Refrigeration working condition self-adaptive energy-saving intelligent control method based on artificial intelligence

Abstract

The present application relates to refrigeration working condition control technical field, especially to a kind of refrigeration working condition self-adaptive energy-saving intelligent control method based on artificial intelligence, the method first obtains target refrigeration space real-time monitoring data, is extracted flow field-thermal load coupling characteristic value by coupling characteristic analysis;Subsequently, the corresponding energy-saving optimization control model is matched, and initial control parameter is called from preset library;Then the dynamic response process of refrigeration space is simulated using initial parameter;Finally, initial parameter is dynamically adjusted according to simulation result, and optimal control strategy is generated to implement adaptive energy-saving control.The present application effectively solves the problems of response lag, rough adjustment and high energy consumption of traditional control under complex and variable working conditions by deeply integrating the coupling characteristics of unsteady flow field and transient thermal load, combining offline knowledge base and online simulation correction mechanism, significantly improves temperature uniformity and cold matching accuracy, and realizes efficient, stable and low-consumption operation of refrigeration system.

Description

Technical Field

[0001] This invention relates to the field of refrigeration operating condition control technology, and in particular to an artificial intelligence-based adaptive energy-saving intelligent control method for refrigeration operating conditions. Background Technology

[0002] With the rapid development of cold chain logistics and low-temperature storage technologies, energy efficiency optimization and temperature stability control of refrigeration systems have become important research directions in the field of industrial energy conservation. Traditional refrigeration systems mostly adopt PID control based on fixed setpoints or simple start-stop control strategies, which are difficult to cope with the frequently changing heat loads and complex unsteady flow field environments in refrigeration operations. Especially when goods are frequently moved in and out, doors are frequently opened and closed, or the ambient temperature and humidity fluctuate greatly, problems such as uneven distribution of cold energy, large temperature fluctuations, and frequent compressor start-stops can easily occur inside the refrigerated space, leading to increased system energy consumption, decreased food preservation quality, and shortened equipment lifespan.

[0003] In recent years, some studies have attempted to introduce intelligent control methods such as fuzzy control, model predictive control, and neural networks to improve system adaptability. However, most methods only focus on a single dimension of heat load change or airflow field, lacking effective modeling and coordinated control of the strong coupling between flow field dynamics and transient heat load. In fact, during refrigeration, airflow conditions (such as velocity distribution and turbulence intensity) directly affect cold transfer efficiency and temperature uniformity, while changes in heat sources, in turn, disturb the flow field structure, forming complex unsteady coupling relationships. Existing control strategies often ignore this coupling mechanism, resulting in limited control accuracy and insufficient exploitation of energy-saving potential.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an adaptive energy-saving intelligent control method for refrigeration conditions based on artificial intelligence. This method aims to solve the technical problem that existing refrigeration control technologies lack effective modeling and coordinated regulation of the strong coupling mechanism between unsteady flow fields and transient heat loads, which makes it difficult to balance temperature uniformity, real-time control, and system energy efficiency under complex and variable conditions.

[0006] To achieve the above objectives, the present invention provides an adaptive energy-saving intelligent control method for refrigeration operating conditions based on artificial intelligence, the method comprising:

[0007] Real-time operating condition monitoring data of the target cold storage space is acquired, and coupling characteristic analysis is performed on the monitoring data to obtain the coupling characteristic values ​​of flow field-heat load.

[0008] Match the energy-saving optimization control model corresponding to the coupling feature value, so as to call the initial control parameters from the preset control parameter library through the energy-saving optimization control model;

[0009] Simulate the dynamic response process corresponding to the target cold storage space based on the initial control parameters;

[0010] The initial control parameters are adjusted based on the dynamic response, and adaptive energy-saving control of the refrigeration condition is performed based on the adjusted control parameters.

[0011] Optionally, the monitoring data includes flow field characteristic parameters and heat load characteristic parameters. The process of acquiring real-time operating condition monitoring data of the target cold storage space and performing coupling characteristic analysis on the monitoring data to obtain flow field-heat load coupling characteristic values ​​includes:

[0012] The flow field characteristic parameters and heat load characteristic parameters of the target cold storage space are obtained, wherein the flow field characteristic parameters include flow velocity distribution parameters and turbulence fluctuation parameters, and the heat load characteristic parameters include instantaneous rate of change parameters and heat leakage increment parameters;

[0013] The particle image velocimetry method is used to extract the coordinates of the feature nodes from the coupling feature parameters, and the flow field transport equation and the heat load coupling equation are constructed based on the coordinates of the feature nodes to obtain the coupling coefficient parameters.

[0014] The coupling feature vector is obtained based on the coupling coefficient parameter, and the coupling feature vector is normalized to obtain the coupling feature value. The coupling feature vector includes the flow field pulsation intensity, the heat load fluctuation amplitude, and the transport lag phase difference.

[0015] Optionally, the step of calling initial control parameters from a preset control parameter library through the energy-saving optimization control model includes:

[0016] Obtain the parameter features corresponding to each combination of control parameters in the control parameter library, and calculate the similarity between each parameter feature and the coupled feature value through the mapping relationship preset in the energy-saving optimization control model;

[0017] When the similarity exceeds a preset threshold, a preset energy-saving optimization rule is invoked to optimize the current parameter combination and generate the initial control parameters corresponding to the optimized parameter combination.

[0018] Optionally, the step of simulating the dynamic response process corresponding to the target cold storage space based on the initial control parameters includes:

[0019] A virtual cold storage space thermodynamic flow field model is constructed based on the discrete compressor frequency and fan speed control points of the initial control parameters. The virtual cold storage space thermodynamic flow field model is then meshed using hexahedral mesh elements to obtain a meshed model.

[0020] The SIMPLE algorithm is used to establish the flow field-thermal coupled transport equation, and the temperature field data and velocity field data of the mesh model are obtained by calculating the flow field-thermal coupled transport equation and the CFD solver.

[0021] The temperature field data and flow velocity field data are subjected to feature extraction by a three-layer convolutional neural network to obtain the transient change feature mapping relationship between the temperature field data and the flow velocity field data;

[0022] The real-time operating data collected by the sensor is processed according to the feature mapping relationship to obtain the temperature fluctuation data and cold supply and demand difference data of the target cold storage space during the control process.

[0023] By fusing the temperature fluctuation data and the cooling supply-demand difference data using a Kalman filter, the actual heat load change curve and the actual flow field distribution curve are obtained.

[0024] The reliability of the initial control parameters is verified based on the actual heat load change curve, the actual flow field distribution curve, the temperature field data, and the flow velocity field data.

[0025] Optionally, adjusting the initial control parameters based on the dynamic response and performing adaptive energy-saving control of the refrigeration condition based on the adjusted control parameters includes:

[0026] When verifying the reliability of the initial control parameters, a fuzzy evaluator based on confidence intervals is used to calculate the temperature uniformity evaluation parameters and system energy consumption evaluation parameters corresponding to the initial control parameters, and a sequence of parameters to be adjusted is generated based on the comparison results of the temperature uniformity evaluation parameters and the system energy consumption evaluation parameters with preset thresholds respectively.

[0027] A gradient boosting tree regressor is trained based on the sequence of parameters to be adjusted. The parameter adjustment amount is predicted by the gradient boosting tree regressor. The parameter values ​​of the initial control parameters are adjusted according to the parameter adjustment amount to obtain the discrete control point coordinates of the updated control parameters.

[0028] The coordinates of the discrete control points for the updated control parameters are fitted using NURBS curves, and the speed planning is dynamically adjusted based on the NURBS curves under the constraints of the maximum allowable temperature fluctuation and the maximum load rate of the compressor to obtain the speed planning curve.

[0029] The curvature values ​​at each discrete point of the speed planning curve are calculated. Transition points are inserted in the curvature abrupt change region by spline interpolation to obtain the adjusted control parameters. Adaptive energy-saving control of the refrigeration condition is then performed based on the adjusted control parameters.

[0030] Optionally, before acquiring the real-time operating condition monitoring data of the target cold storage space, the method further includes:

[0031] The enclosure structure parameters and cargo stacking parameters of the cold storage are obtained to construct an offline simulation model of the unsteady flow field coupled transient heat load corresponding to the cold storage. The offline simulation model is analyzed to obtain the temperature distribution cloud map and energy consumption distribution cloud map corresponding to the cold storage conditions under each preset initial control parameter.

[0032] The discrete points of each initial control parameter in the temperature distribution cloud map and the gradient of change in the energy consumption distribution cloud map are extracted respectively, and the temperature change rate and energy consumption change rate corresponding to each initial control parameter at each discrete point are calculated to obtain the dynamic response curve of heat load under different initial control parameters for the refrigeration condition.

[0033] Based on the dynamic response curve of the heat load, multiple first candidate control parameters are selected from the initial control parameters;

[0034] The control parameters in each of the first candidate control parameters are optimized according to the preset parameter optimization algorithm to obtain several optimized control parameter combinations and save them to the preset control parameter library.

[0035] Optionally, the enclosure structure parameters include thermal conductivity and heat storage coefficient, and the cargo stacking parameters include porosity and volumetric heat capacity. The process of obtaining the enclosure structure parameters and cargo stacking parameters of the cold storage facility to construct an offline simulation model of the unsteady flow field coupled with transient heat load corresponding to the cold storage facility, and analyzing the offline simulation model to obtain the temperature distribution cloud map and energy consumption distribution cloud map corresponding to each preset initial control parameter under the cold storage conditions, includes:

[0036] The thermal conductivity, heat storage coefficient, and cargo stacking porosity of the enclosure structure of the cold storage are obtained in order to calculate the flow field resistance matrix and heat permeation matrix corresponding to the cold storage space. The flow field resistance matrix is ​​determined by the porosity and stacking height, and the heat permeation matrix is ​​determined by the thermal conductivity and heat storage coefficient.

[0037] The nodal thermal disturbance is obtained by the flow field resistance matrix and the thermal penetration matrix and mapped to the mesh nodes to establish a finite volume interpolation equation. The mesh nodes are divided into hexahedral elements and the mesh size is reduced to one-third of the basic mesh size in the near-wall region.

[0038] The flow field gradient value is calculated based on the finite volume interpolation equation to obtain the flow field-thermal coupling control equation, which is the offline simulation model of unsteady flow field coupled transient thermal load; wherein, the flow field gradient value is determined by the velocity gradient vector and the temperature gradient vector;

[0039] Based on the flow field-thermal coupling control equations, a set of nonlinear cold energy supply and demand equations is established to obtain temperature distribution cloud maps and energy consumption distribution cloud maps corresponding to each preset initial control parameter under the refrigeration condition; wherein, the set of nonlinear cold energy supply and demand equations is solved using the SIMPLEC algorithm until the residual is less than the preset value.

[0040] Optionally, calculating the rate of temperature change and rate of energy consumption change corresponding to each of the initial control parameters at each discrete point to obtain the dynamic response curve of heat load under different initial control parameters for the refrigeration condition includes:

[0041] Obtain the control sequence corresponding to each of the initial control parameters, and obtain the coordinate dataset of discrete operating points on the control sequence; wherein, different initial control parameters correspond to different compressor operating frequencies and fan speeds;

[0042] The instantaneous cooling output value is obtained by fitting the discrete operating points with a Bézier curve; wherein the instantaneous cooling output value is calculated through the mapping relationship between control parameters and cooling output;

[0043] A thermodynamic equation is established based on the instantaneous cooling output value, and the heat load change rate at discrete points is obtained by solving the Euler integral formula; wherein the heat load change rate is obtained by integrating the heat load over time.

[0044] A state equation is constructed based on the rate of change of heat load at the discrete points and the temperature fluctuation on the temperature distribution cloud map. The temperature dynamic change function is obtained by solving the least squares method, and the dynamic response curve of heat load is obtained based on the temperature dynamic change function.

[0045] Optionally, the step of selecting multiple first candidate control parameters from a plurality of initial control parameters based on the dynamic response curve of the heat load includes:

[0046] The temperature fluctuation amplitude corresponding to each of the initial control parameters is determined by the dynamic response curve of the heat load. Constraint equations are constructed based on the preset maximum allowable temperature fluctuation of the refrigeration condition and the current temperature fluctuation amplitude corresponding to each of the initial control parameters, so as to obtain the constraint optimization parameters through conjugate gradient iteration calculation.

[0047] Based on the constraint optimization parameters, a multi-objective optimization function is established with a temperature uniformity objective function and an energy consumption objective function, and the mapping relationship between the parameter vector and the constraint response vector corresponding to each initial control parameter is obtained through a support vector machine with radial basis function kernel.

[0048] Based on the mapping relationship, a set of constrained optimization equations is constructed using KKT conditions, and the optimal candidate control parameters and Lagrange multipliers corresponding to each initial control parameter are obtained by solving the sequential quadratic programming method.

