Central air-conditioning system energy consumption prediction method based on WTD hybrid algorithm

Abstract

The invention discloses a central air-conditioning system energy consumption prediction method based on a WTD hybrid algorithm, and relates to the technical field of central air-conditioning systems, and the method comprises the following steps: S1, carrying out the wavelet transform decomposition processing of original cold load data, carrying out the denoising through a soft threshold function, and reconstructing the data into a low-frequency trend component and a high-frequency detail component; s2, inputting the denoised data into a hybrid model comprising a Transform module and an LSTM (Long Short Term Memory) module, extracting global correlation characteristics of input variables by the Transform module through a multi-head self-attention mechanism, and enhancing time sequence information by using position coding; the LSTM module models a long-term dependency relationship of a time sequence through a gating mechanism of an input gate, a forgetting gate, and an output gate. According to the invention, through the improved wavelet transform decomposition technology, in combination with the Daubechies wavelet basis and the adaptive soft threshold function, accurate filtering of noise components and effective signal reconstruction are realized, and interference of non-stationarity on model input is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of central air-conditioning systems, and in particular to a method for predicting energy consumption of central air-conditioning systems based on a WTD hybrid algorithm. Background Art

[0002] With the continuous growth of building energy consumption, central air conditioning systems, as a major component of building energy consumption, have become increasingly important for energy-saving optimization and operational control. Traditional energy consumption prediction methods are mostly based on statistical models or shallow machine learning algorithms. These methods have significant limitations when processing cooling load data with strong nonlinearity, multi-scale characteristics, and complex time dependencies. In recent years, deep learning technology has been introduced into the field of time series prediction, but existing technologies still face the following key issues:

[0003] Existing wavelet denoising methods mostly use fixed threshold strategies, which are difficult to adaptively distinguish between valid information and noise, resulting in distortion of reconstructed data or information loss, affecting the accuracy of subsequent modeling.

[0004] A single model cannot take into account both the global long-term dependency and local nonlinear dynamic characteristics of time series. Existing hybrid models mostly use simple series or parallel structures, which do not achieve deep collaboration between modules, resulting in low feature fusion efficiency.

[0005] The cooling load of central air conditioners is affected by multiple factors such as season, weather, and operating cycle, and exhibits complex fluctuation characteristics. The existing attention mechanism does not show separation trend and seasonal components, and cannot dynamically adjust the weights of each component according to the actual operating scenario, resulting in poor adaptability of the model to multi-scale temporal characteristics. Summary of the Invention

[0006] The present invention provides a method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm to solve the problems raised in the above background technology.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm includes the following steps:

[0009] S1: The original cooling load data is decomposed by wavelet transform, denoised by soft threshold function and reconstructed into low-frequency trend component and high-frequency detail component;

[0010] S2: The denoised data is input into a hybrid model consisting of a Transformer module and an LSTM module. The Transformer module extracts the global correlation features of the input variables through a multi-head self-attention mechanism and enhances the time series information using position encoding. The LSTM module models the long-term dependency of the time series through a gating mechanism consisting of an input gate, a forget gate, and an output gate.

[0011] S3: Introducing an adaptive seasonality and trend attention mechanism to decompose the cooling load data into trend and seasonality, and dynamically adjust the weights of each component based on the operating cycle;

[0012] S4: The output features of the Transformer module, LSTM module, and ASSA mechanism are weighted and fused through the fully connected layer to generate the cooling load prediction value.

[0013] A further improvement of the technical solution of the present invention is that: the wavelet transform decomposition process uses Daubechies wavelet basis function to decompose the original cooling load data, and filters out high-frequency noise components through a soft threshold function.

[0014] A further improvement of the technical solution of the present invention is that the output of the Transformer module serves as the input of the LSTM module, and the Transformer module includes at least two layers of multi-head self-attention sublayers and feedforward neural network sublayers.

[0015] A further improvement of the technical solution of the present invention is that the ASSA mechanism is implemented by the following steps:

[0016] A1: Decompose the time series into trend component, seasonal component and residual component sub-modules;

[0017] A2: Based on the operation cycle of the cooling load, the contribution of each component is allocated using adaptive attention weights.

[0018] A further improvement of the technical solution of the present invention is that: the fully connected layer uses the ReLU activation function to perform nonlinear transformation on the fused features, and outputs the final prediction value through linear regression.

