set coverage problem based onan ICT infrastructure site selection background is converted into a relatively simple single-target optimization problem, and then an efficient solving algorithm based on variable neighborhood searchis designed according to the problem background, so that the situation that only expert experience can be relied on is changed into the situation that automatic calculation can be conducted dependingon a computer. And due to different neighborhoods and search strategies in variable neighborhood search, compared with other local search algorithms, the algorithm can search a better deployment scheme from ICT infrastructure deployment schemes.

The invention discloses a nonlinear interval uncertainty optimization method based on an affine algorithm. The nonlinear interval uncertainty optimization method comprises the following steps: converting an interval uncertainty quantity into an affine type uncertainty quantity; processing the nonlinear objective function and constraint by using an affine algorithm to obtain an interval of the objective function and constraint at a design variable; converting the non-linear interval uncertainty optimization problem into a single-layer deterministic optimization problem; and adopting a genetic algorithm to solve the optimization problem. According to the method, the nonlinear interval uncertainty optimization problem can be converted into a single-layer certainty optimization problem, and the solving efficiency is improved.

The invention discloses a method for solving a time-varying continuous algebraic Riccati equation based on an error adaptive coefficient zero neural network, which comprises the following steps: firstly, abstracting and modeling original problems, such as mechanical arm motion planning, image and signal processing and the like, and converting the problems into a mathematical model for solving the time-varying continuous algebraic Riccati equation; converting the mathematical model into a time-varying matrix linear equation problem through derivation and Kronecker product, using an error-based adaptive coefficient zero neural network for iterative solution, performing mapping and adaptive transformation continuously on system residual errors and state variables in the solution process, and adjusting the solution strategy of the network; finally, stopping iteration and outputting the model system when the model system obtained by iteration solution meets predefined conditions and precision. Compared with other traditional methods for solving the time-varying continuous algebraic Riccati equation, the algorithm provided by the invention has higher convergence precision and robustness.

The invention discloses a hyperspectral endmember extraction method and a hyperspectral endmember extraction device based on multi-objective differential evolution of sorting multiple variations. According to the method, a hyperspectral end member extraction problem is converted into a multi-objective optimization problem; conflict among multiple targets is balanced through a sorting multi-variation (mu + lambda) multi-target differential evolution algorithm; the method specifically comprises the following steps: randomly initializing a population through integer coding; generating a variationvector through a scaling factor parameter pool by adopting a multi-variation strategy operation; adopting a binomial crossover operation to generate a test vector through a crossover control parameter pool; selecting the progeny population by combining a rapid non-dominated sorting method and (mu + lambda) selection operation, and carrying out multiple generations of evolution by repeating the variation, crossover and selection operation to obtain a group of non-dominated Pareto solution sets, so that a group of hyperspectral end member extraction results are obtained, and the extraction effect of the hyperspectral end member can be improved.

The invention belongs to the technical field of computer security, and particularly relates to a malicious software classification method based on PE head visualization. According to the malicious software classification method, the PE head of the malicious software can be visualized, and the PE head is converted into the grayscale image, so that the malicious software classification problem is converted into the image classification problem. According to the malicious software classification method, the malicious software is classified by using the grayscale image of the malicious software and the fields in the PE head, and a new method for representing the malicious software by using the selectable head fields in the PE head is provided, so that the malicious software classification method has relatively high classification precision and relatively high classification speed.

The invention relates to a feasible region projection equivalence method and system. The method comprises the steps of establishing constraint conditions of a distributed energy aggregator model according to line parameters of a region where a distributed energy aggregator is located and distributed energy types; defining a time domain coupling feasible region corresponding to a constraint condition of the distributed energy aggregator model, and then converting a time domain coupling feasible region solving problem into an integer linear programming solving problem through dual transformation and a large M method; screening and eliminating redundant constraint conditions in the constraint conditions by adopting a parallel umbrella constraint algorithm, and identifying effective boundary conditions which play a role in the initial time domain coupling feasible region; and depicting an extreme point of the dual space of the feasible region by using an external approximation algorithm to obtain a precise time domain coupling feasible region corresponding to the distributed energy aggregator. According to the technical scheme, the calculation scale is reduced, the calculation efficiency is improved, and meanwhile the precision of the feasible region is effectively improved.

The invention discloses a threat intelligence named entity recognition method based on machine reading understanding, which comprises the following steps: performing sentence segmentation processing on threat intelligence, filtering sentences which do not contain network security professional vocabularies based on a network security professional lexicon, and obtaining a sentence set after filtering; sentences in the sentence set are taken one by one, and each entity in the sentences is marked with a question and answer pair; training a recognition model by using the labeled sentence set; and performing named entity recognition on the threat intelligence by using the trained recognition model. The threatening intelligence named entity recognition based on machine reading understanding can effectively solve the problems of fuzzy classification of threatening intelligence entities and entity nesting; the built problem is provided with entity hidden information, so that the recognition accuracy can be effectively improved; according to the method, entity recognition is converted into a classification matching problem from a sequence labeling problem, so that one sentence with a plurality of entities can generate a plurality of training samples, and the requirement on the number of sentences is reduced.

The invention provides a lane line detection method, electronic equipment and a vehicle, and the method comprises the steps: inputting a rasterized vehicle foresight picture into a pre-trained model based on deep learning, converting a lane line instance segmentation problem into two sub-tasks: category perception prediction and segmentation mask prediction of grid units, and if a lane line falls in some grid units, the grid units are responsible for predicting classification information and segmentation information of the lane line at the same time; and the classification information and the segmentation information output by the model are simply processed to directly output an instance segmentation result of each lane line due to the fact that the grid units in the picture correspond to lane line instances; and therefore, the complex post-processing step is omitted, a large number of computing resources are saved, and the method can be deployed on low-cost vehicle-mounted embedded equipment on a large scale.

The present invention proposes a layout analysis method and system for automatic classification of test paper content. The method includes: obtaining an input document image; extracting connected components of the document image to form an original set of connected components; The connected components are classified into text and non-text, and the first set of text connected components and the set of non-text connected components are obtained; for each connected component in the set of non-text connected components, the text components are detected and segmented, and the connected components of the non-text classification are obtained. The text components in the connected components, and add this component to the first text connected component set to obtain the second text connected component set; for each connected component in the second text connected component set, carry out the classification of printed characters and handwritten characters ; Output the classification result of the document image content. By adopting the method of the invention, the classification problem of elements is transformed into a global optimization problem for solving the maximum joint probability of all elements, so that the overall classification accuracy rate can be improved.

The invention provides a data processing method in learning modeling, and belongs to the technical field of Python machine learning modeling, and aims to convert a regression problem with a large amount of required information into a multi-classification problem with a small amount of required information by utilizing log transformation and equal-width standardization on the basis of a power law relationship and aiming at a y variable conforming to long tail distribution, so that subsequent modeling work can be carried out smoothly.