[0049] If the norm difference between the optimal candidate control parameter and the initial parameter corresponding to the initial control parameter is less than a preset convergence threshold, then the optimal candidate control parameter is stored as the first candidate control parameter.

[0050] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an artificial intelligence-based adaptive energy-saving intelligent control program for refrigeration conditions. When the artificial intelligence-based adaptive energy-saving intelligent control program for refrigeration conditions is executed by a processor, it implements the steps of the artificial intelligence-based adaptive energy-saving intelligent control method for refrigeration conditions as described above.

[0051] This invention provides an artificial intelligence-based adaptive energy-saving intelligent control method for refrigeration conditions. The method achieves precise perception and optimized regulation of the complex dynamic characteristics of refrigeration conditions by constructing a flow field-heat load coupling characteristic analysis and an intelligent adaptive control closed loop. Its advantages include: effectively overcoming the problems of lag and coarse adjustment in traditional control strategies under the coupling of unsteady flow fields and transient heat loads; significantly improving temperature field uniformity and the accuracy of cold energy supply and demand matching; reducing ineffective energy consumption and frequent start-stop of key equipment such as compressors; and thus significantly improving the overall energy efficiency of the system while ensuring refrigeration quality. It also possesses strong adaptability to various operating conditions and real-time optimization capabilities, making it suitable for diverse and complex actual refrigeration operating environments. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating an embodiment of the AI-based adaptive energy-saving intelligent control method for refrigeration conditions according to the present invention.

[0053] Figure 2 This is a flowchart illustrating another embodiment of the AI-based adaptive energy-saving intelligent control method for refrigeration conditions according to the present invention.

[0054] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0055] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0056] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the AI-based adaptive energy-saving intelligent control method for refrigeration conditions according to the present invention.

[0057] In one embodiment, the AI-based adaptive energy-saving intelligent control method for refrigeration conditions includes:

[0058] Step S100: Obtain real-time operating condition monitoring data of the target cold storage space, and perform coupling characteristic analysis on the monitoring data to obtain the coupling characteristic value of flow field-heat load.

[0059] The target cold storage space can be a closed or semi-closed low-temperature storage area requiring temperature control and energy efficiency optimization. It can be used as the target of control methods, and its internal thermodynamic and hydrodynamic states directly affect food preservation quality and system energy consumption. In an exemplary embodiment, the target cold storage space may include, but is not limited to, cold storage areas, refrigerated transport compartments, and supermarket refrigerated display areas. Real-time operating condition monitoring data can be a multi-dimensional collection of sensor information reflecting the current operating status of the target cold storage space. This data can provide raw input for subsequent coupled feature analysis, characterizing the real-time status of unsteady flow fields and transient heat loads. Furthermore, real-time operating condition monitoring data can be collected in real time through temperature sensors, wind speed sensors, humidity sensors, door open / close status detectors, and compressor operating status monitoring modules deployed within the cold storage space. For example, real-time operating condition monitoring data may include, but is not limited to, temperature field distribution data, air velocity vector data, and heat source change event data. The coupling characteristic values ​​of the flow field and heat load can be high-dimensional feature vectors extracted from real-time operating condition monitoring data using artificial intelligence algorithms. These vectors characterize the dynamic interaction between airflow state and heat load, and can be used as a unified state representation of the system's dynamic characteristics, supporting subsequent model matching and parameter calling. In a specific embodiment, the coupling characteristic values ​​of the flow field and heat load can be used to perform spatiotemporal correlation modeling of multi-source monitoring data based on deep neural networks or graph neural networks, quantifying the impact of flow field structure on heat transfer efficiency and the feedback of thermal disturbances on flow field reconstruction. For example, the coupling characteristic values ​​of the flow field and heat load may include, but are not limited to, flow field disturbance sensitivity characteristics, heat load response delay characteristics, and cold energy transport efficiency characteristics.

[0060] Step S200: Match the energy-saving optimization control model corresponding to the coupling feature value, so as to call the initial control parameters from the preset control parameter library through the energy-saving optimization control model.

[0061] The energy-saving optimization control model can be a pre-trained artificial intelligence model used to map coupling feature values ​​to the optimal control strategy. It can be used to achieve intelligent decision-making from complex states to control parameters, replacing traditional fixed-rule control logic. In an exemplary embodiment, the energy-saving optimization control model can be trained offline through reinforcement learning or transfer learning on a large amount of historical operating data and in a simulation environment, possessing the ability to generalize mapping to different coupling feature values. Furthermore, the energy-saving optimization control model can receive flow field-heat load coupling feature values ​​as input and output an index or weight pointing to a preset control parameter library. Exemplary examples include, but are not limited to, deep Q-network models, long short-term memory network controllers, and graph attention strategy networks. The preset control parameter library can be a data structure storing verified and effective control parameter combinations under various typical operating conditions. It can be used to provide a callable set of parameter candidates for the energy-saving optimization control model, avoiding the computational overhead of online real-time optimization. In a specific embodiment, the preset control parameter library can be constructed through historical operating data clustering, simulation optimization, or expert experience summarization, with each entry corresponding to a recommended parameter set under a specific coupling feature interval. For example, the preset control parameter library may include, but is not limited to, compressor start / stop threshold groups, fan speed gear groups, evaporator liquid supply adjustment coefficient groups, etc.

[0062] The initial control parameters can be a set of control commands initially selected by the energy-saving optimization control model from the control parameter library. These parameters can serve as the starting point for subsequent dynamic response simulation and parameter adjustment, reflecting the initial response strategy to the current operating condition. Furthermore, the initial control parameters can be the input conditions for dynamic response process simulation. Matching the energy-saving optimization control model corresponding to the coupled feature values ​​can involve inputting the coupled feature values ​​into the energy-saving optimization control model, triggering the model's internal mapping logic to determine the applicable control strategy. In a specific embodiment, matching the energy-saving optimization control model corresponding to the coupled feature values ​​can be achieved by weighting the outputs of multiple sub-models through a soft voting mechanism in model ensemble, or by using nearest neighbor retrieval to match models corresponding to similar historical operating conditions in the feature space, thereby enabling intelligent transformation from complex states to control intentions. The initial control parameters can be retrieved from the preset control parameter library by the energy-saving optimization control model outputting indices or weights, and then selecting or interpolating the initial control parameters from the control parameter library. Furthermore, by calling initial control parameters from a preset control parameter library through the energy-saving optimization control model, it is possible to directly retrieve the parameter group corresponding to the nearest feature cluster or to perform weighted fusion of multiple parameter groups based on feature similarity, thereby quickly generating a preliminary control scheme that conforms to the characteristics of the current operating conditions.

[0063] Step S300: Simulate the dynamic response process of the target cold storage space based on the initial control parameters.

[0064] The dynamic response process can be a predicted trajectory of the evolution of the temperature field, flow field, and equipment state within the target cold storage space over time, given initial control parameters. This can be used to predict the actual effect of control actions and identify potential over-adjustment, under-adjustment, or oscillation risks. In an exemplary embodiment, the dynamic response process can use a lightweight physical model or a data-driven surrogate model to perform forward simulation of the cold storage system, simulating system behavior over a future period. Furthermore, the simulation results of the dynamic response process can be used to guide the adjustment of the initial control parameters. For example, the dynamic response process may include, but is not limited to, temperature uniformity evolution curves, compressor start-stop frequency predictions, and cold energy supply-demand deviation sequences. Simulating the dynamic response process corresponding to the target cold storage space based on the initial control parameters can involve using the initial control parameters as input to drive the surrogate model to simulate the system's state evolution over a future period. In a specific embodiment, simulating the dynamic response process corresponding to the target cold storage space based on the initial control parameters can be achieved by using a reduced-order physical model for rapid thermal-fluid coupling simulation or by calling a pre-trained recurrent neural network for sequence prediction, thereby allowing for prediction of the control effect before execution and avoiding blind adjustments.

[0065] Step S400: Adjust the initial control parameters according to the dynamic response, and perform adaptive energy-saving control of the refrigeration condition according to the adjusted control parameters.

[0066] The adjusted control parameters can be the final execution parameters formed by correcting the initial control parameters based on the simulation results of the dynamic response process. These parameters ensure consistency between the actual control actions and the system's physical response, improving control accuracy and stability. The refrigeration operating condition refers to the comprehensive operating state of the target refrigerated space at a specific moment, including environmental disturbances, load changes, and equipment responses. It can be used as the direct control object for adaptive energy-saving control, and its stability and energy efficiency level are the ultimate manifestation of the method's effectiveness. Adjusting the initial control parameters based on the dynamic response can involve comparing the simulated dynamic response with the desired target (such as temperature uniformity or energy minimization) and correcting the control parameters through optimization algorithms. In a specific embodiment, adjusting the initial control parameters based on the dynamic response can be achieved by fine-tuning parameters based on gradient backpropagation or proportionally correcting the fan speed or compressor duty cycle using heuristic rules. This allows for forward-looking optimization of the control parameters and improves execution accuracy. Adaptive energy-saving control of the refrigeration operating condition based on the adjusted control parameters can involve sending the final determined control parameters to actuators such as the compressor, fan, and expansion valve for closed-loop control. Furthermore, adaptive energy-saving control of refrigeration conditions based on the adjusted control parameters can be implemented through closed-loop regulation by the actuator, thereby achieving precise matching between cooling supply and dynamic heat demand, and suppressing temperature fluctuations and frequent equipment start-ups and shutdowns.

[0067] For example, in the scenario of a supermarket freezer operating with frequent door openings, the AI-based adaptive energy-saving intelligent control method for refrigeration conditions in this embodiment can be as follows: The system collects temperature and wind speed data at multiple points inside the freezer in real time, identifies the coupling state of weakened local eddies and intensified heat intrusion through coupling feature analysis, and generates corresponding coupling feature values; the energy-saving optimization control model calls a high wind speed compensation strategy from the parameter library as initial parameters accordingly; the system then simulates the temperature recovery process over the next 5 minutes under this strategy, finds that there is still a temperature lag in the corner area, and automatically increases the speed of the side wall fan and extends the cooling cycle to form adjusted control parameters; finally, the parameters are executed to prevent the compressor from frequently starting due to short-term high temperatures while maintaining food quality.

[0068] In one embodiment, the monitoring data includes flow field characteristic parameters and heat load characteristic parameters. Real-time operating condition monitoring data of the target cold storage space is acquired, and coupled characteristic analysis is performed on the monitoring data to obtain coupled characteristic values ​​of the flow field and heat load, including:

[0069] The flow field characteristic parameters and heat load characteristic parameters of the target cold storage space are obtained. The flow field characteristic parameters include velocity distribution parameters and turbulence fluctuation parameters, and the heat load characteristic parameters include instantaneous rate of change parameters and heat leakage increment parameters.

[0070] The flow field characteristic parameters can be a set of quantitative indicators describing the airflow state within the cold storage space, serving as the fluid dynamics basis for characterizing cold energy transport capacity and temperature uniformity. Further, these flow field characteristic parameters may include, but are not limited to, velocity distribution parameters, turbulence fluctuation parameters, and vorticity intensity parameters. The heat load characteristic parameters can be a set of quantitative indicators reflecting the dynamic characteristics of thermal disturbances within and at the boundaries of the cold storage space, used to characterize the impact of transient heat demand changes on system energy efficiency and temperature stability. In an exemplary embodiment, the heat load characteristic parameters may include, but are not limited to, instantaneous rate of change parameters, heat leakage increment parameters, and cargo heat release rate parameters. Velocity distribution parameters can be spatial distribution data of the magnitude and direction of air velocity at various locations within the cold storage space, used to assess the cold air coverage area and the risk of dead zones. Turbulence fluctuation parameters can be statistical quantities characterizing the intensity of unsteady velocity fluctuations in airflow, used to reflect cold energy mixing efficiency and local temperature fluctuation potential. The instantaneous rate of change parameters can be an indicator of the rate at which heat load changes abruptly per unit time, used to identify the intensity of thermal shock caused by events such as door opening or cargo entry / exit. The heat leakage increment parameter can be the additional heat inflow caused by heat transfer or sealing failure of the building envelope, and can be used to quantify the additional impact of environmental temperature and humidity fluctuations on system load.

[0071] The particle image velocimetry method is used to extract the coordinates of feature nodes from the coupling feature parameters, and the flow field transport equation and the heat load coupling equation are constructed based on the coordinates of the feature nodes to obtain the coupling coefficient parameters.

