Deep learning-based radiation cooling system environmental parameter prediction method

Abstract

The invention relates to the technical field of building energy conservation and intelligent control, and provides a radiation cooling system environmental parameter prediction method based on deep learning, which comprises the following steps: collecting multi-source characteristic data in the operation process of a radiation cooling system; preprocessing the multi-source feature data, dividing the preprocessed data into a training set, a verification set and a test set, and constructing a training sample sequence; constructing a prediction model; training the prediction model, and performing parameter adjustment optimization; the performance of the prediction model is evaluated, and hyper-parameter fine tuning is carried out; selecting an optimal prediction model by comparing prediction precision under different hyper-parameter combinations; and utilizing the optimal prediction model to predict the indoor temperature, the dew point temperature and the radiation tail end surface temperature at the future moment. According to the method, the advantages of the 1DCNN and the LSTM are comprehensively utilized, depth feature extraction and long-time dependence modeling of complex multivariable time series data are realized, and the prediction precision is improved; and the energy consumption fluctuation and the operation risk are reduced.

Description

Technical Field

[0001] This invention relates to the technical field of environmental parameter prediction methods for radiant cooling systems based on deep learning, and particularly to a method for predicting environmental parameters of radiant cooling systems based on deep learning. Background Technology

[0002] In recent years, with the in-depth implementation of the "dual carbon" strategy, building energy consumption has become a key focus of energy conservation and emission reduction; radiant cooling systems have been widely used in public and residential buildings due to their comfort and energy-saving potential.

[0003] However, this system faces the following technical challenges during operation: radiant cooling terminals are easily affected by changes in indoor temperature and humidity; if the dew point temperature and terminal surface temperature cannot be predicted in advance, condensation may occur, leading to operational risks. The system's operating conditions are complex and variable; traditional prediction methods such as linear regression and BP neural networks can only utilize shallow information and are unable to capture the nonlinear coupling relationships between multiple variables, resulting in insufficient prediction accuracy. Furthermore, existing methods have short prediction time domains, failing to meet the advance control requirements of radiant cooling systems in preventing condensation and optimizing energy consumption. Therefore, there is an urgent need to propose a high-precision, long-time-domain prediction method that can adapt to complex operating conditions to provide technical support for the optimization of radiant cooling system operation. Summary of the Invention

[0004] The purpose of this invention is to provide a deep learning-based method for predicting environmental parameters of radiative cooling systems, which can overcome the shortcomings of existing prediction methods in terms of accuracy, robustness, and prediction time domain.

[0005] This invention is implemented as follows: a deep learning-based method for predicting environmental parameters of a radiative cooling system, the method comprising: Collect multi-source characteristic data during the operation of the radiant cooling system; The multi-source feature data is preprocessed, including data cleaning, outlier detection and repair, and normalization. The preprocessed data is divided into training set, validation set and test set in a ratio of 8:1:1, and a training sample sequence is constructed using a sliding window method. Construct a prediction model that combines a one-dimensional convolutional neural network and a long short-term memory network; The prediction model is trained using the training set and validated using the validation set to monitor model convergence. The performance of the prediction model is evaluated using test set data, and the hyperparameters are fine-tuned based on the evaluation results. Then, training is restarted until the prediction model performs optimally. By comparing the prediction accuracy under different combinations of hyperparameters, the optimal prediction model is selected. The optimal prediction model is used to predict the indoor temperature, dew point temperature, and surface temperature of the radiant terminal at future times.

[0006] Preferably, the multi-source feature data includes: indoor temperature, indoor dew point temperature, capillary network radiant terminal surface temperature, and the start / stop status of the radiant system and the fresh air system. The capillary network radiant terminal surface temperature includes the wall surface temperature and ceiling surface temperature where the capillary network terminal is embedded.

[0007] Preferably, the normalization process is as follows: perform a linear transformation on the input data to scale the original data to the [0,1] interval, and the calculation formula is as follows: , In the formula: The maximum value in the input sequence; It is the minimum value in the input sequence.

[0008] Preferably, the preprocessing includes constructing a sliding window sequence, wherein the window length is L time steps and the sampling period is T. Further, the window length L and the sampling period T are set according to engineering needs.

[0009] Preferably, the prediction model includes a convolutional layer, an LSTM layer, and a fully connected output layer. The convolutional layer is used to extract local temporal features, the LSTM layer is used to model long-term dependencies, and the prediction result is finally output through the fully connected layer.

[0010] Preferably, training the prediction model using a training set specifically includes: Mean squared error (MSE) is used as the loss function. Backpropagation and gradient descent are used for parameter updates. Mini-batch gradient descent based on Adam optimization is employed for neural network training. The MSE calculation formula is as follows: , In the formula: N is the number of data points in each training round; Provide the actual input for the target; Output for the current model.

