48 results about "Computational probability" patented technology

Systems and techniques for are described herein. A content passage is received from a corpus of training data comprising labeled training data and unlabeled training data. A query is received from a query hierarchy for a classification domain. The content passage and the query are embedded to form a passage-query pair. A predicted result for the passage-query pair is generated based on a calculated probability of the predicted result being within an answer threshold. A passage-query-result triplet is generated that comprises the passage-query pair and the predicted result according to the query hierarchy for the classification domain. Vectors of the content classification large language model are updated using the passage-query-result triplet.

The invention discloses an electric power work order visual management early warning detection system and method, and the system comprises a data preprocessing module which is used for preprocessing historical work order data; the first model training module is used for training a first model based on the preprocessed historical work order data to extract features; the feature fusion module is used for fusing features extracted by the first model to construct a prediction model; the real-time data processing module is used for inputting the real-time data into the prediction model, extracting the most probable early warning label and calculating a probability difference; the result output and model optimization module is used for outputting a result and iteratively optimizing a prediction model according to real-time data; wherein the prediction model adopts a uniflow model to fuse text, audio and structured features, and optimization is carried out through a Transform layer and cross entropy loss. According to the system, the data quality is improved, text semantics are accurately captured, time sequence and local features are mined, real-time early warning is achieved, a model is dynamically optimized, and efficient operation and maintenance of power enterprises are assisted.

The invention relates to a missing mode robustness enhancement training method based on multi-mode probability shape-preserving distillation, and the method comprises the following steps: obtaining a first multi-mode training set with a complete mode, defining an auxiliary indication vector to represent whether the mode is missing or not, and constructing a second multi-mode training set with mode missing; respectively mapping each input mode in the second multi-mode training set into independent D-dimensional normal distribution, and fusing to obtain fused D-dimensional normal distribution; end-to-end distillation is performed based on a self-distillation framework, in a training stage, a teacher model processes a first multi-modal training set and provides complete modal representation, and a student model processes a second multi-modal training set and provides complete modal representation supervision. And on the basis of the independent D-dimensional normal distribution of each input mode, the D-dimensional normal distribution and the complete mode representation are fused to calculate probability extremum optimization loss and geometric consistency optimization loss, and a loss function is constructed to carry out model training. Compared with the prior art, the method has the advantages that the effectiveness of information migration can be improved, and the processing capacity of the model to the missing mode is improved.

The invention provides a game situation deduction and path reasoning method and device based on a dynamic knowledge graph, and a storage medium, and the method comprises the steps: collecting multi-source game data, and obtaining a multi-modal data set; establishing a dynamic knowledge graph according to the multi-modal data set; based on the dynamic knowledge graph, utilizing a Bayesian network to carry out probability dependency relationship modeling to obtain a probability graph model; calculating the confidence of each path in the probability graph model according to parameters corresponding to a set initial event, and generating a high-confidence reasoning chain set; performing situation evolution prediction on the key indexes in the high-confidence reasoning chain set by using a pre-trained time sequence model to obtain a plurality of situation prediction results; performing multi-scene simulation on each situation prediction result, and determining evaluation information of each situation prediction result; and generating a Pareto optimal strategy path set according to the evaluation information of each situation prediction result. By means of the scheme, the intelligent level of game situation prediction can be improved, and more reliable and efficient technical support is provided for various complex game decision-making scenes.

The invention belongs to the technical field of path planning, and particularly provides a multi-agent path planning method based on an attention mechanism, and the method comprises the following steps: S1, selecting path planning training data, and constructing a training set; s2, respectively designing an encoder and a decoder by using an attention mechanism, and constructing a path planning model based on an encoder-decoder structure; s3, designing a reinforcement learning framework based on a rollback reference, and constructing a multi-agent task-oriented reward function; and S4, inputting the data set obtained in the S1 into the path planning model constructed in the S2, training the path planning model by using the reinforcement learning framework designed in the S3, and applying the trained model to multi-agent autonomous path planning. According to the multi-agent path planning method, node high-dimensional feature representation is obtained through an encoder, probability distribution is calculated through a decoder, the model is trained through a reinforcement learning framework combined with a rollback reference, the multi-agent path planning cost is reduced, and the path planning quality is improved.

