CN111310317A - Prediction method and device of granary space-time temperature field based on big data and interpolation prediction - Google Patents

Abstract

The invention discloses a granary space-time temperature field prediction method and device based on big data and interpolation prediction. The method includes: reading grain condition data and warehouse information; intercepting data at different time points as data basis; 3D reconstruction of the temperature in the warehouse; selecting a section to be predicted, and predicting the temperature time series of the section; , determine whether the coordinate position is an outer coordinate point; input the sample data as the input sequence; initialize, set the training period and accuracy; calculate the actual output of the network layer; calculate the error and correct the weights and thresholds; Whether the training accuracy reaches the preset accuracy or the training period reaches the preset period, the temperature value is predicted; the predicted temperature field map of the section is obtained by spatial interpolation of the predicted discrete temperature points of the section. The invention accurately predicts the change trend of the temperature field in the silo, directly and effectively reflects the internal temperature information of the granary grain pile, and ensures the quality of grain storage.

Description

Prediction method and device of granary space-time temperature field based on big data and interpolation prediction

technical field

The invention relates to a temperature field prediction method in the technical field of grain storage, in particular to a granary space-time temperature field prediction method based on big data and interpolation prediction, and also relates to a granary space-time temperature field prediction method based on big data and interpolation prediction using the method Temperature field prediction device.

Background technique

At present, many large grain depots have been built in China, and the capacity of a single warehouse is several times that of a single warehouse built in the past. Mildew occurs in the grain warehouse, and the problem of insect pests is also more serious. Grain is a special and complex living body, and the change law of the internal temperature field of the grain pile has also become extremely complex. Therefore, accurately grasping the temperature distribution of the granary and reasonably analyzing and predicting the temperature change trend of the granary are important methods to predict the safety state of grain storage. one.

However, the current grain temperature prediction scheme in the granary is aimed at the temperature prediction of a single point. For a complex storage environment, the temperature prediction of a single point is not enough to show the temperature change trend of the whole granary, and it cannot reflect the overall temperature in the large warehouse environment. Temperature distribution, local temperature anomalies in the warehouse may not be predicted, resulting in a large deviation between the theoretical prediction and the actual situation, the changes in the hot and cold cores of the granary cannot be displayed, and the grain management manager makes a wrong judgment.

SUMMARY OF THE INVENTION

In order to solve the technical problem that the grain temperature prediction scheme of the existing granary cannot predict the specific temperature distribution of the whole silo, and the mathematical modeling analysis of each granary is impractical, the present invention provides a granary based on big data and interpolation prediction. Space-time temperature field prediction method and device.

The present invention adopts the following technical solutions to realize: a method for predicting the granary space-time temperature field based on big data and interpolation prediction, which comprises the following steps:

Step S1, read grain condition data and corresponding warehouse information from a grain warehouse database;

Step S2, intercepting data at different time points according to the warehouse state in the warehouse information;

In step S3, 3D space reconstruction is performed on the temperature, and position information corresponding to the temperature point is obtained;

Step S4, select the section to be predicted, and first perform the time series prediction of the temperature of the section;

Step S5, judging whether the temperature points at different positions of the cross section are outer layer temperature points;

When the temperature point of the position is the temperature point of the outer layer, step S6 is performed;

Step S6, select variables including meteorological data, internal temperature and internal humidity, and adjacent temperature points as sample data;

When the temperature point of the position is not the temperature point of the outer layer, step S7 is performed;

Step S7, selecting variables including adjacent temperature points, internal temperature, internal humidity and moisture as sample data;

Step S8, input the sample data as the input sequence of the input layer, and select the number of hidden layer nodes;

Step S9, initialize the weights and thresholds, and set the training period and precision;

Step S10, calculating the actual output of the network layer;

Step S11, calculate the error, and correct the weight and the threshold;

Step S12, performing BP neural network training, and judging whether the training accuracy has reached a preset accuracy, or whether the training period has reached a preset period;

When the training accuracy reaches the preset accuracy or/and the training period reaches the preset period, step S13 is performed;

Step S13, step S13 predicts the temperature values corresponding to all the coordinates of the section;

When the training accuracy does not reach the preset accuracy and the training period does not reach the preset period, step S10 is performed;

Step S14, performing spatial interpolation according to the predicted point data to obtain the temperature field map of the cross section.

