Split type AI linkage multi-module heat pump cluster capacity expansion equipment and cooperative control method

Abstract

The invention discloses split type AI linkage multi-module heat pump cluster capacity expansion equipment and a cooperative control method, and relates to the technical field of heat pump energy saving.The capacity expansion equipment comprises a split type multi-module framework which is based on split type design and comprises a plurality of independent basic heat pump modules and extensible heat pump modules which form a multi-module cluster, standard mechanical interfaces, electrical interfaces and communication protocols are uniformly configured; and the full-dimension cluster sensing module is used for acquiring parameters of an environment where the cluster is located and operation parameters of all split modules in the cluster in real time, performing normalization processing on acquired multi-source heterogeneous data, and generating a standardized data set. The AI cooperative linkage control module integrates multi-dimensional data, a global optimal control scheme is generated through an algorithm, the response speed is high, the control precision is high, environment and load fluctuation can be rapidly adapted, the equipment meets the large-scale energy supply requirement, the operation and maintenance cost is reduced, and the equipment is adapted to complex and changeable application scenes.

Description

Technical Field

[0001] This invention relates to the field of heat pump energy-saving technology, specifically to a split-type AI-linked multi-module heat pump cluster expansion equipment and collaborative control method. Background Technology

[0002] A heat pump is an energy transfer device that efficiently utilizes low-grade heat energy. Its core function is to break the natural law that heat "spontaneously flows from high temperature to low temperature". By consuming a small amount of electrical or thermal energy as a driving force, it extracts heat from low-temperature heat sources in the environment such as air, water, and soil, raises the temperature, and then transfers it to the high-temperature end that needs heating. It can also operate in reverse to achieve cooling. It does not generate heat, but rather transports heat, thus its energy efficiency is far higher than that of traditional electric heating equipment. Heat pump expansion equipment is an auxiliary / upgrade component that enhances the heating or cooling capacity or applicable range of a heat pump system. Its core is to adapt the heat pump to larger loads or more complex operating conditions through technological modifications. Existing technologies lack intelligent control logic. The control of multi-module heat pumps is mostly decentralized independent control or simple linkage control, lacking the ability to collaboratively analyze multi-dimensional data (environmental data, module operation data, load data, heat exchange effect data). It cannot generate a globally optimal control strategy through intelligent algorithms, resulting in low control accuracy, slow response speed, difficulty in adapting to complex environments and dynamic load changes, and inability to achieve overall optimization of system energy supply efficiency and operational stability. In response, we propose a split-type AI-linked multi-module heat pump cluster expansion equipment and collaborative control method. Summary of the Invention

[0003] To address the aforementioned technical challenges, this solution provides a modular heat pump cluster expansion device with split-type AI linkage and a collaborative control method. This technical solution resolves the issue of the inability to generate a globally optimal control strategy through intelligent algorithms.

[0004] To achieve the above objectives, the technical solution adopted by this invention is: a split-type AI-linked multi-module heat pump cluster expansion equipment, the expansion equipment comprising:

[0005] The split-type multi-module architecture is based on a split design and includes several independent basic heat pump modules and expandable heat pump modules to form a multi-module cluster, with unified configuration of standardized mechanical interfaces, electrical interfaces and communication protocols.

[0006] The all-dimensional cluster perception module collects environmental parameters of the cluster and operating parameters of all sub-modules within the cluster in real time. It normalizes the collected multi-source heterogeneous data to generate a standardized dataset and interacts with the AI ​​collaborative control module in real time based on a high-speed communication link.

[0007] The cluster collaborative heat exchange module, based on a standardized dataset, intelligently analyzes the overall operating status and environmental adaptability of the multi-module cluster, and determines whether the overall cooling / heating effect of the cluster under the current environmental parameters meets the human body temperature requirements; if it does, it maintains the collaborative operation of each individual module; if it does not, it outputs adjustment suggestions.

[0008] The cluster load throttling module, based on a standardized dataset and the operating parameters of each sub-module, judges the structural load of all modules in the cluster in real time, identifies modules with peak and low loads, and dynamically adjusts the branch flow of each sub-module.

[0009] The AI-powered collaborative control module integrates the strategy suggestions from the cluster collaborative heat exchange module with the load judgment results from the cluster load throttling module. Through multi-dimensional collaborative analysis using AI algorithms, it generates targeted control electrical signals for collaborative control.

[0010] Preferably, the mechanical structure is a quick-release flange and sealing buckle structure; the electrical structure is a three-phase five-wire interface to prevent mis-insertion; the communication interface is the Modbus TCP / Profinet industrial communication protocol; the basic heat pump module is a cluster carrier unit, and the rated power of a single basic module is uniformly 5kW; the expandable heat pump module is compatible with the basic module in terms of structural dimensions, interface specifications and core parameters.

[0011] Preferably, environmental parameters are collected in real time by sensors, and the data includes ambient temperature and relative humidity. For all separate modules in the cluster, each module has an independent built-in sensing unit to collect core operating parameters in real time, including the operating parameters of the compressor, heat exchanger and throttling components. The operating parameters include energy consumption data and branch fluid parameters. Multi-source heterogeneous data normalization processing uses the min-max normalization algorithm to map data of different dimensions to the [0, 1] interval and unify the data format to obtain a dataset. Based on the weighted algorithm, ambient temperature and relative humidity are integrated into a comprehensive value of environmental parameters.

[0012] Preferably, the specific analysis steps within the cluster collaborative heat exchange module include:

[0013] Based on standardized datasets, environmental feature dimensions, cluster thermal output dimensions, and human comfort correlation dimensions are extracted. A three-dimensional feature space is constructed with environmental parameters as the X-axis, overall cluster load as the Y-axis, and cluster heat exchange efficiency COP value as the Z-axis. Based on historical operating data, the DBSCAN density clustering algorithm is used to divide the clusters into three types: comfortable and efficient clusters, sub-comfortable and adaptable clusters, and uncomfortable and adjusting clusters.

[0014] The system calculates the coordinates of the current operating condition in three-dimensional space in real time. By calculating the distance between this point and the comfort and efficiency cluster, if the distance is less than or equal to the cluster radius (which is determined based on historical comfort conditions), the current heat exchange effect is determined to meet the requirements. If the distance exceeds the radius and falls into the discomfort adjustment cluster, the requirements are determined to be unmet. The system then adjusts the individual modules that do not meet the requirements based on the AI ​​collaborative linkage control module.

