Intelligent management and control method and system for heat pump energy system

Abstract

The invention discloses an intelligent management and control method and system for a heat pump energy system. The intelligent management and control method comprises the steps that S1-S4, indoor and outdoor environment parameters of a target building, working parameters of energy supply equipment, underground soil environment parameters and total energy consumption of the heat pump energy system are collected; s5, predicting indoor and outdoor environment parameters and soil environment parameters of the target building in a period of time in the future based on the collected data by adopting an RNN prediction model; and S6, on the basis of the decomposed multi-objective evolutionary algorithm and the prediction result, working parameters of the heat pump energy system are optimized to achieve the purposes of minimizing total energy consumption, minimizing the working frequency of the heat pump and minimizing the power of the circulating water pump, and the working parameters of the heat pump energy system are output to control the heat pump energy system. According to the method and system, accurate prediction and intelligent adjustment can be achieved, the energy utilization efficiency is improved, the operation cost of the heat pump energy system is reduced, and the stability of the heat pump energy system is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management and control systems, and specifically to an intelligent control method and system for a heat pump energy system, which are used to improve the comfort level of a building's indoor environment, energy efficiency, and the stability of a heat pump energy system. Background Art

[0002] At present, residents in southern China generally rely on air conditioners and electric heaters for household heating during the cold winter season. Although this practice is convenient to operate, it is accompanied by a series of problems. First, the high-power operation of air conditioners and electric heaters puts families who use them for a long time under considerable economic pressure. Secondly, the energy utilization efficiency in the household heating mode is low, which is not conducive to energy conservation and emission reduction, and poses a challenge to environmental protection. In addition, there are potential safety hazards in the use of household heating equipment, such as equipment aging and operating errors, which may trigger safety accidents such as fire or electric shock.

[0003] It is worth noting that compared with the northern regions, the climate conditions in some areas of the south are quite similar to those in some areas of the north, and the heating needs are also common. Therefore, for these southern regions, the implementation of centralized heating has many advantages. Centralized heating can significantly improve energy utilization efficiency and reduce energy waste through centralized management and allocation of energy. At the same time, it can also reduce the investment cost of household heating equipment, further reducing the economic burden on residents.

[0004] With the continuous improvement of residents' quality of life, the application of five-constant centralized systems such as ceiling radiant heating and cooling has become more and more widespread in southern China. The system uses radiation to transfer heat, ensuring the uniform distribution of indoor temperature and greatly improving the comfort of heating and cooling. However, the traditional five-constant centralized system control method is relatively simple, mainly relying on manual setting of system control points, resulting in low energy utilization and high operation and maintenance costs.

[0005] In order to improve the energy efficiency and stability of the Wuheng centralized system, intelligent management and control technology has received increasing attention. At present, the comprehensive control and monitoring technology of the Wuheng heat pump system has made significant progress, but it still faces some challenges. Specifically, the existing intelligent management and control technology still lacks advanced algorithms for coupling intelligent simulation operations with operating status, which makes the system unable to accurately grasp the internal and external environment of the building and the energy consumption demand and supply capacity of each link of the system, thus affecting the stability and energy efficiency of the system. Therefore, how to achieve intelligent and precise control of the heat pump energy system has become a technical problem that needs to be overcome urgently. Summary of the invention

[0006] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0007] In order to solve the above technical problems, the present invention proposes an intelligent control method and system for a heat pump energy system. The method and system of the present invention can realize accurate prediction and intelligent regulation of the heat pump energy system, improve energy utilization efficiency, reduce the operating cost of the heat pump energy system, and enhance the stability of the heat pump energy system.

[0008] In order to achieve these purposes and other advantages according to the present invention, a method for intelligent management and control of a heat pump energy system is provided, comprising the following steps:

[0009] S1. Collect indoor and outdoor environmental parameters of the target building;

[0010] S2. Collecting the working parameters of energy supply equipment;

[0011] S3. Collect soil environmental parameters under the target building;

[0012] S4, collecting the total energy consumption of the heat pump energy system;

[0013] S5, using the RNN prediction model, based on the data collected in steps S1 to S4, predict the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building in the future;

[0014] S6. Adopting a decomposition-based multi-objective evolutionary algorithm, based on the prediction result of step S5, the operating parameters of the heat pump energy system are optimized to minimize the total energy consumption, minimize the operating frequency of the heat pump, and minimize the power of the circulating water pump. The operating parameters of the heat pump energy system are output to perform energy management and control on the heat pump energy system.

[0015] Preferably, the indoor environmental parameters include indoor temperature, humidity, carbon dioxide concentration and indoor area, and the outdoor environmental parameters include outdoor temperature, humidity, wind speed and sunshine intensity;

[0016] The energy supply equipment includes a heat pump host, a circulating water pump, a ground source heat pump buried pipe heat exchanger and an energy storage device. The operating parameters include the inlet and outlet water temperature, compressor frequency and power of the heat pump host, the flow rate, head and power of the circulating water pump, the inlet and outlet water temperature and flow rate of the ground source heat pump buried pipe heat exchanger, the storage capacity and charging and discharging power of the energy storage device;

[0017] The soil environmental parameters include soil temperature, soil moisture, buried pipe water temperature, and soil thermal conductivity;

[0018] When training the RNN prediction model, regularization terms L1 and L2 are added to the loss function, where regularization term L1 is constructed based on the sum of the absolute values ​​of the weights corresponding to all the above parameters, and regularization term L2 is constructed based on the sum of the squares of the weights corresponding to the above parameters. At the same time, the Dropout technology is used in the hidden layer of the RNN model. During the training process, the outputs of some neurons in the hidden layer are randomly set to 0 with a set dropout probability, and the dropout probability ranges from 0 to 1.

[0019] Preferably, the processing of all the above parameters further includes:

[0020] Collect the heat pump system operation data of all the above parameters, pre-process the data for accuracy and completeness, and clarify the energy efficiency target values: energy consumption and energy efficiency ratio;

[0021] Using Python's NumPy and pandas libraries, the Pearson correlation coefficient between the preprocessed data and the energy efficiency target value is calculated using the numpy.corrcoef function, and the Spearman rank correlation coefficient is calculated using the scipy.stats.spearmanr function to evaluate the linear and nonlinear correlation between the processed data and the energy efficiency target value;

[0022] Screen out processed data with high linear or nonlinear correlation with energy efficiency target values: set processed data with an absolute value of Pearson correlation coefficient greater than 0.4-0.5 and an absolute value of Spearman rank correlation coefficient greater than 0.3-0.4 as key focus features to form a feature subset;

[0023] Use Python's scikit-learn library to build a random forest regression model, and divide the above feature subsets into training and test sets at a ratio of 80:20. Use the RandomForestRegressor class to initialize the model, set n_estimators to 100-120 and max_depth to 10-12, and use the fit method to train the model on the training set. After training, use the feature_importances_ attribute to obtain the importance score of each feature in the random forest model.

[0024] The linear and nonlinear correlation results of each feature's score and the above energy efficiency target value are mutually verified to further determine the importance of the feature to energy efficiency prediction.

[0025] Preferably, further determining the importance of the feature to the energy efficiency prediction further includes:

[0026] Use the xgboost library to build the XGBoost model. First, convert the above feature subset into DMatrix format, then initialize the XGBRegressor model, set learning_rate to 0.1-0.2, n_estimators to 150-180, train the model through the fit method, and after the training, use the get_score method to obtain the split gain of each feature.