The invention reads the data in the database first, then intercepts the data at different time points, then reconstructs the temperature in 3D space, then selects the section and predicts the plane temperature of the section, and then judges whether the temperature points at different positions in the section are the outer layer temperature point, and use the selected variables as sample data to realize the preprocessing process of the granary data, and then rearrange the data to serve as the training set and test set in the neural training network, and then carry out the neural network training to predict the plane temperature Finally, the predicted plane temperature points are interpolated to predict the real temperature field prediction map in the warehouse, which solves the problem that the existing grain temperature prediction scheme of the granary cannot predict the specific temperature distribution of the whole warehouse. Mathematical modeling analyzes unrealistic technical problems, obtains the future space-time variation trend of the temperature field in the silo, and intuitively and efficiently reflects the internal temperature information of the granary grain stack, and has universality and scalability, which can be applied Technical effects on various warehousing sites.

As a further improvement of the above scheme, the formula of the space-time prediction model is:

In the formula, z _t+1 (ε) is the interpolation value at the position of ε in the space, and z _t+1 (a _i ,b _i ,c _i ) is the known position of (a _i ,b _i , _ci ) Interpolation value, (a _i , b _i , c _i ) is the row and column layer in the granary, corresponding to the actual position of the temperature sensor in the granary; θ[] represents the time series prediction process based on real data.

As a further improvement of the above scheme, before the BP neural network training, data preprocessing is also performed on the temperature data: different temperature points are corresponding to each position of the granary, and the data is arranged based on the on-site cable arrangement; The data sequence formula of the space temperature point is:

X(a,b,c)=[x1( _a ,b,c), _x2 (a,b,c),..., _xt (a,b,c)]

In the formula, X(a,b,c) is the data sequence of the temperature points in the a-th row, b-column, c-layer, and c-layer in the granary, and x _t (a,b,c) is the temperature point in the a-th row, b-column, b-layer, and c-layer in the granary. data, t is the number of sampled data, and also represents the time series.

As a further improvement of the above scheme, let l be the layer number of the hidden layer of the BP neural network, n be the layer number of the input layer of the BP neural network, and m is the layer number of the output layer of the BP neural network. The output formula of the hidden layer of the BP neural network is:

In the formula, x _i is the input value of the input layer of the BP neural network, ω _ij is the weight from the input layer to the hidden layer, α _j is the bias from the input layer to the hidden layer ; g(x) is the excitation function and satisfies:

As a further improvement of the above scheme, the output formula of the output layer of the BP neural network is:

ξ为1-10的常数；ω_jk为所述隐含层到所述输出层的权重，β_k为所述隐含层到所述输出层的偏置。In the formula,

ξ is a constant of 1-10; ω _jk is the weight from the hidden layer to the output layer, and β _k is the bias from the hidden layer to the output layer.

Further, in step S11, the error calculation formula is:

In the formula, y _k is the expected output.

Still further, the correction formula of the weights from the hidden layer to the output layer is:

The correction formula of the weights from the input layer to the hidden layer is:

In the formula, e _k =y _k -o _k , η is the learning rate.

Still further, the correction formula of the bias from the hidden layer to the output layer is:

The correction formula of the bias from the input layer to the hidden layer is:

In the formula, e _k =y _k -o _k , η is the learning rate.

As a further improvement of the above scheme, the spatial interpolation formula is:

In the formula, z(ε) is the value at the point ε that needs to be predicted in the space, z(a _i , b _i , c _i ) is the predicted value at the ith position of the actual temperature sensor in the granary, and w is the number of measured values. , λ _i is the weight coefficient, satisfying:

The present invention also provides a granary space-time temperature field prediction device based on big data and interpolation prediction, which applies any of the above-mentioned big data and interpolation prediction-based granary space-time temperature field prediction methods, including:

A data reading module, which is used to read grain condition data and corresponding warehouse information from a grain warehouse database;

a data interception module, which is used for intercepting data at different time points according to the warehouse state in the warehouse information;

a reconstruction module, which is used to reconstruct the temperature in 3D space and obtain the position information corresponding to the temperature point;

a section selection module, which is used to select a section in an orientation, and firstly perform the time series prediction of the temperature of the section;

a temperature point judgment module, which is used for judging whether the temperature points at different positions in the cross section are the outer layer temperature points;

A sample data selection module 1, which is used to select variables including meteorological data, internal temperature and internal humidity, and adjacent temperature points as sample data when the temperature point at the location is the outer temperature point;

The second sample data selection module, which is used to select variables including adjacent temperature points, internal temperature and internal humidity, and moisture as sample data when the temperature point at the location is not the outer temperature point;

a sample data input module, which is used to input the sample data as the input sequence of the input layer, and select the number of hidden layer nodes;