[0015] Preferably, the DBSCAN density clustering algorithm is combined with historical operating data of the heat pump cluster for calibration. The calibration logic is grouped and calibrated according to operating condition type. The algorithm executes the standardization of each set of environmental parameters, cluster load, and COP value in the historical operating data into a three-dimensional coordinate point Pi = (Xi, Yi, Zi), where X = comprehensive environmental parameter value, Y = overall cluster load, Z = COP value, and Pi is the i-th point in the three-dimensional coordinates. As a sample, a historical sample set D = {P1, P2, ..., Pm} is constructed, where m is the total number of historical operating conditions. The sample set D is traversed, and for each unlabeled sample Pi, the number of samples in its ε-neighborhood is calculated, where ε is the neighborhood radius. If the number of samples in the neighborhood ≥ MinPts, where MinPts is the minimum number of samples in the neighborhood, it indicates the minimum number of sample points contained within the ε-neighborhood. Pi is labeled as the core point, and all directly reachable core points and boundary points within its neighborhood are grouped into the same cluster. The three types of clusters are then divided.

[0016] Preferably, the distance calculation formula uses Mahalanobis distance to calculate the distance between the current working condition coordinate point and the comfort and efficiency cluster. The formula expression is as follows:

[0017] dM=(P-μ)TΣ-1(P-μ)

[0018] Where dM is the Mahalanobis distance between the current three-dimensional coordinate point P and the center μ of the comfort and efficiency cluster; P = (X, Y, Z) is the standardized three-dimensional coordinate point of the current operating condition, and the minimum value of M is < 3; (P-μ)T is the transpose of the difference vector P-μ, which converts the column vector into a row vector to satisfy the matrix multiplication dimension; μ = (μX, μY, μZ) is the three-dimensional center coordinate of the comfort and efficiency cluster; μX is the mean of the X values ​​of all samples in the cluster; μY and μZ are calculated based on historical comfort operating condition samples; Σ is the 3×3 covariance matrix of the historical samples of the comfort and efficiency cluster, which describes the correlation between the three dimensions of environment, load, and COP.

[0019] Preferably, the specific analysis steps within the cluster load throttling module are as follows:

[0020] Through multi-dimensional load quantification analysis, combined with dynamic thresholds and algorithm verification, the system identifies peak, normal, and low-load modules. Based on the goal of load balancing, it implements differentiated flow adjustments under different load scenarios, gradually throttling and shifting load in peak modules, precisely increasing flow and improving efficiency in low-load modules, and adapting to the total flow before further optimization when the overall load fluctuates. The system is then regulated based on an AI-based collaborative control module.

[0021] Preferably, the multi-dimensional load quantification analysis integrates standardized datasets with outdoor temperature and humidity data, and makes a comprehensive judgment based on historical correlation characteristics; the dynamic threshold is not a fixed value, but is calibrated in real time according to the historical load distribution pattern and changes in the overall cluster load and environment; the algorithm verification adopts a sliding window to analyze the load trend.

[0022] The multi-dimensional load quantification analysis data extracts the corresponding features of historical load and each parameter from a standardized dataset, collects outdoor temperature and humidity in real time through distributed sensors, and fuses them according to the parameter influence weights to comprehensively lock the load data of each module.

[0023] The dynamic threshold determines the basic threshold range by statistically analyzing the load distribution range of different seasons and time periods in history. It then makes minor adjustments to the basic threshold based on the deviation between the current scenario and historical scenarios.

[0024] The collaborative control method for split-type AI-linked multi-module heat pump cluster expansion equipment, the collaborative control steps are as follows:

[0025] S1. Multi-source data integration, synchronous data reception, including cooling / heating effect judgment results and parameter adjustment suggestions output by the cluster collaborative heat exchange module, and load status and flow adjustment information of each individual module fed back by the cluster load throttling module, to build a unified data storage pool;

[0026] S2, AI collaborative analysis, based on a preset AI collaborative algorithm, performs hierarchical collaborative analysis to calculate the deviation amplitude between cooling / heating effect and human thermal comfort needs; combined with load balancing target, it comprehensively weighs the output information; and matches the corresponding control strategy to execute steady-state maintenance, load adjustment, heat exchange and load collaborative regulation for different scenarios.

[0027] S3. Control signal generation and output: Convert the control strategy into electrical signal commands and transmit them in a directional manner to the controllers of each separate module and the actuators of the branch electric regulating valves for coordinated control.

[0028] Preferably, the specific steps of AI collaborative analysis in step S2 are as follows:

[0029] The cooling / heating effect assessment results of the associated cluster collaborative heat exchange modules are combined with human thermal comfort standards to quantify the deviation amplitude between the current heat exchange effect and the demand.

[0030] By integrating the load status and flow adjustment information of each individual module fed back by the cluster load throttling module, the types of load imbalance are analyzed, and the modules that need to be focused on regulation and the current flow allocation bottlenecks are identified.

[0031] The weight values ​​are determined based on the preset priority of human body temperature comfort needs > load balance > energy efficiency optimization;

[0032] The deviation level and load imbalance type are combined to form multiple scenario categories, and corresponding control strategies are matched for different combination scenarios.

[0033] In step S3, the control strategy is converted into electrical signal commands by breaking down the control strategy into modular power and valve opening quantification parameters, encoding them according to the three-phase five-wire interface standard and Modbus TCP / Profinet protocol, and then distributing them to the target components and outputting signals in stages.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] The standardized mechanical and electrical interfaces and communication protocols proposed in this invention enable plug-and-play modules that can flexibly adapt to different load scenarios, are easy to expand, and have strong compatibility. The all-dimensional sensing module covers the environment and all module operating parameters, and forms a standardized dataset after normalization processing, providing reliable support for intelligent decision-making. The cluster collaborative heat exchange module predicts the cooling / heating effect and dynamically adapts to environmental conditions to ensure human comfort. The load throttling module accurately identifies the load distribution and dynamically adjusts the flow rate to avoid module overload or idleness, improve energy efficiency, and extend equipment life. The AI ​​collaborative linkage control module integrates multi-dimensional data and generates a globally optimal control scheme through algorithms, with fast response speed and high control accuracy. It can quickly adapt to environmental and load fluctuations. The equipment not only meets the needs of large-scale energy supply but also reduces operation and maintenance costs, adapts to complex and ever-changing application scenarios, and achieves overall optimization of energy supply efficiency, stability, and economy. Attached Figure Description

[0036] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0037] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0038] Reference Figure 1 As shown, the split-type AI-linked multi-module heat pump cluster expansion equipment includes:

[0039] The split-type multi-module architecture is based on a split design and includes several independent basic heat pump modules and expandable heat pump modules to form a multi-module cluster, with unified configuration of standardized mechanical interfaces, electrical interfaces and communication protocols.

