Construction method, system, device and application of temperature data interpolation model

Abstract

The invention provides a construction method, system, device and application of a temperature data interpolation model. The method comprises the following steps: S1, acquiring temperature time sequence data of each detection node; s2, de-noising the temperature time sequence data by using a convolutional neural network to obtain de-noised temperature time sequence data; and S3, according to the denoised temperature time sequence data, constructing an interpolation model by using a kringing training objective function and a back propagation neural network objective function. According to the method, the purpose of denoising the original temperature collection data is achieved through a deep self-encoding means, the interpolation model is constructed by combining the advantages of Kriging interpolation and neural network interpolation, the precision of the interpolation data is improved, and temperature data at any moment can be obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of material data collection, and in particular relates to a method, system, device and application for constructing a temperature data interpolation model. Background Art

[0002] In the process of asphalt production and construction, the measurement of aggregate temperature is very important. Aggregate temperature that is too high or too low will have a certain impact on the quality of asphalt, so it is necessary to monitor and control the aggregate temperature in real time. Aggregate temperature sensor is an instrument used to measure aggregate temperature. It can convert the temperature of aggregate into electrical signals, which are then recorded and analyzed through the data acquisition system.

[0003] Since the collection of temperature data is periodic, all temperature data are discrete values. In addition, due to the factors of the sensor, some data will be missing, making it impossible for the existing data to accurately describe the temperature change trend of the asphalt pile, and thus unable to accurately monitor the asphalt pile. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method, system, device and application for constructing a temperature data interpolation model, which can obtain temperature data at any time.

[0005] The present invention is implemented as follows: a method for constructing a temperature data interpolation model comprises the following steps: S1. Obtain the temperature time series data of each detection node; S2. De-noising the temperature time series data using a convolutional neural network to obtain de-noised temperature time series data; S3. Based on the denoised temperature time series data, an interpolation model is constructed using the kringing training objective function and the back propagation neural network objective function.

[0006] Furthermore, the S2 comprises the following steps: S201, obtaining the deep features of the temperature time series data through an autoencoder model; S202: reconstruct the temperature time series according to the deep features and the self-decoding model to obtain denoised temperature time series data.

[0007] Furthermore, the S201 includes the following steps: S20101. Input the temperature time series data into the first LSTM layer to obtain temperature short-term dependency features; S20102, inputting the short-term temperature dependence feature into the second LSTM layer to obtain the long-term temperature dependence feature; S20103, inputting the temperature long-term dependence feature into the first convolutional layer for dimensionality increase processing to obtain a high-dimensional temperature long-term dependence feature; S20104. Input the high-dimensional temperature long-term dependency feature into the Dropout layer for random inactivation processing to obtain deep features.

[0008] Further, the S202 includes the following steps: S20201, inputting the depth feature into the second convolutional layer to perform dimensionality reduction processing to obtain a reduced-dimensionality temperature feature; S20202, inputting the dimension-reduced temperature features into the third LSTM layer for feature extraction to obtain a converted feature sequence; S20203. Input the feature sequence into the fourth LSTM layer to perform feature restoration to obtain denoised temperature time series data.

[0009] Furthermore, the S3 comprises the following steps: S301, constructing an interpolation function by fusing the kringing training objective function and the back propagation neural network objective function; S302: Train the interpolation function using samples, back propagation neural network node functions and target functions to obtain an interpolation model.

[0010] Furthermore, the expression of the interpolation function is: ; In the above formula, represents the temperature interpolation function result of the i-th sensor, Represents the value of the sample data variation function; Represents the variance function value calculated by the output of the reverse neural network; represents the separation distance of k groups of sensor node pairs; express The number of points of all sensor nodes separated; y represents The average value of represents the temperature time series data of the i-th sensor, express The average value of represents the maximum value of the estimated value, Indicates the minimum value of the estimate; Indicates that the neural network output is greater than The number of nodes, Indicates that the neural network output is less than The number of nodes.

