Thermal environment prediction model and thermal environment prediction method based on agricultural facilities

Abstract

The present invention provides a thermal environment prediction model and thermal environment prediction method based on agricultural facilities. The model includes: a first memory module, which is used to extract features of time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment, and the time series information includes: environmental parameters and target parameters of the thermal environment; a second memory module, which is used to extract short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features; an autoregressive module, which is used to determine the initial prediction results based on the historical information of the target parameters; and an output module, which is used to output the final prediction results of the thermal environment based on the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction results. The present invention is used to solve the defect that the prior art does not combine the features of the thermal environment of agricultural facilities, resulting in poor results in the long-term prediction task of the thermal environment.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural facilities, and in particular to a thermal environment prediction model and a thermal environment prediction method based on agricultural facilities. Background Art

[0002] The prediction methods for thermal environment in facility agriculture (greenhouse, factory-scale fish and vegetable symbiosis, soilless cultivation, recirculating aquaculture) can be divided into mechanism model prediction, machine learning model prediction and time series model prediction. Since the latter two methods do not need to pay attention to the physical laws and principles in the system, they are called "black box models". The following will introduce in detail the application of these methods in mechanism prediction modeling.

[0003] The mechanism model uses biological and physical principles to quantitatively analyze the relevant factors in the system, and establishes the equilibrium equation through the law of conservation of energy and mass. The mechanism model includes static models and dynamic models: the static model is also called the steady-state model; the dynamic model is usually described by differential equations, which describes the law of change of the system. The earliest static model of the greenhouse environment was established by scholar Businger, laying the foundation for later research. Although the static model is easy to implement, its accuracy is low. Therefore, Japanese scholars Takakura and others established the first relatively complete dynamic model for an unheated single-layer glass greenhouse, which comprehensively describes the heat and moisture transfer process in the greenhouse.

[0004] As the mechanism model continues to improve, it also brings convenience to the prediction and regulation of agricultural facility environment. However, since the mechanism model uses a large number of parameters and physical variables, the agricultural system is often time-varying. Therefore, it is difficult to tune in development and practice, and more and more researchers have begun to pay attention to data-based models. In addition, with the development of artificial intelligence, computing performance and data productivity have been further improved, which has promoted the application of machine learning models in agricultural environmental modeling.

[0005] Although existing technologies can achieve time series prediction through machine learning, they do not combine the characteristics of the thermal environment of agricultural facilities and cannot achieve satisfactory results in the long-term prediction of the thermal environment. Therefore, how to achieve accurate prediction of the thermal environment of agricultural facilities is an important issue that needs to be solved in the industry. Summary of the invention

[0006] The present invention provides a thermal environment prediction model and a thermal environment prediction method based on agricultural facilities, so as to solve the defect that the prior art does not combine the characteristics of the thermal environment of agricultural facilities, resulting in poor effect in the long-term thermal environment prediction task, and realize accurate prediction of the thermal environment.

[0007] The present invention provides a thermal environment prediction model based on agricultural facilities, comprising: a first memory module, used for extracting features of time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment, wherein the time series information comprises: environmental parameters and target parameters of the thermal environment;

[0008] A second memory module, used for extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features;

[0009] An autoregressive module, used to determine an initial prediction result based on historical information of the target parameter;

[0010] An output module is used to output a final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result.

[0011] According to a thermal environment prediction model based on agricultural facilities provided by the present invention, the second memory module includes: an attention unit and at least two second memory units;

[0012] The attention unit is used to connect the first memory module and the second memory module, to adjust the long-term spatiotemporal features, and to transmit the adjusted long-term spatiotemporal features to the second memory unit.

[0013] According to a thermal environment prediction model based on agricultural facilities provided by the present invention, the first memory module includes at least two first memory units;

[0014] The attention unit is specifically used to calculate the attention weight corresponding to each first memory unit based on the long-term spatiotemporal features and the short-term spatiotemporal features after obtaining the long-term spatiotemporal features, adjust the long-term spatiotemporal features based on the attention weights, and transmit the adjusted long-term spatiotemporal features to the second memory unit.

[0015] According to a thermal environment prediction model based on agricultural facilities provided by the present invention, the second memory unit is used to determine the short-term spatiotemporal characteristics based on the adjusted long-term spatiotemporal characteristics and the historical information.

[0016] According to a thermal environment prediction model based on agricultural facilities provided by the present invention, the second memory module is a recurrent neural network RNN, and the second memory unit includes: a long short-term memory network LSTM.