[0072] The feature node coordinates can be the spatial locations of key flow structures identified from flow field images using particle image velocimetry, providing a representative set of discrete points for constructing the physical equations. In one specific embodiment, the feature node coordinates may include, but are not limited to, vortex center coordinates, shear layer boundary coordinates, and recirculation zone vertex coordinates. Further, the feature node coordinates can be obtained by using displacement images of tracer particles under continuous laser irradiation, calculating the local velocity field using a cross-correlation algorithm, and extracting the coordinates of feature points such as the vortex center and recirculation zone boundary. The flow field transport equation can be a partial differential equation describing the convection-diffusion process of cold energy in airflow, used to model the dynamic evolution of the temperature field with the flow field. For example, the flow field transport equation can be solved simultaneously with the thermal load coupling equation to jointly determine the coupling coefficient parameters. The thermal load coupling equation can be a mathematical model describing how external thermal disturbances affect the flow field structure through boundary conditions or source terms, used to characterize the feedback mechanism of thermal load changes on flow field reconstruction.

[0073] The particle image velocimetry method is used to extract feature node coordinates from coupled feature parameters. This can be achieved by scattering tracer particles within a refrigerated space, using dual-pulse laser illumination and high-speed camera imaging to calculate the particle displacement field and identify the locations of flow feature structures. Furthermore, extracting feature node coordinates from coupled feature parameters using particle image velocimetry can be achieved by using time-series PIV to extract dynamic vortex structure trajectories and combining intrinsic orthogonal decomposition to identify spatial nodes corresponding to dominant modes. This yields spatially significant feature points to support subsequent equation construction. The flow field transport equations and thermal load coupling equations are constructed based on the feature node coordinates. This can be done by establishing a set of local coupling equations based on the conservation of mass, momentum, and energy, with the feature nodes as discrete control volume centers. Further, the flow field transport equations and thermal load coupling equations can be constructed by using the finite volume method to build control equations around the feature nodes and using graph neural networks to map node relationships to differential operators. This allows for explicit modeling of the dynamic interaction mechanism between cold transport and thermal disturbance.

[0074] The coupling feature vector is obtained based on the coupling coefficient parameter, and the coupling feature vector is normalized to obtain the coupling feature value. The coupling feature vector includes the flow field pulsation intensity, the heat load fluctuation amplitude, and the transport lag phase difference.

[0075] The coupling coefficient parameter can be a scalar or tensor parameter, obtained by simultaneously solving the flow field transport equation and the heat load coupling equation, characterizing the intensity of their interaction. It can serve as a direct basis for generating the coupling eigenvector. The coupling eigenvector can be a vector representation derived from the coupling coefficient parameter, containing information on multidimensional coupling effects. It can be used to structurally express the core dynamic characteristics of the flow field and heat load interaction. For example, the coupling eigenvector includes, but is not limited to, flow field fluctuation intensity, heat load fluctuation amplitude, and transport lag phase difference. The flow field fluctuation intensity can be a dimensionless measure of the root mean square value of the turbulent fluctuation velocity, which can be used to reflect the intensity of local cold mixing. The heat load fluctuation amplitude can be the maximum deviation of the heat load from its steady-state value, which can be used to characterize the potential impact of thermal disturbances on system stability. The transport lag phase difference can be the phase shift between the occurrence of the thermal disturbance and the peak value of the flow field response, which can be used to reveal the dynamic delay characteristics in the cold transport process.

[0076] Taking the frequent opening of cold storage doors leading to heat intrusion as an example, the AI-based adaptive energy-saving intelligent control method for refrigeration conditions in this embodiment can be as follows: when the cold storage door is opened, hot and humid air from outside rushes in. The system simultaneously collects wind speed and temperature gradient data at multiple points inside to obtain flow velocity distribution parameters and instantaneous rate of change parameters. The coordinates of the center of the thermal plume vortex formed in the doorway area are identified through a preset PIV system. Based on these coordinates, a local flow field transport equation and a heat load coupling equation are constructed to obtain the thermally induced flow field distortion coefficient. This generates a coupled feature vector containing high turbulence pulsation intensity, large heat load fluctuation amplitude, and significant transport lag phase difference. After normalization, the vector is input into the energy-saving optimization control model to trigger a strategy of enhancing sidewall air supply in advance and extending the refrigeration cycle, effectively suppressing temperature rise and compressor overload.

[0077] In one embodiment, the initial control parameters are retrieved from a preset control parameter library through an energy-saving optimization control model, including:

[0078] Obtain the parameter features corresponding to each combination of control parameters in the control parameter library, and calculate the similarity between each parameter feature and the coupled feature value through the pre-set mapping relationship in the energy-saving optimization control model;

[0079] When the similarity exceeds a preset threshold, the preset energy-saving optimization rules are invoked to optimize the current parameter combination and generate the initial control parameters corresponding to the optimized parameter combination.

[0080] The control parameter library can be a data structure storing verified and effective control parameter combinations under various typical operating conditions. It can provide a callable set of parameter candidates for the energy-saving optimization control model, avoiding the computational overhead of online real-time optimization. A control parameter combination can be a complete set of control command configurations, including coordinated setpoints for actuators such as compressors, fans, and expansion valves. It can represent a collaborative working mode of equipment under a specific operating strategy. Furthermore, control parameter combinations can include, but are not limited to, rapid cooling combinations, uniform temperature control combinations, and low-frequency start-stop combinations. Parameter features can be quantifiable representation vectors formed by abstracting control parameter combinations, used to describe the semantic applicability of the combination under coupled operating conditions. In this embodiment, parameter features can be mapped from the original parameter combinations to a feature space isomorphic to the coupled feature values ​​through an embedded network or feature encoder. For example, parameter features can include features based on historical response statistics, features based on physical model inversion, and prototype features based on cluster centers. The mapping relationship can be a mathematical or algorithmic mechanism preset within the energy-saving optimization control model, used to measure the semantic correlation between parameter features and coupled feature values. It can be used to achieve accurate matching between multi-dimensional coupled states and historical control strategies. In one specific embodiment, the mapping relationship can be learned offline by learning the correspondence between different operating condition characteristics and effective control strategies, forming a distance metric or similarity function.

[0081] The preset threshold can be a boundary value used to determine whether the parameter features and coupling feature values ​​are sufficiently similar. It can be used to filter out potentially applicable control parameter combinations, avoiding ineffective optimization calculations. The energy-saving optimization rules can be a set of preset energy efficiency-oriented constraints or adjustment logic used to refine successfully matched parameter combinations. In this embodiment, the energy-saving optimization rules can be formed based on thermodynamic optimization criteria, equipment wear models, or reinforcement learning strategies. For example, energy-saving optimization rules may include compressor start-stop interval extension rules, fan speed gradient smoothing rules, and cooling capacity distribution equalization rules. The optimized parameter combination can be a control parameter combination corrected by the energy-saving optimization rules, possessing higher energy efficiency adaptability. It can be used as a direct source for generating initial control parameters, improving the quality of the control starting point.

[0082] Taking the switching between low-load and high-disturbance operating conditions at night and during the day in a cold storage facility as an example, the AI-based adaptive energy-saving intelligent control method for cold storage conditions in this embodiment can be as follows: When the system detects a large influx of goods into the warehouse in the early morning, resulting in a sudden increase in heat load and frequent door openings, coupling feature analysis generates high-disturbance coupling feature values; the control unit traverses the parameter features of each control parameter combination in the parameter library, and through mapping relationship calculation, it finds that the feature of a certain rapid cooling combination has a similarity of 0.82 with the current coupling feature value, exceeding the preset threshold of 0.75; then, the energy-saving optimization rules are invoked to moderately reduce the compressor starting frequency while maintaining high wind speed to avoid current surges, generating an optimized parameter combination; this combination is converted into initial control parameters for subsequent simulation and execution, ensuring both temperature recovery speed and reducing equipment stress.

[0083] In one embodiment, the dynamic response process corresponding to the target cold storage space is simulated based on initial control parameters, including:

[0084] A virtual cold storage space thermodynamic flow field model is constructed based on the discrete compressor frequency and fan speed control points of the initial control parameters. The virtual cold storage space thermodynamic flow field model is then meshed using hexahedral mesh elements to obtain a meshed model.

[0085] The discrete compressor frequency can be one of a limited number of selectable operating frequency ranges for the compressor in variable frequency control mode. It serves as a key component of the initial control parameters, determining the cooling capacity output of the refrigeration system. The fan speed control point can be one of several preset selectable speed settings in the fan control system. It can be used to adjust the intensity and distribution of airflow, affecting cooling capacity transport efficiency and temperature uniformity. The virtual cold storage space thermodynamic flow field model can be a digital twin model built based on the initial control parameters to simulate the coupling behavior of airflow and heat transfer within the cold storage space. It can provide a high-fidelity simulation environment for dynamic response processes and support explicit calculations of unsteady coupling mechanisms. Furthermore, the virtual cold storage space thermodynamic flow field model can be based on the target cold storage space geometry, using the discrete compressor frequency and fan speed control point as boundary condition inputs to establish a physical model including key components such as evaporators, cargo stacking, and door areas. The hexahedral mesh element can be a regular hexahedral computational unit used for spatial discretization of the virtual cold storage space thermodynamic flow field model. It can improve the numerical stability and computational efficiency of CFD solutions, and is particularly suitable for meshing regular geometric regions. The mesh partitioning model can be a virtual cold storage space thermodynamic flow field model discretized by hexahedral mesh elements, which can be used to provide the computational domain basis for the numerical solution of the flow field-thermal coupled transport equation.

[0086] The SIMPLE algorithm was used to establish the flow field-thermal coupled transport equation, and the temperature field data and velocity field data of the meshed model were obtained by using the flow field-thermal coupled transport equation and CFD solver.

[0087] The SIMPLE algorithm is a semi-implicit pressure correction algorithm for solving the coupled Navier-Stokes equations and energy equations of incompressible fluids. It can be used to achieve iterative coupling of flow and temperature fields, ensuring consistency between mass and energy conservation. The flow-thermal coupled transport equations can be a set of partial differential equations describing the interaction between airflow and heat transfer in a cold storage space. They can be used to accurately characterize the physical processes of cold transport and turbulent mixing under unsteady conditions. Furthermore, the flow-thermal coupled transport equations can consist of continuity, momentum, and energy equations, which are solved by coupling using the SIMPLE algorithm. The CFD solver is a computational fluid dynamics software module for numerically solving the flow-thermal coupled transport equations. It can output high spatiotemporal resolution temperature and velocity field data to support subsequent feature extraction. The temperature field data can be the set of temperature values ​​at each node in the meshed model at the simulation time step. It can be used to reflect the cold distribution state and evaluate temperature uniformity and fluctuation trends. Velocity field data can be the set of air velocity vectors of each node in the meshed model at the simulation time step. It can be used to characterize the airflow structure and affect the local heat transfer efficiency and cooling coverage.

[0088] A three-layer convolutional neural network is used to extract features from temperature field data and flow velocity field data to obtain the transient change feature mapping relationship between temperature field data and flow velocity field data.

[0089] The three-layer convolutional neural network (CNN) can be a deep neural network structure containing three convolutional layers, used to process spatially correlated field data. It can be used to establish a nonlinear transformation from raw field data to transient feature mapping. Furthermore, the three-layer CNN can extract features layer by layer from three-dimensional spatial snapshots of temperature and velocity fields, capturing coupling patterns such as local gradients, vortex structures, and thermal boundary layers. The transient feature mapping can be a high-dimensional correlation representation of the evolutionary laws of temperature and velocity fields in the time dimension, learned by the three-layer CNN. This can be used as a bridge connecting simulation data and real-time sensor data, supporting dynamic interpretation of actual operating states. Extracting features from temperature and velocity field data using a three-layer CNN to obtain the transient feature mapping between them can be achieved by using field data from multiple time steps as input tensors and extracting joint spatial-temporal features through convolutional layers. Furthermore, this operation can be achieved by using 3D convolution kernels to process both spatial and temporal dimensions simultaneously, and by employing a dual-stream network to process and fuse temperature and velocity fields separately, thereby uncovering coupled dynamic patterns that are difficult to identify in traditional CFD post-processing.

[0090] The real-time operating data collected by the sensor is processed according to the feature mapping relationship to obtain the temperature fluctuation data and the cold supply and demand difference data of the target cold storage space during the control process.

[0091] Temperature fluctuation data can be a dynamic sequence of temperature deviations from the set value within the target cold storage space, derived from real-time operating data based on feature mapping relationships. This data can be used to quantify the temperature stability level during the control process. Cooling capacity supply-demand difference data can be an instantaneous deviation sequence between the cooling capacity of the refrigeration system and the actual heat load demand, calculated based on feature mapping relationships. This data can reflect the accuracy of cooling capacity matching and guide control parameter adjustments. Processing the real-time operating data collected by sensors according to feature mapping relationships to obtain temperature fluctuation data and cooling capacity supply-demand difference data of the target cold storage space during the control process can project sparse sensor data onto the feature mapping space and inversely deduce the overall dynamic deviation index, thereby enabling intelligent inference from limited measurement points to global performance indicators.

