CN117391125B - Data processing method and system based on neural network - Google Patents
Data processing method and system based on neural network Download PDFInfo
- Publication number
- CN117391125B CN117391125B CN202311678079.9A CN202311678079A CN117391125B CN 117391125 B CN117391125 B CN 117391125B CN 202311678079 A CN202311678079 A CN 202311678079A CN 117391125 B CN117391125 B CN 117391125B
- Authority
- CN
- China
- Prior art keywords
- data
- detection
- time
- time point
- detection data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 29
- 238000003672 processing method Methods 0.000 title claims abstract description 16
- 238000001514 detection method Methods 0.000 claims abstract description 548
- 230000004927 fusion Effects 0.000 claims description 24
- 238000000034 method Methods 0.000 claims description 23
- 238000002372 labelling Methods 0.000 claims description 7
- 210000005036 nerve Anatomy 0.000 claims description 6
- 125000004122 cyclic group Chemical group 0.000 claims description 4
- 230000001537 neural effect Effects 0.000 claims description 4
- 238000011084 recovery Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 2
- 241000283690 Bos taurus Species 0.000 description 3
- 241000283073 Equus caballus Species 0.000 description 3
- 241000282326 Felis catus Species 0.000 description 3
- 241001465754 Metazoa Species 0.000 description 3
- 241001494479 Pecora Species 0.000 description 3
- 241000009328 Perro Species 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 229910052729 chemical element Inorganic materials 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000017260 vegetative to reproductive phase transition of meristem Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a data processing method and system based on a neural network, which belong to the technical field of data processing, and are used for detecting category relations based on a plurality of detection data sets to obtain a relation probability set, and predicting data of middle points of intervals of time points through the data processing network based on the detection data sets and the relation probability set to obtain a plurality of addition data and a plurality of addition time points. And arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set. The change condition of the predicted data with smaller predicted time point distance is more in line with the change condition of the detected data, the prediction accuracy is improved, and the technical effect of improving the data detection accuracy is achieved.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a data processing method and system based on a neural network.
Background
At present, the application of big data is wider, and the method has the advantages of rapidness, convenience and high flexibility for data detection. The big data can not only improve the efficiency of people for utilizing the data and forecast unknown conditions, but also remarkably improve the detection accuracy.
However, since many data are difficult to detect at present, the acquired detection data are small, and the current detection method of big data cannot be satisfied, so that the accuracy for detection is not accurate enough. The random addition of data only increases the negative sample, and the prediction accuracy of the positive sample cannot be improved. The lack of data results in an inaccurate test which is already a significant cause of the inaccuracy of the test. Therefore, how to obtain a small amount of detection data in real life and expand the detection data into a large amount of data is needed to meet the requirement of improving the accuracy of data detection, which is a problem that people need to solve.
Disclosure of Invention
The invention aims to provide a data processing method and system based on a neural network, which are used for solving the problems in the prior art.
In a first aspect, an embodiment of the present invention provides a data processing method based on a neural network, including:
obtaining detection data; the detection data are data which are detected at a plurality of time points and are related to a detection target; the detection data comprises a plurality of data categories; the data categories represent different kinds of data that can be used for detection;
arranging the detection data from small to large according to time to obtain a plurality of detection data sets; the detection data set comprises a plurality of detection data and a plurality of detection time points; one detection data set corresponds to one data category; one detection data corresponds to one detection time point;
Based on a plurality of detection data sets, category relation detection is carried out, and a relation probability set is obtained; the set of relationship probabilities includes a plurality of relationship probabilities; the relationship probability represents the degree to which the class of data affects detection; one relationship probability corresponds to one data category;
based on the detection data set and the relation probability set, predicting data of the middle points of the intervals of the time points through a data processing network to obtain a plurality of addition data and a plurality of addition time points; one addition data corresponds to one addition time point;
and arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set.
Optionally, the detecting the category relation based on the plurality of detection data sets to obtain a relation probability set includes:
obtaining boundary data and boundary time according to the detection data set; the boundary data are detection data of which the corresponding detection target is at the boundary; the boundary time is the time corresponding to the boundary data;
subtracting the detection data in the detection data set from the boundary data respectively to obtain a lowest boundary data difference value and a lowest boundary time; the lowest boundary data difference value is larger than the difference value between other detection data and boundary data; the lowest boundary time is a detection time point corresponding to detection data corresponding to the lowest boundary data difference value;
Performing curve fitting on the detection data in the plurality of detection data sets for a plurality of times, and extracting information of the change of the detection data along with time to obtain a curve class value and a curve degree value; the curve class value represents the variation type of the detection data with time; the curve degree value represents the change speed of the detection data with time;
obtaining a relation probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value and the lowest boundary time;
and obtaining a plurality of relation probability values according to the plurality of detection data sets, and writing the plurality of relation probability values into the sets to obtain the relation probability sets.
Optionally, the obtaining the relationship probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value and the lowest boundary time includes:
obtaining curve relation probability according to the quotient of the curve degree value divided by the curve class value;
obtaining a total time; the total time represents a time when a difference between time points is greater than a difference of other detection time points in the detection set;
dividing the lowest boundary time by the total time to obtain a time ratio;
The value of the ordinate after the time ratio is input into the normal distribution curve is used as a normal distribution value;
dividing 1 by a normal distribution value to obtain a boundary time relation probability;
dividing the lowest boundary data difference value by the sum of the lowest boundary data difference value and boundary data to obtain boundary data relation probability;
and summing the curve relation probability, the boundary time relation probability and the boundary data relation probability to obtain a relation probability value.
Optionally, the training method of the data processing network includes:
obtaining a plurality of training sets based on the detection data sets; the training set comprises a first-level data set and a second-level data set; the first-level data set comprises detection data with the same interval of time points and detection time points; the values in the second-level data set represent detection data and detection time points in the middle between every two adjacent times in the first-level data set; the values in the first-level data set and the values in the second-level data set are the values contained in the detection data set;
setting detection data in the first-level data set as training data; setting the detection data in the second data set as labeling data;
Inputting the corresponding relation probabilities in the training data and the relation probability set into a data processing network, and predicting detection data of the intermediate time to obtain a prediction data set, wherein the prediction data set comprises a plurality of prediction data;
calculating loss of the prediction data and the labeling data corresponding to the prediction data to obtain a loss value;
and carrying out backward propagation through the loss value to obtain a trained data processing network.
