51results about How to "Improve prediction accuracy" patented technology

A processor integrated circuit has one or more processor cores and a power management controller in a North-Bridge that generates a first power state recommendation for the one or more processor cores. The North-Bridge also receives a second power state recommendation from a South-Bridge integrated circuit. The North-Bridge determines a final power state for the one or more processor cores based on the first and second power state recommendations.

An object of the present invention is to improve prediction accuracy for a performance value of a monitoring target system even if the performance value is substantially different for each of patterns of use. A monitoring data analyzing apparatus includes a regression-model recalculating section 14 configured to combine regression models, which are generated for each of a plurality of groups into which log data including monitoring data in a monitoring target system set as a target of performance management is classified, using the log data belonging to the groups corresponding to the regression models and test target log data, which is the log data set as a target of a performance test, to recalculate the regression models.

A word prediction method that improves the precision accuracy, and a speech recognition method and an apparatus therefor are provided. For the prediction of a sixth word “?”, a partial analysis tree having a modification relationship with the sixth word is predicted. “sara-ni sho-senkyoku no” has two partial analysis trees, “sara-ni” and “sho-senkyoku no”. It is predicted that “sara-ni” does not have a modification relationship with the sixth word, and that “sho-senkyoku no” does. Then, “donyu”, which is the sixth word from “sho-senkyoku no”, is predicted. In this example, since “sara-ni” is not useful information for the prediction of “donyu”, it is preferable that “donyu” be predicted only by “sho-senkyoku no”.

The invention discloses a method to provide input the candidate sentence. The method including: receiving the alphabet string sentence, and the alphabet string to be the candidate syllable of the word after the use of the word referred to the candidate portfolio candidate number of candidates referred to the output sentence after sentence. The invention also open a text input system, including receiver modules, generating units and output modules. This invention provide a number of candidates results sentence, the sentence greatly increase the probability of a correct forecast, the need for more user keyboard, you can output in the candidate window to choose the correct number of results a greatly reduced using a keyboard users election phrase sentence of time for users with a more perfect input and output experience, improve the efficiency of the text input.

The invention discloses a predicting method for short-term electric load of a Seq2seq network based on a multi-layer Bi-GRU, similar daily samples are extracted by using FCM method and the input variables are standardized by the Min max standardization method. The multi-level Seq2seq neural network structure is constructed with Bi GRU neurons as the basic unit. At the same time, the SELU activation function is selected as the output layer activation function of the whole neural network to reduce gradient vanishing and gradient explosion, to realize smooth operation of the whole model in the training process.

The invention discloses an SPARC V8 system structure based classified type mixed branch prediction system. Firstly, a branch target buffer is queried according to PC values of instructions at an instruction fetching stage to obtain branch instruction types; the branch instructions are dispatched to respective prediction modules; a return address stack (RAS) with a dynamic configuration counter is used in skip branch prediction; a complementary prediction method is used in indirect branch prediction; a tag recording correctness of previous branch prediction in a conditional branch target buffer (CBTB) adopts a partial skip three-state conversion algorithm in conditional branch prediction; decoding result information of the instructions are recorded in a prediction information table (PIT) at a decoding stage; a judgment is made at an execution stage; if a prediction result of the branch instructions is that the skip occurs, the result judgment is made by using a skip prediction result arbiter Arbiter_T; and if the prediction result of the branch instructions is that the skip does not occur, the result judgment is made by using a non-skip prediction result arbiter Arbiter_N. Therefore, the instruction delay influence of the branch instructions on an assembly line is eliminated and the execution efficiency of a processor is improved.

A single word and multi-word term integrating system and a method thereof are disclosed, wherein a user uses an input unit to continuously input pinying codes for the system to find combinations of the pinying codes to provide word candidates for the user to choose, wherein the word candidates can be combined into a phrase or a sentence; when the inputted pinying codes are too long or incomplete, there might be a false prediction of a word or a sentence due to an incorrect combination of pinying codes; consequently, the system forcibly determines the pinying codes to be regarded as a single word and does not combine them with the follow-up pinying codes; then the system uses a full sentence prediction result display unit for the user to choose a correct word, thereby improving prediction accuracy.