[0011] Furthermore, the expression of the back propagation neural network node function is: , In the above formula, represents the neural network node output of the i-th sensor, represents the neural network input weights of the i-th sensor input layer and the j-th sensor hidden layer, represents the neural network node threshold of the i-th sensor, and f represents the output mapping; for p samples, each sample includes N input components and M output components, and the neural network node output is calculated using the samples and the back-propagation neural network node function; The expression of the objective function is: ; In the above formula, represents the output of the neural network of the jth sensor corresponding to the pth sample, It represents the temperature interpolation function result of the jth sensor corresponding to the pth sample; when F is less than the set error, the training ends.

[0012] The present invention also provides a system for constructing a temperature data interpolation model, comprising: An acquisition module is used to obtain the temperature time series data of each detection node; A denoising module, used for denoising the temperature time series data using a convolutional neural network to obtain denoised temperature time series data; A building block is used to construct an interpolation model based on the denoised temperature time series data, using the kringing training objective function and the back propagation neural network objective function.

[0013] The present invention also provides a device for constructing a temperature data interpolation model, comprising a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above method.

[0014] The present invention also provides an application of a temperature data interpolation model, in which the collected asphalt stockpile temperature time series data is input into the temperature data interpolation model obtained by the above construction method to obtain asphalt stockpile temperature interpolation data.

[0015] The beneficial effects brought by the present invention are as follows: the present invention achieves the purpose of denoising the original temperature collection data through deep autoencoding means, combines the advantages of Kriging interpolation and neural network interpolation to construct an interpolation model, improves the accuracy of the interpolation data, and realizes the acquisition of temperature data at any time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the method in the present invention; Figure 2 is a flow chart of S2 in the present invention; Figure 3 This is a flow chart of S3 in the present invention. DETAILED DESCRIPTION

[0017] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0018] like Figure 1 As shown, a method for constructing a temperature data interpolation model includes the following steps: For any sampling period, S1. Obtain the temperature time series data of each detection node within the sampling period. The temperature time series data comes from the temperature sensors of different detection nodes. The measurement range of the temperature sensor is -10℃ to +60℃, with an accuracy of ±0.1℃. Due to the failure of the sensor itself and factors such as the acquisition cycle, some data will be missing, which will bring inconvenience to the temperature monitoring.

[0019] S2. De-noising the temperature time series data using a convolutional neural network to obtain de-noised temperature time series data. Specifically, Figure 2 As shown, S2 includes the following steps: S201, obtaining the deep features of the temperature time series data through an autoencoder model. More specifically, S201 includes the following steps: S20101. Input the temperature time series data into the first LSTM layer to obtain the temperature short-term dependency feature.

[0020] S20102. Input the short-term temperature dependence feature into the second LSTM layer to obtain the long-term temperature dependence feature.

[0021] S20103. Input the temperature long-term dependence feature into the first convolutional layer for dimensionality increase processing to obtain a high-dimensional temperature long-term dependence feature.

[0022] S20104. Input the high-dimensional temperature long-term dependency feature into the Dropout layer for random inactivation processing to obtain deep features.

[0023] S202: reconstruct the temperature time series according to the deep features and the self-decoding model to obtain denoised temperature time series data. More specifically, S202 includes the following steps: S20201. Input the depth feature into the second convolutional layer for dimensionality reduction processing to obtain a reduced-dimensional temperature feature.

[0024] S20202, inputting the dimension-reduced temperature features into the third LSTM layer for feature extraction to obtain a converted feature sequence; S20203. Input the feature sequence into the fourth LSTM layer to perform feature restoration to obtain denoised temperature time series data.