Optionally, the inputting the training data and the corresponding relation probability in the relation probability set into a data processing network, predicting the detection data of the middle time point, and obtaining a predicted data set includes:
constructing a two-dimensional array by using the training data and the corresponding relation probabilities in the relation probability set; the rows of the two-dimensional array represent the detection data and the relation probability of a time point; the column of the two-dimensional array represents the relation probability or one of the detection data of a plurality of time points;
carrying out one-dimensional convolution on the two-dimensional array in the direction of the time point by taking the step length as 1 to obtain a data time feature vector; the length of the data time feature vector is the product of the number of columns of training data and the number of convolution kernel channels;
inputting the data time feature vector into a first neural network to obtain a first neural vector;
Inputting the first nerve vector into a recovery nerve network to obtain a prediction data set; the number of values in the prediction data set is the number of columns of training data minus 1.
Optionally, the obtaining a plurality of training sets based on the detection data set includes:
subtracting the detection time points in the detection data set to obtain a plurality of time point distances;
obtaining a first time point interval; the first time point interval represents a time point interval greater than other time point intervals;
writing the detection data with the time point spacing of the detection data set as the first time point spacing and the detection time points into the first time point spacing data set;
obtaining a second set of points in time; the second time point set is a detection time point with the difference of the detection time points in the first time point interval data set being one half of the time point interval;
writing detection data and detection time points corresponding to a second time point set in the detection data set into a second data set of the first training set to obtain a second time set;
adding the detection time point in the second-stage time set to the detection time point of the first time point interval data set to obtain a first fusion time point, and inputting the first fusion time point and detection data corresponding to the first fusion time point into the first-stage data set of the first training set;
Subtracting the detection time point of the first time point interval data set from the detection time point in the second-stage time set to obtain a second fusion time point, and inputting the second fusion time point and detection data corresponding to the second fusion time point into the first-stage data set of the first training set; and obtaining a plurality of training sets by obtaining detection data corresponding to one half of the time point spacing of the last training set for a plurality of times.
Optionally, the obtaining a plurality of training sets by obtaining detection data corresponding to a half time point interval of a previous training set multiple times includes:
obtaining first-level data of a second training set; the first-level data of the second training set are detection data with the time point spacing being one half of the first time point spacing in the detection data set;
the second level data set of the first training set is different from the detection data in the first level data of the second training set.
Optionally, the data processing network sequentially inputs the training sets with the time point intervals from large to small into the data processing network until the loss value is smaller than a first threshold value;
and crossing the training set with large time point spacing and the training set with small time point spacing into a data processing network until the loss value is smaller than a second threshold value to obtain a trained data processing network.
Optionally, the predicting, based on the detection data set and the relation probability set, data of a midpoint of a distance between time points through a data processing network, to obtain a plurality of added data and a plurality of added time points includes:
obtaining a first detection time point interval; the first detection time point distance is that the difference value between two adjacent detection time points in the detection set is larger than the difference value between two other adjacent detection time points;
obtaining a first detection data set according to the first detection time point interval; the first detection data set comprises detection time points with a first detection time point interval and corresponding detection data;
inputting the first detection data set into a data processing network, and predicting a plurality of data to obtain a first addition time data set; the first addition time data set comprises a plurality of predicted time points with half of the time point intervals and corresponding predicted data; the predicted data are data obtained by data processing network prediction;
adding the first adding time data set into the detection set to obtain a first adding set;
obtaining a new adding set by finding the adding set for a plurality of times according to the detection data with fixed time point spacing and inputting a data processing network until the set time point spacing is reached, so as to obtain a plurality of adding data and a plurality of adding time points; the detection data set and the first addition set are one of the addition sets in the cyclic process.
In a second aspect, an embodiment of the present invention provides a data processing system based on a neural network, including:
the acquisition module is used for: obtaining detection data; the detection data are data which are detected at a plurality of time points and are related to a detection target; the detection data comprises a plurality of data categories; the data categories represent different kinds of data that can be used for detection; arranging the detection data from small to large according to time to obtain a plurality of detection data sets; the detection data set comprises a plurality of detection data and a plurality of detection time points; one detection data set corresponds to one data category; one detection data corresponds to one detection time point;
and a relation module: based on a plurality of detection data sets, category relation detection is carried out, and a relation probability set is obtained; the set of relationship probabilities includes a plurality of relationship probabilities; the relationship probability represents the degree to which the class of data affects detection; one relationship probability corresponds to one data category;
and a prediction module: based on the detection data set and the relation probability set, predicting data of the middle points of the intervals of the time points through a data processing network to obtain a plurality of addition data and a plurality of addition time points; one addition data corresponds to one addition time point;
And (3) an arrangement module: and arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set. Compared with the prior art, the embodiment of the invention achieves the following beneficial effects:
the embodiment of the invention also provides a data processing method and a system based on the neural network, wherein the method comprises the following steps: detection data is obtained. The detection data is data related to a detection target detected at a plurality of time points. The detection data includes a plurality of data categories. The data categories represent different kinds of data that can be used for detection. And arranging the detection data from small to large according to time to obtain a plurality of detection data sets. The detection data set includes a plurality of detection data and a plurality of detection time points. One set of detection data corresponds to one class of data. One detection data corresponds to one detection time point. Based on the plurality of detection data sets, category relation detection is carried out, and a relation probability set is obtained. The set of relationship probabilities includes a plurality of relationship probabilities. The relationship probability represents the extent to which the class of data affects detection. One relationship probability corresponds to one data category. And based on the detection data set and the relation probability set, predicting data of the middle points of the intervals of the time points through a data processing network to obtain a plurality of addition data and a plurality of addition time points. One addition data corresponds to one addition time point. And arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set.
The invention expands a small amount of detection data which is difficult to detect into a large amount of data. When the data processing network is trained and used for detection, the same detection data set is adopted, so that the change condition of the predicted data with smaller predicted time point spacing is more consistent with the change condition of the detection data. And the accuracy of prediction is improved by judging the relation and curve conditions of different types of data. The addition of the prediction data to the detection data having different detection time points is performed at the same detection time point pitch. The used data set after data processing can be used for training other networks capable of detecting information later, the detection network can be trained more accurately under big data, the accuracy of the detection network is improved, and the accuracy of data detection is further improved. In conclusion, the invention solves the technical problem of low data detection accuracy, and achieves the technical effect of improving the data detection accuracy.
Drawings
Fig. 1 is a flowchart of a data processing method based on a neural network according to an embodiment of the present invention.
Fig. 2 is a schematic block diagram of an electronic device according to an embodiment of the present invention.
The marks in the figure: a bus 500; a receiver 501; a processor 502; a transmitter 503; a memory 504; bus interface 505.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings.
