A Method for Optimizing and Predicting the Performance of a Heat Pump Cycle

Abstract

The present invention relates to the field of optimizing and predicting the performance of a heat pump cycle, and particularly to a method for optimizing and predicting the performance of a heat pump cycle. The method includes: obtaining different data in the heat pump cycle system in real time through a sensor array to generate a multi-dimensional spatio-temporal data matrix; constructing an adaptive spatio-temporal convolutional neural network prediction model based on the multi-dimensional spatio-temporal data matrix to obtain a prediction result; predicting power consumption and coefficient of performance based on the adaptive spatio-temporal convolutional neural network prediction model to form an initial state vector; adopting an adaptive recursive optimization algorithm based on meta-learning to obtain the power consumption and coefficient of performance in the optimal state; dynamically adjusting the condensation temperature and evaporation temperature of the heat pump cycle system according to the power consumption and coefficient of performance in the optimal state. This solves the technical problem that the existing methods for predicting and optimizing the performance of a heat pump cycle system are difficult to adapt to the complex dynamic changes during the operation of the heat pump cycle system and lack the joint consideration from the perspectives of time and space.

Description

Technical Field

[0001] The invention relates to the field of heat pump cycle performance optimization and prediction, and in particular to a heat pump cycle performance optimization and prediction method. Background Art

[0002] Heat pump technology has gradually received attention from various countries and has been promoted in many application fields. In recent years, with the increase in low-carbon energy-saving needs and the development of refrigerant technology, heat pump technology has gradually expanded from low-temperature applications to high-temperature drying, industrial heating and other fields, further promoting the popularization of heat pump technology on a global scale. However, although traditional heat pump systems have been able to save energy and reduce emissions to a certain extent, there are still many problems in practical applications, such as low heat pump cycle efficiency, improper selection of working fluids, and insufficient waste heat recovery. How to optimize each link of the heat pump cycle and combine it with modern prediction technology to achieve efficient operation of the heat pump system has become a hot topic of current research. As a result, heat pump cycle performance optimization and prediction methods have emerged.

[0003] The research on heat pump cycle performance optimization and prediction methods is of great significance to improving the energy efficiency of heat pump systems and reducing energy consumption. With the improvement of environmental protection requirements and the increase in demand for high-efficiency energy saving, heat pump technology will usher in more extensive applications and development in the future.

[0004] However, the existing heat pump cycle performance optimization and prediction methods have the following technical problems: the prediction and optimization methods of heat pump cycle system performance rely on fixed algorithms or models, which are difficult to adapt to the complex dynamic changes in the operation of the heat pump cycle system, and lack joint considerations from the perspectives of time and space, resulting in the inability to accurately predict the future operating status of the heat pump cycle system; traditional recursive optimization methods have low learning efficiency and difficulty in fast convergence when facing different working conditions and environmental changes, resulting in the heat pump cycle system being unable to make adaptive adjustments in a timely manner, especially when environmental conditions change frequently, and slow response will lead to low energy efficiency of the heat pump cycle system. Summary of the invention

[0005] The present invention provides a method for optimizing and predicting the performance of a heat pump cycle, so as to solve the problems that the existing prediction and optimization methods for the performance of a heat pump cycle system rely on fixed algorithms or models, are difficult to adapt to the complex dynamic changes during the operation of the heat pump cycle system, lack joint considerations from the perspectives of time and space, and are unable to accurately predict the future operating state of the heat pump cycle system; the traditional recursive optimization method has low learning efficiency and is difficult to converge quickly when facing different working conditions and environmental changes, resulting in the heat pump cycle system being unable to make adaptive adjustments in a timely manner, especially when environmental conditions change frequently, and the slow response leads to the technical problem of low energy efficiency of the heat pump cycle system.