While a production apparatus is operating, a computation unit of an analysis device performs main predictive computation on an acceptability in real time by using one or more operating predictors selected from among a plurality of predictors. When the main predictive computation is not performed, the computation unit performs subsidiary predictive computation on the acceptability by using one or more subsidiary predictors that are not selected from among the plurality of predictors. The computation unit performs processing of interchanging the operating predictors and the subsidiary predictors based on prediction accuracies of the operating predictors and prediction accuracies of the subsidiary predictors.

The invention discloses a depth belief network-based link prediction method. The method comprises the steps that firstly a training data collection module performs random sampling in a given network structure to obtain a training edge set, a verification edge set and a test edge set; a network node characteristic representation module generates a characteristic representation of each network node by using a deepwalk algorithm in a network processed through the training data collection module; an edge characteristic representation generation module calculates a characteristic representation of each edge in the training edge set, the verification edge set and the test edge set, and performs normalization processing on eigenvectors of generated edges to meet the requirements of a depth belief network on input data; and finally a depth belief network training module constructs a depth belief network structure and loads the training edge set, the verification edge set and the test edge set to perform training. According to the method, the prediction accuracy higher than that of a conventional link prediction algorithm can be achieved; and the method has universality for networks with various structure characteristics.

The invention relates to a federated learning algorithm for bearing fault diagnosis, which is characterized in that the algorithm runs on a plurality of local nodes and an aggregation node, and comprises the following steps of: 1, converging data of a sensor network by each local node, the data of the sensor are preprocessed in a time-sharing, partitioning, sampling and normalizing manner; 2, training the preprocessed data by adopting a convolutional neural network model; 3, after training is completed, whether the aggregation condition is met or not is judged according to the improved aggregation strategy, and if yes, the round of training is ended; then, calculating an F1 score of the local model; finally, performing homomorphic encryption on the model parameters, the F1 score and the total number of samples, and sending to an aggregation node; and 4, after receiving the information sent by all the local nodes, the aggregation node decrypts the information, then weights and aggregates all the local models according to an F1 score weighting strategy to obtain a new initial model, and sends the new initial model to the local nodes.

The invention relates to a mine car transportation time pre-judgment method for an in-use mine car unmanned transportation system. The mine car transportation time pre-judgment method comprises the steps: 1, building a first time pre-judgment model for offline operation by using historical data; 2, based on the online data, establishing a second time pre-judgment model; and 3, integrating the results of the model M_1 and the model M_2 to obtain a pre-judgment value. The mine car transportation time pre-judgment method utilizes the equipment advantages and convenience of the in-use mine car unmanned transportation system of an end edge cloud framework, and applying the equipment advantages and convenience to service of the mine car unmanned transportation system, thus finally laying a foundation for achieving the purpose of improving the system operation efficiency.

The invention discloses a fruit sugar degree detection method and system based on a genetic algorithm and an extreme learning machine. The method comprises the steps of: obtaining and preprocessing anoriginal near infrared spectrum of a fruit to be detected; screening out the optimal characteristic wavelength by using a genetic algorithm; inputting the optimal characteristic wavelength into a trained extreme learning machine prediction model, outputting soluble solid content information of fruits, and further obtaining fruit sugar degree information, wherein the extreme learning machine prediction model is established based on the corresponding relationship between the original near infrared spectrum of the fruit and the corresponding soluble solid content value. Wavelength is screened based on the genetic algorithm, a correlation coefficient of a predicted value and an actual value of a dependent variable in interactive verification of an extreme learning machine method is used as afitness function of the genetic algorithm, and the most appropriate wavelength is selected from 1557 spectral wavelengths of an original spectrum by using the genetic algorithm, so that the predictionprecision of the fruit sugar degree is greatly improved.