[0040] The all-dimensional cluster perception module collects environmental parameters of the cluster and operating parameters of all sub-modules within the cluster in real time. It normalizes the collected multi-source heterogeneous data to generate a standardized dataset and interacts with the AI ​​collaborative control module in real time based on a high-speed communication link.

[0041] The cluster collaborative heat exchange module, based on a standardized dataset, intelligently analyzes the overall operating status and environmental adaptability of the multi-module cluster, and determines whether the overall cooling / heating effect of the cluster under the current environmental parameters meets the human body temperature requirements; if it does, it maintains the collaborative operation of each individual module; if it does not, it outputs adjustment suggestions.

[0042] The cluster load throttling module, based on a standardized dataset and the operating parameters of each sub-module, judges the structural load of all modules in the cluster in real time, identifies modules with peak and low loads, and dynamically adjusts the branch flow of each sub-module.

[0043] The AI-powered collaborative control module integrates the strategy suggestions from the cluster collaborative heat exchange module with the load judgment results from the cluster load throttling module. Through multi-dimensional collaborative analysis using AI algorithms, it generates targeted control electrical signals for collaborative control.

[0044] The mechanical structure features a quick-release flange and sealing buckle; the electrical structure uses a three-phase five-wire interface to prevent mis-insertion; the communication interface uses the Modbus TCP / Profinet industrial communication protocol; the basic heat pump module is a cluster-bearing unit, with a uniform rated power of 5kW for each basic module; the expandable heat pump module is compatible with the basic module in terms of structural dimensions, interface specifications, and core parameters.

[0045] This application features a quick-release flange and sealing snap-fit ​​structure, balancing ease of installation with reliable sealing. It allows for rapid module assembly and disassembly, improving expansion efficiency and flexibility in later maintenance. The anti-misconnection three-phase five-wire interface avoids equipment damage or safety hazards caused by wiring errors, ensuring safe and stable electrical connections and lowering the construction threshold. The Modbus TCP / Profinet industrial communication protocol has strong compatibility, enabling high-speed and stable communication between multiple modules and control units, adapting to industrial-grade application scenarios and ensuring zero-delay data transmission.

[0046] Environmental parameters are collected in real time by sensors, including ambient temperature and relative humidity. For all modules in the cluster, each module has an independent built-in sensing unit to collect core operating parameters in real time, including the operating parameters of the compressor, heat exchanger and throttling components. The operating parameters include energy consumption data and branch fluid parameters. Multi-source heterogeneous data normalization processing uses the min-max normalization algorithm to map data of different dimensions to the [0, 1] interval and unify the data format to obtain a dataset. Based on a weighted algorithm, ambient temperature and relative humidity are integrated into a comprehensive value of environmental parameters.

[0047] This application enables rapid module assembly and disassembly with robust sealing, significantly improving installation, expansion, and maintenance efficiency while ensuring fluid transmission sealing. The anti-misplugging three-phase five-wire interface structurally prevents wiring errors, eliminating equipment damage and safety hazards, lowering the operational threshold, and ensuring safe and stable electrical connections. The Modbus TCP / Profinet industrial communication protocol offers strong compatibility and stable transmission, ensuring high-speed data interaction between multiple modules and control units, making it suitable for industrial-grade application scenarios.

[0048] The specific analysis steps within the cluster-based collaborative heat exchange module include:

[0049] Based on standardized datasets, environmental feature dimensions, cluster thermal output dimensions, and human comfort correlation dimensions are extracted. A three-dimensional feature space is constructed with environmental parameters as the X-axis, overall cluster load as the Y-axis, and cluster heat exchange efficiency COP value as the Z-axis. Based on historical operating data, the DBSCAN density clustering algorithm is used to divide the clusters into three types: comfortable and efficient clusters, sub-comfortable and adaptable clusters, and uncomfortable and adjusting clusters.

[0050] The system calculates the coordinates of the current operating condition in three-dimensional space in real time. By calculating the distance between this point and the comfort and efficiency cluster, if the distance is less than or equal to the cluster radius (which is determined based on historical comfort conditions), the current heat exchange effect is determined to meet the requirements. If the distance exceeds the radius and falls into the discomfort adjustment cluster, the requirements are determined to be unmet. The system then adjusts the individual modules that do not meet the requirements based on the AI ​​collaborative linkage control module.

[0051] Based on a standardized dataset, three core feature dimensions are extracted: environmental feature dimension, which uses the comprehensive value of environmental parameters integrated by the weighted algorithm, which is calculated by weighting the ambient temperature and relative humidity according to the human comfort weight coefficient, the weight coefficient is calibrated by combining the human thermal comfort model with the actual application scenario; cluster thermal output dimension, which is defined as the overall cooling / heating capacity of the cluster, which is calculated by converting the rated power, operating load ratio and real-time output efficiency of all individual modules, with the unit uniformly set to kW; and human comfort related dimension, which is the cluster heat exchange energy efficiency COP value, which is calculated as the ratio of the cluster's real-time total cooling / heating capacity to the cluster's total energy consumption, including the total energy consumption of all module compressors, fans, pumps and other core components.

[0052] The three-dimensional feature space is constructed with the environmental feature dimension as the X-axis, the cluster thermal output dimension as the Y-axis, and the cluster heat exchange efficiency COP value as the Z-axis. The historical operation data must cover valid operation data of different seasons, different time periods, and different environmental conditions throughout the year. The collection period is no less than one full year, and abnormal operating condition data such as voltage fluctuations and equipment failures must be removed to ensure data validity.

[0053] Based on preprocessed historical operating data, the DBSCAN density clustering algorithm was used for cluster analysis. The key parameters of the algorithm were set as follows: the minimum number of points (MinPts) was determined according to the historical data sample size at a ratio of 1% to 3%, but not less than 5; the cluster radius (ε) was determined through statistical analysis of the K-nearest neighbor distances of historical comfort condition data, specifically by calculating the average nearest neighbor distance of all historical comfort condition coordinate points, taking 1.2 to 1.5 times this distance as the initial cluster radius, and then fixing it after three rounds of cross-validation optimization; finally, three types of clusters were formed, including:

[0054] A comfortable and efficient cluster must simultaneously meet the thermal comfort requirements of the human body and the energy efficiency threshold of the cluster.

[0055] Sub-comfort clusters, where the operating conditions within the cluster meet the requirements for human thermal comfort but the COP value is lower than the above-mentioned energy efficiency threshold, or the COP value meets the standard but the PMV value is in the range of -1.0 to -0.5 or +0.5 to +1.0;

[0056] Uncomfortable adjustment clusters are those where the PMV value within the cluster exceeds the range of -1.0 to +1.0, or the COP value is lower than 60% of the design rating.