[0006] A method for optimizing and predicting the performance of a heat pump cycle specifically includes the following technical solutions:

[0007] A method for optimizing and predicting the performance of a heat pump cycle includes the following steps:

[0008] S1: Obtain different data in the heat pump cycle system in real time to generate a multi-dimensional spatio-temporal data matrix; based on the multi-dimensional spatio-temporal data matrix, obtain prediction results including power consumption and energy efficiency ratio through an adaptive spatio-temporal convolutional neural network prediction model;

[0009] S2: Generate an initial state vector based on the prediction results; adopt an adaptive recursive optimization algorithm based on meta-learning to obtain the power consumption and energy efficiency ratio in the optimal state; dynamically adjust the condensation temperature and evaporation temperature of the heat pump cycle system according to the power consumption and energy efficiency ratio in the optimal state.

[0010] Preferably, the S1 specifically includes:

[0011] The adaptive spatio-temporal convolutional neural network prediction model captures the dynamic correlation features between multi-dimensional spatio-temporal data in the multi-dimensional spatio-temporal data matrix through spatio-temporal convolutional operations, and processes each feature through a weighting operation to obtain a weighted output result; and performs an adaptive non-linear activation on the weighted output result.

[0012] Preferably, the S1 specifically includes:

[0013] The spatio-temporal convolutional operation starts from the input multi-dimensional spatio-temporal data matrix, extracts data features layer by layer through convolutional operations. In each layer of spatio-temporal convolutional operation, a fixed-length time segment is selected through a sliding window method, and the multi-dimensional spatio-temporal data within each time segment is processed to obtain an output result matrix.

[0014] Preferably, the S1 specifically includes:

[0015] The weighting operation obtains a weighted output result by weighting the output result matrix and adding a bias term. The calculation formula for the weighted output result is:

[0016]

[0017] Among them, Z(t + Δt) represents the weighted output result at t + Δt; Y L (t + Δt) represents the feature matrix after spatio-temporal convolutional operation in the L-th layer at t + Δt; represents the weight matrix of the L-th layer; b L represents the bias term of the L-th layer; respectively represent weighted summation of the input feature dimension i and the output feature dimension j. n is the number of input features, and m is the number of output features.

[0018] Preferably, the S2 specifically includes:

[0019] The meta - learning - based adaptive recursive optimization algorithm consists of a recursive optimization inner loop and a meta - learning - based outer loop. Through the collaborative work of the inner and outer loops, the learning rate is dynamically adjusted; and the objective function is used to quantify the balance between the energy efficiency ratio and power consumption of the heat pump cycle system.

[0020] Preferably, the S2 specifically includes:

[0021] Perform gradient calculation on the objective function. By calculating the partial derivatives of power consumption and energy efficiency ratio, the gradient vector in the recursive optimization process is obtained. The formula is as follows:

[0022]

[0023] Among them, F(t + Δt) represents the value of the objective function at t + Δt; represents the gradient vector of the objective function at t + Δt; COP(t + Δt) represents the energy efficiency ratio at t + Δt; P(t + Δt) represents the power consumption at t + Δt; represents the partial derivative of the objective function with respect to power consumption; represents the partial derivative of the objective function with respect to energy efficiency ratio.

[0024] Preferably, the S2 specifically includes:

[0025] The meta - learning optimization framework performs adaptive recursive optimization through the collaborative work of the inner and outer loops. In the inner loop, the state of the heat pump cycle system is updated according to the gradient vector of the objective function; the outer loop introduces the loss function of meta - learning to adjust the learning rate.

[0026] Preferably, the S2 specifically includes:

[0027] Set a convergence threshold. When the change in the state of the heat pump cycle system is less than the convergence threshold, the recursive optimization process stops, and the optimized state vector is obtained, including the power consumption and energy efficiency ratio in the optimal state.

[0028] The beneficial effects of the technical solution of the present invention are:

[0029] 1. Real - time obtain data of the heat pump cycle system through the sensor array and generate a multi - dimensional spatio - temporal data matrix, effectively capturing the dynamic operation state of the heat pump cycle system and ensuring the timeliness and accuracy of the data.