The invention discloses a deep learning model optimization method based on data defense, which can find a relatively effective defense means when coping with different countermeasure sample attack methods. In order to optimize the model for the adversarial sample attack method, the data-level defense strategy mainly comprises the steps that adversarial samples are injected into a training data setin the training stage and then the model is retrained, or the samples are modified in the prediction stage, reconstruction is conducted, and the converted adversarial samples are input into an original model to be predicted. An open source adversarial sample generation tool is used to generate adversarial samples for the to-be-tested model and the target data set, and success rates of the model on the specified data set are compared before and after the adversarial samples are generated.

The embodiment of the invention provides a message processing method and device, network equipment and a storage medium. The message processing method includes: receiving a network message sent basedon the access network, wherein the network message comprises a to-be-processed IPID value; if the IPID value to be processed is not in at least one IPID value range corresponding to the access network, screening out a target IPID value range meeting a turnover judgment condition; and according to the maximum historical IPID value in the target IPID value range and the to-be-processed IPID value, determining whether the to-be-processed IPID value is an IPID value obtained by flipping according to a set rule, and if the to-be-processed IPID value is the IPID value obtained by flipping, updatingthe target IPID value range. Through the technical scheme provided by the embodiment of the invention, the accuracy of quantity statistics of the IPID value range is improved, so that the accuracy ofa judgment result of whether the access network is shared or not is improved, and the accuracy of the determined quantity of the access user equipment in the access network is improved.

An equipment maintenance prediction system and an operation method for the equipment maintenance prediction system are provided. The operation method includes steps of: configuring the processor to configure the factor decision module to select one of a plurality of parameter types as a decision parameter type according to a key parameter type, wherein the decision parameter type and the key parameter type are most correlative; configuring the processor to configure the prediction module to generate a prediction model according to a part of a plurality of historical sensing values of the decision parameter type and formulate a maintenance alerting condition according to a part of a plurality of historical sensing values of the key parameter type; and configuring the processor to configure the maintenance alerting module to monitor and alert according to the maintenance alerting condition.

The invention belongs to the field of plasma physics, and particularly relates to a Tokamak plasma large fracture prediction algorithm based on a deep neural network, which comprises the following steps: preparing a training data set, performing neural network model creation and prediction calculation, then performing model parameter training, and inputting parameters into a neural network model for calculation after the neural network model training is completed to acquire a real-time fracture possibility value. A customized and optimized neural network model is carried out according to the data characteristics of the fusion device, the model can simply and conveniently access different types of control and diagnosis signals, the problem that a standard neural network model limits a data source is solved, the neural network is more suitable for processing long-sequence, multi-modal and multi-noise-label fusion data. And finally, the effect that the prediction accuracy is 30ms ahead of time and 96.1% of the fracture prediction task is realized.

Embodiments of the present disclosure provide a disk capacity predicting method, apparatus, equipment and non-volatile computer storage medium. On the one hand, the change data of the disk capacity is obtained according to the historical capacity data of the disk; then the target inflection point in the historical capacity data is obtained according to the change data of the disk capacity; and then the linear relationship between the time and disk capacity is obtained according to the historical capacity data after the target inflection point. Therefore, technical solutions provided by embodiments of the present disclosure may achieve improvement of accuracy of prediction of disk capacity trend and reduce the cost needed in disk capacity prediction.

The invention belongs to the field of big data analysis, and discloses a risk probability calculation method and device and computer equipment. The method comprises the following steps: obtaining fragmented data on networks of an enterprise and an associated enterprise associated with the enterprise; wherein the obtained data is rich in dimension and carrying out multiple preprocessing projects ofdata, then, analyzing and converting risk conduction in an enterprise association relationship by using an infectious disease model, the association risk of the enterprise is reflected in a numeralization mode. Finally, the prediction accuracy of the enterprise debt default risk prediction model is improved by using a multi-layer model fusion mode.