[0057] The operational data of heat pump clusters are highly scenario-dependent (e.g., the sample distribution differs greatly between cooling season and heating season, and between high daytime load and low nighttime load). If a uniform ε and MinPts are used, it will lead to clustering distortion. Therefore, it is necessary to first group the historical operational data according to the operating condition dimension, and then independently calibrate the parameters for each group of data. The specific steps are as follows:

[0058] Based on the environmental parameters and cluster load defined in the claims, the historical sample set D is divided into operating condition groups according to the following three core dimensions to ensure the consistency of the operating scenarios of samples within each group:

[0059] Dimension 1, Seasonal Operating Conditions, is divided into Cooling Season (Ambient Temperature ≥ 25℃), Transition Season (10℃ < Ambient Temperature < 25℃), and Heating Season (Ambient Temperature ≤ 10℃).

[0060] Dimension 2, Time Period Load, divided into peak load periods (e.g., 9:00-18:00, Y) i ≥70% of rated total load), low load periods (e.g., 22:00-6:00, Y) i≤30% of rated total load), during normal load periods (30% < Y) i <70%);

[0061] Dimension 3, Comfort Needs, based on human thermal comfort standards (such as ISO 7730), categorized by X i (Comprehensive value of environmental parameters) corresponding to the comfort zone grouping (e.g., X) i ∈[0.3,0.6] represents a comfortable environment, X i <0.3 or X i >0.6 indicates an uncomfortable environment;

[0062] This results in multiple sub-sample sets, such as the "cooling season - peak load - comfortable environment" sub-sample set and the "heating season - low load - uncomfortable environment" sub-sample set. Each sub-sample set performs subsequent parameter calibration independently.

[0063] Calibration method for minimum sample size in MinPts

[0064] The core function of MinPts is to distinguish between core points and noise points. It needs to be determined by combining the three-dimensional feature space of the heat pump cluster sample dimension and the density of the subsample set. The specific rules are as follows:

[0065] The base value is set as follows: since the dimension of the three-dimensional feature space (X = comprehensive environmental value, Y = cluster load, Z = COP value) is 3, according to the general principle of DBSCAN parameter design, MinPts > feature dimension, the base value of MinPts is set to feature dimension + 1 = 4, that is, at least 4 samples must be contained in the ε neighborhood to become a core point, so as to avoid a single sample being isolated and clustered.

[0066] Subsample set density adjustment: For each subsample set divided in step 1, calculate the number of samples per unit feature space. The sample density ρ = number of subsamples / volume of the three-dimensional feature space of the subsample set, and adjust MinPts according to the density.

[0067] If ρ≥0.8, for high-density sample sets, such as during peak cooling seasons: MinPts=base value+2=6, to avoid too many core points leading to cluster fusion;

[0068] If 0.3≤ρ<0.8, for medium-density sample sets, such as during the transitional season: MinPts=baseline value=4, the integrity and independence of balanced clusters;

[0069] If ρ < 0.3, for low-density sample sets, such as during low-load periods in the heating season: MinPts = base value - 1 = 3, to avoid insufficient core points due to sparse samples, which would prevent the formation of effective clusters;

[0070] Boundary validation ensures that MinPts ≤ 10% of the number of subsample sets. For example, if a subsample set contains only 20 samples, the maximum MinPts is 2 to avoid the lack of core points due to excessively large MinPts. If it exceeds the limit, it is corrected by rounding down by 10% of the number of subsample sets.

[0071] Calibration method for ε-neighborhood radius

[0072] ε determines whether samples belong to the same cluster, and needs to be determined based on the sample distribution density of the sub-sample set and the clustering objective (distinguishing between comfortable / subcomfortable / uncomfortable clusters). A combination of k-distance graph method and clustering effect feedback is used:

[0073] Initial determination of ε based on k-distance graph

[0074] For each subset of samples, plot the k-distance graph as follows (k = MinPts, i.e., match with MinPts);

[0075] For each sample P in the subset i Calculate its Euclidean distance with all other samples. Since the three-dimensional coordinates have been normalized to [0, 1], the Euclidean distance can directly reflect the sample similarity.

[0076] For each P i Sort all Euclidean distances in ascending order, and take the k-th distance value, which is the distance to the k-th nearest neighbor sample, denoted as d. k (P i );

[0077] d of all samples k (P i Sort the samples in descending order and plot the sample numbers -d. k (P i The curve, i.e., the k-distance plot;

[0078] Find the inflection point of the curve (i.e., d). k (P i The location where the value suddenly increases significantly, and the corresponding value of d at that inflection point. k (P i The value is the initial value of ε;

[0079] ε-feedback optimization based on clustering objectives

[0080] After initially determining ε, it is necessary to verify whether the clustering results meet the partitioning objectives of comfortable and efficient clusters, subcomfortable and well-fitting clusters, and uncomfortable and adjusting clusters. This can be optimized through the following two indicators:

[0081] For intra-cluster consistency verification, calculate the coefficient of variation (CV = standard deviation / mean) of the Z-value (COP value) of all samples in the comfortable and efficient cluster. If CV > 0.2 (indicating that the intra-cluster COP value difference is too large and does not meet the definition of efficiency), then decrease ε (decreasing by 0.05 each time, the normalized value) until CV ≤ 0.2.

[0082] The inter-cluster discrimination verification is calculated by the difference in mean Z-value between the comfortable and efficient cluster and the uncomfortable adjustment cluster (ΔZ = mean Z-comfort - mean Z-uncomfort). If ΔZ < 0.15 (indicating that the energy efficiency difference between the two clusters is not significant and cannot be distinguished), then ε is increased (increased by 0.03 each time) until ΔZ ≥ 0.15.

[0083] The boundary constraint of ε ensures that ε∈[0.1,0.4] (based on the sample distribution characteristics of the normalized three-dimensional space, too small a value is easy to generate a large number of noise points, and too large a value is easy to merge clusters of different working conditions). If the value exceeds the range after optimization, it is necessary to return to step 1 to readjust MinPts and then calibrate ε.

[0084] To ensure parameter stability, a verification process is added to prevent calibration results from being merely adapted to historical data.

[0085] The historical sample set is divided into a calibration set (for calibration of the above parameters) and a validation set (for testing) in a 7:3 ratio. If the proportion of the comfortable and efficient cluster in the clustering results of the validation set deviates from that of the calibration set by ≤10%, and the clustering accuracy (matching degree of comfort condition samples by manual annotation) is ≥85%, then the parameters are valid.