[0030] 2. Through spatio-temporal convolution operations, it is possible to simultaneously capture the dynamic correlation features of multi-dimensional spatio-temporal data in both the temporal and spatial dimensions, effectively identify key features, screen out the features that contribute most to the prediction of the heat pump cycle system through weighted operations, and combine with non-linear processing based on an adaptive activation function, significantly improving the feature expression ability of the adaptive spatio-temporal convolutional neural network prediction model, thereby accurately predicting the future state of the heat pump cycle system, enabling the heat pump cycle system to have extremely strong adaptability and response ability under complex working conditions, and effectively reducing the prediction error.

[0031] 3. By performing weighted sum and linear combination on the feature matrix output by the spatio-temporal convolution operation, and then introducing the Sigmoid function and hyperbolic tangent activation function to convert the linear output into a non-linear form, it can not only capture the complex relationships and non-linear trends in multi-dimensional spatio-temporal data, but also adaptively adjust the performance of the adaptive spatio-temporal convolutional neural network prediction model in different environments, making the adaptive spatio-temporal convolutional neural network prediction model have extremely strong robustness to changes in multi-dimensional spatio-temporal data, being able to quickly adapt to changes in working conditions, and improving the prediction accuracy and the stability of the heat pump cycle system.

[0032] 4. Through an adaptive recursive optimization algorithm based on meta-learning, the dynamic optimization of the heat pump cycle system performance is realized. By using the cooperation of internal and external loops, the state of the heat pump cycle system is quickly adjusted through recursive optimization, and the learning rate is adjusted based on the historical working condition performance to ensure the rapid convergence of the recursive optimization process. The objective function comprehensively considers the balance between power consumption and energy efficiency ratio, and the calculation of the gradient vector can accurately guide the direction of recursive optimization, thereby effectively reducing power consumption, improving the energy efficiency ratio of the heat pump cycle system, enabling the heat pump cycle system to always maintain the best performance under changing working conditions, and significantly improving the performance and economy of the heat pump cycle system.

[0033] 5. The meta-learning loss function combines the relationship between power consumption and energy efficiency ratio, and further enhances the recursive optimization effect through an exponential function. It not only reduces the computational complexity of the recursive optimization process, but also improves the response speed of the heat pump cycle system, enabling the recursive optimization process to converge within fewer iterations, improving the optimization efficiency, and also ensuring that the heat pump cycle system can be quickly adjusted under real-time working conditions, avoiding delays or performance degradation caused by excessive iterations.

[0034] 6. By setting convergence conditions, it is ensured that when the change in the state of the heat pump cycle system is less than a preset threshold, the recursive optimization process stops in a timely manner, avoiding waste of computational resources caused by an overly long recursive optimization process, ensuring that after reaching the optimal state, the power consumption and energy efficiency ratio of the heat pump cycle system can be quickly and accurately adjusted. The efficient stopping mechanism further improves the overall efficiency of the heat pump cycle system optimization process, making the optimization operation not over-execute after reaching the target, thereby saving resources of the heat pump cycle system and extending the service life of the equipment. Brief Description of the Drawings

[0035] Figure 1 It is a flowchart of a method for optimizing and predicting the performance of a heat pump cycle according to the present invention. Detailed Embodiments

[0036] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0038] The following specifically describes the specific solution of a method for optimizing and predicting the performance of a heat pump cycle provided by the present invention with reference to the accompanying drawings.

[0039] Referring to the attached Figure 1 , which shows a flowchart of a method for optimizing and predicting the performance of a heat pump cycle provided by an embodiment of the present invention. The method includes the following steps:

[0040] S1. Real-time obtain different data in the heat pump cycle system to generate a multi-dimensional spatio-temporal data matrix; based on the multi-dimensional spatio-temporal data matrix, obtain prediction results through an adaptive spatio-temporal convolutional neural network prediction model, including power consumption and coefficient of performance;

[0041] Real-time obtain different data in the heat pump cycle system through a sensor array, such as the inlet and outlet temperatures of the heat pump, power consumption, coefficient of performance, and evaporator temperature, to generate a multi-dimensional spatio-temporal data matrix X(t);

[0042] Based on the multi-dimensional spatio-temporal data matrix X(t), construct an adaptive spatio-temporal convolutional neural network prediction model to obtain prediction results, including power consumption P(t+Δt) and coefficient of performance COP(t+Δt);