Example 1
As shown in fig. 1, an embodiment of the present invention provides a data processing method based on a neural network, where the method includes:
s101: detection data is obtained. The detection data are data related to the detection target at a plurality of time points. The detection data includes a plurality of data categories. The data categories represent different kinds of data that can be used for detection. The intervals of the time points at which the detection data are detected may be the same or different.
The time points of detection corresponding to part of the detection data in this embodiment are 0,4,6,8, 12, 20. The time point interval of the detection points is 4,2,2,4,8.
Wherein, the detection target can be any kind of thing which is not easy to detect.
The detection data can be network data, agricultural data, vehicle data, pollution data and the like. The detection data can be used for various detections. In the present embodiment, the detection data used is a VOC data set (Visual Object Class, VOC), and the detection target in the present embodiment is set as a vehicle.
The data category is detected data of various categories related to one detection target, and can be used for establishing a relation between the data and judging the influence of the sensitivity degree. The detection target as in the present embodiment is a vehicle, and the adopted voc data set includes 4 related categories of the position of a bicycle, the position of an automobile, the position of a motorcycle, the position of a bus, and the other 16: the location of the person, the location of the animal (bird, cat, cow, dog, horse, sheep), the location of the vehicle (airplane, boat, train), the location of the furniture (bottle, chair, dining table, plant), the location of the television, and the location of the display. If other detection targets can be adopted, such as crop growth data, the detection targets comprise different types of detection data such as plant height of crops, flowering and fruiting conditions of the crops, watering and fertilizing conditions of the crops, chemical element content in soil of the crops and the like.
S102: and arranging the detection data from small to large according to time to obtain a plurality of detection data sets. The detection data set includes a plurality of detection data and a plurality of detection time points. One set of detection data corresponds to one class of data. One detection data corresponds to one detection time point.
The storage mode in the detection data set is to store one data and store a corresponding event, and in this embodiment, the lower left corner of the image is taken as a zero point, and the lower left corner to the lower right corner are taken as abscissa. From the lower left corner to the upper left corner, the first value 300-312-24-56 indicates that the position of the abscissa of the center point is 300, and the position of the ordinate of the center point is 312.
The voc data set used in this embodiment includes 20 categories of bicycle position, car position, motorcycle position, bus position, person position, animal (bird, cat, cow, dog, horse, sheep) position, vehicle (plane, ship, train) position, furniture (bottle, chair, dining table, plant) position, television position, and display position, so there are 20 detection data sets. Including multiple cases where multiple points in time pass by an area.
S103: based on the plurality of detection data sets, category relation detection is carried out, and a relation probability set is obtained. The set of relationship probabilities includes a plurality of relationship probabilities. The relationship probability represents the extent to which the class of data affects detection. One relationship probability corresponds to one data category.
Wherein the values in the set of relationships represent the extent to which the class of data affects detection. The length of the relation set is the same as the number of the data categories.
Wherein the value in the relation set corresponding to the data category which affects the detection is set as the relation probability, and the value in the relation set corresponding to the data category which does not affect the detection is set as 0. The relationship probability represents the extent to which the class of data affects detection.
In this embodiment, the value in the relationship set corresponding to 4 data categories of bicycle, car, motorcycle, bus is set as the relationship probability, and the value in the relationship set corresponding to human, animal (bird, cat, cow, dog, horse, sheep), vehicle (airplane, ship, train), furniture (bottle, chair, dining table, plant), television, and display is set as 0.
S104: based on the plurality of detection data sets and the relation probability set, predicting data of the middle point of the time point distance through a data processing network to obtain a plurality of addition data and a plurality of addition time points.
The time points of detection corresponding to part of the detection data in this embodiment are 0,4,6,8, 12, 20, and the unit is seconds. The time point interval of the detection points is 4,2,2,4,8. The addition data are predicted detection data of the 2 nd second, the 10 th second, the 14 th second, the 16 th second and the 18 th second.
S105: and arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set.
The data in the use data set comprises detection data and addition data, and does not comprise a detection time point and an addition time point. The time points of the data in the usage data set are equally spaced.
In this embodiment, the data in the data set includes detection data of 0 th second, detection data of 2 nd second, detection data of 4 th second, detection data of 6 th second, detection data of 8 th second, detection data of 10 th second, detection data of 12 th second, detection data of 14 th second, detection data of 16 th second, detection data of 18 th second, and detection data of 20 th second.
Optionally, the detecting the category relation based on the detection data of the plurality of data categories to obtain a relation probability set includes:
and obtaining boundary data and boundary time according to the detection data set. The boundary data is detection data of which the corresponding detection target is at the boundary. The boundary time is the time corresponding to the boundary data.
The boundary data in this embodiment is a position where the vehicle detection is least accurate. As in the present embodiment, the position where the vehicle detects least accurately is typically the boundary position.
The boundary time is the time with the lowest influence of the detection data.
And subtracting the detection data in the detection data set from the boundary data respectively to obtain a lowest boundary data difference value and a lowest boundary time. The difference value of the lowest boundary data of the boundary is larger than the difference value of other detection data and boundary data. And the lowest boundary time is a detection time point corresponding to the detection data corresponding to the lowest boundary data difference value.
Wherein the boundary lowest boundary data difference value represents the difference and the time of the difference between the boundary data and the data which is the most different from the boundary data which affects the detection lowest.
And performing curve fitting on the detection data in the plurality of detection data sets for a plurality of times, extracting information of the change of the detection data along with time, and obtaining a curve class value and a curve degree value. The curve class value represents a variation class of the detection data over time. The curve degree value represents a time-varying change speed of the detection data.
Wherein, a curve class value of 1 corresponding to the detection data in the detection data set indicates that the accuracy of detection of the detection data is approximately increased along with the increase of time. A curve class value of 2 indicates that the accuracy of detection of the detection data is substantially reduced with an increase in time. A curve class value of 3 indicates that the accuracy of detection of the detection data increases with time as the curve changes. A curve class value of 4 indicates that the detection data increases with time and the accuracy of detection varies irregularly.
Wherein the curve class value is 1 or the curve class value is 2, and the curve degree value represents the slope. And if the curve degree value is 0 when the curve type value is 2, and if the curve degree value is 3, the curve change value is lower as the unknown coefficient is more, the curve degree value is sequentially set from 1 to 9 according to the number of the unknown coefficients, and if the unknown coefficient is more than 10 or equal to 10, the curve degree value is set to be 10. Because the more complex the curve, the faster the description fluctuates.
And obtaining a relation probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value and the lowest boundary time.