[0086] Real-time operating condition adaptation: During operation, if the real-time operating condition (such as sudden high load) does not belong to the divided operating condition group, the system automatically matches the ε and MinPts of the most similar operating condition group (similarity is determined by "the smallest Euclidean distance between environmental parameter X and load Y"). The parameters are recalibrated once every 30 days based on newly accumulated operating data to ensure adaptation to changes in operating conditions.

[0087] The DBSCAN density clustering algorithm is combined with historical operating data of the heat pump cluster for calibration; the calibration logic is grouped according to operating condition type; the algorithm execution standardizes the three-dimensional coordinates P of each group of environmental parameters, cluster load and COP value from the historical operating data. i =(X i ,Y i Z i X = Comprehensive environmental parameter value, Y = Overall cluster load, Z = COP value, P i Let P be the i-th point in the three-dimensional coordinate system; as a sample, construct a historical sample set D = {P1, P2, ..., P...} m}, where m is the total number of historical operating conditions; iterate through the sample set D, and for each unlabeled sample P iCalculate the number of samples in its ε-neighborhood, where ε is the neighborhood radius. If the number of samples in the neighborhood is greater than or equal to MinPts, then the minimum number of samples in the neighborhood is determined. MinPts is the minimum number of samples in the neighborhood, representing the minimum number of sample points contained within the ε-neighborhood. Label P. i With P as the core point i Expanding from the center, all directly reachable core points and boundary points within its neighborhood are grouped into the same cluster; the three types of clusters are then divided.

[0088] The distance calculation formula uses Mahalanobis distance to calculate the distance between the current working condition coordinate point and the comfort and efficiency cluster. The formula expression is:

[0089]

[0090] Where d M Let P be the Mahalanobis distance between the current 3D coordinate point P and the center μ of the comfort and efficiency cluster; P = (X, Y, Z) is the standardized 3D coordinate point of the current operating condition, and the minimum value of M is < 3; (P - μ) T The transpose of the difference vector P-μ converts the column vector to a row vector, satisfying the matrix multiplication dimension, μ = (μ X ,μ Y ,μ Z ) represents the three-dimensional center coordinates of the comfortable and efficient cluster, μ X Let μ be the mean of all sample X values ​​within the cluster. Y μ Z Similarly, based on historical comfort condition samples, Σ is calculated as a 3×3 covariance matrix of the historical samples of the comfort and efficiency cluster, describing the correlation between the three dimensions of environment, load, and COP.

[0091] The specific analysis steps within the cluster load throttling module are as follows:

[0092] Through multi-dimensional load quantification analysis, combined with dynamic thresholds and algorithm verification, the system identifies peak, normal, and low-load modules. Based on the goal of load balancing, it implements differentiated flow adjustments under different load scenarios, gradually throttling and shifting load in peak modules, precisely increasing flow and improving efficiency in low-load modules, and adapting to the total flow before further optimization when the overall load fluctuates. The system is then regulated based on an AI-based collaborative control module.

[0093] Multi-dimensional load quantification analysis integrates standardized datasets with outdoor temperature and humidity data, and makes comprehensive judgments based on historical correlation characteristics; the dynamic threshold is not a fixed value, but is calibrated in real time based on historical load distribution patterns and changes in the overall cluster load and environment; the algorithm verification uses a sliding window to analyze load trends.

[0094] The multi-dimensional load quantification analysis data extracts the corresponding features of historical load and each parameter from a standardized dataset, collects outdoor temperature and humidity in real time through distributed sensors, and fuses them according to the parameter influence weights to comprehensively lock the load data of each module.

[0095] The dynamic threshold determines the basic threshold range by statistically analyzing the load distribution range of different seasons and time periods in history. It then makes minor adjustments to the basic threshold based on the deviation between the current scenario and historical scenarios.

[0096] The method for determining the influence weight of parameters, and the implementation logic of fusion based on the influence weight of parameters;

[0097] Key parameters affecting the load are extracted from the standardized dataset, specifically including three categories: first, environmental parameters, namely ambient temperature and relative humidity; second, operational parameters, namely compressor exhaust temperature of each module, inlet and outlet water temperature difference of heat exchanger, and branch fluid flow rate; and third, historical correlation parameters, namely historical load values ​​of the same period.

[0098] The objective weight calculation based on grey relational analysis quantifies the objective correlation between each parameter and the current load using grey relational metrics. The steps are as follows:

[0099] The first step is data preprocessing, which normalizes all parameters to the range of 0 to 1 according to existing requirements, eliminating the influence of different units of measurement.

[0100] The second step is to construct a sequence, using the current load as the reference sequence and various influencing parameters as comparison sequences.

[0101] The third step is to calculate the grey relational degree. First, calculate the correlation coefficient, where the resolution coefficient is taken as 0.5, which is commonly used in engineering. It is calculated by combining the global minimum difference, the global maximum difference, and the difference of a single sample. Then, take the average of the correlation coefficients of all samples to obtain the correlation degree of each parameter.

[0102] The fourth step is objective weight normalization, which processes the correlation between each parameter to make the sum of the weights of all parameters equal to 1, thus obtaining the objective weights.

[0103] Subjective weight correction based on the analytic hierarchy process;

[0104] The objective weights are adjusted based on practical experience in heat pump engineering, and the steps are as follows:

[0105] The first step is to construct a judgment matrix. Three to five engineers in the field of heat pumps are invited to compare various parameters in pairs according to their importance. The comparison criteria are divided into five levels: equally important, slightly important, obviously important, strongly important, and extremely important, which form the corresponding judgment matrix.

[0106] The second step is a consistency check. The consistency index of the judgment matrix is ​​calculated. If the index is less than 0.1, the consistency requirement is met and the matrix is ​​retained; otherwise, the judgment matrix is ​​readjusted.

[0107] The third step is to calculate the subjective weights by extracting the eigenvector corresponding to the largest eigenvalue in the judgment matrix and then normalizing it to obtain the subjective weights.

[0108] The fourth step is to determine the combined weights, which are then combined according to the ratio of 70% objective weights and 30% subjective weights to obtain the final weights of each parameter (the total weights are 1).

[0109] Multiply the normalized values ​​of each parameter by their corresponding final weights, and then sum all the product results to obtain the fused load quantization value.

[0110] The collaborative control method for split-type AI-linked multi-module heat pump cluster expansion equipment, the collaborative control steps are as follows:

[0111] S1. Multi-source data integration, synchronous data reception, including cooling / heating effect judgment results and parameter adjustment suggestions output by the cluster collaborative heat exchange module, and load status and flow adjustment information of each individual module fed back by the cluster load throttling module, to build a unified data storage pool;

[0112] S2, AI collaborative analysis, based on a preset AI collaborative algorithm, performs hierarchical collaborative analysis to calculate the deviation amplitude between cooling / heating effect and human thermal comfort needs; combined with load balancing target, it comprehensively weighs the output information; and matches the corresponding control strategy to execute steady-state maintenance, load adjustment, heat exchange and load collaborative regulation for different scenarios.