Initial setting module, which is used to initialize weights, thresholds, and set training period and accuracy;

A calculation module, which is used to calculate the actual output of the network layer;

a correction module, which is used for calculating the error and correcting the weight and the threshold;

A training judgment module, which is used for BP neural network training, and judges whether the training accuracy reaches a preset accuracy, or whether the training period reaches a preset period; when the training accuracy does not reach the preset accuracy and the When the training period does not reach the preset period, the training judgment module drives the calculation module to perform calculation;

a prediction module, configured to predict the temperature values corresponding to all the coordinates of the section when the training accuracy reaches the preset accuracy or/and the training period reaches the preset period; and

The temperature field acquisition module is used for performing spatial interpolation according to the predicted point data to obtain the temperature field map of the cross section.

Compared with the grain temperature prediction scheme of the existing granary, the granary space-time temperature field prediction method and device based on big data and interpolation prediction of the present invention have the following beneficial effects:

The method for predicting the granary space-time temperature field based on big data and interpolation prediction first reads the data in the database, then intercepts the data at different time points, then reconstructs the temperature in 3D space, and then selects the section and predicts the plane temperature of the section. , and then judge whether the temperature points at different positions in the section are the outer temperature points, and use the selected variables as sample data to realize the preprocessing process of the granary data, and then rearrange the data to be used as the neural training network. The training set and the test set, followed by neural network training to predict the plane temperature point, and finally the predicted plane temperature point is interpolated to predict the real temperature field prediction map in the warehouse, which can accurately predict the temperature field change in the warehouse Trend, intuitively and effectively reflect the internal temperature information of the grain stack in the granary, and achieves the universality and scalability of the model, which can be applied to various storage sites.

Moreover, the granary space-time temperature field prediction method uses the BP neural network method to predict the plane temperature points, and then provides effective data for the temperature field prediction map, and uses the kriging interpolation method to perform spatial interpolation to fit a two-dimensional temperature field. , to establish a space-time temperature field prediction model, which can intuitively and clearly reflect the internal temperature information of the grain pile in the granary, and provides a theoretical basis for the accuracy of the sampling of suspicious points. In the case of a limited number of sensors, this method can reasonably analyze the temperature in the granary in space and time under the premise of reasonable arrangement of temperature sensors, which greatly improves the visualization of grain information, and increases the reliability at the same time. So as to ensure the quality of food storage.

The beneficial effect of the granary space-time temperature field prediction device based on big data and interpolation prediction is the same as the beneficial effect of the above-mentioned granary space-time temperature field prediction method, which will not be repeated here.

Description of drawings

FIG. 1 is a flowchart of a method for predicting a granary space-time temperature field based on big data and interpolation prediction according to Embodiment 1 of the present invention.

FIG. 2 is a comparison diagram of temperature prediction of coordinates (1, 1, 1) in the experiment of the prediction method of the granary space-time temperature field based on big data and interpolation prediction according to Embodiment 2 of the present invention.

FIG. 3 is a comparison diagram of temperature prediction of coordinates (4, 3, 2) in the experiment of the method for predicting the granary space-time temperature field based on big data and interpolation prediction according to Embodiment 2 of the present invention.

4 is a simulation comparison diagram of the temperature field at the previous moment and the temperature field predicted at this moment in the experiment of the prediction method for the granary space-time temperature field based on big data and interpolation prediction according to Embodiment 2 of the present invention.

5 is a simulation comparison diagram of the actual temperature field and the predicted temperature field at the current moment in the experiment of the prediction method of the granary space-time temperature field based on big data and interpolation prediction according to Embodiment 2 of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

Example 1

Referring to FIG. 1, this embodiment provides a method for predicting a granary space-time temperature field based on big data and interpolation prediction. The temperature points in the warehouse are predicted in all directions, and then the spatial interpolation method is used to interpolate the temperature points in the warehouse, and the temperature in the warehouse is analyzed in the form of a temperature field map. In this embodiment, the granary space-time temperature field prediction method adopts BP neural network and kriging interpolation method to establish a granary space-time temperature field model, and includes the following steps.

Step S1, read grain condition data and corresponding warehouse information from a grain warehouse database. The granary database can be an existing database, and stores various information of the granary, such as grain condition data, warehouse information, management information, and the like.

Step S2, according to the warehouse state in the warehouse information, intercept data at different time points. In this embodiment, the space-time temperature field model needs to first perform time series prediction on the temperature points in the granary. Therefore, in this step, data at different time points are intercepted to provide data support for subsequent processing.