[0011] Preferably, the loss function of the prediction model is the mean squared error: when the target is the dew point temperature, samples closer to the dew condensation threshold are given higher weights.

[0012] Preferably, the prediction model is trained using the Adam optimizer, combined with mini-batch training, learning rate decay, and early stopping strategies to avoid overfitting.

[0013] Preferably, the optimal model is selected by comparing the prediction results under different hyperparameter combinations, and the model with the smallest average prediction error on the test set is selected. The hyperparameter combinations include convolution kernel size, number of LSTM units, and learning rate.

[0014] Preferably, the prediction model supports single-task prediction and multi-task joint prediction. The multi-task structure shares underlying features and sets dedicated output heads to predict different targets.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention combines the advantages of 1DCNN and LSTM to achieve deep feature extraction and long-term dependency modeling of complex multivariate time series data, thereby improving prediction accuracy. The prediction results can be directly used as input for the optimized control of radiant cooling systems, providing technical support for reducing energy consumption fluctuations and operational risks. It has strong universality and scalability and can be extended to other building environment control and energy-saving scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the neural network training process of the present invention; Figure 2 This is a schematic diagram of the prediction model structure based on 1DCNN-LSTM of the present invention. Detailed Implementation

[0018] To better understand the technical content of this invention, the technical solutions of this invention are further described and explained below with reference to specific embodiments, but are not limited thereto. The technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] In the description of the embodiments of the present invention, it should be noted that if terms such as "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first," "second," and "third" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] refer to Figures 1 to 2 The deep learning-based method for predicting environmental parameters of a radiative cooling system includes: Step 1: Data Collection Acquire multi-source characteristic data during the operation of the radiant cooling system, including indoor temperature, humidity, dew point temperature, radiant terminal surface temperature, and related operating status parameters.

[0022] Step 2: Data Preprocessing The original data was cleaned, missing values ​​were filled, and normalized. The preprocessed data was divided into training set, validation set and test set in a ratio of 8:1:1. Training sample sequence was constructed using a sliding window method.

[0023] The normalization method is as follows: perform a linear transformation on the input data to scale the original data to the [0,1] interval. The calculation formula is as follows: ; In the formula: The maximum value in the input sequence; It is the minimum value in the input sequence.

[0024] Step 3: Prediction Model Construction A 1DCNN-LSTM hybrid architecture is adopted, and its structure is as follows: Figure 2As shown, the hierarchical connection relationship of convolutional layers, LSTM layers, and fully connected layers is illustrated. First, short-term local temporal features are extracted through one-dimensional convolutional layers. Then, the convolutional features are fed into one or more LSTM layers to capture medium- to long-term time dependencies. Finally, the fully connected layers map the features to the desired output. Independent models can be trained for different prediction targets (indoor temperature, dew point temperature, wall surface temperature, ceiling surface temperature), or a multi-task structure with shared bottom-level features and dedicated output layers can be used. The hyperparameters of the prediction model are determined through cross-validation and manual parameter tuning.

[0025] The Rectified Linear Unit (ReLU) is used as the activation function for each layer of the 1DCNN-LSTM neural network to calculate the activity value of the basic unit of the neural network. The activation function is defined by the following formula: ; The ReLU function is a left-saturating function with a derivative of 1 when x>0, which alleviates the gradient vanishing problem in neural networks to some extent and accelerates the convergence speed of gradient descent.

[0026] Step 4: Predictive Model Training and Optimization The training set data is input into the prediction model for training. The training steps for the neural network prediction model are as follows: Step 1): The 1DCNN layer extracts features from the input; Step 2): Calculate the predicted value according to the LSTM forward propagation formula; Step 3): The 1DCNN layer extracts features from the input; the output of the fully connected layer is calculated. ; Step 4): Compare with the actual value Compare and calculate the loss function; Step 5): Backpropagate the loss and update the parameters of each neural network layer; Step 6): The Adam optimizer adaptively adjusts the learning rate for each parameter.

[0027] Loss function: Mean squared error (MSE) is used as the basic regression loss. To enhance the prediction robustness of the model under key safety constraints (such as dew point threshold), an auxiliary penalty term (e.g., a higher weighting for dew point temperature prediction error) can be introduced during training. The Adam optimizer is used for mini-batch gradient descent training, combined with early stopping and learning rate decay strategies to avoid overfitting. Data augmentation and time series resampling can be employed during training to improve generalization.

[0028] Adam optimization achieves faster convergence and better generalization by calculating different adaptive learning rates for each parameter.