The invention provides an urban power distribution network probabilistic load flow calculation method for large-scale access of demand side resources, and belongs to the technical field of power system power distribution network analysis and optimization. The method comprises the following steps: constructing a dynamic Copula model in combination with space-time sequence analysis, introducing a Markov chain, and carrying out multi-dimensional uncertainty coupling modeling; a probabilistic power flow-multi-objective optimization model is constructed, and energy router cooperative control is carried out based on distributed model predictive control; sparse expansion is carried out by adopting error feedback adaptive sampling and sparse Bayesian learning in combination with polynomial chaos expansion, and probability power flow is calculated through OpenDSS; establishing a three-phase probabilistic power flow model and a DG-EV-DR collaborative probability model; and evaluating the extreme scene risk, and visualizing the result through a digital twin platform. According to the method, the accuracy and practicability of distribution network probabilistic load flow calculation in a demand side resource large-scale access scene are effectively improved, and a risk quantification basis is provided for power grid dispatching.

The invention provides a deep neural network model fingerprint generation and verification method based on decision consistency, and the method comprises the steps: inputting an original sample to a source model, generating a boundary sample through a gradient-based optimization method, carrying out the feature shielding of the boundary sample through a random binary mask matrix, so as to generate a fingerprint sample set, the size of the mask matrix is dynamically adjusted according to the feature complexity of the data set; respectively inputting the fingerprint sample set into a source model and a suspicious model, obtaining output probabilities of boundary samples and fingerprint samples of the two types of models, calculating a probability difference, and respectively calculating feature attribution weight matrixes of the two types of models through regularization linear regression; carrying out binarization processing on the weight matrix according to positive and negative of elements; obtaining binary weight matrixes corresponding to the source model and the suspicious model, and calculating element-by-element decision consistency similarity of the source model and the suspicious model; and when the decision consistency similarity does not exceed a preset threshold, determining the consistency of the two models.

Provided is a method for correcting bias introduced by weighted training in machine learning, comprising: labeling the number of examples of each class in weighted data used by a machine learning classifier; adding a weight to training data, and calculating a weight wij of each data example j in class i of the training data according to a user-given data weighting method; calculating a mean weight wi for examples of each class; conducting classification and logistic regression against features of the examples in the weighted data and labels corresponding to the features; after training, when calculating probabilities Pw(i) of class after machine training using the machine learning classifier, correcting the probabilities P(i) by applying a deweighting formula to obtain corrected probabilities P(i); making a classification decision based on the corrected probabilities P(i). The method improves the accuracy of classifiers in assigning probabilities to new data in machine learning applications.

The invention provides a method for automatically labeling work order types on the basis of agents, which comprises the following steps of: performing punctuation standardization processing on an original work order, and converting spoken and non-standardized work order texts into segmented word segments conforming to field specifications in combination with word segmentation in a power field dictionary; unifying and normalizing the segmented word segments through a preset synonym mapping table to obtain a standardized text sequence; the standardized text sequence is input into a bidirectional encoder expression model to output a semantic vector sequence, and deep semantic understanding of the work order text is achieved; related external information is called to be coded into a feature vector, and then the feature vector is fused with the semantic vector sequence through an attention mechanism to generate an enhanced semantic vector; multi-dimensional label prediction tasks are executed in parallel based on a multi-task learning architecture, and multi-class labels and probability distribution are output; the confidence coefficient is obtained by calculating the maximum value of the probability distribution, and the preset process is executed, so that automation and quality management and control of label generation are realized, and the problem of low efficiency of manual power work order processing in the prior art is solved.

