A ground source heat pump air conditioner intelligent energy-saving control method and system

Abstract

The application provides a ground source heat pump air conditioner intelligent energy-saving control method and system, belongs to the technical field of building energy saving and intelligent control, collects historical load data and environmental parameters of building operation, and constructs a time sequence feature set; an initial quantum population is generated based on Tent chaotic mapping, population observation and updating are carried out in combination with an improved quantum genetic algorithm, and differential variation and crossover operation of individual fitness are carried out by using an improved self-adaptive differential evolution algorithm, so that the hyperparameters of the IADE-IQGA-BiLSTM model are optimized, multiple rounds of iterative training are carried out by using a training set, and an early stopping strategy is implemented by using a verification set; the trained model is used for load prediction of a test set, the prediction result is inversely normalized into an actual load value, and a performance evaluation index is calculated; the IADE-IQGA-BiLSTM model is packaged as a RESTful API interface and deployed on an intelligent energy-saving control software to receive real-time data flow and regularly generate future load prediction; based on the load prediction result, the water pump frequency and the heat exchanger shape of the ground source heat pump system are dynamically adjusted.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation and intelligent control technology, specifically to a smart energy-saving control method and system for ground source heat pump air conditioning. Background Technology

[0002] Ground source heat pump (GSHP) systems are widely used in building heating and cooling load regulation due to their high efficiency, energy saving, and environmental friendliness. However, the following challenges exist in actual operation: (1) drastic load fluctuations lead to a decrease in system energy efficiency (COP); (2) energy dispatch is lagging and it is difficult to adapt to environmental and load changes in real time; (3) traditional prediction methods (such as multiple regression and BP neural networks) have insufficient accuracy and limited generalization ability in complex time series prediction. In recent years, deep learning models (such as LSTM) have performed well in time series prediction, but parameter optimization, model stability, and real-time performance still need to be improved. To this end, this invention proposes a BiLSTM model that combines IADE and IQGA optimization, which is integrated into intelligent energy-saving control software and combined with the hardware design of the ground source heat pump system to achieve integrated prediction and control operation, thereby improving system energy efficiency and operational stability. Summary of the Invention

[0003] To address the above problems, this invention proposes a smart energy-saving control method for ground source heat pump air conditioning, comprising the following steps: Step S1: Collect historical load data and environmental parameters of building operation, remove outliers and align time series data, and construct a time series feature set; Step S2: Divide the time series feature set into a training set, a validation set, and a test set, and use the min-max normalization method to standardize the input and output data; Step S3: Generate an initial quantum population based on the Tent chaotic mapping, observe and update the population using an improved quantum genetic algorithm, and perform differential mutation and crossover operations on individual fitness using an improved adaptive differential evolution algorithm to optimize the hyperparameters of the IADE-IQGA-BiLSTM model and obtain the optimal hyperparameter configuration. Step S4: Construct the IADE-IQGA-BiLSTM model based on the optimal hyperparameter configuration, perform multiple rounds of iterative training using the training set, and implement an early stopping strategy using the validation set to prevent overfitting; Step S5: Use the trained IADE-IQGA-BiLSTM model to predict the load on the test set, and inversely normalize the prediction results to the actual load values ​​to calculate the performance evaluation index. Step S6: Encapsulate the IADE-IQGA-BiLSTM model into a RESTful API interface and deploy it in the smart energy-saving control software to receive real-time data streams and periodically generate future load forecasts; Step S7: Based on the load forecast results, dynamically adjust the pump frequency and heat exchanger status of the ground source heat pump system.

[0004] Furthermore, step S3, which generates the initial quantum population based on the Tent chaotic map, includes: generating a chaotic sequence through the Tent chaotic map and linearly mapping the chaotic sequence to the value space of each hyperparameter to achieve population initialization.

[0005] Furthermore, the encoding of BiLSTM hyperparameters in step S3 includes: taking the learning rate, batch size, window length, number of hidden layer neurons, and number of BiLSTM layers as hyperparameters to be optimized, and combining them into an individual vector.

[0006] Furthermore, the improved quantum genetic algorithm in step S3 includes: using qubits to represent individual genes, obtaining deterministic observation values ​​through observation probability amplitude, and updating the qubits using a quantum rotation gate, wherein the rotation angle is dynamically adjusted according to the fitness relationship between the current observation value and the optimal solution.

[0007] Furthermore, the improved adaptive differential evolution algorithm in step S3 includes: performing differential mutation operation on individuals, randomly selecting three distinct individual vectors from the current population to generate a mutation vector; and introducing a mandatory crossover bit when performing the crossover operation to ensure that at least one bit is inherited from the mutated solution.