And obtaining a plurality of relation probability values according to the plurality of detection data sets, and writing the plurality of relation probability values into the sets to obtain the relation probability sets. By the method, the influence of a plurality of data types on the detection value is generally detected, and the detection data is input into a network capable of detecting, so that the accuracy of detection is judged. The invention can directly judge the relation of the detection data changing along with time without inputting a network. And finding out the detection condition corresponding to each category according to the boundary and curve conditions, thereby judging the relationship between the data category and the detection condition. And judging and obtaining curve conditions of data of different categories.
Optionally, the obtaining the relationship probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value and the lowest boundary time includes:
and obtaining the curve relation probability according to the quotient of the curve degree value divided by the curve class value.
The curve relation probability represents a curve with a curve degree smaller as the curve class value is larger, and the relation between time and detection accuracy is larger. The curve relation probability represents a curve with a curve degree larger as the curve class value is smaller, and the relation between time and detection accuracy is smaller.
The total time is obtained. The total time represents a time when a difference between time points is greater than a difference of other detection time points in the detection set.
Dividing the lowest boundary time by the total time to obtain a time ratio.
And taking the value of the ordinate after the time ratio is input into the normal distribution curve as the normal distribution value.
Dividing 1 by a normal distribution value to obtain the boundary time relation probability.
Wherein the lowest boundary time indicates low curve fluctuation on both sides.
And dividing the lowest boundary data difference value by the sum of the lowest boundary data difference value and boundary data to obtain the boundary data relation probability.
And adding the lowest boundary data difference value with the boundary data to obtain an added value, and dividing the lowest boundary data difference value by the added value to obtain the boundary data relation probability.
The larger the lowest data difference value is, the larger the degree of dissimilarity between the lowest data difference value and other detection data is, the larger the influence of detection is, the smaller the lowest data difference value is, the smaller the degree of dissimilarity between the lowest data difference value and other detection data is, and the smaller the influence of detection is.
And summing the curve relation probability, the boundary time relation probability and the boundary data relation probability to obtain a relation probability value.
Optionally, the training method of the data processing network includes:
and obtaining a plurality of training sets based on the detection data sets. The training set includes a first level data set and a second level data set. The first-level data set comprises detection data and detection time points with the same interval of the time points. The values in the second-level data set represent detection data and detection time points in the middle between every two adjacent times in the first-level data set. The values in the first level data set and the values in the second level data set are values contained in the detection data set.
Firstly training a network according to detection data with the time point spacing being larger than other time point spacing, obtaining a predicted value with the time point spacing being one half, and solving the loss with a smaller value in a detection set. And continuing to calculate the loss by using the detection data with the time point spacing of one half in the detection data set as training data.
The detection data with the time interval larger than the other time intervals are not detection data with all the time intervals larger than the other time intervals in the detection data set. Only detection data including a time-point interval of one-half is in the detection set, and detection of the time-point interval is also in the detection set.
And setting the detection data in the first-level data set as training data. And setting the detection data in the second data set as labeling data.
Inputting the corresponding relation probabilities in the training data and the relation probability set into a data processing network, predicting the detection data of the time point in the middle to obtain a prediction data set, wherein the prediction data set comprises a plurality of prediction data.
And obtaining the loss by the prediction data and the labeling data corresponding to the prediction data, and obtaining a loss value.
Wherein the loss is calculated by a cross entropy loss function.
And carrying out backward propagation through the loss value to obtain a trained data processing network.
According to the method, firstly, a network is trained according to the detection data with the maximum time point interval, a predicted value with the time interval of one half is obtained, and the loss is calculated from the value with the time interval of one half in the detection set. And continuing to calculate the loss by taking the detection data with the time interval of one half in the detection set as training data. Through many times, the detection data is used as not only training data but also labeling data, so that the training data is saved, and the network can be trained under the condition of less data. Optionally, the inputting the training data and the corresponding relation probability in the relation probability set into a data processing network, predicting the detection data of the middle time point, and obtaining a predicted data set includes:
and constructing a two-dimensional array by the training data and the corresponding relation probabilities in the relation probability set. The rows of the two-dimensional array represent the detection data and the relation probability for a point in time. The column of the two-dimensional array represents a probability of a relationship or one of the detection data for a plurality of points in time.
Carrying out one-dimensional convolution on the two-dimensional array in the direction of the time point by taking the step length as 1 to obtain a data time feature vector; the length of the data time feature vector is the product of the number of columns of training data and the number of convolution kernel channels.
In this embodiment, the number of convolution kernels is 1024. One-dimensional convolution is carried out by using a convolution kernel of 2x1x1024 with a step length of 1 to obtain a data time feature vector of 1x1x 1024.
And inputting the data time feature vector into a first neural network to obtain a first neural vector.
Wherein the first neural network is a deep neural network (Deep Nueral Network, DNN).
Wherein the first neural vector represents a characteristic of the detection data between points in time.
And inputting the first nerve vector into a recovery nerve network to obtain a prediction data set. The number of values in the prediction data set is the number of columns of training data minus 1.
Wherein the restoration neural network is a deep neural network (Deep Nueral Network, DNN). The number of neurons of the output layer of the recovery neural network is the number of columns of training data minus 1.
By the method, the detection data is firstly subjected to feature extraction, then two values are converted to obtain a value, the value of which is reduced by 1 is obtained, the feature of the change of the relation probability and the detection data at the time point can be fused, and then the detection data between two adjacent time points is restored.
Optionally, the obtaining a plurality of training sets based on the detection data set includes:
And subtracting the detection time points in the detection data set to obtain a plurality of time point distances.
The time points of detection corresponding to part of the detection data in this embodiment are 0,4,6,8, 12, 20. The time point spacing is 20-0=20, 20-4=16, 20-6=14, 20-8=12-0=12, 20-12=12-4=8-0=8, 12-6=6-0=6, 12-8=8-4=4-0=4, 8-6=6-4=2.
A first time point spacing is obtained. The first time point interval represents a time point interval that is greater than other time point intervals.
In this embodiment, the first time interval is 20.
And writing the detection data with the time point spacing of the detection data set as the first time point spacing and the detection time points into the first time point spacing data set.
Wherein the first time point spacing dataset comprises all detection data in the detection dataset that differ by a first interval.
In this embodiment, the first time point interval is 20. The first time point interval data set includes detection data corresponding to the 0 th second and the 20 th second.
A second set of points in time is obtained. The second time point set is the detection time point with the difference of the detection time points in the first time point interval data set being one half time point interval.
Wherein, the detection time point and the corresponding detection data in the second time point set do not necessarily exist in the detection data set.
And writing the detection data and the detection time point corresponding to the second time point set in the detection data set into a second data set of the first training set to obtain a second time set.