[0043] The purpose of the adaptive spatio-temporal convolutional neural network prediction model is to process multi-dimensional spatio-temporal data to ensure accurate prediction of the future state of the heat pump cycle system. It captures the dynamic correlation features between multi-dimensional spatio-temporal data in the multi-dimensional spatio-temporal data matrix at different times and spatial positions through spatio-temporal convolution operations, processes each feature through a weighting operation, can effectively screen out the features that contribute most to the prediction of the state of the heat pump cycle system, and further enhances the non-linear expression ability by non-linearly processing the weighted output result through an adaptive activation function;

[0044] The spatio-temporal convolution operation starts from the input multi-dimensional spatio-temporal data matrix X(t), and extracts data features layer by layer through multiple layers of spatio-temporal convolution operations. In each layer of spatio-temporal convolution operation, a fixed-length time segment is selected by means of a sliding window, and the multi-dimensional spatio-temporal data within each time segment is processed to ensure that the dynamic changes of multi-dimensional spatio-temporal data over time can be recognized, and the dependencies between different time points can be understood. While processing the time dimension, the convolution kernel also slides in the multi-dimensional spatio-temporal data matrix X(t) at each moment to analyze the mutual relationships between different multi-dimensional spatio-temporal data. The two-dimensional convolution method can effectively capture the complex relationships between temporal dynamics and multi-dimensional spatio-temporal data;

[0045] The formula for the l-th layer of spatio-temporal convolution operation is as follows:

[0046]

[0047] where, Y l (t + Δt) represents the output result matrix after the l-th layer of spatio-temporal convolution operation at t + Δt, that is, the feature matrix after the l-th layer of spatio-temporal convolution operation; denotes the summation over all time points of the time step τ from 0 to Δt; Y l-1 (t - τ) represents the output result matrix after the (l - 1)-th layer of spatio-temporal convolution operation at t - τ, where Y0 = X(t), that is, the output result matrix of the first layer with the input being the multi-dimensional spatio-temporal data matrix; * represents the convolution operator; K l (t, Δt) represents the spatio-temporal convolution kernel of the l-th layer of spatio-temporal convolution operation. As the time t and the time window Δt change, the convolution kernel slides on the multi-dimensional spatio-temporal data matrix X(t) to extract features;

[0048] By performing spatio-temporal convolution operations on the input multi-dimensional spatio-temporal data in the time and space dimensions through the convolution kernel, the complex relationships between multi-dimensional spatio-temporal data are extracted, not only capturing the change trends of data in the time series, but also analyzing the correlations between different data;

[0049] After multiple layers of spatio-temporal convolution operations, the output is the output result matrix Y L (t + Δt) after the L-th layer of spatio-temporal convolution operation at t + Δt;

[0050] The weighted operation is to weight the output result matrix after the L-th layer of spatio-temporal convolution operation at t + Δt and add a bias term to obtain a weighted output result, which can make full use of the features extracted from each layer and weight them according to the importance of the features to ensure that the adaptive spatio-temporal convolution neural network prediction model can make the most reasonable judgment when predicting the performance of the heat pump cycle system;

[0051] Specifically, since different features play different roles in the prediction of the adaptive spatio-temporal convolutional neural network prediction model, by assigning weights to each feature, it is ensured that the most critical features can be accurately extracted. After the weighted processing of each feature is completed, weighted features are obtained; the weighted features are linearly combined to capture the interactions and influences between the weighted features. The introduction of the bias term is to enhance the flexibility of the adaptive spatio-temporal convolutional neural network prediction model and avoid the situation of over-simplification or neglect of input features;

[0052] The calculation formula for the weighted output result is:

[0053]

[0054] Among them, Z(t + Δt) represents the weighted output result at t + Δt; represents the weight matrix of the spatio-temporal convolutional operation in the L-th layer. Each weight value determines the influence of the i-th input feature on the j-th output feature; b L represents the bias term of the spatio-temporal convolutional operation in the L-th layer, which helps the adaptive spatio-temporal convolutional neural network prediction model fit the data. Especially when the input feature is zero, the bias term can ensure that the adaptive spatio-temporal convolutional neural network has a non-zero output, enhancing the flexibility of the adaptive spatio-temporal convolutional neural network prediction model; respectively represent the weighted summation of the dimensions i of the input feature and the dimension of the output feature. n is the number of input features, and m is the number of output features;