[0008] Furthermore, the BiLSTM model constructed in step S4 adopts a bidirectional structure, including forward LSTM and backward LSTM, which respectively extract past and future information of the time series, and concatenate the hidden states in the two directions to obtain a bidirectional hidden vector, and generate the prediction result through linear transformation of the output layer.

[0009] Furthermore, in step S4, the early stopping strategy includes: monitoring the validation set loss; when the validation set loss does not decrease for several consecutive rounds and the decrease is less than a preset threshold, immediately stopping the training and rolling back the model parameters to the optimal weights when the validation set loss is the lowest.

[0010] Furthermore, the dynamic adjustment of the pump frequency in step S7 includes: based on the deviation between the predicted load and the reference load, combined with the outdoor temperature change rate, calculating the adjustment amount of the pump frequency through the adjustment coefficient, and generating control commands.

[0011] Furthermore, in step S7, the heat exchanger is turned on when the predicted load exceeds the preset load threshold, and turned off when the predicted load is lower than the load threshold, thus realizing on-demand start-up and shutdown.

[0012] This invention also proposes a smart energy-saving control system for ground source heat pump air conditioning, used to implement the above-mentioned smart energy-saving control method for ground source heat pump air conditioning, characterized in that it includes: The data management module is used to receive sensor data and perform data cleaning and feature extraction. The core prediction module is used to run the IADE-IQGA-BiLSTM model and output predicted loads. .

[0013] The control optimization module, based on the prediction results, calls the linear programming solver to optimize the control parameters. The communication module uses the Modbus TCP protocol to communicate with the PLC and sensors. The visualization module, based on a web interface, displays real-time load curves, energy consumption statistics, and prediction errors.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The prediction accuracy is significantly improved. The BiLSTM hyperparameters are optimized by the IADE-IQGA algorithm and the population diversity is enhanced by the Tent chaotic mapping. The RMSE is reduced by 25% compared with the traditional LSTM and the MAPE is consistently below 4%. It accurately captures the load fluctuation pattern and provides a reliable basis for control decision-making.

[0015] With outstanding energy-saving effect, the pump frequency and heat exchanger status are dynamically adjusted based on load forecast, reducing energy consumption by 15-20% and improving system COP by 0.8-1.2, taking into account both energy minimization and indoor comfort, and significantly reducing operating costs.

[0016] The software adopts a modular design and Docker containerized deployment, supports edge computing, and controls the PLC to issue control commands. The response time is less than 200ms, which meets the requirements of real-time control and is compatible with different hardware environments.

[0017] The system is highly adaptable. Through training with multi-dimensional environmental parameters and load data, it can flexibly adapt to various types of buildings such as office buildings, residential buildings, and commercial buildings, and can cope with load changes in different climate regions.

[0018] It boasts excellent scalability, with its core algorithms and control architecture being universally applicable. It can be seamlessly extended to air source heat pump systems and is also suitable for heating and cooling supply scenarios in industrial production, making it widely applicable. Attached Figure Description

[0019] Figure 1 This is a structural diagram of a ground source heat pump air conditioning system; Figure 2 This is a diagram of the intelligent energy-saving control software architecture. Figure 3This is a curve comparing the predicted load with the actual load. Detailed Implementation

[0020] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0021] Example 1 The intelligent energy-saving control method for ground source heat pump air conditioning specifically includes the following steps: Step 1: Data acquisition and preprocessing.

[0022] Historical load data and environmental parameters related to building operation were collected, including outdoor temperature, humidity, historical load, and water pump frequency. Outliers were removed, and a unified time series feature set was constructed. The data is shown in Table 1. Table 1 Step 2: Data partitioning and normalization.

[0023] The dataset is divided into training, validation and test sets, and the input and output data are standardized using the min-max normalization method.

[0024] Step 3: Parameter optimization based on chaotic mapping and improved intelligent algorithms.

[0025] The initial quantum population is generated using the Tent chaotic mapping method, and observation and population update are performed using the IQGA algorithm. The IADE algorithm is used to perform differential mutation and crossover operations on individual fitness, and finally the optimal BiLSTM hyperparameter configuration is obtained.

[0026] 1) Tent chaotic mapping (used for population initialization and parameter perturbation) ; in, , These are the state values ​​of the system at steps n and n+1, respectively; These are boundary parameters used to control the shape of the mapping. This is the initial state.