In step S3, the temperature is reconstructed in 3D space, and the position information corresponding to the temperature point is obtained.

Step S4, select a section in an azimuth, and predict the plane temperature of the section. In this embodiment, the final temperature field that needs to be displayed is to select a certain section of the granary as the temperature display surface, so as to predict more comprehensive and abundant temperature data.

Step S5, it is judged whether the temperature points at different positions in the cross section are outer layer temperature points.

When the temperature point at the position is the outer layer temperature point, step S6 is performed.

In step S6, variables including meteorological data, internal temperature and internal humidity, and adjacent temperature points are selected as sample data.

When the temperature point at the position is not the outer layer temperature point, step S7 is performed.

In step S7, variables including adjacent temperature points, internal temperature, internal humidity, and moisture are selected as sample data.

In this embodiment, the method for time series prediction is based on a BP neural network, which is composed of an input layer, a hidden layer and an output layer. The input sequence of the input layer is the temperature data and humidity data of the granary. The data taken from the database needs to be preprocessed and then used as the input sequence of the neural network. In this embodiment, it is also necessary to perform data preprocessing on the temperature data: different temperature points are corresponding to each position of the granary, and the data is arranged based on the on-site cable arrangement; the data sequence formula of the spatial temperature points for:

X(a,b,c)=[x1( _a ,b,c), _x2 (a,b,c),..., _xt (a,b,c)]

In the formula, X(a,b,c) is the data sequence of the temperature points in the a-th row, b-column, c-layer, and c-layer in the granary, and x _t (a,b,c) is the temperature point in the a-th row, b-column, b-layer, and c-layer in the granary. data, t is the number of sampled data. Since the fetched data is timing data, t is also time sequence.

After the temperature data is processed, combined with the humidity data, the BP neural network is trained. The training process consists of two stages, the forward propagation of the signal and the back propagation of the error. The previous stage starts from the input layer, terminates the output layer, and passes through the hidden layer in the middle. The latter stage is just the opposite, starting from the output layer, terminating the input layer, and passing through the hidden layer in the middle, but in the process of propagation, the weights and biases in the propagation need to be adjusted separately.

Step S8, input the sample data as the input sequence of the input layer, and select the number of hidden layer nodes.

In this embodiment, the signal is propagated forward, and the output formula of the hidden layer of the BP neural network is:

In the formula, x _i is the input value of the input layer of the BP neural network, ω _ij is the weight from the input layer to the hidden layer, and α _j is the bias from the input layer to the hidden layer. g(x) is the excitation function and satisfies:

In this embodiment, the output formula of the output layer of the BP neural network is:

n为输入层的层数，m为输出层的层数，ξ为1-10的常数。ω_jk为隐含层到输出层的权重，β_k为隐含层到输出层的偏置。In the formula, l is the number of hidden layers, satisfying:

n is the number of layers of the input layer, m is the number of layers of the output layer, and ξ is a constant of 1-10. ω _jk is the weight from the hidden layer to the output layer, and β _k is the bias from the hidden layer to the output layer.

Step S9, initialize the weights and thresholds, and set the training period and precision.

In step S10, the actual output of the network layer is calculated.

Step S11, calculate the error, and correct the weight and threshold.

In this embodiment, the error propagates backward, so the error calculation formula is:

In the formula, y _k is the expected output. Wherein, i=1,2,...,n; j=1,2,...l; k=1,2,...m.

In this embodiment, by using the gradient descent method to adjust the weights, the correction formula of the weights from the hidden layer to the output layer is:

The correction formula for the weights from the input layer to the hidden layer is:

In the formula, e _k =y _k -o _k , η is the learning rate.

And, the correction formula for the bias from the hidden layer to the output layer is:

The correction formula for the bias from the input layer to the hidden layer is:

In the formula, e _k =y _k -o _k , η is the learning rate.

Step S12, perform BP neural network training, and judge whether the training precision reaches a preset precision, or judge whether the training period reaches a preset period.

When the training precision reaches the preset precision or/and the training period reaches the preset period, step S13 is performed.

In step S13, the temperature values corresponding to all the coordinates of the section are predicted.

When the training accuracy does not reach the preset accuracy and the training period does not reach the preset period, step S10 is performed.

After the predicted value of the temperature point of the granary is obtained, the next step is to perform spatial interpolation on the discrete temperature points of the granary, so as to obtain the estimated temperature value around the temperature measurement point of the granary.

In step S14, according to the predicted point data, spatial interpolation is performed to obtain the temperature field map of the cross-section.