[0092] By fusing temperature fluctuation data and cooling supply-demand difference data using a Kalman filter, the actual heat load change curve and the actual flow field distribution curve are obtained.

[0093] The Kalman filter can be a recursive state estimation algorithm used to fuse noisy multi-source observation data and estimate the true state of the system. This can improve the accuracy of state estimation and suppress the interference of sensor noise on control decisions. Furthermore, the Kalman filter can use temperature fluctuation data and cooling supply-demand difference data as input, combined with the system dynamic model, to recursively estimate the actual heat load and flow field state without noise interference. The actual heat load variation curve can be a continuous curve reflecting the actual thermal disturbance intensity of the target cold storage space over time, obtained after fusion by the Kalman filter. This can be used as a key basis for verifying whether the initial control parameters meet the heat demand. The actual flow field distribution curve can be a spatiotemporal distribution curve characterizing the actual evolution trend of the airflow structure within the cold storage space, obtained after fusion by the Kalman filter. This can be used to evaluate whether the current fan control strategy maintains effective cooling capacity transport. Verifying the reliability of initial control parameters based on actual heat load change curves, actual flow field distribution curves, temperature field data, and velocity field data can involve comparing the consistency between simulation predictions and fusion estimation results, determining whether there is over-adjustment, under-adjustment, or oscillation risk, thereby forming a closed-loop verification mechanism for control parameters to ensure the rationality of the strategy before execution.

[0094] Taking the disturbance condition of bulk goods entering a cold storage facility as an example, the AI-based adaptive energy-saving intelligent control method for cold storage conditions in this embodiment can be as follows: When a cold storage facility receives a large number of ambient temperature goods in a short period of time, the system constructs a virtual cold storage space thermodynamic flow field model based on the high-frequency compressor setting and high fan speed setting in the initial control parameters. After hexahedral meshing, the simulation by the CFD solver reveals that there is heat accumulation in the central region. A three-layer convolutional neural network extracts the coupling characteristics between this heat accumulation and the low-speed recirculation zone, forming a transient mapping relationship. This mapping is used to process real-time temperature sensor data and derive the cold load supply and demand difference data. The Kalman filter fuses this data with temperature fluctuation data to estimate that the actual heat load peak is higher than expected and that there is a dead zone in the flow field distribution. Based on this, the system determines that the initial control parameters are insufficient to cope with the current disturbance, triggers the parameter adjustment process, enhances the side air supply in advance, and extends the cooling cycle.

[0095] In one embodiment, the initial control parameters are adjusted based on the dynamic response, and adaptive energy-saving control of the refrigeration operation is performed based on the adjusted control parameters, including:

[0096] When verifying the reliability of the initial control parameters, a fuzzy evaluator based on confidence intervals is used to calculate the temperature uniformity evaluation parameters and system energy consumption evaluation parameters corresponding to the initial control parameters. Based on the comparison results of the temperature uniformity evaluation parameters and system energy consumption evaluation parameters with preset thresholds, a sequence of parameters to be adjusted is generated.

[0097] The confidence interval-based fuzzy evaluator can be a multi-index comprehensive evaluation module combining statistical confidence intervals and fuzzy logic, used to quantify the performance of control parameters under uncertain environments. In this embodiment, the confidence interval-based fuzzy evaluator can input temperature field simulation data and energy consumption prediction values ​​into a fuzzy rule base, while introducing confidence intervals to reflect data reliability, and outputting a normalized comprehensive evaluation value. Temperature uniformity evaluation parameters can be quantitative indicators characterizing the dispersion of temperature distribution within the refrigerated space. System energy consumption evaluation parameters can reflect the energy consumption level of the refrigeration system per unit time or per unit of cooling capacity under the current control strategy. In a specific embodiment, system energy consumption evaluation parameters can be used to measure the energy-saving effect of the control scheme and support energy efficiency optimization decisions. For example, system energy consumption evaluation parameters can include, but are not limited to, compressor power consumption ratio parameters, fan auxiliary energy consumption parameters, and start-stop loss conversion parameters. The preset threshold can be a performance acceptance boundary value set for temperature uniformity and system energy consumption. The sequence of parameters to be adjusted can be a structured parameter list generated by comparing the evaluation parameters with the preset threshold, indicating the direction and magnitude of optimization. Furthermore, the sequence of parameters to be adjusted can provide the supervision signals needed for training and prediction of the gradient boosting tree regressor.

[0098] A gradient boosting tree regressor is trained based on the sequence of parameters to be adjusted. The parameter adjustment amount is predicted by the gradient boosting tree regressor. The parameter values ​​of the initial control parameters are adjusted according to the parameter adjustment amount to obtain the discrete control point coordinates of the updated control parameters.

[0099] The gradient boosting tree regressor can be a nonlinear regression model based on ensemble learning, which improves prediction accuracy by iteratively fitting the residuals. In this embodiment, the gradient boosting tree regressor can use the historical sequence of parameters to be adjusted and the corresponding actual adjustment amount as the training set to construct multiple decision trees for weighted ensemble. The parameter adjustment amount can be the specific numerical increment or scaling factor output by the gradient boosting tree regressor used to correct the initial control parameters. The discrete control point coordinates of the updated control parameters can be the set of discrete setpoints of control variables such as compressor frequency and fan speed on the time axis obtained after correction by the parameter adjustment amount. In this embodiment, the discrete control point coordinates of the updated control parameters can be used as the original input for subsequent trajectory smoothing and velocity planning.

[0100] The coordinates of discrete control points for the updated control parameters are fitted using NURBS curves, and the speed planning is dynamically adjusted based on the NURBS curves under the constraints of maximum allowable temperature fluctuation and maximum compressor load rate to obtain the speed planning curve.

[0101] The NURBS curve can be a non-uniform rational B-spline curve, a mathematical tool used for high-order continuous smooth fitting of discrete control points. In an exemplary embodiment, the NURBS curve can preserve key control characteristics while improving the geometric continuity of the trajectory, avoiding abrupt changes in the actuator. The maximum permissible temperature fluctuation can be the maximum tolerance for temperature deviation from the set value allowed to ensure refrigeration quality. In this embodiment, the maximum permissible temperature fluctuation serves as a hard constraint for dynamic speed adjustment planning, preventing excessive energy saving at the expense of preservation quality. The compressor maximum load rate constraint can be the maximum power or current percentage limit that the compressor can withstand within its safe operating range. Furthermore, the compressor maximum load rate constraint can prevent control commands from causing equipment overload, ensuring hardware lifespan and system stability. Dynamic speed adjustment planning can be an optimization process that, under physical constraints, time-parameterizes the NURBS fitting curve to determine the execution rate at each moment. In a specific embodiment, dynamic speed adjustment planning can balance response speed and execution smoothness, avoiding rapid acceleration or deceleration. The speed planning curve can be a continuous function describing the rate of change of control parameters over time, generated after dynamic speed adjustment planning. In this embodiment, the speed planning curve can guide the actuator to complete parameter switching at a smooth rhythm.

[0102] The coordinates of discrete control points for the updated control parameters are fitted using NURBS curves. This can be done by using time as a parameter and the discrete control points as control vertices to construct a smooth NURBS trajectory. Furthermore, this operation can be achieved by adjusting the node vectors to enhance the fitting accuracy during critical time periods and introducing weights to control the interpolation accuracy of specific points, thereby eliminating step changes between discrete points and improving trajectory continuity. Based on the NURBS curves, dynamic speed planning is performed under the constraints of maximum allowable temperature fluctuation and maximum compressor load rate. The speed planning curve is obtained by solving a time parameterization problem that satisfies both temperature and load constraints based on the NURBS geometric trajectory, determining the execution rate at each time step. Further, this operation can be achieved by using time scaling for speed remapping and using optimal control theory to solve for the minimum time trajectory, ensuring that the control action is both fast and within limits.

[0103] The curvature values ​​at each discrete point of the speed planning curve are calculated. Transition points are inserted in the curvature change region by spline interpolation to obtain the adjusted control parameters. Adaptive energy-saving control of the refrigeration operation is then performed based on the adjusted control parameters.

[0104] The curvature values ​​at each discrete point can be a measure of the geometric curvature of each sampling point on the velocity planning curve. In this embodiment, the curvature values ​​at each discrete point can be used to identify trajectory discontinuities that may cause execution shocks. Spline interpolation can be a mathematical method that constructs a smooth curve between known data points using a piecewise polynomial function. Furthermore, spline interpolation can insert intermediate points in curvature abrupt change regions to improve the second-order continuity of the trajectory. Curvature abrupt change regions can be local intervals on the velocity planning curve where the curvature undergoes a significant jump. In a specific embodiment, curvature abrupt change regions can identify key locations where trajectory refinement is needed to prevent mechanical stress or control oscillations in the actuator. Transition points can be intermediate control points added in curvature abrupt change regions through spline interpolation. In this embodiment, transition points can ensure that the adjusted control parameters have C² continuity in the time domain, ensuring smooth execution. The adjusted control parameters can be the final executable control instruction sequence formed after NURBS fitting, velocity planning, and transition point insertion. Furthermore, the adjusted control parameters can be used to drive actuators such as compressors and fans to achieve high-precision, low-impact adaptive energy-saving control.

[0105] For example, in the scenario of low-load energy-saving operation of cold storage at night, the AI-based adaptive energy-saving intelligent control method for cold storage conditions in this embodiment can be as follows: After the system verifies the reliability of the initial control parameters, the fuzzy evaluator finds that the temperature uniformity meets the standard but the energy consumption is too high, and generates a sequence of parameters to be adjusted, mainly to reduce the fan speed; the gradient boosting tree regressor predicts that the air volume should be reduced by 15%; the updated discrete control points are fitted by NURBS to form a smooth deceleration curve; speed planning is performed under the constraints of not exceeding the maximum allowable temperature fluctuation of 0.5℃ and the compressor load rate of 80%, resulting in a slow-decreasing speed curve; the curvature calculation finds a slight inflection point in the middle, and two transition points are inserted by cubic spline interpolation to finally form a smooth fan speed control command; after execution, the system maintains temperature uniformity while reducing auxiliary energy consumption by 20%, and there is no equipment start-up or shutdown.

[0106] In one embodiment, before acquiring real-time operating condition monitoring data of the target cold storage space, the method further includes:

[0107] The enclosure structure parameters and cargo stacking parameters of the cold storage are obtained to construct an offline simulation model of the unsteady flow field coupled with transient heat load corresponding to the cold storage. The offline simulation model is analyzed to obtain the temperature distribution cloud map and energy consumption distribution cloud map corresponding to each preset initial control parameter under the cold storage conditions.

[0108] Extract the discrete points of each initial control parameter in the temperature distribution cloud map and the gradient of change in the energy consumption distribution cloud map, and calculate the temperature change rate and energy consumption change rate corresponding to each initial control parameter at each discrete point to obtain the dynamic response curve of heat load under different initial control parameters for refrigeration conditions.

[0109] Based on the dynamic response curve of heat load, several first candidate control parameters are selected from several initial control parameters;

[0110] The control parameters in each first candidate control parameter are optimized according to the preset parameter optimization algorithm to obtain several optimized control parameter combinations and save them to the preset control parameter library.

[0111] The enclosure parameters of the cold storage facility can be a set of physical parameters describing the thermal properties of the building itself. These parameters can be used as external boundary conditions for the offline simulation model, determining the rate of thermal intrusion and cold leakage characteristics. In an exemplary embodiment, the enclosure parameters can be extracted from design drawings or obtained through on-site measurements, including wall thermal conductivity, door and window insulation performance, and the volume and surface area of ​​the storage unit. Cargo stacking parameters can be configuration information reflecting the layout and thermal properties of goods within the cold storage facility. These parameters can be used as key inputs to determine the internal heat source distribution and airflow resistance, influencing the reconstruction of the local temperature and flow fields. For example, cargo stacking parameters are provided by the warehouse management system or manually entered, including cargo type, stacking density, stacking height, gap ventilation rate, and specific heat capacity. The offline simulation model of unsteady flow field coupled with transient heat load can be a high-fidelity digital twin model constructed by integrating computational fluid dynamics and transient heat transfer theory. This model can be used to pre-simulate the dynamic behavior of the system under different control strategies before control deployment, explicitly analyzing the bidirectional coupling mechanism between the flow field and heat load. In this embodiment, the unsteady flow field coupled transient heat load offline simulation model establishes a three-dimensional geometry and mesh based on the enclosure structure parameters and cargo stacking parameters, and performs multiphysics field solutions by coupling the Navier-Stokes equations and energy conservation equations. Furthermore, the unsteady flow field coupled transient heat load offline simulation model receives initial control parameters as excitation input and outputs temperature distribution cloud maps and energy consumption distribution cloud maps. The temperature distribution cloud map can be a visualized numerical distribution map of the temperature field inside the cold storage space under specific initial control parameters, which can be used to assess temperature uniformity and the risk of local hot spots. In a specific embodiment, the temperature distribution cloud map is obtained by interpolating the temperature values ​​of each grid node output by the offline simulation model during time steps to obtain a continuous field.