[0113] S3. Control signal generation and output: Convert the control strategy into electrical signal commands and transmit them in a directional manner to the controllers of each separate module and the actuators of the branch electric regulating valves for coordinated control.

[0114] The specific steps of AI collaborative analysis in step S2 are as follows:

[0115] The cooling / heating effect assessment results of the associated cluster collaborative heat exchange modules are combined with human thermal comfort standards to quantify the deviation amplitude between the current heat exchange effect and the demand.

[0116] By integrating the load status and flow adjustment information of each individual module fed back by the cluster load throttling module, the types of load imbalance are analyzed, and the modules that need to be focused on regulation and the current flow allocation bottlenecks are identified.

[0117] The weight values ​​are determined based on the preset priority of human body temperature comfort needs > load balance > energy efficiency optimization;

[0118] The deviation level and load imbalance type are combined to form multiple scenario categories, and corresponding control strategies are matched for different combination scenarios.

[0119] In step S3, the control strategy is converted into electrical signal commands by breaking down the control strategy into modular power and valve opening quantification parameters, encoding them according to the three-phase five-wire interface standard and Modbus TCP / Profinet protocol, and then distributing them to the target components and outputting signals in stages.

[0120] Methods for determining window size and sliding step size in sliding window analysis

[0121] Determining the window size: The window size refers to the number of historical samples participating in the load trend analysis, and is set in three categories according to the load characteristics of different time periods:

[0122] During peak load periods (e.g., 9:00-18:00, when the load reaches or exceeds 70% of the rated total load), the load fluctuates frequently. The window size is set to 12, which corresponds to 12 samples collected every 10 minutes within 2 hours, balancing trend capture and real-time performance.

[0123] During normal load periods (load is between 30% and 70% of the rated total load): the load fluctuation is moderate, the window size is 8, corresponding to 8 10-minute samples within 1 hour and 20 minutes;

[0124] During low-load periods (e.g., 22:00-6:00, when the load does not exceed 30% of the rated total load): the load fluctuation is gentle, and the window size is set to 4, corresponding to four 10-minute samples within 40 minutes, to avoid the trend being lagging due to an excessively large window.

[0125] Special scenario adaptation: If the ambient temperature changes by 5°C or more within 1 hour, the current window size will be automatically reduced by 50% to improve the trend response speed.

[0126] The sliding step size is determined by the number of samples the window moves each time, and is set in conjunction with the window size and data acquisition frequency.

[0127] The data acquisition frequency is fixed at once every 10 minutes, which meets the requirements for effective fluctuation capture of heat pump operating parameters;

[0128] The sliding step size is one-third of the window size (to ensure reasonable overlap when the window moves and avoid trend breakage): the step size is 4 during peak hours (moving the window once every 40 minutes), the step size is 3 during normal hours (moving the window once every 30 minutes), and the step size is 2 during low-load hours (moving the window once every 20 minutes).

[0129] The load trend verification logic uses a sliding window to calculate the rate of change of the average load within the window to determine the trend: if the rate of change of the average load for two consecutive windows exceeds 5%, it is determined to be an upward or downward trend in load; if the rate of change does not exceed 3%, it is determined to be a stable load, and the subsequent flow adjustment strategy is matched accordingly.

[0130] The specific logic of AI collaborative analysis includes:

[0131] Deviation amplitude quantification, calculation of the difference between heat exchange effect and demand;

[0132] The standard for human thermal comfort is clearly defined, and the internationally recognized ISO 7730 standard is adopted to transform human thermal comfort needs into quantifiable indicators. The core comfort indicator is the predicted average vote value, and its standard comfort range is -0.5 to 0.5. The auxiliary indicator is the cluster heat exchange energy efficiency, and the comfortable and efficient range is an energy efficiency value of not less than 3.0.

[0133] The deviation magnitude calculation logic quantifies two dimensions: the deviation of the predicted average voting value and the energy efficiency deviation, and takes the more severe deviation level as the final deviation:

[0134] Predicted average vote deviation: When the predicted average vote is within the comfortable range of -0.5 to 0.5, the deviation is 0; when the predicted average vote is greater than 0.5 (too hot), the deviation is the current predicted average vote minus 0.5; when the predicted average vote is less than -0.5 (too cold), the deviation is 0.5 minus the current predicted average vote. The deviation levels are as follows: 0 is no deviation, 0.1 to 0.3 is slight deviation, 0.4 to 0.6 is moderate deviation, and more than 0.6 is severe deviation.

[0135] Energy efficiency deviation: When the energy efficiency value is not lower than 3.0, the deviation is 0; when the energy efficiency value is lower than 3.0, the deviation is (3.0 minus the current energy efficiency value) divided by 3.0 and then multiplied by 100% (i.e., the relative deviation rate). The deviation levels are divided into: 0 is no deviation, 5% to 15% is slight deviation, 16% to 30% is moderate deviation, and more than 30% is severe deviation.

[0136] The final deviation level is determined by taking the higher value: if the deviation of the predicted average vote value is moderate and the deviation of energy efficiency is slight, the final deviation level is moderate; if there is no deviation in either, it is determined to be no deviation.

[0137] Based on the load data of the cluster load throttling module, load imbalance is divided into four categories according to the two dimensions of degree of imbalance and scope of impact:

[0138] Local imbalance: Mild imbalance is when the load of a single module exceeds the rated load by 10% or is lower than the rated load by 20%; Moderate imbalance is when the load of a single module exceeds the rated load by 20% or is lower than the rated load by 30%; Severe imbalance is when the load of a single module exceeds the rated load by 30% or is lower than the rated load by 40%.

[0139] Overall Imbalance: Mild imbalance is when the average load exceeds the rated total load by 10% or is lower than the rated total load by 15%; moderate imbalance is when the average load exceeds the rated total load by 20% or is lower than the rated total load by 25%; severe imbalance is when the average load exceeds the rated total load by 30% or is lower than the rated total load by 35%.