In this embodiment, the spatiotemporal analysis of the temperature field is used in the storage of the granary, the spatial reconstruction of the granary is performed, and the data of each temperature point is corresponding to each position of the granary, and the cable arrangement on site is used as the standard , the row and column layers of the granary are represented by a, b, and c, the temperature data is predicted in time series, and then the spatial interpolation estimation is carried out. The formula for the space-time prediction model is:

In the formula, z _t+1 (ε) is the interpolation value at the position of ε in the space, and z _t+1 (a _i ,b _i ,c _i ) is the known position of (a _i ,b _i , _ci ) Interpolation value, (a _i , b _i , c _i ) is the row and column layer in the granary, corresponding to the actual position of the temperature sensor in the granary; θ[] represents the time series prediction process based on real data, which is related to the previous granary temperature data.

代表粮仓的各个温度点预测数据。接下来对平面分布的离散温度点进行插值。在本实施例中，平面中的温度点相当于观测点，周围空白就是未采样点，未采样点的值是邻近观测值的线性加权平均，而权重是由拟合的变异函数决定。因此，空间插值公式为：The final temperature field to be displayed in this model is to select a certain section of the granary as the temperature display surface, select points on the current section to perform space-time prediction analysis, and then perform fitting to obtain a two-dimensional map of the temperature field. In this embodiment, for the trained BP neural network, input temperature data for prediction, and the predicted value output by the neural network

Represents the forecast data for each temperature point of the granary. Next, the discrete temperature points of the planar distribution are interpolated. In this embodiment, the temperature point in the plane corresponds to the observation point, the surrounding blank is the unsampled point, the value of the unsampled point is a linear weighted average of adjacent observation values, and the weight is determined by the fitted variogram. Therefore, the spatial interpolation formula is:

λ_i的取值直接决定了待估计值的精度，由于克里金插值法的依据是无偏最优估计，则λ_i满足

待估计值只与相临近位置的已知值有关，则需满足z(s₀)的协方差Cov(s_i,s_i)存在，λ_i的取值需将增量的方差存在且最小，使得估计数据与已知数据相关性最强从而使得方差最小。In the formula, z(ε) is the value at the point ε to be predicted in the space, and z(ε) is obtained by the linear combination of w observed sample values. Here, ε is any position in the granary space, and is not limited to the above-mentioned temperature points (a, b, c). z(a _i , b _i , c _i ) is the predicted value at the ith position of the actual temperature sensor in the granary, w is the number of measured values, and λ _i is the weight coefficient, which satisfies:

The value of λ _i directly determines the accuracy of the value to be estimated. Since the basis of the kriging interpolation method is the unbiased optimal estimation, then λ _i satisfies

The value to be estimated is only related to the known value of the adjacent position, then the covariance Cov(s _i , s _i ) of z(s ₀ ) needs to exist, and the value of λ _i needs to exist and minimize the variance of the increment, The estimated data is most correlated with the known data and the variance is minimized.

To sum up, compared with the existing grain temperature prediction scheme for grain silos, the method for predicting the space-time temperature field of grain silos based on big data and interpolation prediction in this embodiment has the following advantages:

The method for predicting the granary space-time temperature field based on big data and interpolation prediction first reads the data in the database, then intercepts the data at different time points, then reconstructs the temperature in 3D space, and then selects the section and predicts the plane temperature of the section. , and then judge whether the temperature points at different positions in the section are the outer temperature points, and use the selected variables as sample data to realize the preprocessing process of the granary data, and then rearrange the data to be used as the neural training network. The training set and test set are then trained by neural network to predict the plane temperature point, and finally the predicted plane temperature point is interpolated to predict the real temperature field prediction map in the warehouse, which can accurately predict the temperature field in the warehouse. The change trend can intuitively and effectively reflect the internal temperature information of the grain stack in the granary, and the model is universal and extensible, and can be applied to various storage sites.

Moreover, the granary space-time temperature field prediction method uses the BP neural network method to predict the plane temperature points, and then provides effective data for the temperature field prediction map, and uses the kriging interpolation method to perform spatial interpolation to fit a two-dimensional temperature field. , to establish a space-time temperature field prediction model, which can intuitively and clearly reflect the internal temperature information of the grain pile in the granary, and provides a theoretical basis for the accuracy of the sampling of suspicious points. In the case of a limited number of sensors, this method can reasonably analyze the temperature in the granary in space and time under the premise of reasonable arrangement of temperature sensors, which greatly improves the visualization of grain information, and increases the reliability at the same time. So as to ensure the quality of food storage.