[0112] An energy consumption distribution cloud map can be a numerical map reflecting the spatiotemporal distribution of energy consumption in various components or regions of a system under specific control parameters. It can be used to identify high-energy-consumption areas and ineffective cold energy delivery paths. In this embodiment, the energy consumption distribution cloud map is generated in real time by integrating energy consumption models of subsystems such as compressor power and fan power consumption through a simulation model. Discrete points can be representative spatial sampling points selected from the temperature distribution cloud map. They can be used as spatial benchmarks for calculating the temperature change rate, covering key areas such as corners, door areas, and stacking gaps. Furthermore, discrete points can include, but are not limited to, cargo center monitoring points, airflow dead zones, and door gap heat intrusion sensitive points. The change gradient can be the spatial derivative vector of the energy consumption difference between adjacent regions in the energy consumption distribution cloud map. It can be used to characterize the degree of unevenness in energy consumption distribution and areas with optimization potential. The temperature change rate can be the rate of temperature change over time at a specific discrete point. It can be used to quantify the dynamic impact of control parameters on local temperature stability. In this embodiment, the temperature change rate is obtained by calculating the temperature value at that point in the simulation time series by difference or fitting the derivative.

[0113] The energy consumption change rate can be the rate of change of total system energy consumption or local energy consumption over time or control actions, and can be used to evaluate the marginal energy efficiency benefits of adjusting control parameters. The heat load dynamic response curve can be a multi-dimensional response trajectory characterizing the dynamic performance of control parameters, formed by combining the temperature change rate and energy consumption change rate. It can be used as a core basis for screening candidate control parameters, reflecting the balance between stability and energy efficiency of the control strategy. In a specific embodiment, the heat load dynamic response curve is generated by aligning and normalizing the temperature change rate and corresponding energy consumption change rate at each discrete point over time. Further, the heat load dynamic response curve is used to screen the first candidate control parameters. The first candidate control parameters can be a relatively superior subset initially selected from all initial control parameters based on the heat load dynamic response curve, which can be used to narrow the subsequent optimization search space and improve the efficiency of parameter library construction. In this embodiment, the first candidate control parameters serve as input to a preset parameter optimization algorithm. The preset parameter optimization algorithm can be a calculation program for fine-tuning the first candidate control parameters, which can be used to further explore energy-saving potential while maintaining temperature quality. In one exemplary embodiment, a pre-configured parameter optimization algorithm is embedded in the system's multi-objective optimization engine, supporting constraint handling and Pareto front search. The optimized combination of several control parameters can be a set of control strategies with higher overall performance generated by the parameter optimization algorithm. This set can be used as a high-quality strategy injected into a pre-configured control parameter library to support accurate matching during the online phase.

[0114] Taking the pre-deployment of control strategies for newly built cold storage as an example, the AI-based adaptive energy-saving intelligent control method for refrigeration conditions in this embodiment can be implemented before a newly built fresh food cold storage is put into operation. The system first obtains the parameters of its polyurethane sandwich panel walls and the shelf stacking scheme, and constructs an offline simulation model of unsteady flow field coupled transient heat load. Then, it simulates 100 sets of initial control parameters to generate corresponding temperature and energy consumption cloud maps. From these, it extracts the temperature change rate of 10 discrete points such as the door area and the stacking center, as well as the energy consumption gradient of the fan area, and merges them to form a dynamic response curve of heat load. Based on the curve, 20 sets of first candidate control parameters are selected, and then optimized by a multi-objective particle swarm optimization algorithm to finally obtain 8 sets of control parameter combinations that take into account both rapid cooling and low energy consumption, and stores them in the control parameter library. When running online, the system can quickly match one of the sets according to the real-time coupling feature values, avoiding energy waste caused by blind debugging in the early stage.

[0115] In one embodiment, the enclosure structure parameters include thermal conductivity and heat storage coefficient, and the cargo stacking parameters include porosity and volumetric heat capacity. The enclosure structure parameters and cargo stacking parameters of the cold storage are obtained to construct an offline simulation model of the unsteady flow field coupled with transient heat load corresponding to the cold storage. The offline simulation model is then analyzed to obtain temperature distribution cloud maps and energy consumption distribution cloud maps corresponding to each preset initial control parameter under the cold storage conditions, including:

[0116] The thermal conductivity, heat storage coefficient, and cargo stacking porosity of the enclosure structure of the cold storage are obtained in order to calculate the flow field resistance matrix and heat permeation matrix corresponding to the cold storage space. The flow field resistance matrix is ​​determined by the porosity and stacking height, and the heat permeation matrix is ​​determined by the thermal conductivity and heat storage coefficient.

[0117] Among these, thermal conductivity is a physical property characterizing the heat transfer capacity of the enclosure material. It can be used to calculate steady-state and unsteady-state heat flow through the cold storage walls and is a key input for constructing the heat permeability matrix. Heat storage coefficient is a comprehensive thermophysical parameter reflecting the enclosure structure's ability to absorb and release heat under periodic temperature fluctuations. It can be used to determine the cold storage's buffering capacity against environmental temperature fluctuations and influences the phase and amplitude of transient heat loads. Porosity is the proportion of void volume to total volume in the cargo stacking area. It directly affects airflow resistance and local heat transfer efficiency and is a core component of the flow field resistance matrix. Volumetric heat capacity is the ability of a unit volume of cargo to store heat, determined by the product of density and specific heat capacity. It can be used to determine the dynamic response intensity of cargo as an internal heat source or cold sink and participates in the calculation of nodal thermal disturbances. Obtaining the thermal conductivity, heat storage coefficient, and cargo stacking porosity of the cold storage enclosure structure can be achieved by extracting key thermophysical properties and structural parameters from material databases or measurement reports. Furthermore, this operation can be achieved by calling a standardized building material thermal performance database, performing on-site infrared thermal imaging inversion, or conducting laboratory steady-state heat flux meter tests, thus providing a physical basis for constructing the flow field resistance matrix and heat permeation matrix. Calculating the flow field resistance matrix and heat permeation matrix corresponding to the cold storage space can be achieved by constructing resistance and heat conduction models based on porosity / stacking height and thermal conductivity / heat storage coefficient, respectively, and then discretizing them. Furthermore, this operation can be achieved by using the Darcy-Fochheimer model to construct the flow field resistance matrix and using the thermal network method to separate the heat permeation matrix, thereby enabling a mathematical characterization of the cargo stacking resistance and the thermal dynamics of the enclosure structure.

[0118] The flow field drag matrix can be a sparse matrix describing the local drag effect of cargo stacking on airflow. It can be used to accurately reproduce airflow deceleration, flow around, and dead zone formation phenomena in the stacking area during simulation. In a specific embodiment, the flow field drag matrix may include viscous drag sub-matrices, inertial drag sub-matrices, anisotropic drag sub-matrices, etc. Furthermore, the flow field drag matrix can be constructed based on porosity and stacking height to build a Darcy-Fochheimer drag model and mapped to the momentum equation source terms of the computational grid. The thermal permeability matrix can be a system matrix characterizing the dynamic coupling effect of heat conduction and heat storage in the enclosure structure. It can be used to accurately simulate the time delay and attenuation characteristics of external environmental disturbances transmitted into the storage area. In this embodiment, the thermal permeability matrix may include a high-frequency heat wave attenuation matrix, a low-frequency thermal inertia matrix, a transient impact response matrix, etc. Furthermore, the thermal permeability matrix can be constructed by discretizing the thermal conductivity and heat storage coefficient through the Fourier heat conduction equation and embedding the boundary conditions of the energy equation.

[0119] The thermal disturbance of the nodes is obtained by using the flow field resistance matrix and the thermal penetration matrix and mapped to the mesh nodes to establish the finite volume interpolation equation. The mesh nodes are divided into hexahedral elements and the mesh size is reduced to one-third of the basic mesh size in the near-wall region.

[0120] The nodal thermal disturbance can be the net heat flow disturbance value caused by external heat intrusion and internal heat exchange at the simulation grid node. It can be used as the source term input for the finite volume interpolation equation to drive the evolution of the temperature and flow fields. In an exemplary embodiment, the nodal thermal disturbance can include door zone heat intrusion disturbance, wall conduction disturbance, and cargo breathing thermal disturbance. Furthermore, the nodal thermal disturbance can be mapped from physical disturbances to numerical field source terms through the coupling operation of the flow field resistance matrix and the heat penetration matrix. The grid node can be a spatial location point in the discretized computational domain, used to carry the numerical solution of physical quantities such as temperature and velocity. The hexahedral element can be the basic unit for three-dimensional meshing of the cold storage space using hexahedral geometry, used to provide a structured mesh to improve numerical stability and computational efficiency. The basic grid size can be the standard grid side length used in the main area of ​​the cold storage space, which can be used as a benchmark discretization scale to balance computational accuracy and resource consumption. The near-wall region grid size can be a finer grid size used near the warehouse wall and cargo surface, which can be used to improve the accuracy of wall heat transfer and flow separation simulation. In this embodiment, the mesh size in the near-wall region can be set to one-third of the base mesh size to resolve the thermal boundary layer and the velocity boundary layer.

[0121] Obtaining nodal thermal disturbances by using the flow field drag matrix and heat penetration matrix and mapping them to grid nodes can be achieved by applying the two matrices to the boundary and internal regions, generating heat source terms for the corresponding grid nodes. Furthermore, this operation can be implemented through direct assignment using matrix-vector multiplication, or by using interpolation functions to smoothly transition to internal nodes, thus establishing a mapping bridge between physical disturbances and the numerical field. Establishing finite volume interpolation equations can be achieved by integrating the governing equations on a hexahedral mesh and discretizing them using the Gaussian divergence theorem. Further, this operation can be implemented by using a central difference scheme to handle convection terms, or by using a second-order upwind scheme to improve stability, thus forming an algebraic equation framework suitable for numerical solutions. The finite volume interpolation equations are used to construct an offline simulation model of unsteady flow field coupled with transient thermal load, calculating the flow field gradient values ​​through grid nodes. The calculation formula for the finite volume interpolation equations is as follows:

[0122]

[0123] In the formula, For each face of a discrete mesh element; This represents the total number of faces in the unit, which can be 6, corresponding to a hexahedral mesh. For fluid density; For noodles The velocity vector on; For noodles General physical quantities, such as temperature and momentum; For noodles The area vector; For noodles The generalized diffusion coefficient is related to the thermal conductivity; For noodles Physical quantities The gradient; The source term corresponds to the influence of nodal thermal disturbance and thermal penetration matrix. For grid cells The volume. This equation utilizes the principles of fluid mechanics and heat transfer to accurately calculate the velocity field, temperature field, and heat infiltration process within the cold storage space using a discrete grid (hexahedral elements). Before being put into actual operation, this formula is used to simulate working conditions under different enclosure structures and cargo stacking parameters, generating massive amounts of temperature distribution cloud maps and energy consumption distribution cloud maps. This provides a calculation basis for the preset control parameter library, ensuring that the initial parameters called by the subsequent AI model have a solid physical theoretical foundation, rather than blind guessing.

[0124] The flow field gradient value is calculated based on the finite volume interpolation equation to obtain the flow field-thermal coupling control equation, i.e., the offline simulation model of unsteady flow field coupled transient thermal load; where the flow field gradient value is determined by the velocity gradient vector and the temperature gradient vector.

[0125] The flow field gradient value can be a coupled gradient tensor composed of the velocity gradient vector and the temperature gradient vector, which can be used as the core differential term in constructing the flow-thermal coupling control equations. The velocity gradient vector can be a vector field describing the rate of change of air velocity in all spatial directions, and can be used to determine the generation of fluid shear force and turbulent kinetic energy. The temperature gradient vector can be a vector field describing the rate of change of temperature in all spatial directions, and can be used to drive heat conduction and natural convection effects. The flow field gradient value is calculated based on the finite volume interpolation equations, which can involve local reconstruction or differencing of the discrete solution to obtain the spatial derivatives of velocity and temperature. Furthermore, this operation can be implemented using least-squares gradient reconstruction, surface-averaged gradient calculation based on the Green-Gauss theorem, etc., to extract key differential information of the coupling process. Obtaining the flow-thermal coupling control equations, i.e., the offline simulation model of unsteady flow field coupled transient thermal load, can be achieved by substituting the flow field gradient value into the Navier-Stokes equations and the energy equations to form an unsteady coupled system. Furthermore, this operation can be implemented by directly assembling equations using a fully coupled solver and iteratively updating variables in a step-by-step coupled manner, thereby enabling the construction of a high-fidelity digital twin model. The flow-thermal coupling control equations can be a set of partial differential equations that integrate mass conservation, momentum conservation, and energy conservation, which can be used as the mathematical core of an offline simulation model of unsteady flow field coupled with transient thermal load.