[0027] Compare the linear and nonlinear correlation results of split gain with the above energy efficiency target values, and give higher attention weights to features with high linear and nonlinear correlation and large split gain in the XGBoost model;

[0028] Among them, the results of the random forest regression model and the XGBoost model are combined to calculate the comprehensive weight:

[0029] Let the importance score of the feature in the random forest model be S rf , the split gain of the feature in the XGBoost model is G xgb , and the weight ratio of the random forest model is α: 0<α<1, then the calculation formula of the comprehensive weight W is:

[0030]

[0031] Among them, n is the total number of features and i is the current feature.

[0032] Preferably, the method further includes setting the coefficients of the regularization term L1 and the regularization term L2:

[0033] Set a weight threshold to filter out features with weights greater than the threshold to form the final feature subset

[0034] The final feature subset is divided into a training set and a validation set at a ratio of 80:20. Based on experience, the regularization coefficient λ1 of the regularization term L1 and the regularization coefficient λ2 of the regularization term L2 are preliminarily determined to be in the range of [10 -4 , 1];

[0035] Discrete candidate values ​​are selected within the value range to form a value grid of λ1 and λ2: where both λ1 and λ2 are

[0036] For each combination of λ1 and λ2, the RNN prediction model is trained, and the combination of λ1 and λ2 that makes the model perform best on the validation set is selected as the final regularization coefficient.

[0037] Preferably, training the RNN prediction model includes:

[0038] Initialize the parameters of the RNN prediction model according to the weights

[0039] The mean square error loss function MSE is used, and the loss function including the regularization term is constructed according to the following formula:

[0040] where w i represents the weight corresponding to the i-th feature, and n is the total number of features;

[0041] The stochastic gradient descent algorithm is used to adjust the weight parameters of the model according to the gradient of the loss function, and multiple rounds of training are performed until the preset number of training rounds is reached;

[0042] Among them, the regularization term L1 is

[0043] The regularization term L2 is

[0044] The mean square error loss function MSE is used to measure the difference between the model prediction value and the true value.

[0045] where y j is the true value of the jth sample, is the predicted value of the jth sample, and m is the number of samples.

[0046] Provided is an intelligent management and control system for a heat pump energy system, the heat pump energy system being used to execute the intelligent management and control method, the intelligent management and control system comprising:

[0047] A data acquisition module, which is used to collect indoor and outdoor environmental parameters of the target building, working parameters of energy supply equipment, soil environmental parameters underground of the target building, and total energy consumption of the heat pump energy system;

[0048] A prediction module, which is used to use a recurrent neural network to predict the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building within a period of time in the future based on the data collected by the data collection module;

[0049] The optimization decision module is used to adopt a decomposition-based multi-objective evolutionary algorithm to optimize the working parameters of the heat pump energy system based on the prediction results of the prediction module, so as to minimize the total energy consumption, minimize the operating frequency of the heat pump, and minimize the power of the circulating water pump, and output the working parameters of the heat pump energy system to perform energy management and control of the heat pump energy system.

[0050] Preferably, the data acquisition module includes a sensor network, which is respectively arranged indoors and outdoors, in circulating water pumps, water pipes, indoor fan coils and soil of the target building, for real-time acquisition of various environmental parameters, soil environmental parameters, and working parameters of energy supply equipment.

[0051] Preferably, the intelligent management and control system further includes a user interface module, which is used to display the prediction results, optimized working parameters and current system status, and allows the user to adjust the optimization objectives or set priorities as needed.

[0052] The present invention has at least the following beneficial effects:

[0053] First, by using the recurrent neural network (RNN) as the core algorithm of the prediction module, the characteristics of time series data can be fully utilized to effectively capture the changing trends and periodic laws of environmental parameters (such as indoor temperature, outdoor temperature, humidity, soil temperature, etc.) over time; the cyclic structure of RNN enables it to process input sequences of any length, so that it can make high-precision predictions of environmental parameters in the future based on historical data, providing a reliable basis for optimizing the decision-making module;

[0054] Second, the decomposition-based multi-objective evolutionary algorithm can simultaneously consider multiple optimization objectives (such as minimizing total energy consumption, minimizing heat pump operating frequency, and minimizing circulating water pump power), thereby achieving the optimal configuration of energy use while ensuring system performance; the algorithm searches for the optimal solution in the solution space by simulating natural selection and genetic mechanisms, and can handle complex nonlinear problems and constraints, thereby improving the overall energy efficiency of the heat pump energy system;

[0055] Third, by combining the prediction module with the optimization decision module, this technical solution realizes the intelligent energy management and control of the heat pump energy system; this intelligent energy management and control system can automatically adjust the working parameters of the heat pump energy system according to the real-time collected data and prediction results to adapt to the changing environmental conditions and energy needs, which not only improves the energy efficiency and reliability of the heat pump energy system, but also reduces the cost and risk of manual intervention, providing strong support for the sustainable operation of the building;

[0056] Fourth, it can achieve precise optimization of the working parameters of the heat pump energy system, thereby significantly reducing the total energy consumption of the system, the working frequency of the heat pump and the power of the circulating water pump, thereby achieving the goal of energy conservation and emission reduction, which is of great significance for alleviating the energy crisis, reducing greenhouse gas emissions, and promoting green buildings and sustainable development.

[0057] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of the intelligent control method of the heat pump energy system of the present invention. DETAILED DESCRIPTION

[0059] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0060] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the orientation or positional relationship indicated by the terms is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0061] like Figure 1 As shown, an embodiment of the present invention provides an intelligent management and control method for a heat pump energy system, comprising the following steps:

[0062] S1. Collect indoor and outdoor environmental parameters of the target building; specifically, indoor environmental parameters include indoor temperature, indoor humidity, indoor carbon dioxide concentration and indoor area of ​​the target building; outdoor environmental parameters include outdoor temperature, outdoor humidity, wind speed and sunshine intensity of the target building; indoor environmental parameters reflect the environmental requirements of the target building, and outdoor environmental parameters reflect the external environmental conditions of the target building, reflecting the difficulty or energy consumption of adjusting the outdoor temperature and humidity conditions to the indoor state through the heat pump energy system.

[0063] S2. Collect the working parameters of the energy supply equipment; specifically, the energy supply equipment includes a heat pump host, a circulating water pump, water pipes and indoor fan coils; for example, the heat pump host is the core equipment of the soil source heat pump energy system, responsible for extracting heat from the soil and transferring it to the indoor space. The heat pump host is usually composed of main components such as a compressor, a condenser, an evaporator and an expansion valve. Through the collaboration of these components, heat transfer and transportation are achieved; optionally, the working parameters of the heat pump host include the heat pump operating frequency, set temperature, hysteresis temperature, water discharge temperature, electric heating allowable temperature, electric heating hysteresis temperature, system pressure, etc.; the circulating water pump plays a key role in the soil source heat pump energy system, responsible for circulating water or other heat exchange media between the underground buried pipes and the heat pump host. Through the operation of the circulating water pump, the soil source heat pump energy system can continuously extract or release heat from the soil, thereby maintaining the comfort of the indoor space. ; Optionally, the working parameters of the circulating water pump include flow rate, speed and power; the water pipe is the key part connecting the heat pump host, underground buried pipes and indoor terminal equipment, and is responsible for transmitting heat or cold from underground to the heat pump host, and then transporting heat or cold to the room; Optionally, the working parameters of the water pipe include flow rate, water temperature and flow rate; the indoor fan coil is the indoor terminal equipment in the soil source heat pump energy system, which is responsible for releasing the heat or cold provided by the heat pump host into the indoor space, thereby adjusting the indoor temperature. The indoor fan coil is usually composed of a fan and a heat exchanger, which can efficiently adjust the indoor temperature; Optionally, the working parameters of the indoor fan coil include air volume, wind speed and power; the control system is the brain of the energy supply equipment, responsible for monitoring and adjusting the operating status of the energy supply equipment, and controlling the stable and orderly operation of the energy supply equipment through precise sensor networks and intelligent management and control methods. Preferably, the energy supply equipment working parameters select equipment status parameters (on, off and standby of heat pumps, circulating water pumps and indoor fan coils) and setting parameters (setting parameters of heat pumps, circulating water pumps and indoor fan coils, such as target temperature, flow, pressure, wind speed, etc.).