Example 2

This embodiment provides a method for predicting the space-time temperature field of a granary based on big data and interpolation prediction, which performs a simulation experiment on the basis of Embodiment 1. In the grain situation database, a warehouse under a certain grain situation user is used to predict the temperature field. The data range is the historical data of the last half year, with a total of 328 pieces of data. The maximum training times of BP neural network is set to 5000 times, the learning rate is 0.05, and the training accuracy is 1e-3. The row and column layers of the warehouse are 7 rows, 6 columns and 4 layers.

First, take out the historical data of the warehouse for processing. For the temperature field section to be predicted, according to the position of the temperature point, the corresponding variable data is obtained as the input sequence of the BP neural network. For example, for the prediction of the outer temperature point whose coordinate point is (1,1,1), because it is close to the wall or roof of the granary, the influencing factors are the external temperature and external humidity, the internal temperature and internal humidity, and the adjacent points (1,1 , 2), (1, 2, 1) temperature data; for the central temperature point, such as the temperature prediction of the coordinate point (4, 3, 2), the influence of external temperature and external humidity is small, and the internal temperature and internal humidity, The temperature points of the front, back, left, right, up and down are highly correlated, and the internal temperature and humidity data and adjacent points (4,3,3), (4,3,1), (4,2,2), (4,4, 2), (5,3,2) and (3,3,2) temperature data.

Then use the BP neural network method to analyze and process the data, set the training period and accuracy, train the training set, use the test set for testing, and plot points at different positions in the granary for display. The results are shown in Figure 2 and Figure 3 shown.

After the historical data of the corresponding plane is selected and predicted by the BP neural network, the predicted values of temperature points at different positions in the plane are obtained and compared with the actual ones. As shown in the following table, it is the section of the first row, that is, the comparison table of the actual value prediction value of the temperature point whose coordinates are (1,b,c), please refer to Table 1.

Table 1 Prediction table of section temperature values

Finally, after the predicted temperature value of the current plane is obtained, combined with the actual warehouse information, the spatial distribution of temperature points is restored to the actual scene, and kriging interpolation is performed according to the predicted value of the temperature point to obtain the plane. temperature field distribution. According to this warehouse setting, the warehouse is 40 meters long and 35 meters wide. The distance between rows and columns is 5 meters, the distance between the outermost cable and the surrounding walls is 2.5 meters; the distance between layers is 1.5 meters, and the distance between the topmost temperature measurement point and the top layer is 0.75 meters. The temperature measurement points are arranged from top to bottom, and the coordinates (1, 1) on the section correspond to (0.75, 5.25) in the actual warehouse section diagram, and so on for drawing, as shown in Figure 4 and Figure 5 Show.

Furthermore, the model is evaluated using root mean square error and mean absolute percentage error. For the above experimental data, the error test of the predicted temperature value corresponding to the position of the measurement and control point in the granary and the actual value is carried out, and it is obtained: RMSE=0.1387; MAPE=1.75%. The results show that the temperature field prediction map obtained by the method adopted in the embodiment has a good effect.

Example 3

This embodiment provides a granary space-time temperature field prediction device based on big data and interpolation prediction, which applies the granary space-time temperature field prediction method based on big data and interpolation prediction in Embodiment 1, and includes data reading module, data interception module, reconstruction module, section selection module, temperature point judgment module, sample data selection module 1, sample data selection module 2, sample data input module, initial setting module, calculation module, correction module, training judgment module, Prediction module and temperature field acquisition module.