[0126] A set of nonlinear cooling supply and demand equations is established based on the flow field-thermal coupling control equations to obtain the temperature distribution cloud map and energy consumption distribution cloud map corresponding to each preset initial control parameter under the refrigeration condition; among them, the nonlinear cooling supply and demand equations are solved using the SIMPLEC algorithm until the residual is less than the preset value.

[0127] The nonlinear cooling capacity supply and demand equations can be an extended set of equations that incorporates the dynamic characteristics of the refrigeration system based on the flow-thermal coupling control equations. These equations can be used to predict the temperature and energy consumption responses under different initial control parameters. In a specific embodiment, the nonlinear cooling capacity supply and demand equations may include evaporator cooling capacity output sub-equations, compressor energy efficiency sub-equations, and fan resistance sub-equations. The SIMPLEC algorithm can be an improved semi-implicit algorithm for solving pressure-velocity coupling problems, and can be used to efficiently and stably solve the nonlinear cooling capacity supply and demand equations. The residual can be the imbalance quantity in the discrete form of the control equations during numerical iteration, and can be used as a quantitative indicator to judge the convergence of the solution. The preset value can be a set residual convergence threshold, which can be used to control the simulation termination condition and ensure the physical reliability of the solution.

[0128] A nonlinear cooling supply and demand equation set is established based on the flow-thermal coupling control equations. This can be achieved by introducing a dynamic model of the refrigeration equipment to link cooling supply with heat load demand. Furthermore, this operation can be implemented by embedding the compressor COP curve into the energy equation and coupling the fan characteristic curve to the momentum equation, thereby achieving system-level closed-loop simulation capabilities. The SIMPLEC algorithm is used to solve the nonlinear cooling supply and demand equation set until the residuals are less than a preset value. This can be achieved by iteratively updating the pressure, velocity, and temperature fields until all equation residuals are below a set threshold. Further, this operation can be achieved by combining multigrid to accelerate convergence and using adaptive time step control to maintain stability, thus ensuring numerical solution convergence and good physical consistency. Temperature distribution cloud maps and energy consumption distribution cloud maps corresponding to each preset initial control parameter are obtained under refrigeration conditions. This can be achieved by post-processing the converged numerical solution to generate a visual field map. Further, this operation can be achieved through isosurface rendering, color mapping, and vector overlay techniques, thus providing an intuitive and quantitative evaluation basis for subsequent parameter selection.

[0129] Taking the modeling of a high-density cold storage facility for fruits and vegetables as an example, the AI-based adaptive energy-saving intelligent control method for cold storage conditions in this embodiment can be used to analyze a high-density stacked cold storage facility for fruits and vegetables. The system obtains the thermal conductivity of the polyurethane walls (0.022 W / mK), the heat storage coefficient (1.8 kJ / m²Ks^0.5), and the porosity of the stacked goods (0.35). Based on this, a flow field resistance matrix (considering a stacking height of 4 meters) and a heat penetration matrix are constructed. In a hexahedral mesh, the mesh of the near-wall surface is refined to one-third of the base size, and the thermal disturbance of the nodes is mapped to 200,000 mesh nodes. An interpolation equation is established using the finite volume method to calculate the velocity and temperature gradient vectors, forming a flow field-thermal coupling control equation. The screw compressor and EC fan model are further coupled to establish a nonlinear cold energy supply and demand equation set, and the SIMPLEC algorithm is used to iterate until the residual is less than 1e-4. Finally, temperature cloud maps and energy consumption cloud maps under different fan speed combinations are output, clearly showing the trade-off between the lag in temperature rise at the stacking center under low speed and the surge in fan power consumption under high speed.

[0130] In one embodiment, the rate of temperature change and the rate of energy consumption change corresponding to each initial control parameter at each discrete point are calculated to obtain the dynamic response curve of heat load under different initial control parameters for refrigeration conditions, including:

[0131] Obtain the control sequence corresponding to each initial control parameter, and obtain the coordinate dataset of discrete operating points on the control sequence; where different initial control parameters correspond to different compressor operating frequencies and fan speeds.

[0132] The control sequence can be a time sequence of compressor and fan control commands corresponding to a certain initial control parameter within the simulation time domain, which can be used as a time reference input for dynamic modeling of cooling output. In this embodiment, the control sequence can be expanded from the initial control parameter into execution commands arranged by time steps, including the compressor operating frequency and fan speed at each moment. The coordinate dataset of discrete operating points can be a set of spatial-temporal coordinates of key moments and their corresponding system states selected on the time axis of the control sequence, which can be used to provide spatiotemporal sample points for Bézier curve fitting, supporting continuous modeling of instantaneous cooling output. In an exemplary embodiment, the coordinate dataset of discrete operating points can be formed by time sampling the control sequence and recording the simulation state (such as temperature and wind speed) at the corresponding moments to create multidimensional coordinate points.

[0133] The instantaneous cooling output value is obtained by fitting discrete operating points with a Bézier curve; the instantaneous cooling output value is calculated through the mapping relationship between control parameters and cooling output.

[0134] The Bézier curve can be a parametric smooth curve used for nonlinear interpolation fitting of discrete operating points. It can smooth the nonlinear dynamic process of cooling output changing over time, avoiding distortion caused by step or linear assumptions. In one specific embodiment, the Bézier curve can use discrete operating points as control vertices and construct a continuously differentiable trajectory using Bernstein basis functions. The instantaneous cooling output value can be the actual cooling capacity provided by the control system at any given time, and can be used as an input term in the thermodynamic equation, reflecting the real-time impact of actual control actions on the heat load. In this embodiment, the instantaneous cooling output value can be calculated by fitting a continuous control trajectory using the Bézier curve and then combining it with the nonlinear mapping relationship between compressor frequency and fan speed on cooling capacity. The mapping relationship between control parameters and cooling output can be a functional relationship describing the relationship between control variables such as compressor operating frequency and fan speed and the actual cooling output of the system. It can be used to convert control commands into physically interpretable thermodynamic inputs. In an exemplary embodiment, the mapping relationship between control parameters and cooling output can be obtained by fitting equipment performance test data or modeling manufacturer performance curves, and is typically a nonlinear multivariable function.

[0135] Instantaneous cooling capacity output values ​​are obtained by fitting discrete operating points with Bézier curves. This can be achieved by constructing Bézier curves using discrete operating points as control points and outputting a continuous cooling capacity trajectory parameterized along time. Furthermore, this operation can be implemented by fitting the dynamic start-up and shutdown process of the compressor using cubic Bézier curves or by processing long-period control sequences using piecewise Bézier curve splicing, thus achieving a smooth modeling effect for nonlinear and unsteady-state cooling capacity release processes. Instantaneous cooling capacity output values ​​are calculated through the mapping relationship between control parameters and cooling capacity output. This can be achieved by substituting the control trajectory output from the Bézier curve into a pre-built cooling capacity mapping function to calculate the actual cooling capacity at the corresponding moment. Further, this operation can be implemented by obtaining cooling capacity values ​​through table lookup interpolation and then performing smoothing filtering, or by using a neural network mapping model to predict cooling capacity in real time, thus achieving a differentiable channel from control commands to physical cooling capacity.

[0136] Thermodynamic equations are established based on instantaneous cooling output values, and the heat load change rate at discrete points is obtained by solving the Euler integral formula; wherein, the heat load change rate is obtained by integrating the heat load over time.

[0137] The thermodynamic equation can be a differential equation describing the evolution of heat load over time within the cold storage space, and can be used to characterize the dynamic balance between cooling supply and heat load accumulation. In this embodiment, the thermodynamic equation can be based on the principle of energy conservation, using instantaneous cooling output as input and heat load as the state variable to construct a first-order dynamic model. The Euler integral formula can be a numerical integration method used to approximate the initial value problem of the ordinary differential equation, and can be used to discretize the thermodynamic equation to obtain the heat load change rate at each discrete point. The heat load change rate can be the cumulative rate of change of heat load per unit time, and can be used to quantify the dynamic intensity of system thermal disturbance, serving as a key input to the state equation. In a specific embodiment, the heat load change rate can be obtained by solving the thermodynamic equation using the Euler integral formula and then back-calculating it using the integral relationship of heat load over time.

[0138] Establishing a thermodynamic equation based on instantaneous cooling output can be achieved by using instantaneous cooling as input and constructing a differential equation for the change of heat load over time based on energy conservation. Furthermore, this operation can be implemented by introducing extended model structures such as environmental heat intrusion terms and cargo heat capacity terms, thereby achieving the technical effect of formally describing the dynamic coupling relationship between cooling and heat load. The rate of change of heat load at discrete points is obtained by using the Euler integral formula. This can be achieved by performing a discrete-time step solution on the thermodynamic equation, outputting the rate of change of heat load at each sampling point. Furthermore, this operation can be achieved by using forward Euler for rapid estimation and combining adaptive step sizes to improve integration accuracy, thereby achieving the technical effect of numerically approximating the dynamic characteristics of heat load. Specifically, the Euler integral formula for solving the rate of change of heat load is used to establish the thermodynamic equation based on the instantaneous cooling output and to calculate the rate of change of heat load at discrete points through integration. The calculation formula is as follows:

[0139]

[0140] In the formula, For the next moment The heat load value; For the current moment The heat load value; For time step; This represents the rate of change of the heat load over time at the current moment. This is the instantaneous cooling output value at the current moment, obtained by fitting the Bézier curve. This represents the system's heat capacity. The calculation formula reflects the change in transient heat load over time, particularly how the system temperature changes through integral accumulation when the compressor's cooling output does not match the actual heat load. Combined with the instantaneous cooling output fitted by the Bézier curve, this formula calculates the dynamic response curve of the heat load, helping the system understand how long it takes for the cold storage temperature to stabilize under different control inputs, i.e., the transport lag phase difference. The time-series data it provides is used to verify the accuracy of the offline simulation model, ensuring that the control strategy takes into account the influence of thermal inertia. The rate of change of heat load is obtained by integrating the heat load over time. This can be achieved by numerically integrating the heat load time series to inversely deduce the rate of change, ensuring the consistency of energy conservation. Furthermore, this operation can be verified using trapezoidal integration or higher-order numerical integration, thereby strengthening the constraints of the first law of thermodynamics and improving the physical rationality of the model.

[0141] The state equation is constructed based on the rate of change of heat load at discrete points and the temperature fluctuation on the temperature distribution cloud map. The temperature dynamic change function is obtained by solving the least squares method, and the dynamic response curve of heat load is obtained based on the temperature dynamic change function.

[0142] The temperature fluctuation can be the deviation magnitude of a discrete point in the temperature distribution cloud map relative to a set value. It can be used to characterize local temperature stability and, together with the heat load change rate, constitutes the observed variables of the state equation. The state equation can be a dynamic system model relating the heat load change rate and the temperature fluctuation, and can be used to establish an identifiable mapping between thermodynamic input and temperature response. In an exemplary embodiment, the state equation can use the heat load change rate as input and the temperature fluctuation as output to construct a linear or nonlinear state-space expression. The least squares method can be a parameter estimation method that solves for the optimal parameters by minimizing the sum of squared errors between the observed values ​​and the model predictions. It can be used to identify the coefficients of the temperature dynamic change function from limited observation data, improving the model's generalization ability. The temperature dynamic change function can be an analytical or numerical function describing the evolution of temperature at a discrete point over time. It can be used to provide continuous prediction capabilities for the temperature response and support the generation of the heat load dynamic response curve. In a specific embodiment, the temperature dynamic change function can be obtained by fitting the input-output data pairs in the state equation using the least squares method.

[0143] Constructing a state equation based on the rate of change of heat load at discrete points and the temperature fluctuation on the temperature distribution cloud map can be achieved by using the rate of change of heat load as the input variable and the temperature fluctuation as the output variable to establish a dynamic system model. Furthermore, this operation can be achieved by constructing a linear state-space model for rapid identification and introducing nonlinear terms to capture saturation effects, thus achieving the technical effect of linking thermodynamic driving forces and temperature response to form an identifiable state space. The temperature dynamic change function is obtained by solving the state equation parameters using the least squares method, which can be achieved by fitting the parameters to the state equation using observational data to obtain the temperature response function. Furthermore, this operation can be achieved by using regularized least squares to prevent overfitting and recursive least squares to support online updates, thus achieving data-driven dynamic characteristic identification and reducing reliance on ideal models. The dynamic response curve of the heat load is obtained from the temperature dynamic change function, which can be fused with the corresponding energy consumption index to form a multi-dimensional dynamic performance characterization. Furthermore, this operation can be achieved by superimposing energy consumption weighting factors or introducing multi-objective optimization criteria, thus achieving the technical effect of generating high-fidelity response curves that can be used for parameter screening.