[0140] The matching rules for control strategies and scenario combinations use deviation level + imbalance type as scenario identifiers, and match control strategies according to the priority of human body temperature comfort needs > load balancing > energy efficiency optimization, as follows:

[0141] No deviation and no imbalance: A steady-state holding strategy is adopted to maintain the current power of each module and the opening of the branch valves, and data is collected every 15 minutes to verify the status;

[0142] Slight deviation + local imbalance: Adopt a load fine-tuning strategy, adjust the power of the unbalanced module within 5% and the valve opening within 3%, prioritize the adjustment of low load modules to avoid energy efficiency reduction, and verify whether the deviation has been reduced 5 minutes after adjustment;

[0143] For moderate deviations and local / overall imbalances, a heat exchange-load coordinated control strategy is adopted. First, the comfort deviation is resolved (if it is too hot, the total power of the cluster is reduced by 10%; if it is too cold, the total power is increased by 10%). Then, the load is balanced (the power of the imbalanced module is adjusted within 10%, and the valve opening is adjusted within 5%). The energy efficiency value is monitored simultaneously to ensure that it is not lower than 2.8.

[0144] In cases of severe deviation and overall imbalance, an emergency coordinated control strategy is adopted to prioritize comfort (if it is too hot, shut down 1 to 2 low-load modules; if it is too cold, start the backup expandable module), and then force load balancing (reduce the flow of peak modules by 30% and transfer 20% of the flow to low-load modules). Data is collected every 2 minutes until the deviation drops to below the moderate level.

[0145] Existing technology cases

[0146] Case 1: Multi-module heat pump system in a university library

[0147] Configuration: 16 air source heat pump modules + centralized controller

[0148] Control features: Simple group rotation control + return water temperature control

[0149] Typical drawback: It cannot integrate environmental data such as outdoor temperature and humidity and solar radiation, resulting in frequent defrosting when the temperature drops below -10℃ in winter, causing energy efficiency to drop by 40%.

[0150] By monitoring only the return water temperature as a single parameter and ignoring the differences in the operating status of each module, two compressors burned out due to long-term low-load operation.

[0151] Response lag: Room temperature fluctuates by ±3℃ during sudden load changes, resulting in poor comfort.

[0152] Case 2: Multi-split System for Commercial Complexes

[0153] Configuration: 24 variable frequency multi-split air conditioners + building automation system

[0154] Control characteristics: PID control based on time schedule and zone temperature

[0155] Typical drawbacks: The lack of correlation analysis between environmental data (weather, wind speed) and load data (person flow) leads to a 25% increase in energy consumption during the transition season; the lack of heat exchange effect monitoring results in long-term insufficient cooling in two terminal areas, and pipe blockage was found during maintenance; the rigid control strategy cannot adapt to the differences in customer flow density in different areas of the mall, and the temperature compliance rate in high-customer flow areas is only 82%.

[0156] Case 1 Problem vs. Invention Solution

[0157] Problems with traditional systems: They cannot integrate environmental data such as outdoor temperature and humidity and solar radiation; they require frequent defrosting in winter when temperatures drop below -10°C, resulting in a sharp drop in energy efficiency of 40%.

[0158] Single temperature monitoring makes the system slow to perceive environmental changes and unable to predict defrosting needs.

[0159] Advantages of this invention:

[0160] Real-time collection of multi-dimensional parameters such as ambient temperature and humidity, wind speed, and solar radiation to construct an environmental profile;

[0161] The intelligent defrosting optimization AI algorithm integrates environmental data and module operating parameters to accurately determine the timing of defrosting, reduce ineffective defrosting, and improve energy efficiency by 30%+.

[0162] Dynamic environment adaptation automatically adjusts operating parameters according to environmental changes, ensuring stable and efficient operation even in extreme environments such as -25℃.

[0163] Case 2 Problem vs. Invention Solution

[0164] Traditional system problems include the lack of correlation between environmental and load data, leading to a 25% increase in energy consumption during transitional seasons.

[0165] The control strategy is rigid and cannot adapt to the differences in customer flow density in different areas of the mall; the temperature compliance rate in high-traffic areas is only 82%.

[0166] Advantage of proportion:

[0167] Multi-dimensional data fusion, real-time correlation analysis of environmental parameters, population density and regional temperature, and construction of load-environment mapping model;

[0168] Dynamic zoning control and AI algorithms automatically adjust the operating parameters of each zone module according to passenger flow distribution, increasing the temperature compliance rate to over 98%.

[0169] Adaptive load tracking allows the system to automatically identify load change trends and adjust operating strategies in advance, reducing response time by 70%.

[0170] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A split-type AI-linked multi-module heat pump cluster expansion equipment, characterized in that: Expansion equipment includes: The split-type multi-module architecture is based on a split design and includes several independent basic heat pump modules and expandable heat pump modules to form a multi-module cluster, with unified configuration of standardized mechanical interfaces, electrical interfaces and communication protocols. The all-dimensional cluster perception module collects environmental parameters of the cluster and operating parameters of all sub-modules within the cluster in real time. It normalizes the collected multi-source heterogeneous data to generate a standardized dataset and interacts with the AI ​​collaborative control module in real time based on a high-speed communication link. The cluster collaborative heat exchange module, based on a standardized dataset, intelligently analyzes the overall operating status and environmental adaptability of the multi-module cluster, and determines whether the overall cooling / heating effect of the cluster under the current environmental parameters meets the human body temperature requirements; if it does, it maintains the collaborative operation of each individual module; if it does not, it outputs adjustment suggestions. The cluster load throttling module, based on a standardized dataset and the operating parameters of each sub-module, judges the structural load of all modules in the cluster in real time, identifies modules with peak and low loads, and dynamically adjusts the branch flow of each sub-module. The AI-powered collaborative control module integrates the strategy suggestions from the cluster collaborative heat exchange module with the load judgment results from the cluster load throttling module. Through multi-dimensional collaborative analysis using AI algorithms, it generates targeted control electrical signals for collaborative control.

2. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 1, characterized in that: The mechanical structure features a quick-release flange and sealing buckle; the electrical structure uses a three-phase five-wire interface to prevent mis-insertion; the communication interface uses the Modbus TCP / Profinet industrial communication protocol; the basic heat pump module is a cluster carrying unit, with a uniform rated power of 5kW for each basic module; the expandable heat pump module is compatible with the basic module in terms of structural dimensions, interface specifications, and core parameters.

3. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 1, characterized in that: Environmental parameters are collected in real time by sensors, including ambient temperature and relative humidity. For all modules in the cluster, each module has an independent built-in sensing unit to collect core operating parameters in real time, including the operating parameters of the compressor, heat exchanger and throttling components. The operating parameters include energy consumption data and branch fluid parameters. Multi-source heterogeneous data normalization processing uses the min-max normalization algorithm to map data of different dimensions to the [0, 1] interval and unify the data format to obtain a dataset. Based on a weighted algorithm, ambient temperature and relative humidity are integrated into a comprehensive value of environmental parameters.

4. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 1, characterized in that, The specific analysis steps within the cluster-based collaborative heat exchange module include: Based on standardized datasets, environmental feature dimensions, cluster thermal output dimensions, and human comfort correlation dimensions are extracted. A three-dimensional feature space is constructed with environmental parameters as the X-axis, overall cluster load as the Y-axis, and cluster heat exchange efficiency COP value as the Z-axis. Based on historical operating data, the DBSCAN density clustering algorithm is used to divide the clusters into three types: comfortable and efficient clusters, sub-comfortable and adaptable clusters, and uncomfortable and adjusting clusters. The system calculates the coordinates of the current operating condition in three-dimensional space in real time. By calculating the distance between this point and the comfort and efficiency cluster, if the distance is less than or equal to the cluster radius (which is determined based on historical comfort operating condition statistics), the current heat exchange effect is determined to meet the requirements. If the distance exceeds the radius and falls into the discomfort adjustment cluster, the requirements are determined to be unmet. The system then adjusts the individual modules that do not meet the requirements based on the AI ​​collaborative linkage control module.

5. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 4, characterized in that: The DBSCAN density clustering algorithm is combined with historical operating data of the heat pump cluster for calibration; the calibration logic is grouped according to operating condition type; the algorithm execution standardizes the three-dimensional coordinates P of each group of environmental parameters, cluster load and COP value from the historical operating data. i =(X i ,Y i Z i X = Comprehensive environmental parameter value, Y = Overall cluster load, Z = COP value, P i Let P be the i-th point in the three-dimensional coordinate system; as a sample, construct a historical sample set D = {P1, P2, ..., P...} m }, where m is the total number of historical operating conditions; iterate through the sample set D, and for each unlabeled sample P i Calculate the number of samples in its ε-neighborhood, where ε is the neighborhood radius. If the number of samples in the neighborhood is greater than or equal to MinPts, then the minimum number of samples in the neighborhood is determined. MinPts is the minimum number of samples in the neighborhood, representing the minimum number of sample points contained within the ε-neighborhood. Label P. i With P as the core point i Expanding from the center, all directly reachable core points and boundary points within its neighborhood are grouped into the same cluster; the three types of clusters are then divided.

6. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 4, characterized in that: The distance calculation formula uses Mahalanobis distance to calculate the distance between the current working condition coordinate point and the comfort and efficiency cluster. The formula expression is: Where d M Let P be the Mahalanobis distance between the current 3D coordinate point P and the center μ of the comfort and efficiency cluster; P = (X, Y, Z) is the standardized 3D coordinate point of the current operating condition, and the minimum value of M is < 3; (P - μ) T The transpose of the difference vector P-μ converts the column vector to a row vector, satisfying the matrix multiplication dimension, μ = (μ X ,μ Y ,μ Z ) represents the three-dimensional center coordinates of the comfortable and efficient cluster, μ X Let μ be the mean of all sample X values ​​within the cluster. Y μ Z Similarly, based on historical comfort condition samples, Σ is calculated as a 3×3 covariance matrix of the historical samples of the comfort and efficiency cluster, describing the correlation between the three dimensions of environment, load, and COP.

7. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 1, characterized in that, The specific analysis steps within the cluster load throttling module are as follows: Through multi-dimensional load quantification analysis, combined with dynamic thresholds and algorithm verification, the system identifies peak, normal, and low-load modules. Based on the goal of load balancing, it implements differentiated flow adjustments under different load scenarios, gradually throttling and shifting load in peak modules, precisely increasing flow and improving efficiency in low-load modules, and adapting to the total flow before further optimization when the overall load fluctuates. The system is then regulated based on an AI-based collaborative control module.

8. The split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 7, characterized in that: Multi-dimensional load quantification analysis integrates standardized datasets with outdoor temperature and humidity data, and combines historical correlation characteristics for comprehensive judgment; The dynamic threshold is not a fixed value; it is based on historical load distribution patterns and is calibrated in real time as the overall cluster load and environment change. Algorithm verification uses a sliding window to analyze load trends. The multi-dimensional load quantification analysis data extracts the corresponding features of historical load and each parameter from a standardized dataset, collects outdoor temperature and humidity in real time through distributed sensors, and fuses them according to the parameter influence weights to comprehensively lock the load data of each module. The dynamic threshold determines the basic threshold range by statistically analyzing the load distribution range of different seasons and time periods in history. It then makes minor adjustments to the basic threshold based on the deviation between the current scenario and historical scenarios.

9. A collaborative control method for a split-type AI-linked multi-module heat pump cluster expansion equipment, applied to the split-type AI-linked multi-module heat pump cluster expansion equipment of any one of claims 1 to 8, characterized in that, The steps of coordinated control are as follows: S1. Multi-source data integration, synchronous data reception, including cooling / heating effect judgment results and parameter adjustment suggestions output by the cluster collaborative heat exchange module, and load status and flow adjustment information of each individual module fed back by the cluster load throttling module, to build a unified data storage pool; S2, AI collaborative analysis, based on a preset AI collaborative algorithm, performs hierarchical collaborative analysis to calculate the deviation amplitude between cooling / heating effect and human thermal comfort needs; combined with load balancing target, it comprehensively weighs the output information; and matches the corresponding control strategy to execute steady-state maintenance, load adjustment, heat exchange and load collaborative regulation for different scenarios. S3. Control signal generation and output: Convert the control strategy into electrical signal commands and transmit them in a directional manner to the controllers of each separate module and the actuators of the branch electric regulating valves for coordinated control.

10. The collaborative control method for the split-type AI-linked multi-module heat pump cluster expansion equipment according to claim 9, characterized in that: The specific steps of AI collaborative analysis in step S2 are as follows: The cooling / heating effect assessment results of the associated cluster collaborative heat exchange modules are combined with human thermal comfort standards to quantify the deviation amplitude between the current heat exchange effect and the demand. By integrating the load status and flow adjustment information of each individual module fed back by the cluster load throttling module, the types of load imbalance are analyzed, and the modules that need to be focused on regulation and the current flow allocation bottlenecks are identified. The weight values ​​are determined based on the preset priority of human body temperature comfort needs > load balance > energy efficiency optimization; The deviation level and load imbalance type are combined to form multiple scenario categories, and corresponding control strategies are matched for different combination scenarios. In step S3, the control strategy is converted into electrical signal commands by breaking down the control strategy into modular power and valve opening quantification parameters, encoding them according to the three-phase five-wire interface standard and Modbus TCP / Profinet protocol, and then distributing them to the target components and outputting signals in stages.

Images