Energy Management and Recovery Method and System for Cold Chain Supply Chain

Abstract

The present invention relates to the technical field of cold chain supply chains, and discloses an energy management and recovery method and system for a cold chain supply chain. The method includes: collecting and preprocessing operation parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data; performing feature extraction and local density distribution calculation on the multi-dimensional feature data to obtain energy consumption pattern clustering data; constructing an equipment operation knowledge graph based on the energy consumption pattern clustering data; performing abnormal feature analysis on the equipment operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model to obtain equipment energy consumption abnormal feature data; generating an initial energy recovery plan according to the equipment energy consumption abnormal feature data; applying the initial energy recovery plan to the cold chain system, collecting operation efficiency data and load matching data, and performing online iterative optimization to obtain a target energy management plan. The present invention realizes the dynamic adjustment and continuous optimization of the energy recovery plan, and improves the energy utilization efficiency of the cold chain system.

Description

Technical Field

[0001] The present invention relates to the technical field of cold chain supply chains, and particularly to an energy management and recovery method and system for cold chain supply chains. Background Art

[0002] With the rapid development of the cold chain logistics industry, the scale of cold chain facilities has been continuously expanding, energy consumption has been increasing continuously, and the problem of high energy consumption has become increasingly prominent. The energy management methods of traditional cold chain systems mainly rely on manual experience for adjustment, lacking systematic energy recovery and optimization strategies, resulting in a large amount of waste heat energy being directly discharged into the environment, causing energy waste.

[0003] At present, there are generally problems in the energy management of cold chain systems such as difficult to accurately control equipment operation parameters, low energy recovery efficiency, and poor system load matching. Since the cold chain system involves multiple core components such as compressors, condensers, and evaporators, there are complex energy transfer relationships and thermal coupling characteristics among the components, and traditional single control strategies are difficult to achieve system-level energy optimization management, restricting the improvement of the overall energy efficiency of the cold chain system. In addition, the multi-source heterogeneous data generated during the operation of the cold chain system has not been fully utilized, lacking intelligent data analysis and decision support means. Problems such as untimely monitoring of system operation status, lagging identification of abnormal working conditions, and insufficient optimization of energy recovery schemes lead to unsatisfactory energy management effects and are difficult to meet the urgent needs of modern cold chain logistics for energy conservation and consumption reduction. Summary of the Invention

[0004] The present invention provides an energy management and recovery method and system for cold chain supply chains, which is used to realize the dynamic adjustment and continuous optimization of energy recovery schemes and improve the energy utilization efficiency of cold chain systems.

[0005] In a first aspect, the present invention provides an energy management and recovery method for cold chain supply chains, and the energy management and recovery method for cold chain supply chains includes:

[0006] Collect and preprocess the operation parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data;

[0007] Extract features and calculate the local density distribution of the multi-dimensional feature data to obtain energy consumption pattern clustering data;

[0008] Construct a device operation knowledge graph according to the energy consumption pattern clustering data;

[0009] Perform abnormal feature analysis on the device operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model to obtain device energy consumption abnormal feature data;

[0010] Generate an initial energy recovery scheme according to the device energy consumption abnormal feature data;

[0011] Apply the initial energy recovery scheme to the cold chain system, collect operation efficiency data and load matching data, and perform online iterative optimization to obtain the target energy management scheme.

[0012] In a second aspect, the present invention provides an energy management and recovery system for a cold chain supply chain. The energy management and recovery system for the cold chain supply chain includes:

[0013] A collection module, configured to collect and preprocess operation parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data;

[0014] A calculation module, configured to perform feature extraction and local density distribution calculation on the multi-dimensional feature data to obtain energy consumption pattern clustering data;

[0015] A construction module, configured to construct a device operation knowledge graph according to the energy consumption pattern clustering data;

[0016] An analysis module, configured to perform abnormal feature analysis on the device operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model to obtain device energy consumption abnormal feature data;

[0017] A generation module, configured to generate an initial energy recovery scheme according to the device energy consumption abnormal feature data;

[0018] An optimization module, configured to apply the initial energy recovery scheme to the cold chain system, collect operation efficiency data and load matching data, and perform online iterative optimization to obtain the target energy management scheme.