[0144] Taking the fine control modeling of a variable frequency compressor cold storage as an example, the AI-based adaptive energy-saving intelligent control method for cold storage conditions in this embodiment can be as follows: In a cold storage equipped with a variable frequency compressor and a continuously variable speed fan, the system generates a control sequence containing frequency and speed changes for a set of initial control parameters; 10 discrete operating point points are selected before and after the door opening disturbance, and a coordinate dataset is constructed with time as the abscissa and cooling demand as the ordinate; the dataset is fitted with a cubic Bézier curve to obtain a smooth instantaneous cooling output value; combined with the compressor performance curve mapping relationship, the control trajectory is converted into actual cooling capacity; a thermodynamic dynamic equation is established and the heat load change rate at each point is solved by Euler integral; the state equation is constructed by combining the temperature fluctuation of the cargo center point in the temperature distribution cloud map obtained from the simulation; the temperature dynamic change function is identified by the least squares method, and finally the heat load dynamic response curve under the control parameters is formed for subsequent candidate screening.

[0145] In one embodiment, multiple first candidate control parameters are selected from several initial control parameters based on the dynamic response curve of the heat load, including:

[0146] The temperature fluctuation amplitude corresponding to each initial control parameter is determined by the dynamic response curve of the heat load. Constraint equations are constructed based on the preset maximum allowable temperature fluctuation of the refrigeration condition and the current temperature fluctuation amplitude corresponding to each initial control parameter. Constraint optimization parameters are obtained by iterative calculation using conjugate gradients.

[0147] A multi-objective optimization function is established based on the constraint optimization parameters, including the objective function of temperature uniformity and the objective function of energy consumption. The mapping relationship between the parameter vector and the constraint response vector corresponding to each initial control parameter is obtained through a support vector machine with radial basis function kernel.

[0148] Based on the mapping relationship, a set of constrained optimization equations is constructed using KKT conditions, and the optimal candidate control parameters and Lagrange multipliers corresponding to each initial control parameter are obtained by solving the sequential quadratic programming method.

[0149] If the norm difference between the optimal candidate control parameter and the initial parameter corresponding to the initial control parameter is less than the preset convergence threshold, then the optimal candidate control parameter is stored as the first candidate control parameter.

[0150] The temperature fluctuation amplitude, derived from the dynamic response curve of the heat load, represents the maximum deviation range of temperature from the set value within the refrigerated space under specific initial control parameters. It serves as a core indicator for measuring the impact of the control strategy on temperature stability and is used to construct physical feasibility constraints. In this embodiment, the temperature fluctuation amplitude can be extracted from the time integral extreme value of the discrete point temperature change rate from the dynamic response curve of the heat load or directly calculated from the peak-to-peak value of the simulated temperature sequence. The preset maximum allowable temperature fluctuation can be a temperature fluctuation upper limit threshold set according to the preservation requirements of refrigerated goods. It can be used as a hard constraint to ensure that all candidate control parameters meet the quality standards for food or pharmaceutical storage. In an exemplary embodiment, the preset maximum allowable temperature fluctuation may include, but is not limited to, the allowable fluctuation range for fresh produce, frozen products, and pharmaceutical cold chain products. The constraint equation can be a mathematical inequality expression based on the temperature fluctuation amplitude and the preset maximum allowable temperature fluctuation. It can be used to exclude control parameters that violate temperature quality requirements and ensure the bottom line of refrigerated safety. Furthermore, the constraint equation can be formalized as temperature fluctuation amplitude ≤ preset maximum allowable temperature fluctuation, used to limit the feasible solution space.

[0151] Conjugate gradient iteration can be a numerical iterative algorithm for solving constrained optimization problems. It can efficiently generate optimization starting points that satisfy initial constraints, avoiding convergence difficulties caused by random initialization. In a specific embodiment, conjugate gradient iteration can search for the path that causes the objective function to decrease the fastest along the conjugate direction within the feasible region defined by the constraint equation. The constraint optimization parameters can be intermediate optimization variables that satisfy the constraint equation, calculated by conjugate gradient iteration. They can be used as initialization inputs for multi-objective optimization functions to improve the efficiency and stability of subsequent optimization. The temperature uniformity objective function can be a mathematical function that quantifies the degree of temperature difference at various points in the cold storage space. It can be used to guide the optimization process towards a uniform temperature distribution in the space. In this embodiment, the temperature uniformity objective function can be defined as the standard deviation of temperature at each discrete point, the maximum and minimum temperature difference, or the weighted variance. The energy consumption objective function can be a mathematical function that characterizes the total energy consumption level of the system per unit time or period. It can be used to drive the optimization towards a low-energy-consumption direction. In an exemplary embodiment, the energy consumption objective function can be constructed based on an energy consumption distribution cloud map or simulated cumulative power consumption data, and may include a weighted sum of subsystems such as compressors and fans.

[0152] The multi-objective optimization function can be a joint optimization expression that integrates the temperature uniformity objective function and the energy consumption objective function, which can be used to achieve synergistic optimization of refrigeration quality and energy-saving performance. Furthermore, the multi-objective optimization function can be constructed using methods such as weighted summation, ε-constraint method, or Pareto front modeling. The radial basis function kernel function can be a type of kernel function used in support vector machines (SVMs) to measure high-dimensional nonlinear similarity, enabling the SVM to fit the complex nonlinear mapping between parameter vectors and constraint response vectors. The SVM can be a supervised learning model based on structural risk minimization, which can replace high-cost physical simulations and quickly predict constraint responses during the optimization process. In this embodiment, the SVM can be trained using parameter-response sample pairs generated by offline simulation to learn the implicit mapping from control parameters to system responses. The parameter vector can be a numerical vector form representing a certain initial control parameter, which can be used as input features of the SVM, including control dimensions such as compressor frequency and fan speed. The constraint response vector can be a set of system constraint-related response indicators triggered by the corresponding parameter vector, which can be used as output labels of the SVM, including temperature fluctuation amplitude and energy consumption change rate.

[0153] The mapping relationship can be a nonlinear functional relationship from parameter vectors to constraint response vectors established by support vector machines. It can be used to replace time-consuming simulations in optimization iterations, achieving rapid response prediction. KKT conditions can be a set of mathematical equations used in nonlinear programming to determine the necessary conditions for local optima. They can be used to transform constrained optimization problems into a solvable system of equations, ensuring that the solution satisfies first-order optimality. The constrained optimization equation system can be a system of simultaneous equations constructed based on the mapping relationship between KKT conditions and support vector machines. It can be used to accurately characterize the mathematical structure of multi-objective constrained optimization problems. In this embodiment, the constrained optimization equation system can include zero gradient of the Lagrange function, primal constraints, dual feasibility, and complementary relaxation conditions. Sequential quadratic programming is a classic numerical method for solving nonlinear constrained optimization problems. It can be used to efficiently and stably solve the constrained optimization equation system, obtaining high-quality optimal solutions. In a specific embodiment, sequential quadratic programming can locally approximate the original problem as a quadratic programming subproblem and solve it in each iteration.

[0154] The optimal candidate control parameters can be optimized control parameters that satisfy the KKT conditions, obtained by solving the sequential quadratic programming method. They can represent the local optimal strategy in the neighborhood of the current initial control parameters. Lagrange multipliers can be the dual variables corresponding to the constraints in the KKT conditions. They can be used to reflect the sensitivity and activity of each constraint at the optimal solution, and to verify the effectiveness of the constraints. The initial parameters can be the numerical representation of the original initial control parameters participating in this optimization iteration, and can be used as a benchmark for convergence judgment. The norm difference can be a measure of the vector distance between the optimal candidate control parameters and the initial parameters, and can be used to determine whether the optimization process has produced substantial improvement. In this embodiment, the norm difference can be calculated using the L2 norm. The preset convergence threshold can be a numerical boundary for determining whether the optimization is effective, and can be used to avoid misjudging small perturbations or numerical noise as effective optimization results.

[0155] The temperature fluctuation amplitude corresponding to each initial control parameter is determined by analyzing the time cumulative effect of the temperature change rate in the dynamic response curve of the heat load, or by directly extracting the range of the simulated temperature sequence. Furthermore, this operation can be achieved by extracting the peak-to-peak value of the simulated temperature sequence, thus transforming the dynamic response into a constrained stability index. Constraint equations are constructed based on the preset maximum allowable temperature fluctuation of the refrigeration condition and the current temperature fluctuation amplitude corresponding to each initial control parameter. This can be achieved by establishing an inequality constraint: temperature fluctuation amplitude ≤ preset maximum allowable temperature fluctuation. Further, this operation can be achieved by formalizing the inequality relationship between the temperature fluctuation amplitude and the preset maximum allowable temperature fluctuation, thus ensuring that all candidate solutions meet the rigid requirements of refrigeration quality. Constrained optimization parameters are obtained through conjugate gradient iteration calculation. This can be achieved by performing a conjugate gradient search within the feasible region defined by the constraint equations to find improvement directions that satisfy the constraints. Further, this operation can be achieved by handling inequality constraints using the projected conjugate gradient method and combining it with Armijo line search to ensure reasonable step size, thereby generating high-quality initial optimization values ​​and accelerating subsequent multi-objective solutions.

[0156] A multi-objective optimization function is established based on the constrained optimization parameters, comprising an objective function for temperature uniformity and an objective function for energy consumption. This can be achieved by combining the two objective functions into a unified optimization expression according to weights or priorities. Furthermore, this operation can be implemented by using a weighted sum method to construct a scalarized objective, or by employing an ε-constraint method to transform an objective into a constraint, thereby achieving a synergistic optimization orientation towards quality and energy efficiency. Specifically, the multi-objective optimization function based on confidence intervals is used to select the first candidate control parameter from the initial control parameters, balancing temperature uniformity and energy consumption. The calculation formula for this function is as follows:

[0157]

[0158] Constraints:

[0159]

[0160] In the formula, The control parameter vector includes compressor operating frequency, fan speed, etc.; F(x) is a multi-objective optimization function, the objective of which is to minimize the value of this function. Current control parameters Temperature fluctuation amplitude; Current control parameters The system energy consumption is as follows; and These are the reference temperature fluctuation value and the reference energy consumption value, respectively. and These are weighting coefficients, corresponding to the importance of the temperature uniformity evaluation parameter and the system energy consumption evaluation parameter, respectively. This is a preset maximum allowable temperature fluctuation, i.e., a constraint condition; This represents the physical range of values ​​for the control parameters. It should be understood that refrigeration control faces two conflicting objectives: ensuring temperature uniformity for food safety and reducing system energy consumption to reduce operating costs. This formula, through a weighted sum of squares, unifies these two physical quantities into a quantifiable evaluation index F(x). Using the constraints in the formula, dangerous control parameter combinations that could lead to excessive temperature and damage to goods can be eliminated first. By solving for the minimum value of this function, the system can select the first candidate control parameter that best meets both energy-saving and safety requirements from a vast pool of initial control parameters, thereby enriching the control parameter library and ensuring that the starting point for subsequent online adaptive control is optimal.

[0161] The mapping relationship between the parameter vector and the constraint response vector corresponding to each initial control parameter is obtained through a support vector machine with a radial basis function kernel. This can be achieved by training a support vector regression model using offline simulation data to learn the nonlinear mapping from parameters to response. Furthermore, this operation can be implemented by using cross-validation to select the optimal kernel width and introducing an incremental learning mechanism to update the mapping model, thus replacing high-cost physical simulation and supporting rapid response prediction and gradient estimation. Based on the mapping relationship, a system of constrained optimization equations is constructed using KKT conditions. This can be achieved by substituting the differentiable response function provided by the support vector machine into the KKT conditions, forming a system of equations including the Lagrange function gradient, the original constraints, and complementary conditions. Furthermore, this operation can be achieved by using smooth approximations for non-differentiable points and introducing slack variables to handle strict inequalities, thus transforming the optimization problem into a numerically solvable mathematical system. The optimal candidate control parameters and Lagrange multipliers corresponding to each initial control parameter are obtained by solving a sequential quadratic programming method. This can be achieved by iteratively solving the quadratic programming subproblem derived from the KKT conditions until convergence. Furthermore, this operation can be achieved by using Trust Region SQP to guarantee global convergence and combining sparse matrix techniques to accelerate the solution of large-scale problems, thereby obtaining high-quality control parameters that satisfy the theoretical optimality conditions. Determining whether the norm difference between the optimal candidate control parameter and the initial control parameter is less than a preset convergence threshold can be done by calculating the L2 norm distance between them and comparing it with the threshold. Furthermore, this operation can be implemented by setting a reasonable L2 norm threshold, thereby filtering out optimization results with substantial improvements and avoiding the inclusion of invalid parameters in the database. If the norm difference is less than the preset convergence threshold, the optimal candidate control parameter is stored as the first candidate control parameter. This can be achieved by including the optimal candidate control parameter in the first candidate set only when the optimization brings significant improvement. Furthermore, this operation can be implemented by setting dynamic or static convergence thresholds, thereby ensuring the effectiveness and diversity of strategies in the control parameter database.