[0017] According to a thermal environment prediction model based on agricultural facilities provided by the present invention, the first memory unit includes: a time domain convolutional network TCN.

[0018] According to a thermal environment prediction model based on agricultural facilities provided by the present invention, the output module is specifically used to sum the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction results to obtain the final prediction results.

[0019] The present invention also provides a thermal environment prediction method based on a thermal environment prediction model of agricultural facilities, comprising:

[0020] Extracting features of time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment;

[0021] Extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features;

[0022] Determining an initial prediction result based on the historical information of the target parameter;

[0023] Based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result, a final prediction result of the thermal environment is output.

[0024] The thermal environment prediction method provided by the present invention, extracting the short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features, comprises:

[0025] The long-term spatiotemporal features are adjusted, and the short-term spatiotemporal features are determined based on the adjusted long-term spatiotemporal features and the historical information.

[0026] The thermal environment prediction method provided by the present invention outputs the final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result, including:

[0027] The long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction result are summed to obtain the final prediction result.

[0028] The thermal environment prediction model and thermal environment prediction method based on agricultural facilities provided by the present invention obtain the prediction result of the thermal environment through the thermal environment prediction model, wherein the thermal environment prediction model includes: a first memory module, a second memory module, an autoregressive module and an output module. The present invention uses a first memory module to extract features from the time series information corresponding to the thermal environment to obtain long-term spatiotemporal characteristics of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment. It can be seen that the thermal environment prediction model of the present invention effectively obtains the long-term spatiotemporal characteristics of the thermal environment based on the relevant parameters of the thermal environment; the second memory module extracts the short-term spatiotemporal characteristics of the thermal environment based on the long-term spatiotemporal characteristics of the thermal environment. It can be seen that the present invention obtains the short-term spatiotemporal characteristics of the thermal environment; the autoregressive module is used to determine the initial prediction result based on the historical information of the target parameter; the output module is used to output the final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result. It can be seen that the present invention fully considers the long-term spatiotemporal characteristics and short-term spatiotemporal characteristics of the thermal environment, as well as the spatial relationship between the long-term spatiotemporal characteristics and the short-term spatiotemporal characteristics, so that the present invention can better adapt to the thermal environment prediction of agricultural facilities, and finally realize accurate thermal environment prediction of agricultural facilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 It is one of the structural schematic diagrams of the thermal environment prediction model provided by the present invention;

[0031] Figure 2 This is the second structural schematic diagram of the thermal environment prediction model provided by the present invention;

[0032] Figure 3 This is the third structural schematic diagram of the thermal environment prediction model provided by the present invention;

[0033] Figure 4 It is a schematic flow chart of the thermal environment prediction method provided by the present invention;

[0034] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Combine the following Figure 1-Figure 3 The thermal environment prediction model based on agricultural facilities of the present invention is described.

[0037] The embodiment of the present invention provides a thermal environment prediction model based on agricultural facilities. The structural schematic diagram of the model is shown in FIG. Figure 1 As shown, the model specifically includes:

[0038] The first memory module 101 is used to extract features from the time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment. The time series information includes: environmental parameters and target parameters of the thermal environment.

[0039] The second memory module 102 is used to extract short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features.

[0040] The autoregressive module 103 is used to determine the initial prediction result based on the historical information of the target parameter.

[0041] The output module 104 is used to output the final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result.

[0042] Among them, the long-term spatiotemporal characteristics include: seasonal information, change trend information, spatiotemporal correlation and other information of the thermal environment; the short-term spatiotemporal characteristics include: short-term volatility of the thermal environment, etc.

[0043] Specifically, the present invention is mainly aimed at the prediction task of time series information of thermal environment of agricultural facilities. Among them, environmental parameters include: outdoor temperature, outdoor humidity, soil temperature, soil moisture, atmospheric pressure, indoor light, indoor carbon dioxide (CO2) concentration, indoor temperature, indoor humidity, circulating water temperature, etc. Among them, environmental parameters include: exogenous variables and target variables. The target variable corresponds to the final prediction result.