[0019] In a third aspect, the present invention provides an energy management and recovery device for a cold chain supply chain, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the energy management and recovery device for the cold chain supply chain executes the above-mentioned energy management and recovery method for the cold chain supply chain.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on a computer, the computer is caused to execute the above-mentioned energy management and recovery method for the cold chain supply chain.

[0021] In the technical solution provided by the present invention, through a multi-level data acquisition and preprocessing mechanism, a comprehensive perception of the operating parameters and environmental parameters of the cold chain system is achieved, providing a high-quality data basis for energy management; by adopting a clustering analysis method based on energy consumption distance, the energy consumption pattern characteristics of the system are accurately identified, providing data support for the formulation of energy recovery strategies; combining knowledge graph technology, a time-series correlation model between devices is constructed, deeply mining the energy transfer laws and device state evolution characteristics in the system; through the feature fusion mechanism of the hybrid neural network model, an accurate identification of the abnormal energy consumption state of the device is realized, providing a reliable basis for the generation of energy recovery solutions; based on the operation efficiency evaluation and load matching analysis of multiple time scales, a hierarchical energy management optimization mechanism is established to ensure the stability and efficiency of the system operation; through a closed-loop online iterative optimization strategy, the dynamic adjustment and continuous optimization of the energy recovery solution are realized, significantly improving the energy utilization efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0023] Figure 1 It is a schematic flowchart of the energy management and recovery method for the cold chain supply chain provided by the embodiment of the present application;

[0024] Figure 2 It is a schematic block diagram of the structure of the energy management and recovery system for the cold chain supply chain provided by the embodiment of the present application;

[0025] Figure 3 It is a schematic block diagram of the structure of the energy management and recovery device for the cold chain supply chain provided by the embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0027] The flowcharts shown in the drawings are only illustrative examples, and do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may change based on the actual situation.

[0028] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0029] It should be further understood that the term "and / or" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] The following will, with reference to the accompanying drawings, elaborate on some embodiments of the present application. Without conflict, the following embodiments and the features in the embodiments may be combined with each other.

[0031] Please refer to Figure 1 , Figure 1 , which is a schematic flowchart of the energy management and recovery method for a cold chain supply chain provided by an embodiment of the present application. As Figure 1 shown, the energy management and recovery method for a cold chain supply chain provided by an embodiment of the present application includes steps S100 to S600.

[0032] Step S100: Collect and preprocess the operating parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data;

[0033] It can be understood that the execution subject of the present invention can be an energy management and recovery system for a cold chain supply chain, or a terminal or a server. Specifically, no limitation is made here. The server is taken as the execution subject in the embodiments of the present invention for illustration.

[0034] Specifically, data is collected from the compressor unit to obtain key indicators such as the pressure of the refrigerant, the temperature of the refrigerant, the current of the motor, and the vibration frequency. After the data collection is completed, these data are normalized to ensure that data with different dimensions can be compared under the same standard. After normalization, the operating characteristic data of the compressor unit are obtained, which reflect the real-time operating conditions of the compressor unit and provide basic data support for subsequent energy management. Data is collected on the operating status of the condenser. The collected parameters include key indicators such as the heat exchange pressure, the condensation temperature, the fan speed, and the load power. After the data collection is completed, these data are standardized to ensure the consistency and reliability of the data, forming the operating characteristic data of the condenser, which reflects the important role and operating efficiency of the condenser in the cold chain system. At the same time, data is collected on the operating status of the evaporator, including parameters such as the evaporation pressure, the evaporation temperature, the defrosting status, and the refrigerating capacity. Through dimension unification processing, it is ensured that different types of data remain consistent during the analysis process, and the operating characteristic data of the evaporator are obtained. The operating characteristic data of the compressor unit, the condenser, and the evaporator are subjected to time series alignment processing to ensure effective comparison and analysis of the operating data of different devices in the time dimension, forming a complete device operating characteristic matrix. Data is collected on the temperature, humidity, and door status in the cold storage. To improve the smoothness of the data, the sliding window method is used to smooth the collected environmental parameters to obtain environmental parameter characteristic data. This process can help understand the energy consumption performance of the cold chain system under different environmental conditions. In terms of ensuring the accuracy of the data, local linear regression correction is performed on the outliers in the device operating characteristic matrix to effectively eliminate the noise and interference in the data and obtain more reliable first operating characteristic data. For the missing values in the first operating characteristic data, the method of time series interpolation is used to ensure the integrity and continuity of the data, obtaining the processed second operating characteristic data. The second operating characteristic data and the environmental parameter characteristic data are subjected to feature fusion and dimensionality reduction processing to form multi-dimensional characteristic data. The data from different sources are integrated so that the final obtained data set contains both information on the operating status of the device and can reflect the environmental impact. At the same time, dimensionality reduction processing is to reduce the complexity of the data and improve the efficiency of subsequent analysis, so that useful information can still be obtained quickly even when the data volume is large.