And if the second time point set cannot find the detection data of the time point convolution with the data finding time point interval of one half.
In this embodiment, the detection time points in the time point interval data set include the 0 th second and the 20 th second, and the time point interval of the second time point set is (20-0)/2=20/2=10. The detection time points with the time point interval of 10 do not exist in the detection set. The next time point spacing is obtained as 16. The first time point interval data set includes detection data corresponding to the 4 th second and the 20 th second. The second set of time points is (20-4)/2=16/2=8, 4+8=12. The second data set includes 12 corresponding detection data. The second level time set includes 12 th seconds.
And adding detection data and detection time points corresponding to the detection time points of the first time point interval data set in the second-stage time set to the first-stage data set of the first training set.
And inputting the detection data corresponding to the detection time point of the first time point interval data set subtracted from the detection time point in the second-stage time set and the detection time point into the first-stage data set of the first training set.
In this embodiment, as in 12+8=20, 12-8=4, and the detection data corresponding to the 4 th second and the 20 th second are input into the first-stage data set.
By the method, the value in the second-level data set is the midpoint of the first-level data set. And subtracting every two adjacent values in the current data time set to obtain a phase difference set.
Optionally, the obtaining a plurality of training sets by obtaining detection data corresponding to a half time point interval of a previous training set multiple times includes:
first level data of a second training set is obtained. The first-level data of the second training set is detection data of which the time point interval is one half of the first time point interval in the detection data set.
The second level data set of the first training set may be different from the detection data in the first level data of the second training set.
Optionally, the data processing network sequentially inputs the data processing network by using training sets with time point intervals from large to small until the loss value is smaller than a first threshold value.
Wherein the first threshold is 0.8.
And crossing the training set with large time point spacing and the training set with small time point spacing into a data processing network until the loss value is smaller than a second threshold value to obtain a trained data processing network.
Wherein the second threshold is 0.9.
By the method, the data processing network is more accurate from large to small, and input is intersected, so that the adjustment of some curing parameters can be more accurate in setting up the neural network.
In summary, based on the detection data set, obtaining a plurality of training sets, wherein the plurality of training sets at least comprise a first-stage data set and a second-stage data set; the first-level data set comprises detection data with the same interval of time points and detection time points; the values in the second-level data set represent detection data and detection time points in the middle between every two adjacent times in the first-level data set; the values in the first-level data set and the values in the second-level data set are both in the detection set, and the values in the first-level data set and the values in the second-level data set are the values contained in the detection data set.
s1: acquiring a first training set:
and subtracting the detection time points in the detection data set to obtain a plurality of time point distances.
Obtaining a first time point interval; the first time point interval represents a time point interval that is greater than other time point intervals.
And writing the detection data with the time point spacing of the detection data set as the first time point spacing and the detection time points into the first time point spacing data set.
Obtaining a second set of points in time; the second time point set is the detection time point with the difference of the detection time points in the first time point interval data set being one half time point interval.
And writing the detection data and the detection time point corresponding to the second time point set in the detection data set into a second data set of the first training set to obtain a second time set.
Adding the detection time point in the second-stage time set to the detection time point of the first time point interval data set to obtain a first fusion time point, and inputting the first fusion time point, the detection data corresponding to the first fusion time point and the detection time point into the first-stage data set of the first training set.
Subtracting the detection time point of the first time point interval data set from the detection time point in the second-stage time set to obtain a second fusion time point, and inputting the second fusion time point, detection data corresponding to the second fusion time point and the detection time point into the first-stage data set of the first training set.
s2: obtaining a second time point interval; the second time point interval is the second largest time point interval of the plurality of time point intervals;
and writing the detection data with the time point spacing of the second time point spacing in the detection data set and the detection time points into the second time point spacing data set.
Obtaining a third set of points in time; the third time point set is the detection time point with the difference of the detection time points in the second time point interval data set being one half time point interval.
And writing the detection data and the detection time point corresponding to the third time point set in the detection data set into a second data set of the second training set to obtain a third-level time set.
And adding the detection time point in the third-stage time set with the detection time point of the second time point interval data set to obtain a third fusion time point, and inputting the third fusion time point, detection data corresponding to the third fusion time point and the detection time point into the first-stage data set of the second training set.
Subtracting the detection time point of the second time point interval data set from the detection time point in the third-stage time set to obtain a fourth fusion time point, and inputting the fourth fusion time point, detection data corresponding to the fourth fusion time point and the detection time point into the first-stage data set of the second training set.
Repeating the operations s1 and s2 to obtain a plurality of training sets.
Optionally, the predicting, based on the detection data set and the relation probability set, data of a midpoint of a distance between time points through a data processing network, to obtain a plurality of added data and a plurality of added time points includes:
a first detection time point spacing is obtained. The first detection time point distance is that the difference value between two adjacent detection time points in the detection set is larger than the difference value between two other adjacent detection time points.
In this embodiment, the detection time points corresponding to the partial detection data are 0,4,6,8, 12, 20. 4-0=12-8=4, 6-4=8-6=2, 20-12=8, then the first detection time point interval is 8.
And obtaining a first detection data set according to the first detection time point interval. The first set of detection data includes detection time points and corresponding detection data that differ by a first detection time point spacing.
The detection time in the first detection data set is the detection time point of which the difference value between every two adjacent detection time points is the first detection time point interval.
In this embodiment, the detection time points corresponding to the partial detection data are 0,4,6,8, 12, 20. The first detection time point interval is 8. The detection points of the first detection data set include 12 th and 20 th seconds, and detection data corresponding to 12 th and 20 th seconds.
And inputting the first detection data set into a data processing network, and predicting a plurality of data to obtain a first addition time data set. The first set of addition time data includes predicted points in time and corresponding predicted data at a plurality of point in time intervals of one-half. The predicted data is data predicted by a data processing network.
In this embodiment, the prediction data of 16 th second between 12 th second and 20 th second is predicted.
And adding the first adding time data set into the detection set to obtain a first adding set.
The detection time points in the first adding set are 0,4,6,8, 12, 16, 20 and corresponding detection data.
And obtaining a new adding set by finding the adding set for multiple times according to the detection data with fixed time point spacing and inputting a data processing network until the set time point spacing is reached, so as to obtain a plurality of adding data and a plurality of adding time points. The detection data set and the first addition set are one of the addition sets in the cyclic process.
In this embodiment, the set time interval is 2. The last obtained multiple addition time points in this example were 0,2,4,6,8, 10, 12, 14, 16, 18, 20 and the corresponding 11 addition data.
By the method, the data with unknown detection data in the middle of each adjacent detection time point is judged, and the detection data with the same time interval can be detected.