[0025] S3. Based on the denoised temperature time series data, an interpolation model is constructed using the kringing training objective function and the back propagation neural network objective function. Specifically, Figure 3 As shown, S3 includes the following steps: S301. The kringing training objective function and the back propagation neural network objective function are integrated to construct an interpolation function. Since the spatial correlation of Kriging interpolation is good, but the expression is too smooth, and the accuracy of neural network interpolation is very high, but the spatial structure is weak. Therefore, in order to overcome the shortcomings of these two methods, this patent integrates the kringing training objective function and the back propagation neural network objective function to construct an improved interpolation function, which can have the advantages of both and achieve high-precision estimation of missing data. It should be noted that the kringing training objective function and the back propagation neural network objective function are both existing formulas, so they are not repeated here. The expression of the interpolation function is: ; In the above formula, represents the temperature interpolation function result of the i-th sensor, Represents the value of the sample data variation function; Represents the variance function value calculated by the output of the reverse neural network; represents the separation distance of k groups of sensor node pairs; express The number of points of all sensor nodes separated; y represents The average value of represents the temperature time series data of the i-th sensor, express The average value of represents the maximum value of the estimated value, Indicates the minimum value of the estimate; Indicates that the neural network output is greater than The number of nodes, Indicates that the neural network output is less than The function contains the error and estimated value of the variogram, which can effectively improve the interpolation expression of the neural network.

[0026] S302: Train the interpolation function using samples, back propagation neural network node functions and target functions to obtain an interpolation model.

[0027] Specifically, the expression of the back propagation neural network node function is: , In the above formula, represents the neural network node output of the i-th sensor, represents the neural network input weights of the i-th sensor input layer and the j-th sensor hidden layer, represents the neural network node threshold of the i-th sensor, f represents the output mapping, and the Sigmoid activation function is usually used: Assume that there are p samples, each sample includes N input components and M output components for neural network training, and calculate the output of the neural network node.

[0028] The expression of the objective function is: ; In the above formula, represents the output of the neural network of the jth sensor corresponding to the pth sample, It represents the temperature interpolation function result of the jth sensor corresponding to the pth sample; when F is less than the set error, the training ends.

[0029] In summary, the temperature variation curve of any sensor can be obtained according to the trained temperature interpolation model, and then the temperature value of the sensor at any time can be obtained.

[0030] The present invention also provides an application of the method for constructing the temperature data interpolation model described above in the collection of temperature data of asphalt stockpiles. After the interpolation model is trained, the corresponding temperature data can be obtained by inputting any time, thereby obtaining the temperature data of the entire time series, and then forming a continuous change trend based on the temperature data.

[0031] The present invention also provides a system for constructing a temperature data interpolation model, comprising an acquisition module, a denoising module and a construction module.

[0032] The acquisition module is used to obtain the temperature time series data of each detection node.

[0033] The denoising module is used to denoise the temperature time series data using a convolutional neural network to obtain denoised temperature time series data. Specifically, the denoising module includes an encoder submodule and a decoder submodule, and the encoder submodule is used to obtain the deep features of the temperature time series data through an autoencoding model. The decoder submodule is used to reconstruct the temperature time series according to the deep features and the self-decoding model to obtain the denoised temperature time series data.

[0034] The construction module is used to construct an interpolation model based on the denoised temperature time series data, using the kringing training objective function and the back propagation neural network objective function. The construction module includes a construction submodule and a training submodule. The construction submodule is used to fuse the kringing training objective function and the back propagation neural network objective function to construct an interpolation function. The training submodule is used to train the interpolation function using samples, back propagation neural network node functions and objective functions to obtain an interpolation model.

[0035] The present invention also provides a device for constructing a temperature data interpolation model, comprising a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above method.

[0036] The present invention also provides an application of a temperature data interpolation model, in which the collected asphalt stockpile temperature time series data is input into the temperature data interpolation model obtained by the above construction method to obtain asphalt stockpile temperature interpolation data.