[0044] Below, outdoor temperature, outdoor humidity, soil temperature, soil moisture, atmospheric pressure, indoor light, indoor carbon dioxide (CO2) concentration, indoor temperature, indoor humidity, circulating water temperature, etc. are used as exogenous variables, and circulating water temperature is used as the target variable for explanation:

[0045] Among them, for the exogenous variables and the target variable, u t ∈RN and v t ∈R, the two together constitute the input parameter z of the model t = {u t ;v t The goal of the present invention is to use the environmental parameters of the historical T time steps, that is, the time series information, to predict the target variables of the future H time steps. That is, the thermal environment prediction model establishes a mapping function f(·) between the exogenous variable and the target variable, see formula (1)

[0046]

[0047] Right now

[0048] Y=f(X) (1)

[0049] Among them, the input variable is X, The target variable for prediction is Y, Y = {v T+1 ;v T+2 ;…;v T+H}∈R H , T is a constant, H is a constant

[0050] In a specific embodiment, the first memory module 101 includes at least two first memory units, and the first memory unit includes: a temporal convolutional network (TCN for short).

[0051] Specifically, due to the strong coupling, large inertia and nonlinearity of the thermal environment of facility agriculture, it is necessary to comprehensively consider the multivariate and long time series of data when creating the model. Therefore, the first memory module of the present invention adopts the TCN model. Among them, the TCN model ensures that future information will not be leaked through causal convolution. In addition, the TCN model can effectively obtain spatiotemporal features of arbitrary length through deeper and expanded convolution.

[0052] Therefore, the present invention can capture long-term time series information and extract long-term spatiotemporal characteristics of the thermal environment based on the long-term time series information.

[0053] Specifically, Figure 2As shown, each TCN model includes two causal convolution TCN submodules 201, and each TCN submodule includes: an extended convolution layer 2011, a weight normalization layer 2012, an activation function (Rectified Linear Units, ReLU for short) 2013 and a Dropout layer 2014. Causal convolution means that for a neuron with a time step of t, it can only use the information before the tth time step in the previous layer for convolution, which can effectively ensure the leakage of temporal information. For a convolution kernel f with a size of 1×k, after adding the dilated convolution factor d, the receptive field of the feature map Z increases exponentially compared with the general convolution module, see formula (2). Furthermore, through weight normalization 2012, ReLU activation function 2013 and Dropout layer 2014, the model has good generalization performance. Finally, a residual connection is deployed for the model through 1×1 convolution 202.

[0054]

[0055] Among them, k is a constant, t is a constant, i is a constant, d is the dilation convolution factor, f is the convolution kernel, and Z is the result of each layer of convolution calculation.

[0056] The present invention is based on the above TCN model, extracts the long-term spatiotemporal characteristics of the thermal environment, and the long-term spatiotemporal characteristics at time t are expressed as m i Indicates that the long-term spatiotemporal features output by the TCN model are M = {m1,m2,…,m T}∈R T×p .

[0057] In a specific embodiment, the second memory module 102 includes: an attention unit and at least two second memory units; the attention unit is used to connect the first memory module 101 and the second memory module 102, to adjust the long-term spatiotemporal features, and transmit the adjusted long-term spatiotemporal features to the second memory unit.

[0058] In a specific embodiment, the second memory module 102 is a recurrent neural network (RNN), and the second memory unit includes a long short-term memory network LSTM.

[0059] The present invention adopts the RNN network to effectively capture the short-term volatility in the thermal environment.

[0060] In a specific embodiment, the attention unit is specifically used to calculate the attention weight corresponding to each first memory unit based on the long-term spatiotemporal features and the short-term spatiotemporal features after obtaining the long-term spatiotemporal features, adjust the long-term spatiotemporal features based on the attention weights, and transmit the adjusted long-term spatiotemporal features to the second memory unit.

[0061] Specifically, the present invention effectively integrates long-term spatiotemporal features and short-term spatiotemporal features, so that the RNN network can adaptively select the valuable first memory unit for reasoning. For example, at time t, the output result of the hidden layer unit of the LSTM network is As the query target (Query) of the attention unit, the output result m of the TCN model i ∈R p As the query key of the attention unit, and m i Calculate the attention weight, see formula (3) and formula (4):

[0062]

[0063]

[0064] Among them, the hidden unit output d of the LSTM network t-1 and t-1 , by adding d t-1 and t-1 Splice and get W attn ∈R m×2q , U attn ∈R m×p and v attn ∈R m are the training parameters of the attention unit, As the query target of the attention unit, m i The query key value as the attention unit. Formula (4) is used to calculate the intermediate energy value for each query key value Convert to attention weight

[0065] Among them, the attention weight can reflect the importance of the i-th first memory unit to the model prediction target. After the model adaptively selects the first memory unit, the LSTM network context information c at the t-th time step t The calculation of is shown in formula (5):

[0066]

[0067] Among them, c t is an input parameter of the LSTM network, and T is a constant.