The data reading module is used for reading grain condition data and corresponding warehouse information from a grain warehouse database, which can implement step S1 in the first embodiment. The data interception module is used to intercept data at different time points according to the warehouse state in the warehouse information, and the module can implement step S2 in the first embodiment. The reconstruction module is used to reconstruct the temperature in 3D space and obtain the position information corresponding to the temperature point, and the module can implement step S3 in the first embodiment. The section selection module is used to select a section in an azimuth and predict the plane temperature of the section, and the module can implement step S4 in Embodiment 1. The temperature point judging module is used for judging whether the temperature points at different positions in the cross section are the outer layer temperature points, and the module can implement step S5 in the first embodiment. The sample data selection module 1 is used to select variables including meteorological data, internal temperature and humidity, and adjacent temperature points as sample data when the real-time temperature point is the outer layer temperature point, and this module can implement Step S6 in Embodiment 1 . The second sample data selection module is used to select variables including adjacent temperature points, internal temperature and internal humidity, and moisture as sample data when the real-time temperature point is not the outer temperature point. This module can implement step S7 in Embodiment 1. The sample data input module is used to input the sample data as the input sequence of the input layer, and select the number of nodes in the hidden layer. This module can implement step S8 in the first embodiment. The initial setting module is used for initializing weights, thresholds, and setting training periods and precisions, and this module can implement step S9 in Embodiment 1. The calculation module is used to calculate the actual output of the network layer, and the module can implement step S10 in Embodiment 1. The correction module is used to calculate the error and correct the weight and threshold, and the module can implement step S11 in the first embodiment. The training judging module is used for BP neural network training and judging whether the training accuracy reaches a preset accuracy, or whether the training period reaches a preset period, and the module can implement step S12 in Embodiment 1. When the training accuracy does not reach the preset accuracy and the training period does not reach the preset period, the training judgment module drives the calculation module to perform calculation. The prediction module is used to predict the temperature values corresponding to all the coordinates of the section when the training accuracy reaches the preset accuracy or/and the training period reaches the preset period, and the module can implement step S13 in Embodiment 1. The temperature field acquisition module is used to perform spatial interpolation according to the predicted point data to obtain the temperature field map of the cross-section, and the module can implement step S14 in the first embodiment.

The advantages of the granary space-time temperature field prediction device based on big data and interpolation prediction in this embodiment are the same as the beneficial effects of the above-mentioned granary space-time temperature field prediction method, which will not be repeated here.

Example 4

This embodiment provides a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, the steps of the method for predicting the granary space-time temperature field based on big data and interpolation prediction in Embodiment 1 are implemented.

Example 1 When the method for predicting the air-time temperature field of a granary is applied, it can be applied in the form of software, such as a program designed to run independently, and installed on a computer terminal. The computer terminal can be a computer, a smart phone, a control system, and other Internet of Things. equipment, etc. The method for predicting the space-time temperature field of a granary in Embodiment 1 can also be designed as an embedded running program, which is installed on a computer terminal, such as a single-chip microcomputer.

Example 5

This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by the processor, the steps of the method for predicting the space-time temperature field of a granary based on big data and interpolation prediction in Embodiment 1 are implemented.

When the method for predicting the air-time temperature field of a granary in Embodiment 1 is applied, it can be applied in the form of software, such as a program designed to be run independently by a computer-readable storage medium. , is designed to initiate the entire method program through the external trigger through the U disk.

The above descriptions are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (10)

1. A granary space-time temperature field prediction method based on big data and interpolation prediction is characterized by comprising the following steps:

step S1, reading grain situation data and corresponding warehouse information from a granary database;

step S2, intercepting different time point data according to the warehouse state in the warehouse information;

step S3, performing 3D space reconstruction on the temperature to obtain position information corresponding to the temperature point;

step S4, selecting a section to be predicted, and firstly performing time sequence prediction of the section temperature;

step S5, judging whether the temperature points of different positions of the cross section are outer layer temperature points or not;

when the temperature point of the location is the outer layer temperature point, performing step S6;

step S6, selecting variables including meteorological data, internal temperature and humidity and adjacent temperature points as sample data;

when the temperature point of the location is not the outer layer temperature point, performing step S7;

step S7, selecting variables including adjacent temperature points, internal temperature and humidity and moisture as sample data;

step S8, inputting the sample data into an input sequence of an input layer, and selecting the number of nodes of a hidden layer;

step S9, initializing weight and threshold, and setting training period and precision;

step S10, calculating the actual output of the network layer;

step S11, calculating an error, and correcting the weight and the threshold;

step S12, carrying out BP neural network training, and judging whether the training precision reaches a preset precision or whether the training period reaches a preset period;

when the training precision reaches the preset precision or/and the training period reaches the preset period, executing step S13;

step S13, predicting temperature values corresponding to all coordinates of the section in step S13;

when the training precision does not reach the preset precision and the training period does not reach the preset period, executing step S10;

and step S14, performing spatial interpolation according to the predicted point data to obtain the temperature field diagram of the cross section.

2. The method of predicting a spatio-temporal temperature field of a granary based on big data and interpolation prediction according to claim 1, wherein the formula of the spatio-temporal prediction model is:

in the formula, z_t+1(epsilon) is an interpolation value with a spatial position of epsilon; z is a radical of_t+1(a_i,b_i,c_i) Is a known position of (a)_i,b_i,c_i) The interpolated value of (d); (a)_i,b_i,c_i) The grain bin is a line-row layer in the grain bin and corresponds to the actual position of the temperature sensor in the grain bin; theta 2]Representing a time-sequential prediction process from real data.