The invention discloses an end-to-end sparse trajectory recovery method and device based on road network constraint, electronic equipment and a storage medium, and solves the problem of error accumulation of an existing map matching and trajectory completion two-stage method. The method comprises the following steps: acquiring a sparse trajectory and road network data; and constructing a unified end-to-end deep learning model, and deeply coupling map matching and track completion tasks. The core is that after a candidate road segment set is generated for each GPS point and the probability is calculated, a differentiable virtual road segment embedding mechanism is introduced, and a gradient return channel from track completion to map matching is constructed through probability weighted candidate road segment embedding. The trajectory decoding module fuses the spatio-temporal features with virtual road segment embedding to generate a complete trajectory. The model performs end-to-end optimization through a joint loss function, so that a trajectory reconstruction supervision signal directly guides an upstream candidate selection process. According to the method, error accumulation caused by one-way information transmission in a traditional method is avoided, and the global precision, continuity and robustness of track recovery are remarkably improved.

The invention discloses a named entity recognition method and system based on multi-target learning, and relates to the technical field of natural language processing. The method comprises the steps of obtaining a source text sequence and a target text sequence; according to the source text sequence, the prediction language model and double affine transformation, constructing a span representation matrix to obtain a representation coding layer; according to the span representation matrix, the inter-span relative position information and the matrix region information, span feature representation is constructed by using cavity convolution and a nonlinear activation function, and a span dependency layer is obtained; according to the span feature representation and a multi-layer perception mechanism, calculating a span category probability to obtain a span classification task; and according to the span feature representation, the automatic regression bidirectional transducer and the multi-layer perception mechanism, calculating a probability maximum value, obtaining a boundary generation task, finally obtaining a multi-target learning named entity recognition model, and performing named entity recognition on a target text sequence. According to the method, the problem that span boundary feature representation is fuzzy due to context sharing is solved.

The application discloses a kind of electric information collection intelligent configuration method, it is related to power distribution automation technical field, including, convergence equipment account, topology, historical operation and configuration record are standardized, form data set and candidate template;Extract operation, topology, equipment and alarm features, build comprehensive feature vector;With comprehensive features and candidate template respectively by SVM, random forest and CNN calculate matching probability, form probability matrix;With odds ratio, consistency coefficient and historical success rate fusion calibration, get robust matching probability;Sensitivity detection is carried out to core operation index, together with matching probability determines optimal template, and according to three levels segmentation, in turn, under the parameter of roll-back is sent out.This application determines optimal template by the ingenious combination of the probability calculated by three models, ensures the reliability and accuracy of optimal template.

The application discloses a simulation job scheduling method and system based on a hybrid expert model and a medium, and relates to the field of data processing.In the method, the job parameters of a simulation job to be scheduled and host state information of computing nodes in a cluster are acquired; a feature vector representing job features and cluster states is generated based on the job parameters and the host state information; the feature vector is input into a routing module in the hybrid expert model to obtain the probability and probability distribution of the simulation job relative to a plurality of preset subject experts; the information entropy of the probability distribution is calculated, and the information entropy is compared with a preset information entropy threshold; if it is determined that the information entropy is less than the preset information entropy threshold, a target subject expert with the highest probability is determined from the plurality of preset subject experts to obtain a scheduling decision; and the scheduling decision is sent to a job executor on a target host, so that the job executor executes the simulation job according to the scheduling decision.Implementation of the technical solution provided by the application improves the accuracy of the scheduling decision.

The invention is suitable for the technical field of well drilling safety monitoring, and provides a well drilling overflow prediction method and system based on an expert network model, and the method comprises the following steps: carrying out the Fourier transform of preprocessed multi-dimensional time series data, dividing a frequency range into a plurality of non-overlapping frequency bands, carrying out the frequency domain mask processing, and obtaining a frequency domain mask; recovering the signals into time domain signals corresponding to the frequency bands; respectively inputting the time domain signal of each frequency band into a corresponding expert network model, and outputting a feature vector of the frequency band; global pooling is carried out on feature vectors output by the expert network models, and probability weights are calculated; according to the feature vectors output by the expert network models and the probability weights corresponding to the feature vectors, carrying out weighted fusion to obtain fused feature vectors; and inputting the fusion feature vector into a classification head, and outputting an overflow prediction probability. According to the method, through collaborative learning of frequency domain decomposition and the expert network, random noise and burr interference are effectively suppressed, and the reliability and interpretability of overflow prediction are remarkably improved.