[0068] Specifically, c t That is, the adjusted long-term spatiotemporal characteristics, c t Transfer to the LSTM network.

[0069] In a specific embodiment, the second memory module 102 is specifically used to determine the short-term spatiotemporal features based on the adjusted long-term spatiotemporal features and historical information.

[0070] Specifically, the long-term memory context information c processed by the attention unit t-1 , and the historical information v of the target variable in a short time window t-1 By splicing the information, we can get the input information of the LSTM network, see formula (6) and formula (7):

[0071]

[0072]

[0073] in, and is the training parameter, Indicates v t-1 and c t-1 The splicing result, f LSTM represents the forward propagation process of the LSTM network, and Used to represent the output results of the LSTM network, that is, short-term spatiotemporal features.

[0074] Specifically, Input to LSTM network, output

[0075] In a specific embodiment, the output module 104 is specifically used to sum the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction results to obtain the final prediction results.

[0076] Specifically, the method for obtaining the initial prediction result is shown in formula (8):

[0077]

[0078] in, and b ar are training parameters, According to historical information {v t-T+1 ,v t-T+2 ,…,v t Based on the single-step initial prediction result, the initial prediction result is obtained.

[0079] The autoregressive module of the present invention is used to solve the generalization problem of the prediction model. While the prediction model has a higher accuracy in predicting the thermal environment, it also improves the generalization robustness of the prediction model.

[0080] Specifically, for the hidden results output by the TCN model and the RNN model, the long-term spatiotemporal features extracted by the TCN model and the short-term spatiotemporal features output by the RNN model are projected to the output latitude required for the thermal environment prediction task through a set of affine transformations, see formula (9) and formula (10):

[0081]

[0082]

[0083] Among them, W tcn ∈R H×T , U tcn ∈R p 、b tcn ∈R H , W rnn ∈R H×2q and b rnn ∈R H M={m1,m2,…,m T}∈R T×p represents the long-term spatiotemporal characteristics output by the TCN model, Represents the short-term spatiotemporal features output by the LSTM unit at time t.

[0084] Finally, the local outputs of the TCN model, the RNN model and the autoregressive module 103 are added together to obtain the multi-step prediction result of the thermal environment, as shown in formula (11). In addition, since the model is smooth and differentiable, the parameters in the prediction model can be trained by the back propagation algorithm, as shown in formula (12):

[0085]

[0086]

[0087] Where Θ represents all training parameters in the model, N is the batch size in the mini-batch gradient descent algorithm, and ||·||2 represents the final prediction result predicted by the model and real thermal environment data Y (i) The L2 norm between .

[0088] In addition, after the prediction model training is completed, the model prediction accuracy needs to be verified, as follows:

[0089] In the present invention, different evaluation strategies are proposed from the perspectives of global performance indicators and local performance indicators. Among them, the prediction effect of multiple future time steps is considered as global performance, focusing on the overall prediction accuracy of the model, and evaluated by root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE), see formula (13), formula (14) and formula (15):

[0090] Secondly, the degree of fit of each prediction step (i.e., a single-step evaluation in a multi-step prediction result) is considered as local performance, which effectively reflects the degree of match between the prediction result and the actual production. The relative square root error (RRSE) and the empirical correlation coefficient (CORR) are used to measure the local performance of the model, see formula (16) and formula (17):

[0091] in, is the model prediction value, y is the actual measurement value, then the above indicators are defined as follows:

[0092] Global performance indicators:

[0093]

[0094]

[0095]

[0096] Local performance indicators:

[0097]

[0098]

[0099] Among them, i is a constant, representing any number of data samples, h represents the step size of multi-step prediction of time series in the prediction model, N is a constant, representing the total number of data samples, and Ω represents the set of data samples and multi-step prediction results.

[0100] Among them, for the RMSE, MAE, MAPE and RRSE indicators, the larger the better; for the CORR indicator, the smaller the better.

[0101] Next, through Figure 3 The thermal environment prediction model of the present invention is described in detail:

[0102] First, the exogenous variable u and the target variable v are connected and serialized through the long historical information time window T as the input of the multi-layer TCN model, and the long-term spatiotemporal feature m is obtained. i .

[0103] Then, [m1,m2,…,m T ]and Input to the attention unit and get the attention unit output c t ,in, is the hidden layer output of the LSTM network, i.e., the short-term spatiotemporal features.

[0104] Then, the historical information v of the short time window is t-1 and c t-1 Enter the LSTM network.