[0027] Preferably, ,Will Linear mapping to the value space of each hyperparameter The mapped value p is: .

[0028] 2) BiLSTM Hyperparameter Encoding ; ; N hid n is the number of neurons in the hidden layer. layer The number of BiLSTM layers; the learning rate Ir∈

[10] -5 10 -2 ], window L∈[24,288], batch sizebatch∈{32,64,128}.

[0029] 3) Quantum genetic algorithm Quantum bit representation: the j-th gene of an individual ; in, It is the quantum state of the j-th qubit. and These are the two ground states in quantum computing. and These are two complex numbers, called probability amplitudes.

[0030] Observations: ; Quantum Revolving Door Update: ; Among them, quantum rotation angle , Step size, The lookup strategy is based on the current observation x. j and the optimal solution The combination of these factors determines the rules for adjusting the rotation angle. The current optimal solution x * The fitness value.

[0031] Elite Preservation and Tent Perturbation: Optimal Solution With probability Inject, and then perturb other individuals with Tent noise.

[0032] 4) Adaptive Differential Evolution Algorithm Mutations: It is the mutation vector generated for the i-th individual. , , F is the vector of three distinct individuals (solutions) randomly selected from the current population, where F is the mutation factor.

[0033] Cross operation: Where ui,j is the intermediate solution after crossover of the i-th individual at position j; vi,j is the candidate solution after mutation of the i-th individual at position j; xi,j is the original solution of the i-th individual at position j; rand() is a random number between 0 and 1; CR is the crossover probability, which controls the probability of the crossover operation; jrand is a randomly selected mandatory crossover position, ensuring that at least one position is inherited from the mutated solution, thus avoiding the intermediate solution being completely identical to the original solution.

[0034] Adaptive parameter F: ; F obeys Let be a log-normal distribution with logarithmic mean and 0.1 as logarithmic standard deviation. This is denoted as . ;CR obeys Let f(x) be a normal distribution with mean 0.1 and standard deviation 0.1, denoted as f(x). .

[0035] After a successful iteration, use the set of successful individuals. renew: ; Wherein, SF and SCR are the F and CR sets corresponding to successful individuals in the current iteration. Successful individuals refer to individuals with better fitness after crossover / mutation; c is the update coefficient, which controls the weight of historical parameters and new successful parameters; max(SF) and max(SCR) are the maximum values ​​of F and CR in the successful set.

[0036] The above formula is updated using parameters from successful individuals. , This allows the parameters in subsequent iterations to be adaptively adjusted towards a better direction.

[0037] Step 4: BiLSTM model construction and training.

[0038] A BiLSTM model is constructed based on the optimized parameters, including the number of layers, the number of neurons, the learning rate, and the Dropout ratio. The training set is used for multiple rounds of iterative training, and the validation set is used for early stopping strategy and to prevent overfitting.

[0039] 1) BiLSTM recursion: LSTM unit: In the formula: Forget gate parameters; For the Sigmoid function; Weighting for the forget gate; for The hidden layer outputs a signal at any time; For input signals; Let be the bias matrix of the forget gate. Input gate parameters; For input gate weights; Cell state weights; This is the bias matrix of the input gate; This is the bias matrix for the memory cells; These are the output gate parameters; Cell state weights; Let be the bias matrix of the output gate; for The hidden layer outputs a signal at any time.

[0040] Two-way hidden layer: ; Let be the bidirectional hidden state at time t; This represents the hidden state of the forward LSTM at time t. The hidden state of the inverse LSTM at time t is represented by the bidirectional hidden layer, which concatenates the forward and inverse hidden states to obtain a bidirectional hidden vector containing information about the past and future.

[0041] Output layer: ; is the predicted value for the next H steps, where H=1 indicates the prediction of the next time step; Wy is the weight matrix of the output layer; The bias vector of the output layer is used; the output layer undergoes a linear transformation using the bidirectional hidden states to obtain the final prediction result.

[0042] 2) Objective function and early stopping Loss function: .

[0043] The loss function L measures the error between the predicted value and the true value, and N is the number of samples. This is the predicted value for the nth sample; is the true value of the nth sample; MSE is the mean squared error.

[0044] Early stopping: if the validation set continuous The wheel did not descend, and If so, then stop and roll back to the optimal weight.

[0045] Step 5: Load forecasting and inverse normalization.

[0046] The trained model is used to predict test data. The prediction results are then inversely normalized to restore the actual load values. Performance is evaluated by calculating metrics such as RMSE and MAE.