Example 2
Based on the data processing method based on the neural network, the embodiment of the invention also provides a data processing system based on the neural network, which comprises an acquisition module, a relation module, a prediction module and an arrangement module.
The acquisition module is used for acquiring detection data. The detection data is data related to a detection target detected at a plurality of time points. The detection data includes a plurality of data categories. The data categories represent different kinds of data that can be used for detection. And arranging the detection data from small to large according to time to obtain a plurality of detection data sets. The detection data set includes a plurality of detection data and a plurality of detection time points. One set of detection data corresponds to one class of data. One detection data corresponds to one detection time point.
The relation module is used for detecting category relations based on the plurality of detection data sets to obtain a relation probability set. The set of relationship probabilities includes a plurality of relationship probabilities. The relationship probability represents the extent to which the class of data affects detection. One relationship probability corresponds to one data category.
The prediction module is used for predicting data of the middle point of the interval of the time points through the data processing network based on the detection data set and the relation probability set to obtain a plurality of added data and a plurality of added time points. One addition data corresponds to one addition time point.
The arrangement module is used for arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set.
The specific manner in which the various modules perform the operations in the systems of the above embodiments have been described in detail herein with respect to the embodiments of the method, and will not be described in detail herein.
An embodiment of the present invention further provides an electronic device, as shown in fig. 2, including a memory 504, a processor 502, and a computer program stored in the memory 504 and executable on the processor 502, where the processor 502 implements the steps of any of the above-described data processing methods based on a neural network when executing the program.
Where in FIG. 2 a bus architecture (represented by bus 500), bus 500 may include any number of interconnected buses and bridges, with bus 500 linking together various circuits, including one or more processors, represented by processor 502, and memory, represented by memory 504. Bus 500 may also link together various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., as are well known in the art and, therefore, will not be described further herein. Bus interface 505 provides an interface between bus 500 and receiver 501 and transmitter 503. The receiver 501 and the transmitter 503 may be the same element, i.e. a transceiver, providing a means for communicating with various other apparatus over a transmission medium. The processor 502 is responsible for managing the bus 500 and general processing, while the memory 504 may be used to store data used by the processor 502 in performing operations. The algorithms and displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general-purpose systems may also be used with the teachings herein. The required structure for a construction of such a system is apparent from the description above. In addition, the present invention is not directed to any particular programming language. It will be appreciated that the teachings of the present invention described herein may be implemented in a variety of programming languages, and the above description of specific languages is provided for disclosure of enablement and best mode of the present invention.
In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Claims (8)
1. A data processing method based on a neural network, comprising:
obtaining detection data; the detection data are data which are detected at a plurality of time points and are related to a detection target; the detection data comprises a plurality of data categories; the data categories represent different kinds of data that can be used for detection;
arranging the detection data from small to large according to time to obtain a plurality of detection data sets; the detection data set comprises a plurality of detection data and a plurality of detection time points; one detection data set corresponds to one data category; one detection data corresponds to one detection time point;
based on a plurality of detection data sets, category relation detection is carried out, and a relation probability set is obtained; the set of relationship probabilities includes a plurality of relationship probabilities; the relationship probability represents the degree to which the class of data affects detection; one relationship probability corresponds to one data category;
Based on the detection data set and the relation probability set, predicting data of the middle points of the intervals of the time points through a data processing network to obtain a plurality of addition data and a plurality of addition time points; one addition data corresponds to one addition time point;
arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set;
based on a plurality of detection data sets, performing category relation detection to obtain a relation probability set, including:
obtaining boundary data and boundary time according to the detection data set; the boundary data are detection data of which the corresponding detection target is at the boundary; the boundary time is the time corresponding to the boundary data;
subtracting the detection data in the detection data set from the boundary data respectively to obtain a lowest boundary data difference value and a lowest boundary time; the lowest boundary data difference value is larger than the difference value between other detection data and boundary data; the lowest boundary time is a detection time point corresponding to detection data corresponding to the lowest boundary data difference value;
performing curve fitting on the detection data in the plurality of detection data sets for a plurality of times, and extracting information of the change of the detection data along with time to obtain a curve class value and a curve degree value; the curve class value represents the variation type of the detection data with time; the curve degree value represents the change speed of the detection data with time;
Obtaining a relation probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value and the lowest boundary time;
obtaining a plurality of relation probability values according to a plurality of the detection data sets, and writing the plurality of relation probability values into the sets to obtain a relation probability set;
based on the detection data set and the relation probability set, predicting data of middle points of the intervals of time points through a data processing network to obtain a plurality of added data and a plurality of added time points, wherein the method comprises the following steps of:
obtaining a first detection time point interval; the first detection time point distance is that the difference value between two adjacent detection time points in the detection set is larger than the difference value between two other adjacent detection time points;
obtaining a first detection data set according to the first detection time point interval; the first detection data set comprises detection time points with a first detection time point interval and corresponding detection data;
inputting the first detection data set into a data processing network, and predicting a plurality of data to obtain a first addition time data set; the first addition time data set comprises a plurality of predicted time points with half of the time point intervals and corresponding predicted data; the predicted data are data obtained by data processing network prediction;
Adding the first adding time data set into the detection set to obtain a first adding set;
obtaining a new adding set by finding the adding set for a plurality of times according to the detection data with fixed time point spacing and inputting a data processing network until the set time point spacing is reached, so as to obtain a plurality of adding data and a plurality of adding time points; the detection data set and the first addition set are one of the addition sets in the cyclic process.
2. The method of claim 1, wherein the obtaining the relationship probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value, and the lowest boundary time comprises:
obtaining curve relation probability according to the quotient of the curve degree value divided by the curve class value;
obtaining a total time; the total time represents a time when a difference between time points is greater than a difference of other detection time points in the detection set;
dividing the lowest boundary time by the total time to obtain a time ratio;
the value of the ordinate after the time ratio is input into the normal distribution curve is used as a normal distribution value;
dividing 1 by a normal distribution value to obtain a boundary time relation probability;
Dividing the lowest boundary data difference value by the sum of the lowest boundary data difference value and boundary data to obtain boundary data relation probability;
and summing the curve relation probability, the boundary time relation probability and the boundary data relation probability to obtain a relation probability value.