The invention relates to a deep learning model optimization method based on network addition / modification, in particular to an optimization method integrating a model level in the field of deep learning model optimization, and specifies an optimization strategy for adversarial samples by adopting an evaluation feedback mechanism. A defense strategy is formulated by evaluating a model optimized byusing a defense method and evaluating a feedback mechanism, and an optimal defense means is selected for coping with different attack methods. In order to optimize the model for the adversarial sampleattack method, the defense strategy of the model level is to modify the network, and the structure of the original DNN model is modified in the training stage, or the original model is not changed, and an external model is used as an additional network, so the defended DNN classifier can detect the adversarial sample or identify the adversarial sample as a correct label.

Provided is a learning apparatus 10 including a feature amount generation unit 11 configured to generate a feature amount based on learning data, a division condition generation unit 12 configured to generate a division condition in accordance with the feature amount and a complexity requirement that indicates the number of feature amounts, a learning data division unit 13 configured to divide the learning data into groups based on the division condition, a learning data evaluation unit 14 configured to evaluate a significance of each division condition by using a pre-division group and a post-division group; and a node generation unit 15 configured to, if there is a significance in the division condition of the pre-division and post-division groups, generate a node of a decision tree relating to the division condition.

The invention relates to a model establishing method for predicting foam retention of a draft beer in a shelf life based on a broth and a prediction method thereof. By analyzing relevance between protease A activity in the broth and the foam retention of the draft beer in the shelf life as well as analyzing relevance between foam protein content in the broth and the foam retention of the draft beer in the shelf life, the measured foam protein content of a broth sample, protease A activity and a foam retention value of the draft beer in the shelf life are input in Mintab software, and a linearequation among the foam protein content and the protease A activity in the broth and the foam retention of the draft beer is subjected to fitting by the Mintab software. Through a prediction model ofthe foam retention of the draft beer in the shelf life based on the broth, the foam retention value of the draft beer with a batch in the shelf life can be know in advance, through verification, the foam protein content and protease A activity double factors are measured for prediction, the prediction precision is higher, and the result is more accurate and reliable.

The invention provides electronic equipment and a method for predicting duration time of a battery. With the method and equipment, the prediction precision of the duration time of the battery can be improved, thereby showing the endurance performance of the electronic equipment well. The method comprises: each of a plurality of working states of electronic equipment is monitored and recorded, wherein the multiple working states are different; power consumption information of a battery providing power for the electronic equipment in each working state is monitored and recorded; duration time of each working state is monitored and recorded, wherein the multiple working states of the electronic equipment change within a required time period and duration time of a working state before changing is obtained when the working state of the electronic equipment changes; and on the basis of the duration time of the required time period, power consumption in each working state, and duration time of each working state, duration time of the battery is predicted based on a predetermined algorithm.

A method is described for determining a predisposition of an organism for Crohn's Disease, especially Crohn's Disease of the small intestine. In this context, in a biological specimen of an organism, the presence or the absence of SNP's in at least one gene is determined, which codes for a protein associated with the Writ signaling pathway in Paneth cells. The gene, in this instance, may be selected from TCF4, LRP5, LRP6, GSK3A, GSK3B and TCF7. The present methods and systems also relate to primers and allele-specific probes to prove the presence or the absence of an SNP, diagnostic kits which have at least one such primer or one such allele-specific probe, as well as the use of certain SNP's for determining a predisposition of an organism for Crohn's Disease. The present method and systems also relate to a method for the differential diagnosis of inflammatory bowel diseases, for distinguishing Crohn's Disease and the other respective inflammatory or infectious intestinal diseases.