A method includes obtaining a first input instance and a first class corresponding to the first input instance, from an input instance dataset. The method further includes calculating a probability distance score based on a set of probability values by a distance function generator to transform the first input instance. The method further includes processing the first input instance and the probability distance score by a counterfactuals generator to generate a counterfactual instance. The method further includes executing a machine learning model to process the counterfactual instance and assign a second class, different from the first class, to the counterfactual instance.

The invention discloses a long-tail target detection method based on expanded data feature distribution variance, which solves the problem that inter-class and intra-class feature distribution in a long-tail distribution data set is unbalanced, and comprises the following steps: acquiring the long-tail distribution data set; the method comprises the following steps: extracting original backbone features by adopting ResNet-50-FPN; through a feature cluster compression method, the tail class cluster is more compact by using an adaptive scaling factor, and the separation degree of the tail class cluster and the head class cluster is kept; calculating a probability weight based on the number of category samples, and introducing a dynamic bias term to expand a feature distribution variance; the offset is adjusted in combination with the Euclidean distance, and the dispersion and stability of the sample are balanced; the processed features are used for model training, and the original features are input during testing to obtain a detection result. According to the method, through a cooperative strategy of feature cluster compression and bias term calculation, inter-class discrimination is enhanced, intra-class feature diversity is improved, the long tail problem is effectively relieved, and the detection precision and generalization ability of the model to the tail class are improved.

Systems, methods, and computer-readable media for providing low-latency markets for wagering on sporting events are disclosed. In some embodiments, an outcome matrix may be generated by a plurality of contest simulations utilizing statistical data of a history of sporting events. Outcome data of the outcome matrix may be indicative of probabilities of events occurring during an upcoming contest. The probabilities may be calculated, and markets may be priced based on the calculated probabilities. The priced markets may be provided to users by a graphical user interface by user computing device. Furthermore, users may request user-requested markets by inputting different markets into the GUI. The user-requested markets may then be priced using the outcome data of the outcome matrix and stored calculations thus, providing new markets based on the user-requested markets from the outcome matrix without generating new simulations.

A method to provide a model-assisted inline wafer-level inspection during high volume manufacturing is disclosed. More particularly, a method for using a computational model to generate fingerprint wafer defect maps and then guide wafer selection for inline inspection is disclosed. A computational probability prediction model is disclosed to generate defective die probability estimates with improved accuracy and versatility to guide different wafers for inspection.

The invention discloses an Ethereum network account classification method and device based on a graph neural network, and the method and device can capture a low-order neighborhood mode and a high-order neighborhood mode at the same time, improve the richness and discrimination of node representation, achieve the local adaptive feature transformation, effectively cope with challenges caused by the diversity of feature distribution in a disparate graph, enhance the discrimination capability of account behavior modes, and improve the user experience. Key modes and features in account prediction are better captured, and the prediction accuracy is improved. The method comprises the following steps: (1) representing an account and a transaction behavior as a graph structure; (2) inputting the feature information into a KAN to extract Fourier features of a transaction, extracting a global feature mode of an account, and enhancing periodic information of transaction data; (3) after feature enhancement is obtained, performing neighbor aggregation by using GraphSAGE, and extracting an account low-order neighborhood mode; and (4) inputting the last layer of GraphSAGE for classified output, and calculating probability distribution through softmax.

The application provides a radar signal open set identification method based on SoftMax information entropy, research is carried out on the basis of SoftMax probability distribution, and the probability distribution information entropy is calculated for unknown discrimination; the information entropy is calculated based on the known type SoftMax probability distribution of output; the uncertainty degree of the identification result is reflected through the information entropy; whether the uncertainty degree of the identification result is higher than the preset is judged; if yes, the identification result of the radar signal to be identified is updated as an unknown radar signal category and is output; otherwise, the radar signal category with the highest probability value in the identification result is output. Compared with directly using the maximum probability, the threshold discrimination interval is improved, so that the unknown sample can be more easily identified; the application does not need additional network design and training, is simple to realize, has high real-time performance, has good open set performance, can be applied to any network based on the SoftMax layer classification, has wide application range and does not bring too much time consumption, can be used immediately, and has good practicability and real-time performance.