3. A data processing method based on a neural network according to claim 1, characterized in that the training method of the data processing network comprises:
obtaining a plurality of training sets based on the detection data sets; the training set comprises a first-level data set and a second-level data set; the first-level data set comprises detection data with the same interval of time points and detection time points; the values in the second-level data set represent detection data and detection time points in the middle between every two adjacent times in the first-level data set; the values in the first-level data set and the values in the second-level data set are the values contained in the detection data set;
setting detection data in the first-level data set as training data; setting the detection data in the second data set as labeling data;
inputting the corresponding relation probabilities in the training data and the relation probability set into a data processing network, predicting detection data of a time point in the middle to obtain a prediction data set, wherein the prediction data set comprises a plurality of prediction data;
Calculating loss of the prediction data and the labeling data corresponding to the prediction data to obtain a loss value;
and carrying out backward propagation through the loss value to obtain a trained data processing network.
4. A data processing method based on a neural network according to claim 3, wherein inputting the training data and the corresponding relation probability in the relation probability set into the data processing network, predicting the detection data of the intermediate time point, and obtaining the predicted data set includes:
constructing a two-dimensional array by using the training data and the corresponding relation probabilities in the relation probability set; the rows of the two-dimensional array represent the detection data and the relation probability of a time point; the column of the two-dimensional array represents the relation probability or one of the detection data of a plurality of time points;
carrying out one-dimensional convolution on the two-dimensional array in the direction of the time point by taking the step length as 1 to obtain a data time feature vector; the length of the data time feature vector is the product of the number of columns of training data and the number of convolution kernel channels;
inputting the data time feature vector into a first neural network to obtain a first neural vector;
inputting the first nerve vector into a recovery nerve network to obtain a prediction data set; the number of values in the prediction data set is the number of columns of training data minus 1.
5. A data processing method based on a neural network according to claim 3, wherein the obtaining a plurality of training sets based on the detected data set includes:
subtracting the detection time points in the detection data set to obtain a plurality of time point distances;
obtaining a first time point interval; the first time point interval represents a time point interval greater than other time point intervals;
writing the detection data with the time point spacing of the detection data set as the first time point spacing and the detection time points into the first time point spacing data set;
obtaining a second set of points in time; the second time point set is a detection time point with the difference of the detection time points in the first time point interval data set being one half of the time point interval;
writing detection data and detection time points corresponding to a second time point set in the detection data set into a second data set of the first training set to obtain a second time set;
adding the detection time point in the second-stage time set to the detection time point of the first time point interval data set to obtain a first fusion time point, and inputting the first fusion time point and detection data corresponding to the first fusion time point into the first-stage data set of the first training set;
Subtracting the detection time point of the first time point interval data set from the detection time point in the second-stage time set to obtain a second fusion time point, and inputting the second fusion time point and detection data corresponding to the second fusion time point into the first-stage data set of the first training set; and obtaining a plurality of training sets by obtaining detection data corresponding to one half of the time point spacing of the last training set for a plurality of times.
6. The method for processing data based on a neural network according to claim 5, wherein the obtaining a plurality of training sets by obtaining the detection data corresponding to the half time point interval of the last training set a plurality of times includes:
obtaining first-level data of a second training set; the first-level data of the second training set are detection data with the time point spacing being one half of the first time point spacing in the detection data set;
the second level data set of the first training set is different from the detection data in the first level data of the second training set.
7. The data processing method based on the neural network according to claim 2, wherein the data processing network sequentially inputs the data processing network from a training set with a time point interval from large to small until the loss value is smaller than a first threshold value;
And crossing the training set with large time point spacing and the training set with small time point spacing into a data processing network until the loss value is smaller than a second threshold value to obtain a trained data processing network.
8. A data processing system based on a neural network, comprising:
the acquisition module is used for: obtaining detection data; the detection data are data which are detected at a plurality of time points and are related to a detection target; the detection data comprises a plurality of data categories; the data categories represent different kinds of data that can be used for detection; arranging the detection data from small to large according to time to obtain a plurality of detection data sets; the detection data set comprises a plurality of detection data and a plurality of detection time points; one detection data set corresponds to one data category; one detection data corresponds to one detection time point;
and a relation module: based on a plurality of detection data sets, category relation detection is carried out, and a relation probability set is obtained; the set of relationship probabilities includes a plurality of relationship probabilities; the relationship probability represents the degree to which the class of data affects detection; one relationship probability corresponds to one data category;
and a prediction module: based on the detection data set and the relation probability set, predicting data of the middle points of the intervals of the time points through a data processing network to obtain a plurality of addition data and a plurality of addition time points; one addition data corresponds to one addition time point;
And (3) an arrangement module: arranging the plurality of added data and the plurality of detection data in the detection data set from far to near in time sequence to obtain a use data set;
based on a plurality of detection data sets, performing category relation detection to obtain a relation probability set, including:
obtaining boundary data and boundary time according to the detection data set; the boundary data are detection data of which the corresponding detection target is at the boundary; the boundary time is the time corresponding to the boundary data;
subtracting the detection data in the detection data set from the boundary data respectively to obtain a lowest boundary data difference value and a lowest boundary time; the lowest boundary data difference value is larger than the difference value between other detection data and boundary data; the lowest boundary time is a detection time point corresponding to detection data corresponding to the lowest boundary data difference value;
performing curve fitting on the detection data in the plurality of detection data sets for a plurality of times, and extracting information of the change of the detection data along with time to obtain a curve class value and a curve degree value; the curve class value represents the variation type of the detection data with time; the curve degree value represents the change speed of the detection data with time;
Obtaining a relation probability value based on the curve degree value, the curve class value, the boundary data, the lowest boundary data difference value and the lowest boundary time;