[0037] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a temperature data interpolation model, characterized in that: The steps include: S1. Obtain the temperature time series data of each detection node; S2. Using a convolutional neural network to perform denoising on the temperature time series data to obtain denoised temperature time series data; S3. Based on the denoised temperature time series data, an interpolation model is constructed using the kringing training objective function and the back propagation neural network objective function.

2. The method for constructing a temperature data interpolation model according to claim 1, characterized in that: The S2 comprises the following steps: S201, obtaining the deep features of the temperature time series data through an autoencoder model; S202: reconstruct the temperature time series according to the deep features and the self-decoding model to obtain denoised temperature time series data.

3. The method for constructing a temperature data interpolation model according to claim 2, characterized in that: The S201 includes the following steps: S20101. Input the temperature time series data into the first LSTM layer to obtain temperature short-term dependency features; S20102, inputting the temperature short-term dependence feature into the second LSTM layer to obtain the temperature long-term dependence feature; S20103, inputting the temperature long-term dependence feature into the first convolutional layer for dimensionality increase processing to obtain a high-dimensional temperature long-term dependence feature; S20104. Input the high-dimensional temperature long-term dependency feature into the Dropout layer for random inactivation processing to obtain deep features.

4. The method for constructing a temperature data interpolation model according to claim 2, characterized in that: The S202 comprises the following steps: S20201, inputting the depth feature into the second convolutional layer to perform dimensionality reduction processing to obtain a reduced-dimensionality temperature feature; S20202, inputting the dimension-reduced temperature features into the third LSTM layer for feature extraction to obtain a converted feature sequence; S20203. Input the feature sequence into the fourth LSTM layer to perform feature restoration to obtain denoised temperature time series data.

5. The method for constructing a temperature data interpolation model according to claim 1, characterized in that: The S3 comprises the following steps: S301, constructing an interpolation function by fusing the kringing training objective function and the back propagation neural network objective function; S302: Train the interpolation function using samples, back propagation neural network node functions and target functions to obtain an interpolation model.

6. The method for constructing a temperature data interpolation model according to claim 5, characterized in that: The expression of the interpolation function is: ； In the above formula, represents the temperature interpolation function result of the i-th sensor, Represents the value of the sample data variation function; Represents the variance function value calculated by the output of the reverse neural network; represents the separation distance of k groups of sensor node pairs; express The number of points of all sensor nodes separated; y represents The average value of represents the temperature time series data of the i-th sensor, express The average value of represents the maximum value of the estimated value, Indicates the minimum value of the estimate; Indicates that the neural network output is greater than The number of nodes, Indicates that the neural network output is less than The number of nodes.

7. The method for constructing a temperature data interpolation model according to claim 6, characterized in that: The expression of the back propagation neural network node function is: ， In the above formula, represents the neural network node output of the i-th sensor, represents the neural network input weights of the i-th sensor input layer and the j-th sensor hidden layer, represents the neural network node threshold of the i-th sensor, and f represents the output mapping; for p samples, each sample includes N input components and M output components, and the neural network node output is calculated using the samples and the back-propagation neural network node function; The expression of the objective function is: ； In the above formula, represents the output of the neural network of the jth sensor corresponding to the pth sample, It represents the temperature interpolation function result of the jth sensor corresponding to the pth sample; when F is less than the set error, the training ends.

8. A system for constructing a temperature data interpolation model, characterized in that: include: An acquisition module is used to obtain the temperature time series data of each detection node; A denoising module, used for denoising the temperature time series data using a convolutional neural network to obtain denoised temperature time series data; A building block is used to construct an interpolation model based on the denoised temperature time series data, using the kringing training objective function and the back propagation neural network objective function.

9. A device for constructing a temperature data interpolation model, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the method according to claim 1.

10. An application of a temperature data interpolation model, characterized in that: The collected asphalt stockpile temperature time series data are input into a temperature data interpolation model obtained by the method for constructing a temperature data interpolation model according to claim 1 to obtain asphalt stockpile temperature interpolation data.

Images