[0064] S3. Collect soil environmental parameters underground of the target building. Specifically, soil environmental parameters include soil temperature, water temperature in underground pipes, flow rate and valve opening. After the soil source heat pump energy system is in operation, it will continuously absorb or release heat from the soil, causing changes in soil temperature.

[0065] S4. Collect the total energy consumption of the heat pump energy system; specifically, the total energy consumption is measured by the total power consumption of the entire energy system, which is detected and collected through an electric energy meter.

[0066] During the data collection process, a time synchronization mechanism is used to ensure that the timestamps of all collected data are consistent, which facilitates subsequent analysis and prediction.

[0067] S5. Using the RNN prediction model, based on the data collected in steps S1 to S4, the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building in the future are predicted; using a recurrent neural network (RNN), combined with a long short-term memory (LSTM) unit or a gated recurrent unit (GRU), to enhance the modeling ability of time series data, based on the collected data, the following parameters in the future are predicted with high precision, preferably, indoor temperature value, indoor humidity value and indoor carbon dioxide concentration value, outdoor environmental parameters of the target building outdoor temperature value, outdoor humidity value, wind speed value and sunshine intensity value, soil temperature and water temperature in the underground buried pipe; preferably, before the prediction, it also includes data preprocessing steps, such as outlier detection and processing, missing value filling, etc., to improve the accuracy of the prediction. Preferably, the RNN prediction model can learn and simulate the law of changes in environmental parameters and working parameters of energy supply equipment over time through historical data training to improve the accuracy of the prediction; preferably, when the RNN prediction model is trained through historical data, the model parameters are optimized by cross-validation and grid search methods to improve the accuracy of the prediction. At the same time, the RNN prediction model has online learning capabilities and can be continuously updated and optimized according to actual operating data.

[0068] S6. Adopting a decomposition-based multi-objective evolutionary algorithm, based on the prediction results of step S5 (indoor temperature value, indoor humidity value and indoor carbon dioxide concentration value, outdoor temperature value, outdoor humidity value, wind speed value and sunshine intensity value, soil temperature and water temperature in underground pipes), the working parameters of the heat pump energy system (the on, off and standby state parameters of the heat pump, circulating water pump and indoor fan coil unit, the setting parameters of the heat pump, circulating water pump and indoor fan coil unit, such as target temperature, flow rate, pressure, wind speed, etc.) are optimized to minimize the total energy consumption, minimize the operating frequency of the heat pump, minimize the power of the circulating water pump, and output the working parameters of the heat pump energy system to perform energy management and control on the heat pump energy system. The minimization of total energy consumption aims to reduce energy consumption by adjusting the working parameters of the heat pump energy system, thereby reducing operating costs; by reducing the working frequency of the heat pump, its energy consumption and wear can be reduced, the equipment life can be extended, and noise pollution can be reduced; reducing the power of the circulating water pump can reduce the energy consumption of the water circulation system; specifically, the multi-objective evolutionary algorithm based on decomposition adopts elite retention strategy, dynamic adjustment of weights and other methods in the solution process to maintain the diversity and convergence of the solution set. At the same time, the algorithm has an adaptive adjustment mechanism, which can dynamically adjust parameters according to feedback information during the optimization process. Preferably, according to the output working parameters of the heat pump energy system (such as the working parameters of the heat pump host and the circulating water pump), the working parameters of energy equipment such as the heat pump host and the circulating water pump can be automatically adjusted to achieve the optimization of energy use and improve the stability of the operation of the heat pump energy system.

[0069] This implementation case uses the RNN prediction model to learn and simulate the laws of changes in environmental parameters and working parameters of energy supply equipment over time, so as to achieve accurate prediction and simulation of future environmental parameters of the target heat pump energy system, and provide a data basis for the adjustment of control parameters. By adopting a multi-objective evolutionary algorithm based on decomposition, not only indoor environmental parameters and energy equipment working parameters are considered, but also external environmental data. Under the premise of reducing total energy consumption, the start-stop frequency of energy supply equipment can be reduced, the stable and orderly operation of energy supply equipment can be controlled, and the comfort of the indoor environment can be guaranteed.

[0070] In another technical solution, the indoor environmental parameters include indoor temperature, humidity, carbon dioxide concentration and indoor area, and the outdoor environmental parameters include outdoor temperature, humidity, wind speed and sunshine intensity;

[0071] The energy supply equipment includes a heat pump host, a circulating water pump, a ground source heat pump buried pipe heat exchanger and an energy storage device. The operating parameters include the inlet and outlet water temperature, compressor frequency and power of the heat pump host, the flow rate, head and power of the circulating water pump, the inlet and outlet water temperature and flow rate of the ground source heat pump buried pipe heat exchanger, the storage capacity and charging and discharging power of the energy storage device;

[0072] The soil environmental parameters include soil temperature, soil moisture, buried pipe water temperature, and soil thermal conductivity;

[0073] When training the RNN prediction model, regularization terms L1 and L2 are added to the loss function, where regularization term L1 is constructed based on the sum of the absolute values ​​of the weights corresponding to all the above parameters, and regularization term L2 is constructed based on the sum of the squares of the weights corresponding to the above parameters. At the same time, the Dropout technology is used in the hidden layer of the RNN model. During the training process, the outputs of some neurons in the hidden layer are randomly set to 0 with a set dropout probability, and the dropout probability ranges from 0 to 1.

[0074] This solution mainly solves the problems of unclear definition of various parameters and poor training effect of RNN prediction model. By clarifying the specific content of various parameters, adding regularization terms and using Dropout technology when training the model, the accuracy and generalization ability of the model are improved. Clarifying the content of various parameters makes data collection and processing more standardized and accurate, which is conducive to improving the operation and management level of the system. Adding regularization terms L1 and L2 to the loss function can effectively prevent model overfitting and improve the adaptability and generalization ability of the model to different data. Using Dropout technology in the hidden layer of the RNN model further enhances the robustness of the model, making the model more stable during training and improving the accuracy of prediction.

[0075] Parameter clarification and sensor selection:

[0076] For indoor environmental parameters, high-precision temperature sensors are used, such as platinum resistance temperature sensors, which have an accuracy of ±0.05°C and can accurately measure indoor temperature. Indoor humidity sensors use capacitive humidity sensors with an accuracy of ±2%RH to ensure the accuracy of humidity data. Carbon dioxide concentration sensors can use infrared sensors with a measurement range of 0-5000ppm and an accuracy of ±50ppm to meet the needs of indoor air quality monitoring. Indoor area can be accurately measured using a laser rangefinder or professional area measurement software combined with architectural drawings.

[0077] In terms of outdoor environmental parameter monitoring, ultrasonic wind speed sensors are used with a measurement range of 0-60m / s and an accuracy of ±0.1m / s, which can stably obtain wind speed data. Outdoor temperature and humidity sensors are similar to indoor ones, but they need to have better protection performance to adapt to severe weather conditions. The sunlight intensity sensor uses a photoelectric sensor that can convert light intensity into an electrical signal for measurement. The measurement range can cover 0-2000W / ㎡ with an accuracy of ±5W / ㎡.