[0055] The non-linear processing is to adaptively non-linearly activate the weighted output result by combining the Sigmoid activation function and the hyperbolic tangent function, converting the linear output into a non-linear form. It can flexibly adjust the non-linear mapping process according to the characteristics of the weighted output result, thereby providing the adaptive spatio-temporal convolutional neural network prediction model with powerful feature expression ability and enhancing the robustness and adaptability of the adaptive spatio-temporal convolutional neural network prediction model;

[0056] Based on the adaptive spatio-temporal convolutional neural network prediction model, the prediction result is obtained. The calculation formula is:

[0057]

[0058] Among them, O(t + Δt) represents the prediction result at t + Δt, including the power consumption P(t + Δt) and the coefficient of performance COP(t + Δt); denotes the Sigmoid activation function, which is used to provide smooth non-linear changes; α denotes the hyperparameter that controls the non-linear intensity in the Sigmoid activation function; β denotes the hyperparameter that adjusts the influence of the hyperbolic tangent function; tanh denotes the hyperbolic tangent function, which is used to enhance the non-linear expression ability of the adaptive spatio-temporal convolutional neural network prediction model.

[0059] S2. Generate an initial state vector based on the prediction results; adopt an adaptive recursive optimization algorithm based on meta-learning to obtain the power consumption and energy efficiency ratio of the optimal state; dynamically adjust the condensation temperature and evaporation temperature of the heat pump cycle system according to the power consumption and energy efficiency ratio of the optimal state;

[0060] Based on the power consumption P(t + Δt) and energy efficiency ratio COP(t + Δt) at time t + Δt predicted by the adaptive spatio-temporal convolutional neural network prediction model, an initial state vector S0(t + Δt) is formed as follows:

[0061]

[0062] Adopt an adaptive recursive optimization algorithm based on meta-learning. By combining recursive optimization and meta-learning strategies, it can quickly adapt to the changing environment and achieve performance optimization of the heat pump cycle system, obtaining the power consumption and energy efficiency ratio of the optimal state;

[0063] The adaptive recursive optimization algorithm based on meta-learning consists of two core parts: a recursive optimization inner loop and a meta-learning-based outer loop. Through the collaborative work of the inner and outer loops, the learning rate is dynamically adjusted to achieve fast convergence and efficient optimization, improving the performance of the heat pump cycle system under different working conditions;

[0064] Quantify the balance between the energy efficiency ratio and power consumption of the heat pump cycle system through the objective function, and clarify the target direction of the recursive optimization process;

[0065] The objective function is defined as follows:

[0066]

[0067] Among them, F(t + Δt) represents the value of the objective function at t + Δt; COP(t + Δt) represents the energy efficiency ratio at t + Δt; P(t + Δt) represents the power consumption at t + Δt; denotes the logarithmic adjustment term of the power consumption, which is used to further control the influence of the power consumption. The logarithmic function plays a role in smooth adjustment to avoid the excessive influence of the power consumption on the entire objective function, thereby maintaining the stability of the recursive optimization process;

[0068] To achieve recursive optimization, it is necessary to calculate the gradient of the objective function. By calculating the partial derivatives of power consumption and energy efficiency ratio, the gradient vector in the recursive optimization process is obtained. The gradient vector can find the optimal direction for adjustment in each iteration to achieve the goal of maximizing energy efficiency ratio and minimizing power consumption. The formula is as follows:

[0069]

[0070] Where, represents the gradient vector of the objective function at t + Δt, which is used for state update in recursive optimization and includes the gradients of power consumption and energy efficiency ratio of the heat pump cycle system; represents the partial derivative of the objective function with respect to power consumption, reflecting the impact of changes in power consumption on the performance of the heat pump cycle system; represents the partial derivative of the objective function with respect to energy efficiency ratio, reflecting the impact of changes in energy efficiency ratio on the performance of the heat pump cycle system;