[0105] Furthermore, the historical information of the long-term window [v1, v2, …, v T ] is input to the autoregressive module 103 to obtain the initial prediction result output by the autoregressive module 103.

[0106] Finally, the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction results are respectively input into the output module 104 , and the final prediction results are obtained through the output module 104 .

[0107] In addition, after actual verification in a fish-vegetable symbiosis base, the thermal environment prediction model of the present invention has achieved excellent results in the multi-step prediction of circulating water temperature. In the prediction tasks of 6 hours, 12 hours, and 24 hours, the RMSE index of the present invention has achieved 15.40%, 13.93%, and 22.15% accuracy improvement respectively. In particular, the advantages in the long-term prediction of the thermal environment are more obvious. The multi-scale memory structure of the present invention gives the prediction model time series insight from coarse-grained to fine-grained, and can fully learn the laws in the thermal environment time series data. In addition, the different components used in the present invention can effectively mine the temporal and spatial correlation characteristics in the thermal environment data, so that the prediction model can obtain sufficient temporal and spatial patterns, effectively improving the accuracy of the prediction task.

[0108] The thermal environment prediction model and thermal environment prediction method based on agricultural facilities provided by the present invention obtain the prediction result of the thermal environment through the thermal environment prediction model, wherein the thermal environment prediction model includes: a first memory module, a second memory module, an autoregressive module and an output module. The present invention uses a first memory module to extract features from the time series information corresponding to the thermal environment to obtain long-term spatiotemporal characteristics of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment. It can be seen that the thermal environment prediction model of the present invention effectively obtains the long-term spatiotemporal characteristics of the thermal environment based on the relevant parameters of the thermal environment; the second memory module extracts the short-term spatiotemporal characteristics of the thermal environment based on the long-term spatiotemporal characteristics of the thermal environment. It can be seen that the present invention obtains the short-term spatiotemporal characteristics of the thermal environment; the autoregressive module is used to determine the initial prediction result based on the historical information of the target parameter; the output module is used to output the final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result. It can be seen that the present invention fully considers the long-term spatiotemporal characteristics and short-term spatiotemporal characteristics of the thermal environment, as well as the spatial relationship between the long-term spatiotemporal characteristics and the short-term spatiotemporal characteristics, so that the present invention can better adapt to the thermal environment prediction of agricultural facilities, and finally realize accurate thermal environment prediction of agricultural facilities.

[0109] The embodiment of the present invention provides a thermal environment prediction method based on a thermal environment prediction model of agricultural facilities. The thermal environment prediction method described below corresponds to the thermal environment prediction model described above. The specific implementation of the method is as follows: Figure 4 As shown:

[0110] Step 401 , extracting features from the time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment.

[0111] The time series information includes: environmental parameters and target parameters of the thermal environment.

[0112] Step 402: extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features.

[0113] In a specific embodiment, the long-term spatiotemporal features are adjusted, and the short-term spatiotemporal features are determined based on the adjusted long-term spatiotemporal features and historical information.

[0114] Step 403: Determine an initial prediction result based on the historical information of the target parameter.

[0115] Step 404: outputting a final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result.

[0116] In a specific embodiment, the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction results are summed to obtain the final prediction results.

[0117] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502 and the memory 503 communicate with each other through the communication bus 504. The processor 501 may call the logic instructions in the memory 503 to execute the thermal environment prediction method, which includes: extracting features from the time series information corresponding to the thermal environment to obtain the long-term spatiotemporal features of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment; extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features; determining the initial prediction results based on the historical information of the target parameters; and outputting the final prediction results of the thermal environment based on the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction results.

[0118] In addition, the logic instructions in the above-mentioned memory 503 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0119] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the thermal environment prediction method provided by the above-mentioned methods, and the method includes: extracting features of time series information corresponding to the thermal environment to obtain long-term spatiotemporal characteristics of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment; based on the long-term spatiotemporal characteristics, extracting short-term spatiotemporal characteristics of the thermal environment; based on the historical information of the target parameters, determining the initial prediction result; based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result, outputting the final prediction result of the thermal environment.

[0120] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the thermal environment prediction methods provided above, the methods comprising: extracting features from time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment, wherein the time series information comprises: environmental parameters and target parameters of the thermal environment; extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features; determining an initial prediction result based on historical information of the target parameters; and outputting a final prediction result of the thermal environment based on the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction result.