[0078] In terms of energy supply equipment parameter monitoring, armored thermocouple water temperature sensors are installed at the inlet and outlet of the heat pump main engine, which have fast response speed and measurement accuracy of ±0.1℃. The compressor frequency can be obtained by communicating with the compressor control system, and the power measurement uses a power analyzer with an accuracy of 0.5. The flow sensor of the circulating water pump uses an electromagnetic flowmeter with an accuracy of ±0.5%, and the head is measured by a pressure sensor with an accuracy of ±0.2%. The inlet and outlet water temperature sensors of the ground source heat pump buried pipe heat exchanger also use armored thermocouples, and the flow sensor uses an ultrasonic flowmeter with an accuracy of ±1%. The storage capacity and charging and discharging power of the energy storage equipment are read through its own management system.

[0079] When measuring soil environmental parameters, the soil temperature sensor uses a low-temperature and corrosion-resistant platinum resistance sensor, which is inserted into the soil at an appropriate depth for measurement. The soil moisture sensor uses a frequency domain reflection sensor, which can accurately measure soil moisture. The buried pipe water temperature sensor is installed near the buried pipe to monitor the water temperature in real time. The soil thermal conductivity is measured by a professional thermal probe measuring instrument to ensure the reliability of the data.

[0080] RNN model training details:

[0081] Choose a suitable deep learning framework, such as TensorFlow. When building an RNN model, determine the number of input layer neurons based on the dimension of the input data. For example, if the input contains 10 different parameters, the number of input layer neurons is 10. The number of hidden layer neurons can be determined through multiple experiments, such as setting it to 32, 64, etc., to balance model complexity and performance. The number of output layer neurons is determined according to the prediction target. For example, if there are 5 parameters to be predicted, such as indoor and outdoor temperature, soil temperature, etc., the number of output layer neurons is 5.

[0082] When training the RNN prediction model, add the regularization term strictly in accordance with claim 2. When calculating the regularization term L1, traverse the weights corresponding to all parameters in the model, and add their absolute values ​​to obtain the value of L1; when calculating the regularization term L2, square each weight and sum them. Add these two regularization terms to the loss function in a certain proportion, such as L = MSE + 0.01 * L1 + 0.001 * L2 (the proportional coefficient can be adjusted according to actual conditions).

[0083] Enable Dropout technology in the hidden layer of the RNN model. Taking the dropout probability of 0.5 as an example, during the training process, each neuron in the hidden layer has a 50% probability of being randomly set to 0 at each iteration, which can prevent the model from over-relying on certain neurons and reduce overfitting. When implemented in the framework, it can be achieved through simple function calls, such as using `tf.keras.layers.Dropout(0.5)` in TensorFlow.

[0084] Model evaluation and tuning:

[0085] After training is complete, the model is evaluated using an independent test data set. Multiple error metrics between the predicted value and the true value are calculated. In addition to the root mean square error (RMSE) and mean absolute error (MAE), the mean absolute percentage error (MAPE) can also be calculated to more comprehensively evaluate the prediction accuracy of the model. For example, if the RMSE value is large, it means that the deviation between the model prediction value and the true value is large and further optimization is needed.

[0086] Adjust the model based on the evaluation results. If the model is found to be overfitting, the discard probability of Dropout can be appropriately increased, or the coefficient of the regularization term can be increased; if the model is underfitting, try to increase the number of neurons in the hidden layer, or adjust the model structure, such as increasing the number of hidden layers. At the same time, the training data can also be expanded, such as through data enhancement technology, to generate more training samples and improve the generalization ability of the model. After multiple adjustments and retraining, the model performance meets the actual application requirements.

[0087] In another technical solution, the processing of all the above parameters further includes:

[0088] Collect the heat pump system operation data of all the above parameters, pre-process the data for accuracy and completeness, and clarify the energy efficiency target values: energy consumption and energy efficiency ratio;

[0089] Using Python's NumPy and pandas libraries, the Pearson correlation coefficient between the preprocessed data and the energy efficiency target value is calculated using the numpy.corrcoef function, and the Spearman rank correlation coefficient is calculated using the scipy.stats.spearmanr function to evaluate the linear and nonlinear correlation between the processed data and the energy efficiency target value;

[0090] Screen out processed data with high linear or nonlinear correlation with energy efficiency target values: set processed data with an absolute value of Pearson correlation coefficient greater than 0.4-0.5 and an absolute value of Spearman rank correlation coefficient greater than 0.3-0.4 as key focus features to form a feature subset;

[0091] Use Python's scikit-learn library to build a random forest regression model, and divide the above feature subsets into training and test sets at a ratio of 80:20. Use the RandomForestRegressor class to initialize the model, set n_estimators to 100-120 and max_depth to 10-12, and use the fit method to train the model on the training set. After training, use the feature_importances_ attribute to obtain the importance score of each feature in the random forest model.

[0092] The linear and nonlinear correlation results of each feature's score and the above energy efficiency target value are mutually verified to further determine the importance of the feature to energy efficiency prediction.

[0093] This solution is dedicated to solving the problem of how to screen and process data with high correlation with energy efficiency target values ​​to improve the accuracy of energy efficiency prediction. Through a series of data processing and model building methods, the important features for energy efficiency prediction are determined. Data preprocessing ensures the quality of the data and provides reliable data support for subsequent analysis and model training. By calculating the correlation coefficient to screen out data with high correlation with the energy efficiency target value, key factors can be highlighted, the interference of redundant data on the model can be reduced, and the accuracy of energy efficiency prediction can be improved. Constructing a random forest regression model and obtaining feature importance scores provides a quantitative basis for determining the importance of features to energy efficiency prediction, which helps to manage energy efficiency more scientifically.

[0094] Data collection and preprocessing: Export the historical operation data of the heat pump system containing all parameters from the data storage server. These data should cover operation records in different seasons, different time periods and different working conditions. Use Python's data processing library, such as pandas, to clean the data. Check whether there are missing values ​​in the data. For a small number of missing values, use the mean, median or interpolation method to fill them; for data samples with a large number of missing values, consider directly deleting them. At the same time, check the outliers in the data, identify and process outliers by setting a reasonable data range (such as the normal fluctuation range of parameters such as temperature and humidity), and ensure the accuracy and completeness of the data. Clarify the energy efficiency target value, that is, energy consumption and energy efficiency ratio. Energy consumption can be directly obtained through the total energy consumption data of the system, and the energy efficiency ratio is calculated based on the heating or cooling capacity of the system and the total energy consumption.

[0095] Correlation analysis: Use Python's NumPy and pandas libraries to import preprocessed data. Use the numpy.corrcoef function to calculate the Pearson correlation coefficient between each parameter and the energy efficiency target value. This coefficient is used to measure the degree of linear correlation between two variables. The closer its value is to 1 or -1, the stronger the linear correlation. Use the scipy.stats.spearmanr function to calculate the Spearman rank correlation coefficient, which is used to evaluate the nonlinear correlation between variables. After the calculation is completed, organize the calculation results of the two correlation coefficients into a table to intuitively compare the correlation between each parameter and the energy efficiency target value.

[0096] Feature screening: According to the set threshold, that is, the absolute value of the Pearson correlation coefficient is greater than 0.4-0.5 and the absolute value of the Spearman rank correlation coefficient is greater than 0.3-0.4, the processed data with high linear or nonlinear correlation with the energy efficiency target value are screened out, and these data are taken as key focus features to form a feature subset.