The invention provides a smart grid false data injection attack detection method, a terminal and a storage medium, and relates to the technical field of power system information security. The method comprises the steps that heterogeneous measurement data of a target smart grid are collected, a three-dimensional space-time tensor is constructed based on the heterogeneous measurement data, and the heterogeneous measurement data comprise node measurement data, branch measurement data and network topology data; inputting the three-dimensional space-time tensor into a constructed attack feature extraction model, and outputting a target feature corresponding to the three-dimensional space-time tensor; based on the target features, calculating an attack prediction probability, and based on the attack prediction probability, determining whether the target smart grid is subjected to a false data injection attack; and when the target smart power grid is subjected to the false data injection attack, calculating a probability distribution vector by using the target features, and determining an attack type of the false data injection attack on the target smart power grid according to the probability distribution vector. According to the invention, the accuracy and reliability of detection can be improved.

Systems, methods, and computer-readable media for providing low-latency markets for wagering on sporting events are disclosed. In some embodiments, an outcome matrix may be generated by a plurality of contest simulations utilizing statistical data of a history of sporting events. Outcome data of the outcome matrix may be indicative of probabilities of events occurring during an upcoming contest. The probabilities may be calculated, and markets may be priced based on the calculated probabilities. The priced markets may be provided to users by a graphical user interface by user computing device. Furthermore, users may request user-requested markets by inputting different markets into the GUI. The user-requested markets may then be priced using the outcome data of the outcome matrix and stored calculations thus, providing new markets based on the user-requested markets from the outcome matrix without generating new simulations.

Embodiments of the present application disclose a situation awareness implementation method, device, equipment and medium. The method comprises: taking at least two nodes in a system as target nodes, and obtaining risk level data corresponding to each target node respectively; determining the association relationship between the target nodes; for the nodes with a first association relationship in the target nodes, establishing a conditional probability distribution model based on the conditional probability and the risk level data; for the nodes with a second association relationship in the target nodes, establishing a joint probability distribution model based on the joint probability distribution and the risk level data; and performing situation awareness according to the conditional probability distribution model and the joint probability distribution model to obtain a situation awareness result. The above technical solution is used to analyze the association relationship between the target nodes, different probability distribution models are established according to different association relationships, the algorithm difficulty is reduced through the calculation of the probability distribution, and the technical effects of improving the flexibility of situation awareness modeling and reducing the modeling cost are achieved.

The invention discloses an electricity utilization information acquisition intelligent configuration method, which relates to the technical field of power distribution automation, and comprises the following steps: converging equipment ledger, topology, historical operation and configuration records and standardizing to form a data set and a candidate template; extracting characteristics of operation, topology, equipment, alarm and the like, and constructing a comprehensive characteristic vector; the matching probability is calculated through SVM, random forest and CNN according to the comprehensive features and the candidate template, and a probability matrix is formed; performing fusion calibration according to the advantage ratio, the consistency coefficient and the historical success rate to obtain a robust matching probability; and performing sensitivity detection on the core operation indexes, determining an optimal template according to the matching probability, segmenting according to three levels, and sequentially issuing parameters in a rollback manner. According to the method, the probabilities are calculated through the three models and are ingeniously combined, and the optimal template is determined, so that the reliability and the accuracy of the optimal template are ensured.