[0029] The update rules for the Adam optimization algorithm are as follows: ; In the formula: The gradient of the parameter; , The decay coefficient of the weighted average of the two exponentials; , This is the average value after gradient deviation correction; The updated parameters; The learning rate; It is a constant used to avoid division by zero.

[0030] The update method can be summarized in the following steps: First, calculate the moving average of the current gradient to capture the first-order momentum of the gradient; second, calculate the moving average of the square of the current gradient to estimate the second-order momentum of the gradient. By combining these two momentum estimates, the Adam algorithm can adaptively adjust the learning rate of each parameter to cope with gradient changes of different parameters, thus avoiding the tedious process of manually adjusting the learning rate in traditional gradient descent algorithms.

[0031] Model performance evaluation The performance of the prediction model is evaluated using test set data, and the hyperparameters are fine-tuned based on the evaluation results. Then, training is restarted until the prediction model performs optimally.

[0032] The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are used. 2 () is used as an indicator to evaluate the predictive performance of the model; The specific calculation formula is as follows: ; In the formula: These are the model's predicted values; n represents the true value; n is the sample size. This is the average of the true values.

[0033] Optimal model selection By comparing prediction models with different combinations of hyperparameters such as kernel size and number of LSTM units, the model with the highest prediction accuracy is selected as the optimal prediction model. The hyperparameter combinations include kernel size, number of LSTM units, and learning rate.

[0034] Predictive applications By inputting the characteristic data of the time to be predicted into the optimal prediction model, the predicted values ​​of indoor temperature, dew point temperature and terminal surface temperature for the next 5 to 30 minutes are obtained.

[0035] The embodiments described above are only some embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention. Parts not covered in the present invention are the same as or can be implemented using existing technology.

Claims

1. A method for predicting environmental parameters of a radiative cooling system based on deep learning, characterized in that, The method includes the following: Collect multi-source characteristic data during the operation of the radiant cooling system; The multi-source feature data is preprocessed, including data cleaning, outlier detection and repair, and normalization. The preprocessed data is divided into training set, validation set and test set in a ratio of 8:1:1, and a training sample sequence is constructed using a sliding window method. Construct a prediction model that combines a one-dimensional convolutional neural network and a long short-term memory network; The prediction model is trained using the training set and validated using the validation set to monitor model convergence. The performance of the prediction model is evaluated using test set data, and the hyperparameters are fine-tuned based on the evaluation results. Then, training is restarted until the prediction model performs optimally. By comparing the prediction accuracy under different combinations of hyperparameters, the optimal prediction model is selected. The optimal prediction model is used to predict the indoor temperature, dew point temperature, and surface temperature of the radiant terminal at future times.

2. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The multi-source feature data includes: indoor temperature, indoor dew point temperature, capillary network radiant terminal surface temperature, and the start / stop status of the radiant system and the fresh air system. The capillary network radiant terminal surface temperature includes the surface temperature of the wall and the ceiling where the capillary network terminals are embedded.

3. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The normalization process is as follows: The input data is linearly transformed to scale the original data to the [0,1] interval. The calculation formula is as follows: ， In the formula: The maximum value in the input sequence; It is the minimum value in the input sequence.

4. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The preprocessing includes constructing a sliding window sequence, wherein the window length is L time steps and the sampling period is T. Further, the window length L and the sampling period T are set according to engineering needs.

5. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The prediction model includes convolutional layers, LSTM layers, and a fully connected output layer. The convolutional layers are used to extract local temporal features, the LSTM layers are used to model long-term dependencies, and the prediction results are finally output through the fully connected layer.

6. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The step of training the prediction model using a training set specifically includes: Mean squared error (MSE) is used as the loss function. Backpropagation and gradient descent are used for parameter updates. Mini-batch gradient descent based on Adam optimization is employed for neural network training. The MSE calculation formula is as follows: ， In the formula: N is the number of data points in each training round; Provide the actual input for the target; Output for the current model.

7. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The loss function of the prediction model is the mean squared error: when the target is the dew point temperature, samples closer to the dew point threshold are given higher weights.

8. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The prediction model is trained using the Adam optimizer, combined with mini-batch training, learning rate decay, and early stopping strategies to avoid overfitting.

9. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The optimal model is selected by comparing the prediction results under different combinations of hyperparameters, and the model with the smallest average prediction error on the test set is chosen. The hyperparameter combinations include convolution kernel size, number of LSTM units, and learning rate.

10. The method for predicting environmental parameters of a radiative cooling system based on deep learning as described in claim 1, characterized in that, The prediction model supports single-task prediction and multi-task joint prediction. The multi-task structure shares underlying features and sets up dedicated output heads to predict different targets.

Images