[0097] Random forest regression model construction and training: Use Python's scikit-learn library to build a random forest regression model. First, divide the feature subset into training set and test set in a ratio of 80:20 to ensure that the data distribution of the training set and test set is representative. Use the RandomForestRegressor class to initialize the model, set n_estimators to 100-120, which represents the number of trees in the forest. The more trees there are, the stronger the generalization ability of the model may be, but the computational cost will also increase; set max_depth to 10-12, which limits the maximum depth of the tree to prevent overfitting. Use the fit method to train the model on the training set. During the training process, the model will automatically learn the relationship between the feature and the energy efficiency target value. After the training is completed, use the feature_importances_ attribute to obtain the importance score of each feature in the random forest model. The score reflects the contribution of each feature to the model prediction result.

[0098] In another technical solution, further determining the importance of the feature to the energy efficiency prediction also includes:

[0099] Use the xgboost library to build the XGBoost model. First, convert the above feature subset into DMatrix format, then initialize the XGBRegressor model, set learning_rate to 0.1-0.2, n_estimators to 150-180, train the model through the fit method, and after the training, use the get_score method to obtain the split gain of each feature.

[0100] Compare the linear and nonlinear correlation results of split gain with the above energy efficiency target values, and give higher attention weights to features with high linear and nonlinear correlation and large split gain in the XGBoost model;

[0101] Among them, the results of the random forest regression model and the XGBoost model are combined to calculate the comprehensive weight:

[0102] Let the importance score of the feature in the random forest model be S rf , the split gain of the feature in the XGBoost model is G xgb , and the weight ratio of the random forest model is α: 0<α<1, then the calculation formula of the comprehensive weight W is:

[0103]

[0104] Among them, n is the total number of features and i is the current feature.

[0105] This solution mainly solves how to further accurately determine the importance of features to energy efficiency prediction. By building an XGBoost model and calculating the comprehensive weights based on the results of the random forest regression model, the importance of features can be more accurately evaluated. The XGBoost model is built and the feature split gain is obtained to evaluate the importance of features from another perspective, which complements the results of the random forest regression model. By comparing the split gain and correlation results, higher attention weights are given to important features, making feature evaluation more comprehensive and accurate. The comprehensive weights are calculated, taking into account the results of the two models, providing a more reliable feature importance assessment for energy efficiency prediction, which helps to optimize energy efficiency more accurately.

[0106] XGBoost model construction and training: First, convert the obtained feature subset into DMatrix format, which is a data structure unique to the XGBoost library and can improve the efficiency of model training. Initialize the XGBRegressor model and set learning_rate to 0.1-0.2. This parameter controls the step size of the model update at each iteration. A smaller value can make the model training more stable, but the training time may increase; set n_estimators to 150-180, which also indicates the number of trees in the model. The model is trained through the fit method. During the training process, the model will continuously adjust the tree structure and parameters according to the input feature subset and energy efficiency target value to improve the accuracy of the prediction.

[0107] Feature importance evaluation and weight calculation: After training, use the get_score method to obtain the split gain of each feature. The split gain indicates the contribution of the feature to reducing the uncertainty of the sample (such as information entropy) during the decision tree splitting process. The larger the split gain, the more important the feature. Compare the split gain of each feature in the XGBoost model with the linear and nonlinear correlation results of the calculated energy efficiency target value. For features with high linear and nonlinear correlation and large split gain in the XGBoost model, give higher attention weights. Combine the results of the random forest regression model and the XGBoost model to calculate the comprehensive weight. Assume that the importance score of the feature in the random forest model is S i , the split gain of the feature in the XGBoost model is G i , and the weight of the random forest model is w, according to the formula W i =w×S i +(1-w)×G i , calculate the comprehensive weight of each feature, where n is the total number of features and i is the current feature. Through the comprehensive weight, the importance of each feature to energy efficiency prediction can be evaluated more comprehensively and accurately.

[0108] In another technical solution, the step of setting the coefficients of the regularization term L1 and the regularization term L2 is also included:

[0109] Set a weight threshold to filter out features with weights greater than the threshold to form the final feature subset

[0110] The final feature subset is divided into a training set and a validation set at a ratio of 80:20. Based on experience, the regularization coefficient λ1 of the regularization term L1 and the regularization coefficient λ2 of the regularization term L2 are preliminarily determined to be in the range of [10 -4 , 1];

[0111] Discrete candidate values ​​are selected within the value range to form a value grid of λ1 and λ2: where both λ1 and λ2 are

[0112] For each combination of λ1 and λ2, the RNN prediction model is trained, and the combination of λ1 and λ2 that makes the model perform best on the validation set is selected as the final regularization coefficient.

[0113] This solution aims to solve the problem of how to reasonably set the regularization term L1 and L2 coefficients to optimize the performance of the RNN prediction model, which is achieved by screening features, dividing the data set, and selecting the optimal coefficient combination within the value range. Screening features with large weights to form the final feature subset reduces the complexity of model training and improves training efficiency. Reasonable division of training sets and validation sets provides a scientific evaluation basis for selecting the optimal regularization coefficient. By selecting the optimal coefficient combination within the value range, the performance of the RNN prediction model is effectively optimized, and the accuracy and stability of the model are improved.

[0114] Feature screening and data set partitioning: According to the obtained feature importance score or split gain, set a weight threshold, such as 0.05 (which can be adjusted according to actual conditions). Screen out features with weights greater than the threshold to form the final feature subset. These features are considered to have a key impact on the performance and energy efficiency prediction of the model. The final feature subset is divided into a training set and a validation set at a ratio of 80:20. The training set is used to train the RNN prediction model, and the validation set is used to evaluate the performance of the model under different regularization coefficient combinations to ensure the generalization ability of the model.

[0115] Regularization coefficient value range determination and grid generation: Based on experience, it is preliminarily determined that the value range of the regularization coefficient λ1 of the regularization term L1 and the regularization coefficient λ2 of the regularization term L2 are both in [0.001, 0.1]. Discrete candidate values ​​are selected within this value range, for example, λ1 and λ2 are both 0.001, 0.01, 0.1, forming a value grid for λ1 and λ2.

[0116] Model training and coefficient selection: For each λ1 and λ2 combination, the RNN prediction model is trained according to the method described. During the training process, the model adjusts the weight parameters according to different regularization coefficients to balance the model's fitting ability and generalization ability.

[0117] Use the validation set to evaluate the trained model, and by calculating the loss function value (such as mean square error loss function MSE), accuracy and other indicators on the validation set, select the λ1 and λ2 combination that makes the model perform best on the validation set as the final regularization coefficient. For example, if the loss function value of the model on the validation set is the smallest when λ1=0.01, λ2=0.001, then select this set of coefficients as the final regularization coefficient.

[0118] In another technical solution, training the RNN prediction model includes:

[0119] Initialize the parameters of the RNN prediction model according to the weights

[0120] The mean square error loss function MSE is used, and the loss function including the regularization term is constructed according to the following formula:

[0121] where w i represents the weight corresponding to the i-th feature, and n is the total number of features;

[0122] The stochastic gradient descent algorithm is used to adjust the weight parameters of the model according to the gradient of the loss function, and multiple rounds of training are performed until the preset number of training rounds is reached;

[0123] Among them, the regularization term L1 is

[0124] The regularization term L2 is

[0125] The mean square error loss function MSE is used to measure the difference between the model prediction value and the true value.

[0126] where y j is the true value of the jth sample, is the predicted value of the jth sample, and m is the number of samples.