[0071] The meta - learning optimization framework achieves adaptive recursive optimization through the collaborative work of inner and outer loops. In the inner loop, the state of the heat pump cycle system is updated according to the gradient vector:

[0072]

[0073] Where, S n+1 (t + Δt) represents the state of the heat pump cycle system after the (n + 1) - th iteration at t + Δt; S n (t + Δt) represents the state of the heat pump cycle system after the n - th iteration at t + Δt; h represents the learning rate, that is, the dynamic adjustment parameter in the n - th iteration, which controls the step size of the state update of the heat pump cycle system in each iteration, that is, the speed of update;

[0074] The outer loop adjusts the learning rate γ according to the historical performance of the working conditions, that is, the historical operation data of historical power consumption and historical energy efficiency ratio n :

[0075]

[0076] Where, γ0 represents the learning rate in the initial iteration; γ n-1 represents the learning rate, that is, the dynamic adjustment parameter in the (n - 1) - th iteration; represents the loss function of meta - learning The partial derivative of the learning rate γ n-1 , where the calculation formula of the meta - learning loss function is:

[0077]

[0078] Where, Represents the meta - learning loss function, which is used to measure the optimization effect of the heat pump cycle system; ΔP represents the prediction error of power consumption, that is, the difference between the true value and the predicted value; ΔCOP represents the prediction error of the coefficient of performance (COP), that is, the difference between the true value and the predicted value; P(t + Δt) represents the power consumption at t + Δt; COP(t + Δt) represents the coefficient of performance at t + Δt; Represents the exponential decay term, which combines the product of power consumption and coefficient of performance, and weights the optimization effect through an exponential function;

[0079] By introducing the meta - learning loss function, the complexity of recursive optimization calculation is reduced, and the prediction error can be measured. At the same time, through the exponential decay term, the recursive optimization of power consumption and coefficient of performance is further enhanced, helping the recursive optimization process to be adaptive, thereby improving the response ability of the heat pump cycle system under different working conditions and ensuring that the recursive optimization can converge in very few iterations;

[0080] The set convergence threshold ∈, until the following convergence condition is met:

[0081] |S n+1 (t + Δt)-S n (t + Δt)|<∈

[0082] When the change in the state of the heat pump cycle system is less than the set convergence threshold ∈, the recursive optimization process stops, and the power consumption and coefficient of performance of the optimal heat pump cycle system state are obtained; the set convergence threshold ∈ ensures that the recursive optimization process stops in time after reaching the optimal heat pump cycle system state, ensuring that the recursive optimization process is efficient and not lengthy. The set convergence threshold ∈ can be specifically set according to the specific implementation scenario and is not limited here;

[0083] After the recursive optimization process stops, the optimized state vector is obtained, including the power consumption and coefficient of performance of the optimal state. The optimized state vector is as follows:

[0084]

[0085] Among them, Represents the power consumption of the optimal state; Represents the coefficient of performance of the optimal state;

[0086] According to the power consumption and coefficient of performance of the optimal state, dynamically adjust the condensation temperature T c (t) and the evaporation temperature T e (t), to achieve the optimization of the heat pump cycle performance. The feedback regulation formula is as follows:

[0087]

[0088] Among them, T c(t + Δt) represents the condensation temperature at t + Δt; T c (t) represents the condensation temperature at t; COP(t) represents the coefficient of performance at t;

[0089]

[0090] where, T e (t + Δt) represents the evaporation temperature at t + Δt; T e (t) represents the evaporation temperature at t; P(t) represents the power consumption at t;

[0091] To ensure the stability of the feedback regulation, the following stability conditions are set:

[0092] |T c (t + Δt) - T c (t)| < δ1

[0093] |T e (t + Δt) - T e (t)| < δ2

[0094] where, δ1 and δ2 respectively represent the allowable change ranges of the condensation temperature and the evaporation temperature, which can be specifically set according to the specific implementation scenario and are not limited herein. When the above stability conditions are met, the adjustment of the condensation temperature and the evaporation temperature of the heat pump cycle system is ended, and the performance of the heat pump cycle is optimized.