[0047] Step 6: Encapsulate the IADE-IQGA-BiLSTM model into a RESTful API interface, deploy it on energy-saving software, receive real-time data streams (JSON format), and generate load forecasts for the next hour every 15 minutes. The software architecture adopts a modular design, supports Docker containerized deployment, and runs on edge computing devices (such as NVIDIA Jetson) or industrial computers.

[0048] Step 7: Smart energy-saving control based on prediction results.

[0049] Dynamically adjust the water pump frequency and heat exchanger status: Among them, f pump,t The pump's operating frequency at time t determines its output; f base : The water pump's base frequency, the default operating frequency; Q is an adjustment coefficient used to control the degree of influence of load deviation and temperature change on the frequency. pred,t : Forecast load at time t; Q ref Reference load; This represents the rate of change of outdoor temperature.

[0050] The above formula dynamically adjusts the pump frequency based on the deviation between the predicted load and the reference load, combined with changes in outdoor temperature, to match actual needs.

[0051] ; She,t represents the heat exchanger state at time t, where 1 = on and 0 = off; Qthres represents the load threshold, a pre-set start / stop determination boundary.

[0052] The above formula is used to activate the heat exchanger when the predicted load exceeds a threshold, and shut it down otherwise, achieving on-demand start-up and shutdown. Control commands are issued via PLC, with a response time of less than 200ms.

[0053] A method for controlling pump frequency and heat exchanger opening in a ground source heat pump system based on DDPG is adopted: Overview: This paper describes a Deep Deterministic Policy Gradient (DDPG) algorithm that learns a policy from system observations to control commands in a continuous action space. The policy directly outputs the control quantity (pump frequency). f With heat exchanger opening α The goal is to minimize energy consumption, maintain indoor comfort, and minimize equipment wear. Training is first completed offline in a high-fidelity simulation environment, followed by online deployment and fine-tuning on-site using controlled safety filters and MPC backup.

[0054] Example 2 like Figure 2The diagram shown is a software architecture diagram for intelligent energy-saving control, which includes the following software functional modules: 1. Data Management Module: Receives sensor data, collects outdoor temperature, humidity, water pump frequency, etc., stores it in an SQLite database, and supports data cleaning and feature extraction.

[0055] 2. Prediction Core Module: Runs the IADE-IQGA-BiLSTM model and outputs the predicted load. .

[0056] 3. Control Optimization Module: Based on the prediction results, the linear programming solver is invoked to optimize the control parameters. Constraints: That is, the predicted load at time t The indoor temperature at time t shall not exceed the system's maximum capacity Qcap, ensuring the system meets demand; It needs to be maintained within the set upper and lower limits. Inside.

[0057] Where E is energy consumption, and c1 and c2 are energy consumption coefficients.

[0058] 4. Communication module: Supports Modbus TCP protocol for communication with sensors.

[0059] 5. Visualization module: Based on a web interface (HTML5 + JavaScript), it displays real-time load curves, energy consumption statistics, and prediction errors.

[0060] The software adopts a microservice architecture, and its core functions include: Data stream processing: The Kafka stream processing platform receives real-time data and stores it in the InfluxDB time-series database.

[0061] Model deployment: Convert the trained model into TensorFlowLite, which runs on edge devices, reducing latency.

[0062] Fault tolerance mechanism: Supports data loss retransmission and online model updates.

[0063] HMI Interface: Based on the React framework, it displays load forecast curves, energy consumption trends, and control status, and supports user-defined energy-saving strategies.

[0064] The software supports cross-platform deployment and is compatible with Linux and Windows systems. The minimum hardware requirements are 4GB RAM and a 4-core CPU.

[0065] like Figure 1As shown, the ground source heat pump air conditioning system consists of the following subsystems: 1. Underground heat exchange system: Utilizing vertical U-shaped buried pipes at a depth of 100-150m, with a heat exchange efficiency of 4.5-5.0W / m, and a soil thermal conductivity of [missing information]. .

[0066] 2. Heat pump system: Rated COP 4.2, equipped with variable frequency compressor, power range 10-500kW, suitable for different building sizes.

[0067] 3. Circulating water system: Variable frequency water pump (flow rate 10-100%), equipped with electromagnetic flow meter and pressure sensor to ensure fluid stability.

[0068] 4. Indoor terminal system: fan coil unit (air volume 500-2000m³ / h) or floor radiant heating (heat flux density 80-120W / m²).

[0069] 5. Control and monitoring system: integrates high-precision sensors (temperature ±0.1℃, flow rate ±1%), PLC supports Modbus protocol, and host computer runs energy-saving software.

[0070] This system design combines intelligent forecasting and energy-saving control strategies, which can effectively improve the overall energy efficiency (COP), extend the service life of equipment, and enhance the system's adaptability to complex climate load changes, and has broad prospects for promotion and application.

[0071] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A smart energy-saving control method for ground source heat pump air conditioning, characterized in that, Includes the following steps: Step S1: Collect historical load data and environmental parameters of building operation, remove outliers and align time series data, and construct a time series feature set; Step S2: Divide the time series feature set into a training set, a validation set, and a test set, and use the min-max normalization method to standardize the input and output data; Step S3: Generate an initial quantum population based on the Tent chaotic mapping, observe and update the population using an improved quantum genetic algorithm, and perform differential mutation and crossover operations on individual fitness using an improved adaptive differential evolution algorithm to optimize the hyperparameters of the IADE-IQGA-BiLSTM model and obtain the optimal hyperparameter configuration. Step S4: Construct the IADE-IQGA-BiLSTM model based on the optimal hyperparameter configuration, perform multiple rounds of iterative training using the training set, and implement an early stopping strategy using the validation set to prevent overfitting; Step S5: Use the trained IADE-IQGA-BiLSTM model to predict the load on the test set, and inversely normalize the prediction results to the actual load values ​​to calculate the performance evaluation index. Step S6: Encapsulate the IADE-IQGA-BiLSTM model into a RESTful API interface and deploy it in the smart energy-saving control software to receive real-time data streams and periodically generate future load forecasts; Step S7: Based on the load forecast results, dynamically adjust the pump frequency and heat exchanger status of the ground source heat pump system.

2. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 1, characterized in that, Step S3, which generates the initial quantum population based on the Tent chaotic map, includes: generating a chaotic sequence through the Tent chaotic map and linearly mapping the chaotic sequence to the value space of each hyperparameter to achieve population initialization.

3. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 2, characterized in that, The encoding of BiLSTM hyperparameters in step S3 includes: taking the learning rate, batch size, window length, number of hidden layer neurons, and number of BiLSTM layers as hyperparameters to be optimized, and combining them into an individual vector.

4. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 3, characterized in that, The improved quantum genetic algorithm in step S3 includes: using qubits to represent individual genes, obtaining deterministic observation values ​​through observation probability amplitude, and updating the qubits using a quantum rotation gate, wherein the rotation angle is dynamically adjusted according to the fitness relationship between the current observation value and the optimal solution.

5. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 4, characterized in that, The improved adaptive differential evolution algorithm in step S3 includes: performing differential mutation operations on individuals, randomly selecting three distinct individual vectors from the current population to generate a mutation vector; and introducing a mandatory crossover bit during the crossover operation to ensure that at least one bit is inherited from the mutated solution.

6. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 1, characterized in that, The BiLSTM model constructed in step S4 adopts a bidirectional structure, including forward LSTM and backward LSTM, which extract past and future information of the time series respectively, and concatenate the hidden states in the two directions to obtain a bidirectional hidden vector. The prediction result is generated through linear transformation of the output layer.

7. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 6, characterized in that, In step S4, the early stopping strategy includes: monitoring the validation set loss; when the validation set loss does not decrease for several consecutive rounds and the decrease is less than a preset threshold, immediately stop training and roll back the model parameters to the optimal weights when the validation set loss is the lowest.

8. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 1, characterized in that, The dynamic adjustment of the pump frequency in step S7 includes: based on the deviation between the predicted load and the reference load, combined with the outdoor temperature change rate, calculating the adjustment amount of the pump frequency through the adjustment coefficient, and generating control commands.

9. The intelligent energy-saving control method for ground source heat pump air conditioning according to claim 8, characterized in that, In step S7, the heat exchanger is turned on when the predicted load exceeds the preset load threshold, and turned off when the predicted load is lower than the load threshold, thus realizing on-demand start-up and shutdown.

10. A smart energy-saving control system for ground source heat pump air conditioning, used to implement the smart energy-saving control method for ground source heat pump air conditioning as described in any one of claims 1-9, characterized in that, include: The data management module is used to receive sensor data and perform data cleaning and feature extraction. The core prediction module is used to run the IADE-IQGA-BiLSTM model and output predicted loads. . The control optimization module, based on the prediction results, calls the linear programming solver to optimize the control parameters. The communication module uses the Modbus TCP protocol to communicate with the PLC and sensors. The visualization module, based on a web interface, displays real-time load curves, energy consumption statistics, and prediction errors.

Images