[0019] A further improvement of the technical solution of the present invention is that the input variables include historical cooling load data; environmental variables: dry-bulb temperature, wet-bulb temperature, wind speed, and sunshine intensity; system operating parameters: the number of chillers turned on, the temperature difference between the supply and return water of chilled water, and the fan frequency; and all input variables are normalized by Z-score standardization.

[0020] A further improvement of the technical solution of the present invention is that the training process of the hybrid model adopts mean square error as the loss function and updates the parameters through the Adam optimizer.

[0021] A further improvement of the technical solution of the present invention is that: an electronic device is proposed, including a processor and a memory, the memory storing a computer program, and when the processor executes the computer program, it implements a central air-conditioning system energy consumption prediction method based on the WTD hybrid algorithm as described in any one of claims 1 to 7.

[0022] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0023] 1. The present invention provides a central air-conditioning system energy consumption prediction method based on the WTD hybrid algorithm. Through the improved wavelet transform decomposition technology, combined with the Daubechies wavelet basis and the adaptive soft threshold function, the noise component is accurately filtered out and the effective signal is reconstructed, thereby reducing the interference of non-stationarity on the model input.

[0024] 2. The present invention provides a method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm. By designing a Transformer-LSTM series hybrid architecture, the Transformer's multi-head self-attention mechanism is used to extract the global correlation of input variables, and the local temporal dependency is modeled through the LSTM gating mechanism. The position encoding is combined to enhance the expression of temporal information, thereby realizing the complementary fusion of long-term and short-term features.

[0025] 3. The present invention provides a method for predicting energy consumption of central air-conditioning systems based on the WTD hybrid algorithm. By introducing an adaptive seasonal and trend attention mechanism, the trend, seasonal and residual components are explicitly separated through STL decomposition, and the attention weights of each component are dynamically adjusted based on the operating cycle, thereby enhancing the model's adaptability to complex fluctuations. Combined with the gated fusion strategy, the collaborative efficiency of multi-module features is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the method flow of the present invention

[0027] Figure 2 Schematic diagram of the WTD-Transformer-LSTM network architecture of the present invention. DETAILED DESCRIPTION

[0028] The present invention is described in further detail below in conjunction with the embodiments:

[0029] Example 1

[0030] like Figure 1 、 Figure 2 As shown, the present invention provides a method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm, comprising the following steps:

[0031] S1: The original cooling load data is decomposed by wavelet transform, denoised by soft threshold function and reconstructed into low-frequency trend component and high-frequency detail component;

[0032] S2: The denoised data is input into a hybrid model consisting of a Transformer module and an LSTM module. The Transformer module extracts the global correlation features of the input variables through a multi-head self-attention mechanism and enhances the time series information using positional encoding. The LSTM module models the long-term dependencies of the time series through a gating mechanism consisting of an input gate, a forget gate, and an output gate.

[0033] S3: Introducing an adaptive seasonality and trend attention mechanism to decompose the cooling load data into trend and seasonality, and dynamically adjust the weights of each component based on the operating cycle;

[0034] S4: The output features of the Transformer module, LSTM module, and ASSA mechanism are weighted and fused through the fully connected layer to generate the cooling load prediction value.

[0035] Example 2

[0036] like Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, the wavelet transform decomposition processing uses Daubechies wavelet basis function to decompose the original cooling load data, and filters out high-frequency noise components through the soft threshold function, the output of the Transformer module is used as the input of the LSTM module, and the Transformer module includes at least two layers of multi-head self-attention sublayers and feedforward neural network sublayers.

[0037] In this embodiment, the Daubechies wavelet basis function decomposes the cooling load data into 3-5 layers to obtain the low-frequency trend component {A n} and high frequency detail components {D1, D2...D n}, apply soft threshold denoising to the high frequency component, and the threshold is set as: Where σ is the standard deviation of noise, N is the data length, and the denoised components are reconstructed to generate the smoothed cooling load sequence X clean , the formula is The number of heads in the multi-head self-attention sublayer is set to 4-8, and the key / query / value dimension of each head is d k =d u =d model / h, h is the number of heads, the feedforward neural network sublayer contains two layers of fully connected networks, the activation function is GELU, and the hidden layer dimension is 4d model .

[0038] Example 3

[0039] like Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: the ASSA mechanism is implemented by the following steps:

[0040] A1: Decompose the time series into trend component, seasonal component and residual component sub-modules;

[0041] A2: Based on the cooling load operation cycle, the contribution of each component is allocated using adaptive attention weights;

[0042] The fully connected layer uses the ReLU activation function to perform nonlinear transformation on the fused features and outputs the final predicted value through linear regression. The input variables include historical cooling load data; environmental variables: dry-bulb temperature, wet-bulb temperature, wind speed, and sunshine intensity; system operating parameters: number of chillers on, chilled water supply and return temperature difference, and fan frequency. All input variables are normalized using Z-score standardization. The training process of the hybrid model uses mean square error as the loss function and uses the Adam optimizer for parameter update.

