45results about How to "Prediction results are stable" patented technology

The invention discloses a bus load prediction method. The method comprises the following steps: correcting abnormal values in historical load data by use of a transverse comparison method, and determining key influence factors of bus load by use of a grey relation projection method; putting load curves with similar features in the same category by use of an improved K-Means clustering method to get a plurality of typical load patterns, building a random forest classification model, and establishing the mapping relationship between influence factors and clustering results; for each load pattern, training a plurality of prediction models by use of a multivariate linear regression method; and determining the category of a day under test, and selecting a matching regression model to realize load prediction. A data mining method is introduced to analyze the changing rules of bus load, and a prediction model library is built. Model matching is realized based on the category of a day under test. The accuracy and real-time performance of short-term bus load prediction are improved. More accurate decision support is provided for power grid planning and real-time dispatching.

The invention relates to a method for predicting ammonia process flue gas desulphurization efficiency based on multiple parameters. The method is characterized by comprising the following steps of: selecting four different artificial intelligent computation models and taking parameters acquired in an ammonia process desulphurization system operational process such as multiple groups of flue gas amounts, flow of a circulating pump, the flow of a concentration pump, the ammonia concentration, the concentration of absorption liquid, the liquid-gas ratio, the inlet flue gas temperature, the ammonia consumption, the density of spraying slurry, the pH value of slurry of a spraying tower and the pH value of the slurry of a pre-washing tower as input variables of the four models; respectively training each model, and establishing a non-linear function relationship between four desulphurization parameters and the desulphurization efficiency; then respectively transmitting parameters monitored in real time into the trained artificial intelligent model, and predicting the desulphurization efficiency; and taking the average value of two predicted values in the middle as a final predicted value... The method disclosed by the invention can be used for better predicting the ammonia process desulphurization efficiency and has the characteristics of higher stability and stronger prediction capability compared with single model prediction.

The invention provides a machine learning model training method, device and system, electronic equipment and a computer readable storage medium. The method comprises the steps that service party equipment sends a machine learning model to multiple pieces of training participant equipment, so that the multiple pieces of training participant equipment independently train the machine learning model based on training samples stored in the multiple pieces of training participant equipment, wherein the first loss function used by the training participant equipment for training the machine learning model is used for balancing the distribution of parameters of the machine learning model after the training participant equipment is trained; receiving training results returned by the plurality of training participant devices; aggregating the training results returned by the plurality of training participant devices to obtain a global machine learning model; and training the global machine learning model according to the training sample stored in the service side device. According to the method and the device, the prediction stability of distributed learning can be realized on the basis of realizing intensive utilization of resources for training the machine learning model.

The invention discloses a short-term electrical load on-line predicting method based on the self-adaptation enhancing algorithm in the technical field of electrical load predicting. The short-term electrical load on-line predicting method comprises the step of selecting M factors affecting meteorological data and extracting actual measurement values of factors affecting the meteorological data in past L days to form a meteorological data matrix SL*M, the step of extracting electrical load data of n time points of each day in the past L days to form an electrical load data matrix DL*n, the step of selecting m factors with maximum association with the electrical load data from the factors affecting the meteorological data, serving the m factors as valid constituents and forming a valid meteorological data matrix TL*m according to the actual measurement values of the valid constituents of the past L days, the step of solving a short-term electrical load predicting model according to the valid meteorological data matrix TL*m and the electrical load data matrix DL*n, and the step of carrying out electrical load prediction according to the short-term electrical load predicting model. According to the short-term electrical load on-line predicting method, the effect on the model predicting precision of data noise can be effectively eliminated, and a more accurate and stable predicting result can be obtained.

The invention discloses a lithium battery life estimation method based on IMM-UPF. The method uses a new fusion model interactive multi-model for fusion calculation of different attenuation models. Inconsideration that Kalman filtering may cause a large error because of the non-Gaussian and nonlinear attenuation trend of the lithium battery, the method filters each model by using unscented particle filtering, thereby on the one hand solving the lack of particles in particle filtering during resampling, and on the other hand, achieving a more accurate prediction result than the Kalman filtering. The provided IMM-UPF method is verified by the comparison of simulation results and experimental data. The results show that the method can improve the accuracy of lithium battery life prediction.

The invention provides a neural network architecture search method based on convolution kernel prediction. The method comprises the steps of: constructing a super network for neural network architecture search, wherein the super network comprises a teacher network and a student network, the teacher network is a pre-trained network, and the student network is composed of a plurality of basic units;training the student network by taking a training set and the coding information of a neural network architecture as input, and predicting to obtain an optimal convolution kernel when a loss functionconverges to a minimum value, wherein the loss function is determined according to the loss function of the teacher network and the loss function of the student network; and updating the coding information of the neural network architecture according to the loss function of the student network to obtain an optimal neural network architecture. According to the neural network architecture search method based on convolution kernel prediction, the teacher network is introduced as guidance, and a convolution kernel prediction module in the student network can accurately predict the optimal convolution kernel, so that the search efficiency is greatly improved, and the global optimum of the search result can be ensured.