[0127] This solution mainly solves how to efficiently train the RNN prediction model. By reasonably initializing parameters, constructing loss functions, and using the stochastic gradient descent algorithm for multiple rounds of training, we ensure that the model can accurately predict relevant parameters. Reasonable initialization of model parameters lays the foundation for effective model training. Constructing a loss function containing regularization terms can not only measure the model prediction error, but also prevent overfitting and improve the generalization ability of the model. Using the stochastic gradient descent algorithm for multiple rounds of training enables the model to continuously optimize weight parameters, improve the accuracy of prediction, and achieve better prediction results.

[0128] Model parameter initialization: Initialize the parameters of the RNN prediction model based on historical data and experience. For example, for the weight matrix of the model, you can use a random initialization method, such as using a normal distribution or uniform distribution to generate the initial weight value, but make sure that the weight value is within a reasonable range to avoid the problem of gradient vanishing or gradient exploding in the early stages of model training. For the bias vector, you can initialize it to 0 or a small constant.

[0129] Loss function construction and optimization algorithm selection: Use mean square error loss function MSE, and follow the formula

[0130] Construct a loss function containing a regularization term, where wi represents the weight corresponding to the i-th feature, n is the total number of features, and λ1 and λ2 are the coefficients of the regularization terms L1 and L2, respectively.

[0131] The stochastic gradient descent algorithm (SGD) is used to adjust the weight parameters of the model according to the gradient of the loss function. In each iteration, a small batch of data samples is randomly selected, the gradient of the loss function on these samples is calculated, and then the weight parameters of the model are updated according to the direction and step size (learning rate) of the gradient. The learning rate can be initialized to a small value, such as 0.01, and adjusted according to the convergence of the model during the training process. For example, a learning rate decay strategy is adopted to gradually reduce the learning rate as the number of training rounds increases to ensure that the model can converge more stably in the later stages of training.

[0132] Model training and evaluation: Perform multiple rounds of training until the preset number of training rounds is reached, such as 500 rounds. In each round of training, input the training set data into the model, calculate the loss function value and gradient, and then update the model's weight parameters. After a certain number of training rounds (such as 50 rounds), use the validation set to evaluate the model and calculate indicators such as the loss function value and accuracy on the validation set. Observe the changing trends of these indicators. If the loss function value on the validation set no longer decreases or the accuracy no longer increases in several consecutive rounds of training, it means that the model may have converged and training can be stopped at this time.

[0133] After the training is completed, the model is finally evaluated using the test set. The prediction performance of the model is evaluated by calculating the prediction error on the test set (such as root mean square error RMSE, mean absolute error MAE, etc.) to ensure that the model can accurately predict the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building.

[0134] In another technical solution, the RNN prediction model is trained with historical data to learn and simulate the patterns of changes in indoor and outdoor environmental parameters, soil environmental parameters, and working parameters of energy supply equipment over time.

[0135] One embodiment includes:

[0136] Data Collection:

[0137] Indoor environmental data: collect indoor temperature, indoor humidity and indoor carbon dioxide concentration values ​​for the past year, and record them once every minute;

[0138] Outdoor environmental data: collect outdoor temperature, outdoor humidity, wind speed and sunshine intensity values ​​at the same time, and record them once every hour.

[0139] Soil environmental parameters: For the ground source heat pump energy system, data on soil temperature, water temperature in underground pipes, flow rate and valve opening are collected and recorded every hour.

[0140] Energy supply equipment parameters: record the on, off and standby status parameters of the heat pump, circulating water pump and indoor fan coil unit, and the setting parameters of the heat pump, circulating water pump and indoor fan coil unit, such as target temperature, flow, pressure, wind speed, etc., once every hour.

[0141] Data preprocessing: clean the data, remove outliers and missing values; normalize and scale all data to between 0 and 1 to improve model training efficiency. Prediction model construction: use RNN architecture to build the prediction model. The input layer receives all the preprocessed historical data mentioned above. The hidden layer is designed as 4 layers of LSTM units. Each layer is followed by a Dropout layer to prevent overfitting. The output layer predicts the indoor environment parameters, outdoor environment parameters, soil environment parameters and working parameters of energy supply equipment in the next 24 hours.

[0142] Prediction model training and optimization: 5-fold cross validation was used to evaluate model performance to ensure the generalization ability of the model on different data subsets. The grid search method was used to adjust hyperparameters such as the number of LSTM layers, number of units, learning rate, batch size, etc. to maximize prediction accuracy. The mean square error (MSE) was selected as the loss function, and Adam was used as the optimizer.

[0143] Prediction model verification and deployment: Test the model performance on the validation set to ensure that the prediction error is within an acceptable range, deploy the trained model to the control system of the ground-source heat pump energy system, receive new data in real time, and output prediction results.

[0144] In another technical solution, the decomposition-based multi-objective evolutionary algorithm decomposes the three objectives of minimizing total power consumption, minimizing heat pump operating frequency, and minimizing circulating water pump power into multiple sub-problems, searches for the optimal solution set through a parallel search strategy, and selects the optimal solution from the solution set according to actual needs.

[0145] One embodiment includes:

[0146] T1. Problem definition:

[0147] 1. Determine the decision variables:

[0148] x1: indoor environmental parameters; x2: outdoor environmental parameters; x3: soil environmental parameters; y: energy supply equipment working parameters;

[0149] 2. Determine the objective function:

[0150] Minimizing the total power consumption E can be expressed as a function of decision variables: E = f1(x1, x2, x3, y);

[0151] The heat pump operating frequency V is minimized and can be expressed as a function of the decision variables: V = f2 (x1, x2, x3, y);

[0152] The circulating water pump power W is minimized and can be expressed as a function of decision variables: W = f3 (x1, x2, x3, y);

[0153] 3. Set constraints:

[0154] The soil temperature and water temperature in the outdoor environmental parameter x2 and the soil environmental parameter x3 remain unchanged, and are equal to the predicted value output by the RNN prediction model or the collected value;

[0155] The indoor environmental parameter x1 is set to a specific range value, such as a temperature of 20°C to 26°C, a humidity of 40% to 60%, or is equal to the predicted value output by the RNN prediction model or the collected value;

[0156] The operating parameters of the energy supply equipment are set within the physical limitations of the equipment itself;

[0157] 4. Establish mathematical model:

[0158] The relationship between the total power E, the heat pump operating frequency V, the circulating water pump power W and the decision variables can be expressed as a nonlinear function: E = a1 x1 2 +b1·x22 +c1 x3 2 +d1·y 2 +e1·x1·x2+f1·x1·x3+g1·x1·y+h1·x2·x3+i1·x2·y+j1·x3·y+k1·x1·x2·x3+l1·x1·x2·y+m1·x2·x3·y+n1·x1·x2·x3·y;

[0159] V=a2·x1 2 +b2 x2 2 +c2 x3 2 +d2·y 2 +e2·x1·x2+f2·x1·x3+g2·x1·y+h2·x2·x3+i2·x2·y+j2·x3·y+k2·x1·x2·x3+l2·x1·x2·y+m2·x2·x3·y+n2·x1·x2·x3·y;

[0160] W=a3·x1 2 +b3·x2 2 +c3 x3 2 +d3·y 2 +e3·x1·x2+f3·x1·x3+g3·x1·y+h3·x2·x3+i3·x2·y+j3·x3·y+k3·x1·x2·x3+l3·x1·x2·y+m3·x2·x3·y+n3·x1·x2·x3·y;

[0161] Among them, a1, b1, c1.....n1, a2, b2 , c2, ....n2, a3, b3, c3, ....n3 are the coefficients to be determined, which can be obtained by fitting historical data;

[0162] T2. Algorithm settings:

[0163] 1. Based on the MOEA / D algorithm framework, the nonlinear programming algorithm (particle swarm algorithm) is used to solve the mathematical model and obtain the optimal values ​​of x1, x2, x3, and y;

[0164] 2. Set the fitness function: used to evaluate the quality of each solution, specifically:

[0165] Fitness function of total power consumption E:

[0166] F1(x,y)=E, where x=[x1,x2,x3] represents environmental parameters, and y represents the working parameters of the energy supply equipment. This fitness function directly reflects the total power consumption of the system. The smaller the value, the better the solution.