[0121] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0122] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A thermal environment prediction model based on agricultural facilities, characterized in that: include: The first memory module is used to extract features from the time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment, wherein the long-term spatiotemporal features include: seasonal information, change trend information, and spatiotemporal correlation information of the thermal environment, and the environmental parameters include: outdoor temperature, outdoor humidity, soil temperature, soil moisture, atmospheric pressure, indoor lighting, indoor carbon dioxide concentration, indoor temperature, indoor humidity, and circulating water temperature; A second memory module is used to extract short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features, wherein the short-term spatiotemporal features include: short-term volatility of the thermal environment; An autoregressive module, used to determine an initial prediction result based on historical information of the target parameter; An output module, configured to output a final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result; The second memory module includes: an attention unit and at least two second memory units; The attention unit is used to connect the first memory module and the second memory module, to adjust the long-term spatiotemporal features, and transmit the adjusted long-term spatiotemporal features to the second memory unit; The second memory module is a recurrent neural network RNN, and the second memory unit includes: a long short-term memory network LSTM; The extraction of the short-term spatiotemporal features of the thermal environment specifically includes: , and the historical information of the target variable in a short time window Information splicing is used to obtain the input information of the LSTM network. See the formula: ； ； in, and is the training parameter, express and The splicing result is represents the forward propagation process of the LSTM network, and Used to represent the short-term spatiotemporal features of the output of the LSTM network.

2. The thermal environment prediction model based on agricultural facilities according to claim 1 is characterized in that: The first memory module includes at least two first memory units; The attention unit is specifically used to calculate the attention weight corresponding to each first memory unit based on the long-term spatiotemporal features and the short-term spatiotemporal features after obtaining the long-term spatiotemporal features, adjust the long-term spatiotemporal features based on the attention weights, and transmit the adjusted long-term spatiotemporal features to the second memory unit.

3. The thermal environment prediction model based on agricultural facilities according to claim 2 is characterized in that: The second memory unit is used to determine the short-term spatiotemporal features based on the adjusted long-term spatiotemporal features and the historical information.

4. The thermal environment prediction model based on agricultural facilities according to claim 3 is characterized in that: The first memory unit includes: a time domain convolutional network TCN.

5. The thermal environment prediction model based on agricultural facilities according to claim 1, characterized in that: The output module is specifically used to sum the long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction results to obtain the final prediction results.

6. A thermal environment prediction method based on the thermal environment prediction model based on agricultural facilities according to any one of claims 1 to 5, characterized in that: include: Extracting features from the time series information corresponding to the thermal environment to obtain long-term spatiotemporal features of the thermal environment, wherein the time series information includes: environmental parameters and target parameters of the thermal environment, wherein the long-term spatiotemporal features include: seasonal information, change trend information, and spatiotemporal correlation information of the thermal environment, and the environmental parameters include: outdoor temperature, outdoor humidity, soil temperature, soil moisture, atmospheric pressure, indoor lighting, indoor carbon dioxide concentration, indoor temperature, indoor humidity, and circulating water temperature; Extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features, wherein the short-term spatiotemporal features include: short-term volatility of the thermal environment; extracting short-term spatiotemporal features of the thermal environment based on the long-term spatiotemporal features includes: adjusting the long-term spatiotemporal features, and determining the short-term spatiotemporal features based on the adjusted long-term spatiotemporal features and the historical information; determining an initial prediction result based on the historical information of the target parameter; Outputting a final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result; The second memory module includes: an attention unit and at least two second memory units; The attention unit is used to connect the first memory module and the second memory module, to adjust the long-term spatiotemporal features, and transmit the adjusted long-term spatiotemporal features to the second memory unit; The second memory module is a recurrent neural network RNN, and the second memory unit includes: a long short-term memory network LSTM; The extraction of the short-term spatiotemporal features of the thermal environment specifically includes: , and the historical information of the target variable in a short time window Information splicing is used to obtain the input information of the LSTM network. See the formula: ； ； in, and is the training parameter, express and The splicing result is represents the forward propagation process of the LSTM network, and Used to represent the short-term spatiotemporal features of the output of the LSTM network.

7. The thermal environment prediction method based on the thermal environment prediction model of agricultural facilities according to claim 6 is characterized in that: The outputting a final prediction result of the thermal environment based on the long-term spatiotemporal characteristics, the short-term spatiotemporal characteristics and the initial prediction result comprises: The long-term spatiotemporal features, the short-term spatiotemporal features and the initial prediction result are summed to obtain the final prediction result.

Images