[0043] In this embodiment, the Z-score normalization formula is: Where μ is the mean, σ is the standard deviation, and A1 is the cooling load data X clean STL decomposition was performed with the following parameters: seasonal period: 24 hours, sliding window length of trend component: 7 days, residual component was calculated by R t =X clean -(S t +T t ) calculation; A2 designs the learnable weight matrix W α ∈R 3×d and bias b α , the trend component T t , seasonal component S t , residual component R t Calculate attention weight after splicing: α t =Softmax(W α ·[S t ;T t ; R t ]+b α ), dynamically adjust the contribution weight of each component, for example, increase the weight of the seasonal component during the high temperature period in summer.

[0044] Example 4

[0045] like Figure 1 、 Figure 2As shown, based on Example 1, the present invention provides an electronic device: including a processor and a memory, the memory storing a computer program, and when the processor executes the computer program, a central air-conditioning system energy consumption prediction method based on the WTD hybrid algorithm according to any one of claims 1 to 7 is implemented.

[0046] In this embodiment, the processor is a multi-core CPU or GPU with a memory capacity of ≥16GB. The prediction program is written based on the PyTorch or TensorFlow framework, and communicates with the central air-conditioning control system in real time through the Modbus / TCP protocol. The sensor data is transmitted to the edge computing device in real time. The model performs a prediction every hour and outputs the results to the control terminal. The control terminal adjusts the operating parameters of the chiller according to the predicted values to achieve dynamic energy saving.

[0047] While the present invention has been generally described above, modifications and improvements are readily apparent to those skilled in the art. Therefore, modifications and improvements that do not depart from the spirit of the present invention are intended to be within the scope of the present invention.

Claims

1. A method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm, characterized by: The following steps are involved: S1: The original cooling load data is decomposed by wavelet transform, denoised by soft threshold function and reconstructed into low-frequency trend component and high-frequency detail component; S2: The denoised data is input into a hybrid model consisting of a Transformer module and an LSTM module. The Transformer module extracts the global correlation features of the input variables through a multi-head self-attention mechanism and enhances the time series information using position encoding. The LSTM module models the long-term dependency of the time series through a gating mechanism consisting of an input gate, a forget gate, and an output gate. S3: Introducing an adaptive seasonality and trend attention mechanism to decompose the cooling load data into trend and seasonality, and dynamically adjust the weights of each component based on the operating cycle; S4: The output features of the Transformer module, LSTM module, and ASSA mechanism are weighted and fused through the fully connected layer to generate the cooling load prediction value.

2. The method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm according to claim 1, characterized in that: The wavelet transform decomposition process uses Daubechies wavelet basis function to decompose the original cooling load data, and filters out high-frequency noise components through a soft threshold function.

3. The method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm according to claim 1, characterized in that: The output of the Transformer module serves as the input of the LSTM module, and the Transformer module includes at least two layers of multi-head self-attention sublayers and feedforward neural network sublayers.

4. The method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm according to claim 1, characterized in that: The ASSA mechanism is implemented through the following steps: A1: Decompose the time series into trend component, seasonal component and residual component sub-modules; A2: Based on the operation cycle of the cooling load, the contribution of each component is allocated using adaptive attention weights.

5. The method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm according to claim 1, characterized in that: The fully connected layer uses the ReLU activation function to perform nonlinear transformation on the fused features and outputs the final prediction value through linear regression.

6. The method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm according to claim 1, characterized in that: The input variables include historical cooling load data; environmental variables: dry-bulb temperature, wet-bulb temperature, wind speed, and sunshine intensity; system operating parameters: number of chillers on, chilled water supply and return temperature difference, and fan frequency; and all input variables are normalized using Z-score standardization.

7. The method for predicting energy consumption of a central air-conditioning system based on a WTD hybrid algorithm according to claim 1, characterized in that: The training process of the hybrid model adopts mean square error as the loss function and updates the parameters through the Adam optimizer.

8. According to the WTD hybrid algorithm-based central air-conditioning system energy consumption prediction method of claim 1, an electronic device is proposed, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, it implements a central air-conditioning system energy consumption prediction method based on a WTD hybrid algorithm as described in any one of claims 1 to 7.

Images