[0162] Taking the selection of high-precision temperature control parameters for pharmaceutical cold storage as an example, the AI-based adaptive energy-saving intelligent control method for cold storage conditions in this embodiment can be implemented during the parameter library construction stage of a vaccine storage cold storage. The system extracts the temperature fluctuation amplitude under each initial control parameter from the dynamic response curve of the heat load and finds that some strategies fluctuate by ±2.5℃ after door opening disturbance, exceeding the preset maximum allowable temperature fluctuation of ±1.5℃ for pharmaceutical cold chain. Based on this, constraint equations are constructed, and initial values ​​that satisfy the constraints are generated through conjugate gradient iteration. Subsequently, a multi-objective optimization function with temperature standard deviation and compressor power consumption as dual objectives is established. The parameter-response mapping in 1000 sets of simulation samples is learned using radial basis kernel support vector machine. Based on this mapping, KKT conditions are applied to construct a set of optimization equations, which are then solved using sequential quadratic programming. Finally, only the optimization results with a norm difference greater than 0.3 (preset convergence threshold) are retained as the first candidate control parameters to ensure that the storage strategies have significantly better temperature control performance than the initial schemes.

[0163] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an artificial intelligence-based adaptive energy-saving intelligent control program for refrigeration conditions. When the artificial intelligence-based adaptive energy-saving intelligent control program for refrigeration conditions is executed by a processor, it implements the steps of the artificial intelligence-based adaptive energy-saving intelligent control method for refrigeration conditions as described above.

[0164] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An adaptive energy-saving intelligent control method for refrigeration operating conditions based on artificial intelligence, characterized in that, The method includes: Real-time operating condition monitoring data of the target cold storage space is acquired, and coupling characteristic analysis is performed on the monitoring data to obtain the coupling characteristic values ​​of flow field-heat load. Match the energy-saving optimization control model corresponding to the coupling feature value, so as to call the initial control parameters from the preset control parameter library through the energy-saving optimization control model; Simulate the dynamic response process corresponding to the target cold storage space based on the initial control parameters; The initial control parameters are adjusted based on the dynamic response, and adaptive energy-saving control of the refrigeration condition is performed based on the adjusted control parameters.

2. The adaptive energy-saving intelligent control method for refrigeration operating conditions based on artificial intelligence as described in claim 1, characterized in that, The monitoring data includes flow field characteristic parameters and heat load characteristic parameters. The process involves acquiring real-time operating condition monitoring data of the target cold storage space and performing coupling characteristic analysis on the monitoring data to obtain the coupled characteristic values ​​of the flow field and heat load, including: The flow field characteristic parameters and heat load characteristic parameters of the target cold storage space are obtained, wherein the flow field characteristic parameters include flow velocity distribution parameters and turbulence fluctuation parameters, and the heat load characteristic parameters include instantaneous rate of change parameters and heat leakage increment parameters; The particle image velocimetry method is used to extract the coordinates of the feature nodes from the coupling feature parameters, and the flow field transport equation and the heat load coupling equation are constructed based on the coordinates of the feature nodes to obtain the coupling coefficient parameters; The coupling feature vector is obtained based on the coupling coefficient parameter, and the coupling feature vector is normalized to obtain the coupling feature value. The coupling feature vector includes the flow field pulsation intensity, the heat load fluctuation amplitude, and the transport lag phase difference.

3. The adaptive energy-saving intelligent control method for refrigeration operating conditions based on artificial intelligence as described in claim 1, characterized in that, The step of calling initial control parameters from a preset control parameter library through the energy-saving optimization control model includes: Obtain the parameter features corresponding to each combination of control parameters in the control parameter library, and calculate the similarity between each parameter feature and the coupled feature value through the mapping relationship preset in the energy-saving optimization control model; When the similarity exceeds a preset threshold, a preset energy-saving optimization rule is invoked to optimize the current parameter combination and generate the initial control parameters corresponding to the optimized parameter combination.

4. The adaptive energy-saving intelligent control method for refrigeration operating conditions based on artificial intelligence as described in claim 1, characterized in that, The process of simulating the dynamic response of the target cold storage space based on the initial control parameters includes: A virtual cold storage space thermodynamic flow field model is constructed based on the discrete compressor frequency and fan speed control points of the initial control parameters. The virtual cold storage space thermodynamic flow field model is then meshed using hexahedral mesh elements to obtain a meshed model. The SIMPLE algorithm is used to establish the flow field-thermal coupled transport equation, and the temperature field data and velocity field data of the mesh model are obtained by calculating the flow field-thermal coupled transport equation and the CFD solver. The temperature field data and flow velocity field data are subjected to feature extraction by a three-layer convolutional neural network to obtain the transient change feature mapping relationship between the temperature field data and the flow velocity field data; The real-time operating data collected by the sensor is processed according to the feature mapping relationship to obtain the temperature fluctuation data and cold supply and demand difference data of the target cold storage space during the control process. By fusing the temperature fluctuation data and the cooling supply-demand difference data using a Kalman filter, the actual heat load change curve and the actual flow field distribution curve are obtained. The reliability of the initial control parameters is verified based on the actual heat load change curve, the actual flow field distribution curve, the temperature field data, and the flow velocity field data.

5. The AI-based adaptive energy-saving intelligent control method for refrigeration conditions as described in claim 4, characterized in that, The step of adjusting the initial control parameters based on the dynamic response and performing adaptive energy-saving control of the refrigeration condition based on the adjusted control parameters includes: When verifying the reliability of the initial control parameters, a fuzzy evaluator based on confidence intervals is used to calculate the temperature uniformity evaluation parameters and system energy consumption evaluation parameters corresponding to the initial control parameters, and a sequence of parameters to be adjusted is generated based on the comparison results of the temperature uniformity evaluation parameters and the system energy consumption evaluation parameters with preset thresholds respectively. A gradient boosting tree regressor is trained based on the sequence of parameters to be adjusted. The parameter adjustment amount is predicted by the gradient boosting tree regressor. The parameter values ​​of the initial control parameters are adjusted according to the parameter adjustment amount to obtain the discrete control point coordinates of the updated control parameters. The coordinates of the discrete control points for the updated control parameters are fitted using NURBS curves, and the speed planning is dynamically adjusted based on the NURBS curves under the constraints of the maximum allowable temperature fluctuation and the maximum load rate of the compressor to obtain the speed planning curve. The curvature values ​​at each discrete point of the speed planning curve are calculated. Transition points are inserted in the curvature abrupt change region by spline interpolation to obtain the adjusted control parameters. Adaptive energy-saving control of the refrigeration condition is then performed based on the adjusted control parameters.

6. The adaptive energy-saving intelligent control method for refrigeration operating conditions based on artificial intelligence as described in claim 1, characterized in that, Before acquiring the real-time operating condition monitoring data of the target cold storage space, the method further includes: The enclosure structure parameters and cargo stacking parameters of the cold storage are obtained to construct an offline simulation model of the unsteady flow field coupled transient heat load corresponding to the cold storage. The offline simulation model is analyzed to obtain the temperature distribution cloud map and energy consumption distribution cloud map corresponding to the cold storage conditions under each preset initial control parameter. The discrete points of each initial control parameter in the temperature distribution cloud map and the gradient of change in the energy consumption distribution cloud map are extracted respectively, and the temperature change rate and energy consumption change rate corresponding to each initial control parameter at each discrete point are calculated to obtain the dynamic response curve of heat load under different initial control parameters for the refrigeration condition. Based on the dynamic response curve of the heat load, multiple first candidate control parameters are selected from the initial control parameters; The control parameters in each of the first candidate control parameters are optimized according to the preset parameter optimization algorithm to obtain several optimized control parameter combinations and save them to the preset control parameter library.

7. The artificial intelligence-based adaptive energy-saving intelligent control method for refrigeration conditions as described in claim 6, characterized in that, The enclosure structure parameters include thermal conductivity and heat storage coefficient, and the cargo stacking parameters include porosity and volumetric heat capacity. The process involves acquiring the enclosure structure parameters and cargo stacking parameters of the cold storage facility to construct an offline simulation model of the unsteady flow field coupled with transient heat load corresponding to the cold storage facility. The offline simulation model is then analyzed to obtain temperature distribution cloud maps and energy consumption distribution cloud maps corresponding to the cold storage conditions under each preset initial control parameter, including: The thermal conductivity, heat storage coefficient, and cargo stacking porosity of the enclosure structure of the cold storage are obtained in order to calculate the flow field resistance matrix and heat permeation matrix corresponding to the cold storage space. The flow field resistance matrix is ​​determined by the porosity and stacking height, and the heat permeation matrix is ​​determined by the thermal conductivity and heat storage coefficient. The nodal thermal disturbance is obtained by the flow field resistance matrix and the thermal penetration matrix and mapped to the mesh nodes to establish a finite volume interpolation equation. The mesh nodes are divided into hexahedral elements and the mesh size is reduced to one-third of the basic mesh size in the near-wall region. The flow field gradient value is calculated based on the finite volume interpolation equation to obtain the flow field-thermal coupling control equation, which is the offline simulation model of unsteady flow field coupled transient thermal load; wherein, the flow field gradient value is determined by the velocity gradient vector and the temperature gradient vector; Based on the flow field-thermal coupling control equations, a set of nonlinear cold energy supply and demand equations is established to obtain temperature distribution cloud maps and energy consumption distribution cloud maps corresponding to each preset initial control parameter under the refrigeration condition; wherein, the set of nonlinear cold energy supply and demand equations is solved using the SIMPLEC algorithm until the residual is less than the preset value.

8. The adaptive energy-saving intelligent control method for refrigeration conditions based on artificial intelligence as described in claim 6, characterized in that, The calculation of the temperature change rate and energy consumption change rate corresponding to each of the initial control parameters at each discrete point, to obtain the dynamic response curve of the heat load under different initial control parameters for the refrigeration condition, includes: Obtain the control sequence corresponding to each of the initial control parameters, and obtain the coordinate dataset of discrete operating points on the control sequence; wherein, different initial control parameters correspond to different compressor operating frequencies and fan speeds; The instantaneous cooling output value is obtained by fitting the discrete operating points with a Bézier curve; wherein the instantaneous cooling output value is calculated through the mapping relationship between control parameters and cooling output; A thermodynamic equation is established based on the instantaneous cooling output value, and the heat load change rate at discrete points is obtained by solving the Euler integral formula; wherein the heat load change rate is obtained by integrating the heat load over time. A state equation is constructed based on the rate of change of heat load at the discrete points and the temperature fluctuation on the temperature distribution cloud map. The temperature dynamic change function is obtained by solving the least squares method, and the dynamic response curve of heat load is obtained based on the temperature dynamic change function.

9. The adaptive energy-saving intelligent control method for refrigeration conditions based on artificial intelligence as described in claim 6, characterized in that, The step of selecting multiple first candidate control parameters from a plurality of initial control parameters based on the dynamic response curve of the heat load includes: The temperature fluctuation amplitude corresponding to each of the initial control parameters is determined by the dynamic response curve of the heat load. Constraint equations are constructed based on the preset maximum allowable temperature fluctuation of the refrigeration condition and the current temperature fluctuation amplitude corresponding to each of the initial control parameters, so as to obtain the constraint optimization parameters through conjugate gradient iteration calculation. Based on the constraint optimization parameters, a multi-objective optimization function is established with a temperature uniformity objective function and an energy consumption objective function, and the mapping relationship between the parameter vector and the constraint response vector corresponding to each initial control parameter is obtained through a support vector machine with radial basis function kernel. Based on the mapping relationship, a set of constrained optimization equations is constructed using KKT conditions, and the optimal candidate control parameters and Lagrange multipliers corresponding to each initial control parameter are obtained by solving the sequential quadratic programming method. If the norm difference between the optimal candidate control parameter and the initial parameter corresponding to the initial control parameter is less than a preset convergence threshold, then the optimal candidate control parameter is stored as the first candidate control parameter.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an AI-based adaptive energy-saving intelligent control program for refrigeration conditions. When the AI-based adaptive energy-saving intelligent control program for refrigeration conditions is executed by a processor, it implements the steps of the AI-based adaptive energy-saving intelligent control method for refrigeration conditions as described in any one of claims 1 to 9.

Images