A method for designing a turbomachine vane, in which predefined input parameters are transmitted to a neuronal network system and vane parameters are determined and output by the neuronal network system based on the transmitted input parameters. The neuronal network system has several separate neuronal networks each with an output layer, each of which determines one or more of the vane parameters and outputs same via the output layer. A first neuronal network and a second neuronal network belong to the separate neuronal networks of the neuronal network system and the vane parameter(s) which are determined by the first neuronal network and output via the output layer of said neuronal network differ(s) from the vane parameter(s) that are determined by the second neuronal network and are output via the output layer of said neuronal network.

A method and apparatus of improving prediction accuracy of a branch instruction scheme includes reading an individual instruction in a current set of instructions, fetching the individual instruction when an instruction fetch unit determines that the individual instruction is valid, and allowing the instruction fetch unit to use an index address for the fetched individual instruction. A method and apparatus of improving branch prediction accuracy includes receiving a set of instructions having an assigned address, making a prediction for a branch instruction in the set of instructions using the assigned address, and retaining the assigned address for the branch instruction in the set of instructions.

The invention discloses a method for generating a whole sentence in Chinese character input process. The method comprises the following steps of: dividing syllables of a pinyin string which is input by a user, and comparing each divided syllable with a preset family-name dictionary and a preset first-name dictionary; and when one or two continuous syllables in the pinyin string are matched and positioned in the family-name dictionary, and the one or two continuous syllables which are subsequently adjacent to the matched syllables are positioned in the first-name dictionary, providing personal names corresponding to a family-name syllable and a first-name syllable which are matched and using the personal names as candidate words for generation of the whole sentence, and displaying a generated result of the whole sentence by an interface. The invention also provides a device for generating the whole sentence in the Chinese character input process. By the method and the device, the probability of accurate prediction of the whole sentence when the pinyin string input by the user comprises personal-name pinyin is improved, so a better input experience is provided for the user.

The invention discloses a human body fatigue state prediction method and system based on a fuzzy sensor, and the method comprises: collecting breathing data through a cardiopulmonary function tester,collecting heart rate data through an electrocardio monitor, and collecting a fatigue state recognized by a testee through a questionnaire; normalizing the data to eliminate the influence of dimensions among the indexes; dividing the collected data into a training data set and a test data set, and designing a corresponding fuzzy sensor; implementing the fuzzy sensor, sending the training data setinto the fuzzy sensor for training, learning a group of weight values and storing the weight values; and applying the test data set to the fuzzy sensor to obtain a prediction result. The prediction result is compared with prediction results of other machine learning methods. The fuzzy theory is combined with a traditional linear sensor, the fuzzy sensor method is designed, the fuzzy concept similar to 'fatigue' in life can be classified through the method, and the application range of the classification method in machine learning is expanded.

The invention discloses a ship detection method and system based on adaptive data enhancement. The method comprises the following steps: acquiring a picture data set through a camera and performing data enhancement processing; based on a neural network model, carrying out picture feature extraction processing on the enhanced picture; carrying out multi-scale dimension attention calculation on the feature picture to obtain a feature picture with multi-scale dimensions; performing loss calculation on the feature picture with multiple scales and dimensions through a loss function to obtain a final loss value; updating a neural network model according to the final loss value, and constructing a ship detection model; and detecting the to-be-detected picture based on the ship detection model to obtain a detection result. On the basis of a visible light video detection technology, the detection perception capability of the ship body is improved by constructing the ship detection model, and the ship detection precision can be improved in a special detection environment. The ship detection method and system based on adaptive data enhancement can be widely applied to the technical field of ship detection.

The invention discloses a multi-variable dynamic alignment diagram prediction model and application thereof. Visual analysis is performed based on a computer programming technology, a biological information technology and predictive factor scoring parameters. The dynamic prediction model established by the invention can be used for analyzing the Henry unit density value of the blood clot after aneurysm subarachnoid hemorrhage and the scoring parameter of a serum marker after the electronic computed tomography, and is presented in a web version chart mode. The model is helpful for clinical doctors to carry out early rapid prediction on the delayed cerebral ischemia patients after aneurysm subarachnoid hemorrhage and make clinical decisions according to the obtained results.