[0095] In summary, a method for optimizing and predicting the performance of a heat pump cycle is completed.

[0096] The sequence of the invention embodiments is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0097] Each embodiment in this specification is described in a progressive manner. For the same or similar parts between each embodiment, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

[0098] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for optimizing and predicting heat pump cycle performance, characterized in that: The following steps are involved: S1: Real-time acquisition of different data in the heat pump circulation system to generate a multi-dimensional spatiotemporal data matrix; based on the multi-dimensional spatiotemporal data matrix, the prediction results, including power consumption and energy efficiency ratio, are obtained through an adaptive spatiotemporal convolutional neural network prediction model; the adaptive spatiotemporal convolutional neural network prediction model captures the dynamic correlation characteristics between the multi-dimensional spatiotemporal data in the multi-dimensional spatiotemporal data matrix at different times and spatial positions through spatiotemporal convolution operations, and processes each feature through weighted operations to obtain weighted output results; and the weighted output results are adaptively nonlinearly activated; the spatiotemporal convolution operation starts from the input multi-dimensional spatiotemporal data matrix, extracts data features layer by layer through convolution operations, and in each layer of spatiotemporal convolution operations, selects a fixed length of time segment through a sliding window method, processes the multi-dimensional spatiotemporal data in each time segment, and obtains an output result matrix; the weighted operation obtains a weighted output result by weighting the output result matrix and adding a bias term, and the specific calculation formula is: Among them, Z(t+Δt) represents the weighted output result at t+Δt; Y L (t+Δt) represents the feature matrix of the Lth layer after the spatiotemporal convolution operation at t+Δt; represents the weight matrix of the Lth layer; b L represents the bias term of the Lth layer; They represent the weighted sum of the dimension i of the input feature and the dimension j of the output feature, respectively. n is the number of input features, and m is the number of output features. S2: Generate an initial state vector based on the prediction results; use an adaptive recursive optimization algorithm based on meta-learning to obtain the optimal power consumption and energy efficiency ratio; dynamically adjust the condensing temperature and evaporating temperature of the heat pump circulation system according to the optimal power consumption and energy efficiency ratio.

2. A heat pump cycle performance optimization and prediction method according to claim 1, characterized in that: The S2 specifically includes: The meta-learning-based adaptive recursive optimization algorithm consists of a recursive optimization inner loop and a meta-learning-based outer loop. The learning rate is dynamically adjusted through the collaborative work of the inner and outer loops; and the objective function is used to quantify the balance between the energy efficiency ratio and power consumption of the heat pump circulation system.

3. A heat pump cycle performance optimization and prediction method according to claim 2, characterized in that: The S2 specifically includes: The objective function is gradient-calculated, and the gradient vector in the recursive optimization process is obtained by calculating the partial derivatives of power consumption and energy efficiency ratio. The formula is as follows: Where F(t+Δt) represents the value of the objective function at t+Δt; represents the gradient vector of the objective function at t+Δt; COP(t+Δt) represents the energy efficiency ratio at t+Δt; P(t+Δt) represents the power consumption at t+Δt; represents the partial derivative of the objective function with respect to power consumption; Represents the partial derivative of the objective function with respect to the energy efficiency ratio.

4. A heat pump cycle performance optimization and prediction method according to claim 3, characterized in that: The S2 specifically includes: The meta-learning optimization framework performs adaptive recursive optimization through the collaborative work of inner and outer loops. In the inner loop, the state of the heat pump circulation system is updated according to the gradient vector of the objective function; the outer loop introduces the meta-learning loss function to adjust the learning rate.

5. A heat pump cycle performance optimization and prediction method according to claim 4, characterized in that: The S2 specifically includes: A convergence threshold is set. When the state change of the heat pump circulation system is less than the convergence threshold, the recursive optimization process stops and the optimized state vector is obtained, including the power consumption and energy efficiency ratio of the optimal state.

Images