The invention discloses a radar aircraft system adversarial simulation confidence evaluation method. The method comprises the following steps of automatically encoding index vectors of subsystems of two complex and diverse adversarial parties into low-dimensional feature vectors with fixed dimensions by using a self-encoding model, simplifying index input of a flight target, and inputting the index vectors into a confidence prediction model in a stable form; and based on an attention mechanism of a deep network, guiding training of a confidence evaluation model, using a small amount of radar-aircraft system confrontation actual measurement data as a verification reference, and predicting the confidence of a simulation confrontation system according to the similarity degree of a simulationsystem and an actual measurement system. Therefore, the correlation and influence coefficients generated by the indexes of the input subsystems in the confrontation process are indirectly learned, a weighting matrix does not need to be set manually, and the objectivity of simulation confidence evaluation is improved.

The invention discloses a slope displacement fractal forecasting method improved by a grey theory and relates to the high slope deformation forecasting field of strip mine. The slope displacement fractal forecasting method improved by the grey theory includes that using a grey model (1, 1) to carry out fitting prediction on a fractal dimension sequence according to a fractal theory and a grey forecasting theory to obtain fractal dimensions of next few periods, and using an inversion extrapolation method to inversely calculate the forecast fractal dimensions to obtain a precise slope deformation forecast value. The slope displacement fractal forecasting method improved by the grey theory is suitable for forecasting the high slope deformation of the strip mine and capable of effectively solving the problems of low high mine slope deformation forecasting precision and bad forecasting stability of the traditional fractal forecasting under the condition of small sample monitored data with volatility, and accordingly the forecasting error range is controlled effectively, the reliance of a forecasting model on large quantity of monitored data is reduced, and the practicability of the fractal forecasting is improved in application.

The invention discloses a traffic flow prediction method based on spatio-temporal data embedding. The method comprises the following steps: acquiring historical traffic flow data; performing spatio-temporal data embedding based on historical traffic flow data, wherein the spatio-temporal data embedding comprises the following steps: performing interval representation of traffic flow: determining an interval to which each traffic flow belongs, and converting the determined interval into a corresponding traffic flow interval; generating traffic flow vectors: taking all the traffic flow intervals as input data, and converting the input data into embedded data, namely corresponding traffic flow vectors, by adopting a Word2vec model; time features are extracted based on the traffic flow vector to obtain a node feature matrix, and correlation between the electronic police devices is extracted to obtain a dynamic association graph; and inputting the node feature matrix and the dynamic association graph into a graph convolutional neural network to obtain a prediction result output by the graph convolutional neural network. According to the method, implicit correlation between traffic flows can be quantified and measured, high-level time features and a dynamic association graph are extracted for effective modeling, and accurate and stable traffic flow prediction is obtained.

The invention discloses a state holographic perception data driven power transformer fault prediction method and system. The method comprises the following steps: establishing a historical fault data set of a transformer; the method comprises the following steps: collecting the concentration of gas dissolved in oil in the operation process of a transformer, environment meteorological data reflecting the state of the transformer and other operation data, forming an optimal data set through a sudden change point detection and phase-space reconstruction mode, and building a recent prediction SARIMA model of the concentration of the gas dissolved in the transformer oil; obtaining a prediction result of the concentration change of the dissolved gas in the recent oil of the transformer; based on the relationship between the concentration of the dissolved gas in the oil and other state quantities of the transformer, the concentration of the dissolved gas in the oil changing in real time is obtained; building a transformer fault diagnosis model based on a DBN network; and taking a prediction result of the change of the concentration of the dissolved gas in the oil recently or the concentration of the dissolved gas in the oil changing in real time as an input characteristic quantity of the transformer fault diagnosis model to realize transformer fault prediction. And comprehensive real-time analysis and auxiliary decision making of the operation risk level of the power grid transformer are realized.

The invention discloses a well-to-seismic collaborative sedimentary microfacies division method based on a high-precision sequence framework model, and the method comprises the steps: building a matching relation of well-to-seismic sedimentary facies features based on the lithologic combination of a stratum under a high-precision sequence stratum shelf and the seismic response features of sedimentary body cycle features, and optimizing the seismic attributes sensitive to the types of stratum sedimentary facies; and establishing a three-dimensional data volume for researching sedimentary subfacies and microfacies of a target layer through well-to-seismic collaborative simulation. And verifying the reliability and accuracy of the sedimentary facies three-dimensional data volume through a phase sequence rule, and finally obtaining a sedimentary facies division result suitable for industrial standards. According to the three-dimensional data volume of the sedimentary microfacies, the working efficiency of sedimentary facies division can be greatly improved, the transverse resolution and the longitudinal resolution are high, and the determinacy and the reliability of sedimentary microfacies fine division results are higher.