To shorten the time required for calculating the condition diagnosis of a water treatment base.SOLUTION: The situation diagnosis support system of a water treatment base, comprises a memory memorizing a plurality of unit models, i.e., Bayesian network models corresponding to each of a plurality of water treatment units in the base, a base model generation unit generating the Bayesian network models of the water treatment base, a conditional probability distribution set on the node present in the unit model and a calculation unit calculating a probability distribution based on a probability distribution inputted from another node. The unit model comprises an input port functioning as an assigning point for the probability distribution calculation result, and an output port functioning as an extraction point for the probability distribution calculation result. The base model generation unit connects the output port of the first unit model and the input port of the second unit model to generate the Bayesian network model of the water treatment base. The probability distribution calculation result calculated in the first unit model, is assigned to the second unit model.SELECTED DRAWING: Figure 1

The invention discloses a comparison normalization probability representation method for multi-modal image-text retrieval, which comprises the following steps of: respectively extracting high-order semantic features of an image and a text by integrating a Vision Transform and a BERT as a vision and text encoder; then, probability distribution modeling is carried out on the features, and a many-to-many matching relation between modals is naturally represented; a comparison normalization module is introduced, an embedding space is optimized through feature normalization and comparison learning, and the problem of feature misalignment caused by modal difference is solved; the probability distribution similarity is calculated by adopting KL divergence or Euclidean distance, and the model is trained in combination with a soft contrast loss function and a regularization strategy, so that efficient and robust cross-modal retrieval is finally realized, and the retrieval precision and generalization ability in a complex scene are remarkably improved.

Systems, methods, and computer-readable media for providing low-latency markets for wagering on sporting events are disclosed. In some embodiments, an outcome matrix may be generated by a plurality of contest simulations utilizing statistical data of a history of sporting events. Outcome data of the outcome matrix may be indicative of probabilities of events occurring during an upcoming contest. The probabilities may be calculated, and markets may be priced based on the calculated probabilities. The priced markets may be provided to users by a graphical user interface by user computing device. Furthermore, users may request user-requested markets by inputting different markets into the GUI. The user-requested markets may then be priced using the outcome data of the outcome matrix and stored calculations thus, providing new markets based on the user-requested markets from the outcome matrix without generating new simulations.

The application discloses a power transmission line ground inclination calculation method and device based on information entropy weight, and comprises the following steps: S1, constructing a three-dimensional scene; S2, sampling equidistantly left and right along the direction of the vertical line of the overhead line of each base tower, and obtaining the original slope value of each sampling point; S3, constructing an original slope value matrix R based on the original slope value of all sampling points on the left side or the right side; S4, performing normalization processing on the original slope value matrix R, and obtaining a normalized matrix R ′ ; S5, constructing a probability matrix P; S6, calculating the entropy value of each base tower in the probability matrix P; S7, calculating the slope value weight of each sampling point according to the entropy value; S8, modifying the original slope value by using the slope value weight, and obtaining the ground inclination on the left side or the right side of the tower; S9, selecting the other side of the vertical line of the tower to the overhead line, and repeating the process described in steps S3-S8, so as to calculate the ground inclination of the tower on the other side of each base tower. The application can improve the accuracy of ground inclination calculation.

Some embodiments provide a method for training a machine-trained (MT) network to classify inputs into multiple categories. The method propagates a set of input training items through the MT network to generate a set of corresponding outputs. Each input training item belongs to a category and the output for each input training item includes, for each category, a computed probability of the input belonging to the category. The method computes a value for a loss function based on the generated outputs. The loss function includes a first term based on the computed probabilities of each input belonging to its category and not based on individual computed probabilities of the inputs belonging to other categories and a second term based on the individual computed probabilities of each input belonging to each of the categories. The method uses the computed value for the loss function to train the MT network.

The present application relates to a visual positioning method based on three-dimensional LoD map and neural wireframe alignment. The method comprises: uniformly sampling the four degrees of freedom with the initial posture as the center, generating posture hypotheses in four directions respectively; calculating the straight line alignment cost of the posture hypothesis and the pre-constructed three-dimensional wireframe points according to the neural wireframe alignment method, combining the straight line alignment costs in the four directions in a grid manner to calculate the posture cost volume and then calculating the probability distribution volume; determining the posture sampling range of the next layer according to the probability distribution volume to generate the posture hypothesis of the next layer, and taking the selected posture obtained at the last feature level as the candidate selected posture; mapping the multi-level features, using the mapped features and the three-dimensional wireframe points to design the optimization objective function of the candidate selected posture, and then solving it to obtain the final posture. The use of this method can improve the accuracy of visual positioning and reduce memory.