obtaining a plurality of relation probability values according to a plurality of the detection data sets, and writing the plurality of relation probability values into the sets to obtain a relation probability set;
based on the detection data set and the relation probability set, predicting data of middle points of the intervals of time points through a data processing network to obtain a plurality of added data and a plurality of added time points, wherein the method comprises the following steps of:
obtaining a first detection time point interval; the first detection time point distance is that the difference value between two adjacent detection time points in the detection set is larger than the difference value between two other adjacent detection time points;
obtaining a first detection data set according to the first detection time point interval; the first detection data set comprises detection time points with a first detection time point interval and corresponding detection data;
inputting the first detection data set into a data processing network, and predicting a plurality of data to obtain a first addition time data set; the first addition time data set comprises a plurality of predicted time points with half of the time point intervals and corresponding predicted data; the predicted data are data obtained by data processing network prediction;
Adding the first adding time data set into the detection set to obtain a first adding set;
obtaining a new adding set by finding the adding set for a plurality of times according to the detection data with fixed time point spacing and inputting a data processing network until the set time point spacing is reached, so as to obtain a plurality of adding data and a plurality of adding time points; the detection data set and the first addition set are one of the addition sets in the cyclic process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311678079.9A CN117391125B (en) | 2023-12-08 | 2023-12-08 | Data processing method and system based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311678079.9A CN117391125B (en) | 2023-12-08 | 2023-12-08 | Data processing method and system based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117391125A CN117391125A (en) | 2024-01-12 |
| CN117391125B true CN117391125B (en) | 2024-03-01 |
Family
ID=89466928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311678079.9A Active CN117391125B (en) | 2023-12-08 | 2023-12-08 | Data processing method and system based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117391125B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670569A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670571A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN111325138A (en) * | 2020-02-18 | 2020-06-23 | 中国科学院合肥物质科学研究院 | Road boundary real-time detection method based on point cloud local concave-convex characteristics |
| CN112445832A (en) * | 2019-08-28 | 2021-03-05 | 北京达佳互联信息技术有限公司 | Data anomaly detection method and device, electronic equipment and storage medium |
| CN114118622A (en) * | 2021-12-08 | 2022-03-01 | 北京北大软件工程股份有限公司 | Data trend prediction method and system based on time series |
| CN115147144A (en) * | 2022-06-13 | 2022-10-04 | 阿里巴巴（中国）有限公司 | Data processing method and electronic equipment |
| CN115269108A (en) * | 2021-04-30 | 2022-11-01 | 华为云计算技术有限公司 | Data processing method, device and equipment |
| CN115640337A (en) * | 2022-12-22 | 2023-01-24 | 粤港澳大湾区数字经济研究院(福田) | Abnormal data prediction method, system and related equipment for symbolic graph network |
| CN115831217A (en) * | 2022-11-23 | 2023-03-21 | 四川大学 | Chromatin topological association domain boundary prediction method based on multi-modal fusion |
| CN116340726A (en) * | 2023-03-28 | 2023-06-27 | 广东电网有限责任公司 | Energy economy big data cleaning method, system, equipment and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10055687B2 (en) * | 2014-04-17 | 2018-08-21 | Mark B. Ring | Method for creating predictive knowledge structures from experience in an artificial agent |
| US10509586B2 (en) * | 2018-04-24 | 2019-12-17 | EMC IP Holding Company LLC | System and method for capacity forecasting in backup systems |
| US20220067571A1 (en) * | 2020-08-31 | 2022-03-03 | Mercari, Inc. | Machine-learning prediction or suggestion based on object identification |
| US20220129683A1 (en) * | 2020-10-22 | 2022-04-28 | Mobileye Vision Technologie Ltd. | Selecting data for deep learning |
| US20230305904A1 (en) * | 2022-03-25 | 2023-09-28 | Capital One Services, Llc | Systems and methods for correlating probability models with non-homogenous time dependencies to generate time-specific data processing predictions |
-
2023
- 2023-12-08 CN CN202311678079.9A patent/CN117391125B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670569A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN109670571A (en) * | 2017-10-16 | 2019-04-23 | 优酷网络技术（北京）有限公司 | Neural net prediction method and device |
| CN112445832A (en) * | 2019-08-28 | 2021-03-05 | 北京达佳互联信息技术有限公司 | Data anomaly detection method and device, electronic equipment and storage medium |
| CN111325138A (en) * | 2020-02-18 | 2020-06-23 | 中国科学院合肥物质科学研究院 | Road boundary real-time detection method based on point cloud local concave-convex characteristics |
| CN115269108A (en) * | 2021-04-30 | 2022-11-01 | 华为云计算技术有限公司 | Data processing method, device and equipment |
| CN114118622A (en) * | 2021-12-08 | 2022-03-01 | 北京北大软件工程股份有限公司 | Data trend prediction method and system based on time series |
| CN115147144A (en) * | 2022-06-13 | 2022-10-04 | 阿里巴巴（中国）有限公司 | Data processing method and electronic equipment |
| CN115831217A (en) * | 2022-11-23 | 2023-03-21 | 四川大学 | Chromatin topological association domain boundary prediction method based on multi-modal fusion |
| CN115640337A (en) * | 2022-12-22 | 2023-01-24 | 粤港澳大湾区数字经济研究院(福田) | Abnormal data prediction method, system and related equipment for symbolic graph network |
| CN116340726A (en) * | 2023-03-28 | 2023-06-27 | 广东电网有限责任公司 | Energy economy big data cleaning method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117391125A (en) | 2024-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Handling missing data in near real-time environmental monitoring: A system and a review of selected methods | |
| Fernandes et al. | Sugarcane yield prediction in Brazil using NDVI time series and neural networks ensemble | |
| US20210209467A1 (en) | Method and device for predicting thermal load of electrical system | |
| CN112598248B (en) | Load prediction method, load prediction device, computer equipment and storage medium | |
| CN111126680A (en) | Road section traffic flow prediction method based on time convolution neural network | |
| CN110686633B (en) | Landslide displacement prediction method and device and electronic equipment | |
| CN113496104A (en) | Rainfall forecast correction method and system based on deep learning | |
| CN112330050A (en) | Power system load prediction method considering multiple features based on double-layer XGboost | |
| Runfola et al. | Measuring the temporal instability of land change using the Flow matrix | |
| CN115545334A (en) | Land use type prediction method, land use type prediction device, electronic device, and storage medium | |
| Kim et al. | Untangling the contribution of input parameters to an artificial intelligence PM2. 5 forecast model using the layer-wise relevance propagation method | |
| CN111461923A (en) | Electricity stealing monitoring system and method based on deep convolutional neural network | |
| CN117391125B (en) | Data processing method and system based on neural network | |
| CN116739172A (en) | Method and device for ultra-short-term prediction of offshore wind power based on climbing identification | |
| CN115935822A (en) | Method, system, device and medium for predicting long-term degradation trend of computer parameters | |
| CN112859034B (en) | Natural environment radar echo amplitude model classification method and device | |
| CN115616408A (en) | Battery thermal management data processing method and system | |
| CN114676887A (en) | River water quality prediction method based on graph convolution STG-LSTM | |
| CN112423031B (en) | KPI monitoring method, device and system based on IPTV | |
| Semenenko et al. | Automated system-cognitive analysis of the dependence of export and import of agricultural machinery on its production (the case of Russia) | |
| CN113902187A (en) | Time-of-use electricity price prediction method and device and terminal equipment | |
| Janbain et al. | Use of long short-term memory network (LSTM) in the reconstruction of missing water level data in the River Seine | |
| CN113537459A (en) | Method for predicting humiture of drug storage room | |
| Mohamed et al. | Data Analysis of Agriculture using Data Mining Techniques | |
| CN116486097B (en) | Remote automatic feeding method and system applied to rodent feeding scene |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |