Indoor unit energy-saving control method and system of multi-split air conditioning system

Abstract

The invention discloses an indoor unit energy-saving control method and system of a multi-split system, and relates to the technical field of air conditioner energy-saving control, and the method comprises the following steps: inputting environment situation sensing data into a time sequence attention model, extracting temperature change matrixes of different areas by a convolutional layer, extracting a time dependence matrix of indoor unit air volume by a bidirectional LSTM layer, and obtaining a time dependence matrix of indoor unit air volume; a multi-head self-attention mechanism is used for fusion, and a Gaussian process regression algorithm is synchronously used for generating an air volume prediction demand table; combining the air volume prediction demand table with the building three-dimensional point cloud to construct a dynamic air flow topological graph, performing iterative search on the dynamic air flow topological graph by using a PSO algorithm to obtain an air supply parameter combination, and synchronously mapping the air supply parameter combination into an execution instruction set through an equipment protocol mapping library; according to the method, by integrating the environment situation sensing data and carrying out online incremental updating on the time sequence attention model, the energy-saving control scheme can be dynamically adjusted and optimized according to the actual operation effect, and the energy efficiency performance of the air conditioning equipment is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of air-conditioning energy-saving control, and in particular to an energy-saving control method and system for indoor units of a multi-connected system. Background Art

[0002] As a crucial component of modern building air conditioning solutions, VRF (Variable Refrigerant Flow) systems are widely used in commercial and residential buildings due to their high efficiency and flexibility. With technological advancements, particularly in artificial intelligence, the Internet of Things, and big data analytics, control methods for VRF systems are also evolving.

[0003] However, existing energy-saving control methods for indoor units in VRF systems still have some shortcomings. For one thing, the control logic of existing VRF systems is still based on static setpoints or simple feedback adjustment mechanisms, failing to fully integrate multi-dimensional information such as indoor and outdoor environmental parameters, equipment status, and building structure. Furthermore, existing systems typically employ open-loop control, where control instructions are formulated based on the current state without feedback correction based on actual execution results. This lacks an online model update mechanism based on real-time power consumption data, making it unable to address issues such as equipment aging and sudden environmental changes that can occur during long-term operation, impacting the overall stability and energy efficiency of the system. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an energy-saving control method for internal units in a multi-connected system to solve the problems of weak environmental perception capability and lack of a real-time feedback update mechanism.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an energy-saving control method for indoor units in a multi-connected system, comprising: inputting environmental situation awareness data into a temporal attention model, extracting a temperature change matrix of different regions using a convolutional layer, extracting a time-dependent matrix of indoor unit air volume using a bidirectional LSTM layer, fusing the matrix using a multi-head self-attention mechanism, and simultaneously generating an air volume forecast demand table using a Gaussian process regression algorithm; Combine the air volume forecast demand table with the building's 3D point cloud to construct a dynamic airflow topology map. Use the PSO algorithm to iteratively search the dynamic airflow topology map to obtain air supply parameter combinations. Simultaneously, use the device protocol mapping library to map these air supply parameter combinations into execution instruction sets. The execution instruction set is sent to the device control center, and the variable frequency fan is driven by the PID controller to perform air flow regulation. Real-time device power consumption data is simultaneously collected and input into the time series attention model for incremental update to obtain the optimized time series attention model. According to the optimized temporal attention model, the NSGA-III algorithm is used to solve multiple objectives and generate the final energy-saving solution for the internal machine.

[0007] As a preferred solution of the energy-saving control method for the internal unit of the multi-split system described in the present invention, the environmental situation perception data includes indoor and outdoor environmental parameters, equipment operating status parameters, equipment power consumption data and building three-dimensional point cloud data.

[0008] As a preferred solution of the energy-saving control method for the indoor units of the multi-split system of the present invention, wherein: the air volume forecast demand table is generated using the Gaussian process regression algorithm, and the specific steps are as follows: The convolutional layer and the bidirectional LSTM layer are integrated through a layered stacking strategy to build a temporal attention model, and the environmental situation awareness data is input into the temporal attention model through the MQTT protocol interface; The convolution layer captures local spatiotemporal features through one-dimensional dilated convolution; the local spatiotemporal features are nonlinearly enhanced using residual connections, and the weights are assigned using Softmax normalization to obtain the temperature change matrix; The forward LSTM of the bidirectional LSTM layer performs forward time sequence iteration through a gating mechanism to generate a forward state sequence. The backward LSTM uses a reverse gating mechanism to perform reverse time sequence iteration to obtain a reverse state sequence. The forward state sequence and the reverse state sequence are fused using a weighted average method to generate a time dependency matrix. The multi-head self-attention mechanism is used to project the temperature change matrix and the time dependency matrix, and then spliced ​​through cross-modal channels to generate spatiotemporal fusion features; The Gaussian process regression algorithm is used to perform nonlinear mapping on the spatiotemporal fusion features to generate the wind volume forecast demand table.

[0009] As a preferred solution of the energy-saving control method for the internal unit of the multi-split system of the present invention, wherein: the construction of the dynamic airflow topology diagram specifically includes the following steps: The air volume forecast demand table is gridded and discretized using the spatial coding algorithm to obtain the air volume distribution matrix. The air volume distribution matrix and the building's three-dimensional point cloud are associated with nodes and edge weights are updated using the A* algorithm to generate a dynamic airflow topology map.

[0010] As a preferred solution of the energy-saving control method for indoor units of the multi-split system of the present invention, wherein: the air supply parameter combination is mapped into an execution instruction set through the device protocol mapping library, specifically including the following steps: The PSO algorithm is used to perform multi-dimensional parameter space decomposition on the dynamic airflow topology to obtain the global particle swarm state. Dynamic inertia weight adjustment and neighborhood mutation are used to iteratively search the global particle swarm state to obtain the air supply parameter combination. According to the device protocol mapping library, the air supply parameter combination is encoded into hexadecimal instruction using Modbus function code to obtain the original instruction frame; the original instruction frame is timestamp embedded using clock synchronization method to generate an execution instruction set.

[0011] As a preferred solution of the energy-saving control method for the internal unit of the multi-connected system of the present invention, wherein: the obtaining of the optimized temporal attention model specifically includes the following steps: The execution instruction set is sent to the device control center through the CoAP protocol, and the execution instruction set is prioritized using the priority scheduling algorithm to generate a time-series instruction queue; The PID controller converts the timing instruction queue into PWM duty pulses through proportional-integral-differential operations; the PWM duty pulses drive the variable frequency fan to adjust the air flow by adjusting the armature voltage; The device power consumption data is collected using a current sensing resistor and input into the timing attention model through the RS485 interface; The temporal attention model assigns weights to the device power consumption data through Softmax normalization to obtain a power consumption weight vector. Based on the power consumption weight vector, the GRU parameters are incrementally updated through the sliding window incremental learning method to output the optimized temporal attention model.

[0012] As a preferred solution of the energy-saving control method for the indoor unit of the multi-split system of the present invention, wherein: the generating of the final energy-saving plan for the indoor unit specifically includes the following steps: The optimized temporal attention model is embedded in causal convolution to perform dynamic feature focusing and generate a temporal energy distribution matrix. The NSGA-III algorithm performs multi-objective optimization on the time-series energy distribution matrix through hierarchical non-dominated sorting to obtain the Pareto solution set. The entropy weight method is used to normalize the weights of the Pareto solution set to obtain the optimal energy-saving parameters. 3D rendering is performed using the Unreal Engine engine to generate the final energy-saving solution for the internal unit.

[0013] In a second aspect, the present invention provides an energy-saving control system for an internal unit of a multi-split system, comprising an air volume prediction module, an instruction generation module, a model optimization module, and a solution generation module. The air volume prediction module is used to input environmental situation awareness data into the time series attention model. The convolution layer extracts the temperature change matrix of different regions, and the bidirectional LSTM layer extracts the time dependency matrix of the indoor air volume. The multi-head self-attention mechanism is used to fuse the data, and the Gaussian process regression algorithm is used to generate the air volume forecast demand table. The instruction generation module is used to combine the air volume forecast demand table with the building's three-dimensional point cloud to construct a dynamic airflow topology map. It then uses the PSO algorithm to iteratively search the dynamic airflow topology map to obtain air supply parameter combinations. The air supply parameter combinations are then mapped into an execution instruction set through the device protocol mapping library. The model optimization module is used to send the execution instruction set to the device control center, drive the variable frequency fan to perform air flow regulation through the PID controller, synchronously collect real-time device power consumption data, and input the time series attention model for incremental update to obtain the optimized time series attention model; The solution generation module is used to generate the final energy-saving solution for the internal machine by performing multi-objective solution solving through the NSGA-III algorithm based on the optimized temporal attention model.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the energy-saving control method for the internal unit of the multi-split system as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the energy-saving control method for the internal unit of the multi-split system as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of this invention include significantly enhanced perception and response speed to complex indoor environmental changes by integrating multi-dimensional information such as indoor and outdoor environmental parameters, equipment status, and three-dimensional building point clouds, ensuring more accurate and efficient temperature regulation. Furthermore, by performing online incremental updates to the temporal attention model based on real-time power consumption data, the energy-saving control scheme can be dynamically adjusted and optimized based on actual operating results, significantly improving the stability and energy efficiency of air conditioning equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1The present invention is a flow chart of an energy-saving control method for indoor units in a multi-split system.

[0019] Figure 2 This is a schematic diagram of the energy-saving control system for the indoor units of a multi-split system.

[0020] Figure 3 Flowchart of the process for obtaining a forecast demand table for air volume.

[0021] Figure 4 Flowchart of the dynamic airflow topology construction and optimization process. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an energy-saving control method for indoor units in a multi-split system, comprising the following steps: S1. Input the environmental situation awareness data into the temporal attention model. The convolution layer extracts the temperature change matrix of different areas. The bidirectional LSTM layer extracts the time dependency matrix of the internal air volume. The data is then fused using the multi-head self-attention mechanism. The Gaussian process regression algorithm is then used to generate the air volume forecast demand table.

[0026] The specific steps include: S1.1. Collect environmental situation awareness data, which includes indoor and outdoor environmental parameters, equipment operating status parameters, equipment power consumption data, and building three-dimensional point clouds. Indoor and outdoor environmental parameters are collected through multiple sensors deployed inside and outside the building: Indoors, temperature and humidity sensors are installed in key locations in the room, such as areas with frequent human activity and air conditioning return vents, to collect indoor temperature and humidity data; light sensors are installed near windows and on ceilings with natural light sources to collect light data; Outdoors, temperature and humidity sensors are installed on the exterior walls of the building to collect outdoor temperature and humidity data; wind speed sensors are installed on the roof to collect wind speed data; the collected temperature, humidity, light, and wind speed data are integrated through the MQTT communication protocol to generate indoor and outdoor environmental parameters; Equipment operating status parameters are collected through intelligent sensors and controllers integrated into the VRF system. The built-in pressure, temperature, and frequency sensors in the indoor units continuously monitor key operating parameters, including the compressor's operating frequency, the temperatures of the condenser and evaporator, and the refrigerant's pressure. The microprocessor on the control panel also records the equipment's on / off status and fault alarm information. Key operating parameters and status parameters are then combined using a weighted fusion method to generate equipment operating status parameters. Equipment power consumption data is collected through current sensing devices, which include high-precision energy meters and current transformers. These devices are installed in the power supply lines of the main components of the VRF (multi-split system), such as the compressor and fan motor. The current sensing devices can directly read the electrical parameters of voltage, current, and power factor, and use discrete Fourier transform (DFT) to calculate the power of these electrical parameters to obtain the total energy consumption value. The energy consumption value is accumulated over time using an integration algorithm to obtain equipment power consumption data. The 3D point cloud of a building is collected using a laser scanner. The laser scanner is mounted on the building's exterior wall, scanning the building structure and surrounding environment while emitting a laser beam and receiving the reflected echo signals. Trimble RealWorks is used to filter, register, and stitch the echo signals to generate a high-precision point cloud. For hard-to-reach areas of the building's exterior, drone-mounted cameras are used for oblique photography to generate supplementary point clouds. Professional 3D modeling software (such as Autodesk Revit) is used to stitch and fuse the high-precision point cloud and the supplementary point cloud to ultimately form a complete 3D point cloud of the building. S1.2. Preprocess the collected environmental situation awareness data. Specifically, the environmental situation awareness data is first subjected to sliding window smoothing and nonlinear noise separation using mean filtering to eliminate sensor noise. Then, the K-nearest neighbor interpolation method is used to interpolate missing values ​​in the environmental situation awareness data to obtain a complete spatiotemporal dataset. Subsequently, the complete spatiotemporal dataset is scaled using Min-Max normalization and dimensionality reduction using PCA. Dynamic time warping (DTW) is used to align timestamps to obtain the preprocessed environmental situation awareness data. S1.3. Build and train the temporal attention model. Specifically, in the PyTorch framework, first initialize the convolution layer through a two-layer cascaded Conv1D structure: set the first layer convolution kernel size to 5 and the number of channels to 64; set the second layer convolution kernel size to 3 and the number of channels to 128; and connect batch normalization (BatchNorm1d) and ReLU activation function after the second layer convolution kernel; secondly, use the nn.LSTM parameter to initialize the bidirectional LSTM layer: set the LSTM hidden layer dimension to 256, the number of attention heads of the multi-head attention layer to 4, and connect the maximum pooling and Dropout layers after the multi-head attention layer; use the layered stacking strategy to perform residual connections on the initialized convolution layer and the bidirectional LSTM layer to obtain high-dimensional feature representation; use the fully connected layer to integrate the high-dimensional feature representation to complete the construction of the temporal attention model; Next, the constructed temporal attention model is trained. Specifically, the environmental situation awareness data is input into the temporal attention model through the MQTT protocol interface, and the environmental situation awareness data is divided into a sample set, a training set, and a validation set using the stratified random sampling method. On the sample set, the sliding window is used to segment the temporal segments to generate standardized training samples. On the training set, the Adam optimizer is used to perform gradient backpropagation on the standardized training samples to obtain the parameter update gradient. The MAE loss function is used to iteratively update the parameter update gradient to generate the MAE value. On the validation set, when the MAE value reaches the convergence threshold for 10 consecutive epochs, early stopping is triggered and training is terminated. The trained temporal attention model is saved synchronously using the torch.save function. It should be noted that the convergence threshold is defined based on the sliding average of the MAE value; the value range is [0.001~0.005]; S1.4. Generate spatiotemporal fusion features through the temporal attention model. Specifically, the convolution layer divides the environmental situation awareness data into temporal windows through one-dimensional dilated convolution, and uses causal filling to expand the features to generate a coarse-grained feature map. Based on the coarse-grained feature map, secondary convolution is used to perform feature abstraction and dimensionality transformation to obtain local spatiotemporal features. The local spatiotemporal features are nonlinearly enhanced through gated linear units to obtain a sparse feature matrix. Softmax normalization is used to assign different weights to the sparse feature matrix, and a multi-head attention mechanism is used to perform cross-channel splicing to generate a temperature change matrix. The forward LSTM of the bidirectional LSTM layer extracts temporal features from the environmental situational awareness data through a gating mechanism and aggregates them through a fully connected layer to generate a preliminary hidden state. The Sigmoid activation function is used to assign gated weights to the preliminary hidden state to obtain gated weighted features. The Tanh activation function is used to perform a nonlinear transformation on the gated weighted features to generate candidate memory states. The forget gate and input gate are then used to perform forward temporal iteration to obtain a forward state sequence. The backward LSTM uses a reverse gating mechanism to extract reverse temporal features from environmental situational awareness data and generate a reverse hidden state. The reverse hidden state is nonlinearly transformed using the ReLU activation function to obtain the reverse hidden state. The reverse hidden state is normalized using Layer Normalization to generate a dynamic feature representation. Based on the dynamic feature representation, a Gated Linear Unit is used to perform reverse temporal iteration to generate a reverse state sequence. The weighted average method is used to assign different weights to the forward state sequence and the reverse state sequence. Based on the different weights, dynamic fusion is performed in the fully connected layer to generate a time-dependent matrix. Gated Feature Interaction is used to perform bidirectional interaction between the temperature change matrix and the time dependency matrix, and a multi-head self-attention mechanism is used for multi-granularity weight allocation. Residual connections are simultaneously used for feature enhancement and gradient stabilization, and attention-enhanced features are output. The attention-enhanced features are projected using a linear projection layer and spliced ​​through cross-modal channels to generate spatiotemporal fusion features. S1.5. Use the Gaussian process regression algorithm to perform nonlinear mapping on the spatiotemporal fusion features to generate an air volume forecast demand table. Specifically, the Gaussian process regression algorithm is used to decompose the spatiotemporal fusion features and perform matrix calculations to obtain the predicted mean and variance. Bayesian optimization is then used to perform hyperparameter fitting to obtain the Gaussian prediction distribution. The kernel function is used to perform bandwidth smoothing on the Gaussian prediction distribution to generate an optimized distribution. Maximum likelihood estimation is used to perform probability density decomposition on the optimized distribution, and sliding window filtering is used for noise suppression to generate a smoothed air volume forecast sequence. Based on the 3σ interval covering rule, Gaussian kernel regression is used to map the smoothed air volume forecast sequence into an air volume forecast demand table. It should be noted that the 3σ interval covering rule is defined based on the integral area of ​​the probability density function of the normal distribution; The air volume forecast demand table can not only accurately reflect the future trend of air volume demand changes, but also provide early warning of abnormal fluctuations and provide data support for the scheduling of air-conditioning equipment. S2. Combine the air volume forecast demand table with the building's three-dimensional point cloud to construct a dynamic airflow topology map. Use the PSO algorithm to iteratively search the dynamic airflow topology map to obtain the air supply parameter combination. Simultaneously, map the air supply parameter combination into an execution instruction set through the device protocol mapping library.

[0027] The specific steps include: S2.1. Combine the air volume forecast demand table with the building's three-dimensional point cloud to construct a dynamic airflow topology map. Specifically, the air volume forecast demand table is first encoded using a spatial encoding algorithm according to the octree index to generate air volume-related data. Kriging interpolation is used to reconstruct the air volume-related data into a three-dimensional field to obtain a continuous air volume field. The finite volume method is used to discretize the continuous air volume field into a grid through flux integration to generate an unstructured grid. Gradient descent is used to numerically solve the flow field and achieve residual convergence on the unstructured grid to obtain the air volume distribution matrix. ICP registration is used to spatially align the air volume distribution matrix with the building's three-dimensional point cloud, and the coordinate system is unified using a rigid body transformation matrix to obtain air-solid coupling data. Based on the air-solid coupling data, Delaunay triangulation is used to perform topological connection to obtain the initial airflow network. The A* algorithm is used to search for the optimal path of the initial airflow network to obtain key airflow channels. Graph embedding is used to aggregate node features in the key airflow channels, complete node association, and generate a weighted directed graph. The weighted directed graph is optimized through spectral clustering using the Laplace matrix to achieve edge weight update and generate a dynamic airflow topology map. It should be noted that the rigid body transformation matrix refers to the 4×4 homogeneous coordinate transformation matrix for rotation and translation in three-dimensional space, which is called by the registration_icp function of the Open3D library; the Laplace matrix refers to the characteristic matrix that describes the topological structure in graph theory, which can be calculated and output by the scipy library of the Python engine; S2.2. Use the PSO algorithm to iteratively search the dynamic airflow topology map to obtain the air supply parameter combination. In the specific operation, the PSO algorithm (particle swarm optimization algorithm) is used to perform spectral clustering eigenvalue decomposition on the dynamic airflow topology map to obtain the eigenvalues ​​of the adjacency matrix; the K-means clustering is used to spatially partition the eigenvalues ​​of the adjacency matrix to generate the initial particle distribution; the PCA is used to perform multidimensional parameter space decomposition on the initial particle distribution to generate a search space after dimensionality reduction; in the search space after dimensionality reduction, the particle fitness of the initial particle distribution is verified according to the judgment threshold. For example, the initial particle distribution that exceeds the judgment threshold is defined as verified, and the verified initial particle distribution is integrated through the adaptive weighting method to obtain the global particle swarm state; It should be noted that the tertile range (IQR) of the fitness distribution of the judgment threshold particle swarm is defined in the range of [0.7, 0.9]; Dynamic inertia weight adjustment is used to update the speed and position of the global particle swarm state, and cross-validation is used for weighted sorting to generate an elite particle set. The Gaussian mutation operator is used to perform neighborhood mutation on the elite particle set to obtain enhanced candidate solutions. Based on the enhanced candidate solutions, Pareto sorting is used for iterative search to generate a non-dominated solution set. The non-dominated solution set is iteratively optimized using TOPSIS decision making to generate a combination of air supply parameters. It should be noted that the Gaussian mutation operator refers to a probabilistic perturbation operator based on the normal distribution, which performs neighborhood mutation on the elite particle set by randomly sampling N(μ,σ²), with a value range of [μ-3σ, μ+3σ]; S2.3. Map the air supply parameter combination to an execution instruction set through the device protocol mapping library. Specifically, the device protocol mapping library is called through the OPC UA interface, and the device protocol mapping library is matched with the register address through the function code parser to obtain the Modbus function code. The air supply parameter combination is divided into blocks and packaged according to the Modbus function code to generate structured data. The structured data is hashed and replaced with repeated strings using the LZW dictionary encoding method to generate deduplicated data blocks. The deduplicated data blocks are encoded with hexadecimal instructions using the Modbus RTU encoder to obtain the original instruction frame. The original instruction frame is aligned with the time base using the clock synchronization method to generate a time synchronization instruction frame. The time synchronization instruction frame is compressed in real time using the LZ4 compression algorithm, and the timestamp is embedded using the NTP protocol. The Modbus TCP protocol is used for synchronous encapsulation to generate an execution instruction set. The execution instruction set can not only accurately control the speed and wind direction of the air-conditioning equipment, but also intelligently adjust the supply air temperature to quickly reach a comfortable temperature indoors.

[0028] S3. Send the execution instruction set to the device control center, drive the variable frequency fan through the PID controller to perform air flow regulation, synchronously collect real-time device power consumption data, and input the timing attention model for incremental update to obtain the optimized timing attention model.

[0029] The specific steps include: S3.1. Send the execution instruction set to the device control center, and use the PID controller to drive the variable frequency fan to adjust the air flow. The specific operation is as follows: First, the execution instruction set is binary TLV encoded using the CoAP protocol to obtain the protocol data unit. The block transmission option of Block2 is used to send the protocol data unit in segments to the device control center. At the same time, the retransmission mechanism is used to ensure the reliability of data transmission during network jitter. After receiving the execution instruction set, the control center uses a priority scheduling algorithm to dynamically assign priorities based on QoS tags. It also uses the NTP time server to decompose the UTC timestamps of the execution instruction set to obtain the timestamp difference. Based on the timestamp difference, the queue adjuster promotes the queue position of urgent instructions, and uses the EDF algorithm to perform secondary sorting based on the instruction's effective time window, ultimately outputting a time-sequential instruction queue. It should be noted that the QoS tag refers to the instruction criticality level identifier, which is called by the protocol parsing unit of the device control center; The PID controller is initialized using the Ziegler-Nichols tuning method: the integral and differential components are set to zero, and the proportional coefficient is gradually increased until the output of the PID controller exhibits constant-amplitude oscillations. The critical proportional coefficient and oscillation period at this point are recorded, and the output waveform is monitored using a step response test. Initialization is complete when the output waveform meets the overshoot threshold. It should be noted that the critical proportional coefficient refers to the proportional gain value when the PID controller produces constant amplitude oscillation under pure proportional coefficient control, which is defined based on the closed-loop critical stability of the proportional coefficient; the overshoot threshold is defined based on the first peak amplitude of the step response, and the value range is 15%-20% of the first peak amplitude; Next, the initialized PID controller is used to perform real-time comparison of the set value (SP) and process variable (PV) of the time sequence instruction queue to obtain the current control deviation; the sliding window filter is used to suppress the noise of the current control deviation to generate a deviation signal value, and the proportional-integral-differential operation is used to dynamically compensate the deviation signal value to generate the compensation control amount. The specific mathematical formula is as follows: ; in, Represents the time index, Indicates time The compensation control amount, represents the proportionality coefficient, Indicates the relationship between the set value and the process variable in time The filtered difference, represents the integral coefficient, represents the time step, Indicates the time step between the set value and the process variable The filtered difference, represents the differential coefficient; It should be noted that the set value refers to the target control quantity of the timing instruction queue, which is defined based on the actual control task requirements; the process variable refers to the real-time feedback quantity of the timing instruction queue, which is collected through the signal conditioning circuit; the proportional coefficient is defined based on the actual response speed requirements and has a value range of [0.5-1.2]; the integral coefficient is defined based on the actual steady-state error requirements and has a value range of [0.1-3.0]; the differential coefficient is defined based on the actual overshoot suppression requirements and has a value range of [0.01-0.1]; The output limiter is used to clamp the amplitude of the compensation control quantity to obtain the safety control quantity; the PWM generator is used to linearly map the duty cycle of the safety control quantity to generate PWM duty pulses; The PWM duty pulse is linearly proportionally mapped through a digital low-pass filter to obtain an equivalent voltage command. The equivalent voltage command is converted into phase by a phase modulator to generate an armature drive signal. Based on the armature drive signal, the armature voltage is PWM-chopper modulated using an H-bridge power topology to generate a pulse voltage waveform. The pulse voltage waveform is ripple-suppressed and adjusted using an LC filter circuit to generate a controllable DC voltage. Based on the controllable DC voltage, the brushless motor drives the variable frequency fan to modulate the speed and obtain real-time speed feedback data. Based on the real-time speed feedback data, the proportional control algorithm is used to correct the error of the air flow rate to complete the air flow adjustment. S3.2. Synchronously collect real-time device power consumption data and input it into the time-series attention model for incremental update to obtain an optimized time-series attention model. The specific operation is as follows: During the air flow adjustment process, the current sensing resistor is used to synchronously collect device power consumption data and input it into the time-series attention model through the RS485 interface; The temporal attention model dynamically weights device power consumption data using Softmax normalization to obtain an attention weight matrix. Sliding window sampling is used to randomly sample the attention weight matrix, and the embedding layer performs dimension expansion to obtain a high-dimensional feature vector. A linear transformation layer performs matrix projection on the high-dimensional feature vector to generate a query matrix, a key matrix, and a value matrix. A multi-head attention mechanism is used to assign dynamic weights to the query matrix, the key matrix, and the value matrix, and the power consumption weight vector is obtained through weight aggregation. The power consumption weight vector is robustly enhanced using adversarial training, and stability is optimized through gradient penalty to obtain anti-interference weights. Based on the obtained anti-interference weights, the GRU parameters are fine-tuned online using the sliding window incremental learning method to generate dynamic GRU units. The dynamic GRU units are incrementally updated using momentum gradient descent to obtain adaptive network parameters. The adaptive network parameters are weighted and integrated to output the optimized temporal attention model. It should be noted that GRU parameters refer to the update gate weights, reset gate weights, and candidate hidden state weights in the gated recurrent unit, which are obtained through time backpropagation; The optimized temporal attention model can accurately predict the real-time power consumption of air-conditioning equipment, laying the foundation for maximizing energy efficiency.

[0030] S4. Based on the optimized temporal attention model, the NSGA-III algorithm is used to solve multiple objectives and generate the final energy-saving solution for the internal machine.

[0031] The specific steps include: S4.1. The optimized temporal attention model is embedded in causal convolution to perform dynamic feature focusing and generate a temporal energy distribution matrix. Specifically, the convolution layer of the optimized temporal attention model is first embedded with causal convolution via the dilation rate parameter. The causal convolution is then zero-padding for temporal alignment to generate causal convolution features. The causal convolution features are activated by a gated linear unit (GLU) and then added to the positional encoding to form a spatiotemporal feature basis. It should be noted that positional encoding refers to the sine and cosine function mapping vectors of causal convolution features, which is called by the torch.arange function of PyTorch; Locality-sensitive hashing (LSH) is used to bucket the spatiotemporal feature base to reduce data complexity and generate a set of hash buckets with similar features. A sparse attention mechanism is used to selectively assign weights and associate features to the hash bucket set to achieve dynamic feature focusing and obtain feature representations with high information density. At the same time, strided convolution is used to enhance local features of the feature representations with high information density to generate multi-scale energy features. The feature tensor reorganization method is used to fuse the multi-scale energy features in the spatiotemporal dimension to obtain a time series energy distribution matrix. S4.2, NSGA-III algorithm performs multi-objective optimization on the time series energy distribution matrix through hierarchical non-dominated sorting to obtain the Pareto solution set. In the specific operation, NSGA-III algorithm normalizes the time series energy distribution matrix and samples the reference points to obtain the initial population; the initial population is hierarchically divided into target spaces through non-dominated sorting to obtain the non-dominated frontier level; based on the non-dominated frontier level, the initial population is subjected to multi-objective optimization of polynomial mutation and simulated binary crossover using a multi-objective elite retention strategy to obtain a candidate solution set, and the candidate solution set is subjected to niching allocation to generate a new generation population; the new generation population is subjected to crowding iteration using an environmental selection operator to obtain the Pareto solution set; It should be noted that the environment selection operator is defined based on the target space grid crowding, and its value range is [0,1], where 0 indicates complete overlap of solutions and 1 indicates uniform distribution; S4.3. Use the entropy weight method to normalize the weights of the Pareto solution set to obtain the optimal energy-saving parameters, and perform three-dimensional rendering through the Unreal Engine engine to generate the final indoor unit energy-saving plan. In the specific operation, first, divide the Pareto solution set into energy consumption parameters, comfort parameters, and equipment life index parameters through a sliding window according to the multi-objective dimension; discretize the energy consumption parameters, comfort parameters, and equipment life index parameters using the entropy weight method to generate a standardized decision matrix; apply the weighted summation method to sort and normalize the standardized decision matrix to obtain objective weight coefficients; based on the objective weight coefficients, perform weighted standardization on the Pareto solution set to construct a decision matrix; use the TOPSIS algorithm to measure the bidirectional Euclidean distance between the ideal solution and the negative ideal solution of the decision matrix to obtain a comprehensive closeness index; based on the comprehensive closeness index, verify the validity of the decision matrix: when the comprehensive closeness index is greater than the closeness threshold, the decision matrix is ​​judged to have passed the verification, and converted into parameter format through JSON-LD semantics, and the optimal energy-saving parameters are output simultaneously using the Modbus-TCP protocol; It should be noted that the proximity threshold is defined based on the operating time of the air-conditioning equipment and the real-time load rate, and the value range is [0.78, 0.88]; The optimal energy-saving parameters were imported into the CAD model of the air conditioning equipment using the Unreal Engine Datasmith plug-in, and 3D rendering was performed using Blueprint visual scripting. Energy consumption parameters were dynamically displayed through the material's self-luminous intensity, comfort parameters drove the color changes of the virtual PMV sensor, and equipment life indicators were visually displayed through simulated compressor vibration amplitude and noise particle effects. Finally, Sequencer software was used to output a demonstration animation of the final indoor unit energy-saving solution, allowing real-time adjustment of air supply angle and wind speed to observe the optimization effect. The final indoor unit energy-saving solution can not only effectively reduce the comprehensive energy consumption of air-conditioning equipment, but also more accurately regulate the temperature and humidity of the indoor environment, thereby improving the user's comfort experience and extending the service life of the equipment.

[0032] This embodiment also provides an energy-saving control system for indoor units of a multi-split system, including: an air volume prediction module, an instruction generation module, a model optimization module, and a solution generation module. The air volume prediction module is used to input environmental situation awareness data into the time series attention model. The convolution layer extracts the temperature change matrix of different regions, and the bidirectional LSTM layer extracts the time dependency matrix of the indoor air volume. The multi-head self-attention mechanism is used to fuse the data, and the Gaussian process regression algorithm is used to generate the air volume forecast demand table. The instruction generation module is used to combine the air volume forecast demand table with the building's three-dimensional point cloud to construct a dynamic airflow topology map. It then uses the PSO algorithm to iteratively search the dynamic airflow topology map to obtain air supply parameter combinations. The air supply parameter combinations are then mapped into an execution instruction set through the device protocol mapping library. The model optimization module is used to send the execution instruction set to the device control center, drive the variable frequency fan to perform air flow regulation through the PID controller, synchronously collect real-time device power consumption data, and input the time series attention model for incremental update to obtain the optimized time series attention model; The solution generation module is used to generate the final energy-saving solution for the internal machine by performing multi-objective solution solving through the NSGA-III algorithm based on the optimized temporal attention model.

[0033] This embodiment also provides a computer device suitable for the energy-saving control method of the internal unit of a multi-split system, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the energy-saving control method of the internal unit of the multi-split system proposed in the above embodiment.

[0034] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0035] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the program implements the energy-saving control method for the internal units of the multi-split system proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0036] In summary, this invention significantly enhances the system's perception and response to complex indoor environmental changes by integrating multidimensional information such as indoor and outdoor environmental parameters, equipment status, and three-dimensional building point clouds, ensuring more accurate and efficient temperature regulation. Furthermore, by performing online incremental updates to the temporal attention model based on real-time power consumption data, the energy-saving control scheme can be dynamically adjusted and optimized based on actual operating results, significantly improving the stability and energy efficiency of air conditioning equipment.

[0037] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for controlling energy conservation of indoor units in a multi-split system, characterized by: include, The environmental situational awareness data is input into the temporal attention model. The convolutional layer extracts the temperature change matrix of different regions, and the bidirectional LSTM layer extracts the time dependency matrix of the indoor air volume. The data is then fused using a multi-head self-attention mechanism, and the Gaussian process regression algorithm is used to generate an air volume forecast demand table. Combine the air volume forecast demand table with the building's 3D point cloud to construct a dynamic airflow topology map. Use the PSO algorithm to iteratively search the dynamic airflow topology map to obtain air supply parameter combinations. Simultaneously, use the device protocol mapping library to map these air supply parameter combinations into execution instruction sets. The execution instruction set is sent to the device control center, and the variable frequency fan is driven by the PID controller to perform air flow regulation. Real-time device power consumption data is simultaneously collected and input into the time series attention model for incremental update to obtain the optimized time series attention model. According to the optimized temporal attention model, the NSGA-III algorithm is used to solve multiple objectives and generate the final energy-saving solution for the internal machine.

2. The energy-saving control method for indoor units in a multi-split system according to claim 1, wherein: The environmental situation awareness data includes indoor and outdoor environmental parameters, equipment operating status parameters, equipment power consumption data and building three-dimensional point cloud.

3. The energy-saving control method for indoor units in a multi-split system according to claim 2, wherein: The Gaussian process regression algorithm is used to generate the air volume forecast demand table. The specific steps are as follows: The convolutional layer and the bidirectional LSTM layer are integrated through a layered stacking strategy to build a temporal attention model, and the environmental situation awareness data is input into the temporal attention model through the MQTT protocol interface; The convolution layer captures local spatiotemporal features through one-dimensional dilated convolution; the local spatiotemporal features are nonlinearly enhanced using residual connections, and the weights are assigned using Softmax normalization to obtain the temperature change matrix; The forward LSTM of the bidirectional LSTM layer performs forward temporal iteration through a gating mechanism to generate a forward state sequence; The backward LSTM uses the reverse gating mechanism to perform reverse time sequence iteration to obtain the reverse state sequence; the forward state sequence and the reverse state sequence are fused using the weighted average method to generate a time dependency matrix; The multi-head self-attention mechanism is used to project the temperature change matrix and the time dependency matrix, and then spliced ​​through cross-modal channels to generate spatiotemporal fusion features; The Gaussian process regression algorithm is used to perform nonlinear mapping on the spatiotemporal fusion features to generate the wind volume forecast demand table.

4. The energy-saving control method for indoor units in a multi-split system according to claim 3, wherein: The construction of the dynamic airflow topology diagram specifically includes the following steps: The air volume forecast demand table is gridded and discretized using the spatial coding algorithm to obtain the air volume distribution matrix. The air volume distribution matrix and the building's three-dimensional point cloud are associated with nodes and edge weights are updated using the A* algorithm to generate a dynamic airflow topology map.

5. The energy-saving control method for indoor units in a multi-split system according to claim 4, characterized in that: Mapping the air supply parameter combination into an execution instruction set through the device protocol mapping library specifically includes the following steps: The PSO algorithm is used to perform multi-dimensional parameter space decomposition on the dynamic airflow topology to obtain the global particle swarm state. Dynamic inertia weight adjustment and neighborhood mutation are used to iteratively search the global particle swarm state to obtain the air supply parameter combination. According to the device protocol mapping library, the air supply parameter combination is encoded into hexadecimal instruction using Modbus function code to obtain the original instruction frame; the original instruction frame is timestamp embedded using clock synchronization method to generate an execution instruction set.

6. The energy-saving control method for indoor units in a multi-split system according to claim 5, characterized in that: The step of obtaining the optimized temporal attention model specifically includes the following steps: The execution instruction set is sent to the device control center through the CoAP protocol, and the execution instruction set is prioritized using the priority scheduling algorithm to generate a time-series instruction queue; The PID controller converts the timing instruction queue into PWM duty pulses through proportional-integral-differential operations; the PWM duty pulses drive the variable frequency fan to adjust the air flow by adjusting the armature voltage; The device power consumption data is collected using a current sensing resistor and input into the timing attention model through the RS485 interface; The temporal attention model assigns weights to the device power consumption data through Softmax normalization to obtain a power consumption weight vector. Based on the power consumption weight vector, the GRU parameters are incrementally updated through the sliding window incremental learning method to output the optimized temporal attention model.

7. The energy-saving control method for indoor units in a multi-split system according to claim 6, characterized in that: The generation of the final indoor unit energy saving plan specifically includes the following steps: The optimized temporal attention model is embedded in causal convolution to perform dynamic feature focusing and generate a temporal energy distribution matrix. The NSGA-III algorithm performs multi-objective optimization on the time series energy distribution matrix through hierarchical non-dominated sorting to obtain the Pareto solution set; The entropy weight method is used to normalize the weights of the Pareto solution set to obtain the optimal energy-saving parameters. 3D rendering is performed using the Unreal Engine engine to generate the final energy-saving solution for the internal unit.

8. An energy-saving control system for an internal unit of a multi-split system, based on the energy-saving control method for an internal unit of a multi-split system according to any one of claims 1 to 7, characterized in that: Including air volume prediction module, instruction generation module, model optimization module, and solution generation module; The air volume prediction module is used to input environmental situation awareness data into the time series attention model. The convolution layer extracts the temperature change matrix of different regions, and the bidirectional LSTM layer extracts the time dependency matrix of the indoor air volume. The multi-head self-attention mechanism is used to fuse the data, and the Gaussian process regression algorithm is used to generate the air volume forecast demand table. The instruction generation module is used to combine the air volume forecast demand table with the building's three-dimensional point cloud to construct a dynamic airflow topology map. It then uses the PSO algorithm to iteratively search the dynamic airflow topology map to obtain air supply parameter combinations. The air supply parameter combinations are then mapped into an execution instruction set through the device protocol mapping library. The model optimization module is used to send the execution instruction set to the device control center, drive the variable frequency fan to perform air flow regulation through the PID controller, synchronously collect real-time device power consumption data, and input the time series attention model for incremental update to obtain the optimized time series attention model; The solution generation module is used to generate the final energy-saving solution for the internal machine by performing multi-objective solution solving through the NSGA-III algorithm based on the optimized temporal attention model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the energy-saving control method for the internal units of the multi-split system according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the energy-saving control method for the internal units of a multi-split system according to any one of claims 1 to 7 are implemented.

Images