[0167] Fitness function of heat pump operating frequency V:

[0168] F2(x,y)=V, this fitness function evaluates the operating frequency of the heat pump. Similarly, the smaller the value, the better the solution;

[0169] Fitness function of circulating water pump power W:

[0170] F3(x,y)=W, this fitness function reflects the power consumption of the circulating water pump. The smaller the value, the better the solution.

[0171] In addition, in MOEA / D, since the algorithm is based on decomposition, it decomposes the multi-objective optimization problem into a series of single-objective sub-problems. Each sub-problem will have a weight vector associated with it, which is used to weigh between multiple objectives. The algorithm then solves these sub-problems independently and evaluates the pros and cons of the solutions by combining multiple objectives into a single objective value through aggregation functions. However, in this application, there are already three clear objectives, and each objective has a clear fitness function, which can be used directly to evaluate the pros and cons of the solution without the need for additional aggregation functions. Within the framework of MOEA / D, this is achieved by assigning a weight vector to each sub-problem, which reflects the priority or importance of the objective.

[0172] 3. Select genetic operations (crossover or mutation) to generate new solutions, introduce parallel search strategies, and accelerate algorithm convergence;

[0173] T3. Data preparation:

[0174] Collect historical data, including energy supply equipment operating parameters and environmental data;

[0175] Use data preprocessing techniques, such as data cleaning and normalization, to ensure data quality;

[0176] The RNN prediction model of this application is used to predict future data.

[0177] T4. Algorithm execution:

[0178] Initialize the population, generate a set of initial solutions, iteratively perform genetic operations, generate new solutions and evaluate their fitness, update the solution set, retain the optimal solution, and monitor algorithm performance.

[0179] T5, solution set evaluation and selection: In the solution set, the optimal solution is selected according to the preset priority and speed conditions;

[0180] In the above technical solution, by implementing a decomposition-based multi-objective evolutionary algorithm, the intelligent management and control method can achieve more efficient energy utilization, reduce total power consumption, heat pump operating frequency and circulating water pump power, and improve the comfort of the indoor environment.

[0181] Based on the same inventive concept, the present invention provides an intelligent management and control system for a heat pump energy system, the heat pump energy system is used to execute the intelligent management and control method, and the heat pump energy system includes:

[0182] A data acquisition module is used to collect indoor and outdoor environmental parameters of the target building, working parameters of the energy supply equipment, soil environmental parameters underground of the target building and the total energy consumption of the heat pump energy system; optionally, the indoor environmental parameters are derived from various monitoring probes indoors of the target building, such as temperature and humidity probes, and carbon dioxide concentration probes; the outdoor environmental parameters are derived from various monitoring probes outdoors of the target building, such as temperature and humidity probes, sunshine intensity probes and anemometers; optionally, the working parameters of the energy supply equipment are derived from their respective monitoring instruments, such as ammeters, voltmeters, flow meters, thermometers, etc.; it is also necessary to simultaneously record the specific time of data collection, such as date and time (hours, minutes, etc.); preferably, the data acquisition module includes a sensor network, which is respectively arranged indoors, outdoors, in circulating water pumps, water pipes, indoor fan coils and soil of the target building, for real-time acquisition of various environmental parameters, soil environmental parameters, and working parameters of the energy supply equipment.

[0183] A prediction module, which is used to use a recurrent neural network to predict the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building within a period of time in the future based on the data collected by the data collection module;

[0184] The optimization decision module is used to optimize the working parameters of the heat pump energy system based on the prediction results of the prediction module by using a multi-objective evolutionary algorithm based on decomposition, so as to minimize the total energy consumption, minimize the working frequency of the heat pump, and minimize the power of the circulating water pump, and output the working parameters of the heat pump energy system to perform energy management and control on the heat pump energy system. The output working parameters of the heat pump energy system are preferably the state parameters of the opening, closing and standby of the heat pump, circulating water pump and indoor fan coil, and the setting parameters of the heat pump, circulating water pump and indoor fan coil, such as target temperature, flow, pressure, and wind speed.

[0185] In the above technical solution, by adopting the recurrent neural network (RNN) as the core algorithm of the prediction module, the characteristics of time series data can be fully utilized to effectively capture the changing trends and periodic laws of environmental parameters (such as indoor temperature, outdoor temperature, humidity, soil temperature, etc.) over time; the cyclic structure of RNN enables it to process input sequences of any length, so that it can make high-precision predictions of environmental parameters in the future based on historical data, providing a reliable basis for optimizing the decision-making module;

[0186] The optimization decision module adopts a multi-objective evolutionary algorithm based on decomposition, which can simultaneously consider multiple optimization objectives (such as minimizing total energy consumption, minimizing heat pump operating frequency, and minimizing circulating water pump power), so as to achieve the optimal configuration of energy use while ensuring system performance; the algorithm searches for the optimal solution in the solution space by simulating natural selection and genetic mechanisms, and can handle complex nonlinear problems and constraints, thereby improving the overall energy efficiency of the heat pump energy system;

[0187] By combining the prediction module with the optimization decision module, this technical solution realizes the intelligent energy management and control of the heat pump energy system; this intelligent energy management and control system can automatically adjust the working parameters of the heat pump energy system according to the real-time collected data and prediction results to adapt to the changing environmental conditions and energy needs, which not only improves the energy efficiency and reliability of the heat pump energy system, but also reduces the cost and risk of manual intervention, providing strong support for the sustainable operation of the building;

[0188] Based on this technical solution, it is possible to achieve precise optimization of the operating parameters of the heat pump energy system, thereby significantly reducing the total energy consumption of the system, the operating frequency of the heat pump and the power of the circulating water pump, thereby achieving the goal of energy conservation and emission reduction, which is of great significance for alleviating the energy crisis, reducing greenhouse gas emissions, and promoting green buildings and sustainable development.

[0189] After testing, the prediction module of this intelligent management and control system has a prediction accuracy of up to 90.84%. The combination of the prediction module and the optimization decision-making module reduces the operating cost per square meter and saves energy by up to 41% compared with pure manual control.

[0190] In another technical solution, the intelligent management and control system of the heat pump energy system also includes a user interface module, which is used to display prediction results, optimized working parameters and current system status, and allows users to adjust or set priorities for optimization goals as needed.

[0191] In the above technical solution, the user interface module provides an intuitive and easy-to-use interface, allowing users to clearly view the prediction results, optimized working parameters and the current operating status of each system; users can adjust the optimization goals or set the priorities of different goals through the interface according to actual needs to achieve more personalized energy management and control strategies.

[0192] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0193] The present invention provides an electronic device, comprising:

[0194] one or more processors;

[0195] and a memory in communication with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the intelligent management and control method. The electronic device may be any terminal device including a mobile phone, a laptop computer, a desktop computer, a tablet computer, a car computer, etc.

[0196] The present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the intelligent management and control method is implemented.

