8 results about "Artificial neural network algorithm" patented technology

The invention relates to the technical field of data processing, and particularly discloses an AI-based air quality prediction method and system, and the method comprises the steps: S1, dividing a preset region into independent grids of a preset size, and setting the information of each grid; s2, extracting influence factors of grid air quality, wherein the influence factors comprise influence factors of spatial change and influence factors of time change; s3, constructing a space classifier based on an artificial neural network algorithm, and capturing the influence of space factors on the air quality; constructing a time classifier based on a convolution LSTM algorithm, and learning time periodic features in the time sequence data; and S4, through a collaborative algorithm, according to results of the space classifier and the time classifier, obtaining gridding air quality data with a time sequence. By adopting the technical scheme of the invention, the air quality prediction accuracy of mountainous regions and river valley cities can be effectively improved.

The application discloses a duckweed carbon sink evaluation method, system and equipment suitable for a rice-duckweed symbiotic system, relates to the technical field of agricultural production and ecological environment, and discloses the following technical scheme: after rice is harvested, the information of duckweed in the rice-duckweed symbiotic system is collected by using spectral scanning and laser point cloud technology, and the volume of the duckweed is determined based on the duckweed information; the duckweed biomass in the rice-duckweed symbiotic system is determined according to the duckweed volume; the duckweed carbon sink is estimated by using an artificial neural network algorithm and the duckweed biomass, or the duckweed carbon sink is estimated by using a duckweed carbon sink estimation formula and the duckweed biomass. The application can flexibly evaluate the duckweed carbon sink according to information such as rice agronomic management, soil and climate, and provides good technical support for the management and application of duckweed in a rice field.

The application discloses a kind of molecular potential energy field analysis methods based on multivariate feature information transmission, it is related to molecular dynamics technical field, including steps: obtaining the element characteristic vector and structure characteristic vector of target molecular formula;By undercomplete self-encoder constitutes coding network, and the multivariate characteristic vector formed by element characteristic vector and structure characteristic vector is handled to obtain atom representation;According to atom representation, obtain interatomic distance and aggregate adjacent atoms, and update adjacent aggregated atom representation by continuous convolution filter;According to the updated atom representation, atom potential is predicted by artificial neural network algorithm, and the total potential energy of system is obtained according to the atom potential of each atom in target molecular formula.The application is by the use of general local environment characteristics, and makes atom representation contain more feature information, guarantees the feature description of different material systems, and then improves the generality of model.

The invention relates to the technical field of biological medicine, and discloses a method for constructing a pregnant woman health index evaluation system, which comprises the following steps: collecting original data of a pregnant woman in a perinatal period, cleaning the original data, removing abnormal values, filling missing values, and carrying out standardization processing to obtain preprocessed data; a random forest feature screening technology is adopted, variables with feature importance higher than an average value are selected, a composite perinatal period bad outcome prediction model is established based on an artificial neural network algorithm, and a prediction value output by the prediction model is converted into a pregnant woman health index through linear conversion; according to the method, features screened in the construction process of a composite perinatal period bad outcome prediction model are used as independent variables, a Logistic regression model is used for prediction, model fitting is carried out for different bad outcomes, respective Logistic regression equations are obtained, and a single bad outcome prediction model is constructed; according to the invention, the early warning capability is improved, the clinical management efficiency is enhanced, and the occurrence rate of bad outcomes in the perinatal period is reduced.

The invention discloses a building waterproof layer damage trend prediction method, which is characterized in that a plurality of groups of sensor clusters are dispersedly arranged on a building waterproof layer, whether the building waterproof layer is damaged or not is reflected by monitoring physical property change data of the building waterproof layer through the sensor clusters, and the physical property change data comprise strain, humidity, temperature, optical wavelength and audio frequency; monitoring physical property change data of the building waterproof layer in real time by each group of sensor clusters in a wireless transmission mode, and transmitting the data to analysis equipment; on an analysis device, layout position data of a sensor cluster on a building waterproof layer, acquired physical performance change data and a time sequence are used as input data of an artificial neural network algorithm; and generating damage probability and damage trend data of the building waterproof layer at each layout node position through an artificial neural network algorithm to predict the damage trend of the building waterproof layer. According to the invention, the problem of monitoring hysteresis due to the damage of the waterproof layer of the video monitoring building can be avoided.

The application provides an on-chip training method of an in-memory computing memory artificial neural network, and belongs to the field of artificial neural network algorithm optimization.The application follows the Manhattan rule idea, introduces a probability-based ternary update rule, converts high-precision weight update in an ideal classical error back propagation algorithm BP algorithm into ternary weight update, only applies at most one programming pulse to one device in each training batch, reduces the operation times, the training method converges fast and is stable, the recognition accuracy is high after training, the original BP algorithm is slightly changed, and the performance exceeds the Manhattan and threshold-Manhattan rules from the algorithm perspective;The application can efficiently realize on-chip stochastic gradient descent SGD and mini-batch gradient descent MBGD, does not need to store high-precision weight update values, reduces the additional hardware overhead, and optimizes the design of the inference circuit.

The application provides a method and device for evaluating slow obstacle invasion risk, and a storage medium, and the method comprises the following steps: obtaining obstacle data and influence parameters, wherein the influence parameters comprise a vibration factor of a train, a device maintenance cycle and a wind speed; predicting an obstacle position by using a long short-term memory (LSTM) artificial neural network algorithm based on the obstacle data and the influence parameters; and taking the predicted obstacle position as a reference point, evaluating the risk probability of the obstacle based on a pre-set corresponding relationship between a limit range and a risk probability. The method provided by the application predicts the obstacle position based on the obstacle data and the influence parameters by using the LSTM algorithm, takes the predicted obstacle position as the reference point, evaluates the risk probability of the obstacle based on the pre-set corresponding relationship between the limit range and the risk probability, realizes the prediction of the obstacle, plays a role in early prevention, and is beneficial to early planning of operation.

The application relates to a multi-energy virtual power plant optimization scheduling method and system, belongs to the technical field of power optimization scheduling, and solves the problems of incapability of a single prediction model to meet the demand of multi-energy virtual power plant optimization scheduling, a large data set used in prediction calculation, and high operation cost; the application adopts a random forest algorithm to screen influencing factor indexes of wind and light electric set output, uses an artificial neural network algorithm to predict the wind and light electric set output, further makes up for the deficiency of a single algorithm through a hybrid prediction algorithm, and improves the prediction accuracy; a Monte Carlo method is used for simulation analysis for load prediction of electric vehicles; an optimization scheduling model of the multi-energy virtual power plant is established by taking the minimum virtual power plant operation cost as an objective function, and the problem that only a single distributed energy can be predicted and optimized in the prior art is solved.