The embodiment of the invention provides an acoustic network model training method and device and electronic equipment, and belongs to the technical field of data processing, and the method comprisesthe steps: obtaining a training sample for the training of an acoustic network model, and enabling the training sample to comprise one or more voice files; performing peak detection on voice containedin the voice file by using a BLSTM network so as to divide letters contained in the voice file based on the detected peak to form segmented voice; performing semantic prediction on the segmented voice based on a preset CTC segmentation network to obtain a text prediction result; and after the prediction result meets a preset performance requirement, stopping training the acoustic network model soas to carry out character prediction on the input voice by using the acoustic network model. Through the processing scheme of the invention, the training efficiency of the acoustic network model canbe improved.

The invention discloses a glomerular filtration rate estimation method based on WASP-BAS (Weights and Structure Policy-Beetle Antennae Search). The glomerular filtration rate estimation method based on WASP-BAS includes the steps: dividing experimental data into a data set, a verification set and a test set; taking the number of times that a smaller error cannot be found continuously as a constraint condition, and circulating under the condition that the number of times is not exceeded, i.e., one pruning process; temporarily determining the structure of the neural network after primary pruning, and then simplifying the neural network to reduce the number of hidden layer neurons, namely a secondary pruning process; finally, obtaining the neural network with the determined structure and weight threshold value, and estimating the glomerular filtration rate through seven inputs of the gender, the age, the height, the weight, the albumin, the serum creatinine and the urea of the neural network. According to the glomerular filtration rate estimation method based on WASP-BAS, a secondary pruning part is optimized by using a BAS, and a weight threshold from an input layer to a hidden layerand a weight from the hidden layer to an output layer are used as solution vectors for optimization; and the pruning efficiency of the WASP is higher, and the prediction result is more accurate and more stable.

The invention discloses intelligent water level detection equipment, an intelligent water level monitoring management device and a detection method. The intelligent water level detection equipment iscomposed of a processor, a field programmable gate array, a random access memory, a digital memory card and a network communication module. The field-programmable gate array is preset to train a CNN convolutional neural network based on deep learning and predict a water line. The intelligent water level detection method based on deep learning is implemented through the intelligent water level monitoring management device. The method specifically comprises the following steps: setting a system state as a normal state, capturing an image according to a capturing strategy to perform image cuttingprocessing, predicting a water line, calculating a virtual water level, judging whether the water level exceeds an early warning virtual water level, judging whether to enter a system early warning state for the first time, updating early warning information, generating an early warning record, and sending an early warning short message. According to the invention, the problems of complex detection method and low efficiency of the image-based detection method for the water area in the prior art are solved.

The invention discloses a short-term power sale quantity predicting method based on daily cumulative distributed electric quantity. The method comprises the following steps: step S1, establishing a linear function relation between a daily cumulative distributed electric quantity of a predicted month and a daily cumulative distributed electric quantity of the day in same period last year, taking the daily cumulative distributed electric quantity of the predicted month as a dependent variable, and taking the daily cumulative distributed electric quantity of the day in same period last year as an independent variable; step S2, substituting daily cumulative distributed electric quantity data from the first day to the predicted day of the predicted month and daily cumulative distributed electric quantity data of the day in the same period last year, and calculating model parameters; step S3, substituting the model parameters into the linear function relation to obtain a short-term power sale quantity linear model; and step S4, substituting the daily cumulative distributed electric quantity data of the last day of the month in the same period last year into the short-term power sale quantity linear model to obtain the predicted monthly power demand. The daily cumulative distributed electric quantity data accurately reflect fluctuation situations such as weather and economy of the same month, if the predicted day is close to the end of the month, the daily cumulative distributed electric quantity data used during modeling are more, and the accuracy is higher.

The invention discloses a multi-step wind power forecasting method based on singular spectrum analysis and locality sensitive hashing. The method is charactierzed by decomposing historical wind powerdata of a wind power plant into two independent components through singular spectrum analysis, the independent components being a low-frequency average trend component for reflecting wind energy overall change trend and a high-frequency fluctuation component for reflecting intermittency and fluctuation of wind respectively; reconstructing the two components in a phase space to obtain an average trend section and a fluctuation component section; and finding similar average trend sections of an average trend section to be forecasted through locality sensitive hashing, and carrying out locality predication. To prevent accumulated errors brought by separate forecast of each component and fixed error brought by forecast of only one component, the forecast input is combination of the similar average trend sections and corresponding fluctuation component sections, and finally, a prediction result of wind power output power is obtained. The method is clear in physical significance in wind power plant generation power prediction, short in prediction time and accurate and stable in prediction results; and the prediction results do not rely on prior knowledge of users.

The invention discloses a data completeness test and feature transfer learning method for building load prediction, and the method comprises the following steps: proposing a feature screening method based on diffusion kernel density estimation and maximum correlation minimum redundancy, and determining an optimal feature set under different load prediction models, establishing a data set feature completeness test method through the determination of the correlation between the features of the new feature set and the optimal feature set and the feature distribution similarity; and establishing a load prediction model of feature migration. According to the method, the problem that the building load prediction precision is reduced due to incomplete data features can be solved, the stability of the prediction result can still be kept when the input of the model generates tiny change, and meanwhile, the calculation complexity of the building load prediction model is reduced. The method has the advantages that building load prediction precision is greatly improved, robustness of the load prediction model is enhanced, and calculation time is shortened.