[0197] Through the description of the above implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0198] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. An intelligent control method for a heat pump energy system, characterized in that: The following steps are involved: S1. Collect indoor and outdoor environmental parameters of the target building; S2. Collecting the working parameters of energy supply equipment; S3. Collect soil environmental parameters under the target building; S4, collecting the total energy consumption of the heat pump energy system; S5, using the RNN prediction model, based on the data collected in steps S1 to S4, predict the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building in the future; S6. Adopting a decomposition-based multi-objective evolutionary algorithm, based on the prediction result of step S5, the operating parameters of the heat pump energy system are optimized to minimize the total energy consumption, minimize the operating frequency of the heat pump, and minimize the power of the circulating water pump. The operating parameters of the heat pump energy system are output to perform energy management and control on the heat pump energy system.

2. The intelligent control method of the heat pump energy system according to claim 1, characterized in that: The indoor environmental parameters include indoor temperature, humidity, carbon dioxide concentration and indoor area, and the outdoor environmental parameters include outdoor temperature, humidity, wind speed and sunshine intensity; The energy supply equipment includes a heat pump host, a circulating water pump, a ground source heat pump buried pipe heat exchanger and an energy storage device. The operating parameters include the inlet and outlet water temperature, compressor frequency and power of the heat pump host, the flow rate, head and power of the circulating water pump, the inlet and outlet water temperature and flow rate of the ground source heat pump buried pipe heat exchanger, the storage capacity and charging and discharging power of the energy storage device; The soil environmental parameters include soil temperature, soil moisture, buried pipe water temperature, and soil thermal conductivity; When training the RNN prediction model, regularization terms L1 and L2 are added to the loss function, where regularization term L1 is constructed based on the sum of the absolute values ​​of the weights corresponding to all the above parameters, and regularization term L2 is constructed based on the sum of the squares of the weights corresponding to the above parameters. At the same time, the Dropout technology is used in the hidden layer of the RNN model. During the training process, the outputs of some neurons in the hidden layer are randomly set to 0 with a set dropout probability, and the dropout probability ranges from 0 to 1.

3. The intelligent control method of the heat pump energy system according to claim 2, characterized in that: Processing of all the above parameters also includes: Collect the heat pump system operation data of all the above parameters, pre-process the data for accuracy and completeness, and clarify the energy efficiency target values: energy consumption and energy efficiency ratio; Using Python's NumPy and pandas libraries, the Pearson correlation coefficient between the preprocessed data and the energy efficiency target value is calculated using the numpy.corrcoef function, and the Spearman rank correlation coefficient is calculated using the scipy.stats.spearmanr function to evaluate the linear and nonlinear correlation between the processed data and the energy efficiency target value; Screen out processed data with high linear or nonlinear correlation with energy efficiency target values: set processed data with an absolute value of Pearson correlation coefficient greater than 0.4-0.5 and an absolute value of Spearman rank correlation coefficient greater than 0.3-0.4 as key focus features to form a feature subset; Use Python's scikit-learn library to build a random forest regression model, and divide the above feature subsets into training and test sets at a ratio of 80:

20. Use the RandomForestRegressor class to initialize the model, set n_estimators to 100-120 and max_depth to 10-12, and use the fit method to train the model on the training set. After training, use the feature_importances_ attribute to obtain the importance score of each feature in the random forest model. The linear and nonlinear correlation results of each feature's score and the above energy efficiency target value are mutually verified to further determine the importance of the feature to energy efficiency prediction.

4. The intelligent control method of the heat pump energy system according to claim 3, characterized in that: Further features to determine the importance of energy efficiency prediction include: Use the xgboost library to build the XGBoost model. First, convert the above feature subset into DMatrix format, then initialize the XGBRegressor model, set learning_rate to 0.1-0.2, n_estimators to 150-180, train the model through the fit method, and after the training, use the get_score method to obtain the split gain of each feature. Compare the linear and nonlinear correlation results of split gain with the above energy efficiency target values, and give higher attention weights to features with high linear and nonlinear correlation and large split gain in the XGBoost model; Among them, the results of the random forest regression model and the XGBoost model are combined to calculate the comprehensive weight: Let the importance score of the feature in the random forest model be S rf , the split gain of the feature in the XGBoost model is G xgb , and the weight ratio of the random forest model is α: 0<α<1, then the calculation formula of the comprehensive weight W is: Among them, n is the total number of features and i is the current feature.

5. The intelligent control method of the heat pump energy system according to claim 4, characterized in that: It also includes steps to set the regularization term L1 and regularization term L2 coefficients: Set a weight threshold to filter out features with weights greater than the threshold to form the final feature subset The final feature subset is divided into a training set and a validation set at a ratio of 80:

20. Based on experience, the regularization coefficient λ1 of the regularization term L1 and the regularization coefficient λ2 of the regularization term L2 are preliminarily determined to be in the range of [10 -4 ,1]; Discrete candidate values ​​are selected within the value range to form a value grid of λ1 and λ2: where both λ1 and λ2 are For each combination of λ1 and λ2, the RNN prediction model is trained, and the combination of λ1 and λ2 that makes the model perform best on the validation set is selected as the final regularization coefficient.

6. The intelligent control method of the heat pump energy system according to claim 5, characterized in that: Training the RNN prediction model includes: Initialize the parameters of the RNN prediction model according to the weights The mean square error loss function MSE is used, and the loss function including the regularization term is constructed according to the following formula: where w i represents the weight corresponding to the i-th feature, and n is the total number of features; The stochastic gradient descent algorithm is used to adjust the weight parameters of the model according to the gradient of the loss function, and multiple rounds of training are performed until the preset number of training rounds is reached; Among them, the regularization term L1 is The regularization term L2 is The mean square error loss function MSE is used to measure the difference between the model prediction value and the true value. where y j is the true value of the jth sample, is the predicted value of the jth sample, and m is the number of samples.

7. The intelligent control method of the heat pump energy system according to claim 6, characterized in that: The steps of selecting the combination of λ1 and λ2 that makes the model perform best on the validation set as the final regularization coefficient include: Select the λ1 and λ2 values ​​that minimize the mean square error between the predicted value and the true value; or select the λ1 and λ2 values ​​that have the highest accuracy in energy efficiency prediction.

8. Intelligent management and control system of heat pump energy system, characterized in that: The intelligent management and control system is used to execute the intelligent management and control method according to any one of claims 1 to 6, and the intelligent management and control system includes: A data acquisition module, which is used to collect indoor and outdoor environmental parameters of the target building, working parameters of energy supply equipment, soil environmental parameters underground of the target building, and total energy consumption of the heat pump energy system; A prediction module, which is used to use a recurrent neural network to predict the indoor and outdoor environmental parameters of the target building and the soil environmental parameters under the target building within a period of time in the future based on the data collected by the data collection module; The optimization decision module is used to adopt a decomposition-based multi-objective evolutionary algorithm to optimize the working parameters of the heat pump energy system based on the prediction results of the prediction module, so as to minimize the total energy consumption, minimize the operating frequency of the heat pump, and minimize the power of the circulating water pump, and output the working parameters of the heat pump energy system to perform energy management and control of the heat pump energy system.

9. The intelligent management and control system of the heat pump energy system according to claim 7, characterized in that: The data acquisition module includes a sensor network, which is arranged indoors and outdoors, in circulating water pumps, water pipes, indoor fan coils and soil of the target building, and is used to obtain various environmental parameters, soil environmental parameters, and working parameters of energy supply equipment in real time.

10. The intelligent management and control system of the heat pump energy system according to claim 7, characterized in that: The intelligent management and control system also includes a user interface module, which is used to display the prediction results, optimized working parameters and current system status, and allows the user to adjust the optimization goals or set priorities as needed.

Images