3. The method for predicting the spatiotemporal temperature field of the granary based on the big data and the interpolation prediction as claimed in claim 1, wherein before the training of the BP neural network, the data preprocessing is further performed on the temperature data: respectively corresponding different temperature points to each position of the granary, and carrying out data arrangement by taking field cable arrangement as a basis; the data sequence formula of the space temperature point is as follows:

X(a,b,c)＝[x₁(a,b,c),x₂(a,b,c),…,x_t(a,b,c)]

wherein X (a, b, c) is a data sequence of temperature points of a layer in a row a, a column b and a column c in the granary, and X_t(a, b, c) is the temperature point data of the layer c in the row a, column b and column a in the granary, and t is the number of the sampling dataNumbers, also represent time series.

4. The method according to claim 1, wherein l is the number of layers of the hidden layer of the BP neural network, n is the number of layers of the input layer of the BP neural network, and m is the number of layers of the output layer of the BP neural network. The method is characterized in that the output formula of the hidden layer of the BP neural network is as follows:

in the formula, x_iThe input value of the input layer of the BP neural network is x at different positions_t(a,b,c)；ω_ijα as the weight of the input layer to the hidden layer_jA bias for the input layer to the hidden layer; g (x) is an excitation function and satisfies:

5. the method of predicting the spatiotemporal temperature field of the granary based on big data and interpolation prediction according to claim 4, wherein the output formula of the output layer of the BP neural network is as follows:

in the formula, satisfy:

ξ is a constant number from 1 to 10_jkWeight of the hidden layer to the output layer, β_kIs the biasing of the hidden layer to the output layer.

6. The method of predicting the spatiotemporal temperature field of a granary based on big data and interpolation prediction according to claim 5, wherein in step S12, the error calculation formula is:

in the formula, y_kIs the desired output.

7. The method of claim 6, wherein the formula for modifying the weights from the hidden layer to the output layer is as follows:

the modification formula of the weight from the input layer to the hidden layer is as follows:

in the formula, e_k＝y_k-o_kAnd η is the learning rate.

8. The method of claim 6, wherein the offset from the hidden layer to the output layer is modified by the formula:

the modification formula of the bias from the input layer to the hidden layer is as follows:

in the formula, e_k＝y_k-o_kAnd η is the learning rate.

9. The method of claim 1, wherein the spatial interpolation formula is as follows:

wherein z (epsilon) is a numerical value at a point epsilon to be predicted in space, and z (a)_i,b_i,c_i) The predicted value of the ith position of the actual temperature sensor in the granary is shown, w is the number of measured values, and lambda is_iAs a weight coefficient, satisfy:

10. a prediction device of a granary space-time temperature field based on big data and interpolation prediction, which is applied to the prediction method of the granary space-time temperature field based on big data and interpolation prediction according to any one of claims 1-9, and is characterized by comprising:

the data reading module is used for reading the grain condition data and corresponding warehouse information from a granary database;

the data interception module is used for intercepting different time point data according to the warehouse state in the warehouse information;

the reconstruction module is used for performing 3D space reconstruction on the temperature to obtain position information corresponding to the temperature point;

the section selection module is used for selecting a section in one direction and predicting the time sequence of the temperature of the section;

the temperature point judging module is used for judging whether the temperature points at different positions in the section are outer layer temperature points or not;

the first sample data selection module is used for selecting variables comprising meteorological data, internal temperature and internal humidity and adjacent temperature points as sample data when the real-time temperature point is the outer-layer temperature point;

the second sample data selection module is used for selecting variables including adjacent temperature points, internal temperature humidity and moisture as sample data when the real-time temperature point is not the outer-layer temperature point;

the sample data input module is used for inputting the sample data into an input sequence of an input layer and selecting the number of nodes of a hidden layer;

the initial setting module is used for initializing the weight and the threshold value and setting the training period and the precision;

a calculation module for calculating an actual output of the network layer;

the correcting module is used for calculating errors and correcting the weight and the threshold;

the training judgment module is used for carrying out BP neural network training and judging whether the training precision reaches a preset precision or not or judging whether the training period reaches a preset period or not; when the training precision does not reach the preset precision and the training period does not reach the preset period, the training judgment module drives the calculation module to calculate;

the prediction module is used for predicting temperature values corresponding to all coordinates of the cross section when the training precision reaches the preset precision or/and the training period reaches the preset period; and

and the temperature field acquisition module is used for carrying out spatial interpolation according to the predicted point data to obtain a temperature field diagram of the section.

Images