[0035] Step S200: Feature extraction and local density distribution calculation are performed on the multi-dimensional characteristic data to obtain energy consumption pattern clustering data;

[0036] Specifically, the multi-dimensional feature data is grouped according to device operation characteristics, environmental characteristics, and energy consumption characteristics. The mean and standard deviation are calculated for each group of features to obtain feature statistical parameters, which reflect the distribution and change trends of different features. An energy consumption feature normalization matrix is constructed based on the feature statistical parameters to perform standardization transformation on the multi-dimensional feature data, ensuring that different features are compared under the same standard to obtain standardized feature data. Based on the standardized feature data, a feature weighting matrix is constructed by setting the weight coefficients of device operation characteristics, environmental characteristics, and energy consumption characteristics to obtain weighted feature data, enhancing the attention to important features and ensuring that the energy consumption analysis can reflect the actual situation. Using the Euclidean distance between any two data samples in the weighted feature data and the feature weighting matrix, a sample distance matrix is obtained. To improve the accuracy of clustering, a search radius threshold is set, and the number of neighboring samples of each sample point is counted within this range to identify the density distribution of the samples in the local area, obtaining local density distribution data. According to the local density distribution data, for each sample point, a set of sample points with a larger density value is found, and the distance to the nearest sample point in this set is calculated to generate density distance data. Combining the product of the local density value and the density distance can effectively identify the density aggregation points, obtaining clustering center data, which reflects the main characteristics of the energy consumption pattern. Based on the clustering center data and the sample distance matrix, the remaining sample points are assigned to the nearest clustering center, ensuring that each sample point is reasonably classified into the energy consumption pattern that best matches its characteristics, obtaining the final energy consumption pattern clustering data.

[0037] Step S300: Construct a device operation knowledge graph according to the energy consumption pattern clustering data;

[0038] Specifically, the device information, energy consumption data, and operating parameters in the energy consumption pattern clustering data are classified and encoded. Different types of information are effectively distinguished and represented in a standardized form by encoding, facilitating subsequent graph construction. A node attribute table is constructed, where the nodes represent the devices in the system and their operating characteristics, forming the basic node data. Based on the basic node data, a device composition relationship graph is constructed. The physical composition relationship and operating parameters of the compressor unit, condenser, and evaporator are transformed into physical connection relationships between nodes, obtaining device relationship data. The device relationship data reflects the hardware connection situation between devices and clarifies how the operating parameters of different devices affect each other. Using the device relationship data, an energy consumption transfer path is established, converting the energy flow process in the cold chain system into a directed edge relationship, obtaining energy consumption relationship data. The energy consumption relationship data describes the flow mode of energy between different devices, enabling the energy transfer process to be modeled in the form of a graph, providing an overall perspective on the energy consumption of the cold chain system, and better understanding the interaction between devices and the energy transfer path. According to the energy consumption relationship data, a device state transition link is established, converting the dynamic changes in the device operating state into a time-sequence state sequence, obtaining state migration data. By recording the state changes of the device at different time periods, it is possible to analyze the performance of the device under different operating conditions and how these changes affect the overall energy consumption of the system. Multimodal information integration is performed on the state migration data, mapping physical quantities such as temperature, electricity, and pressure to the dynamic attributes of the corresponding nodes, obtaining node feature data. Based on the node feature data, association rule mining is carried out to construct a time-sequence dependence relationship matrix between nodes, obtaining association rule data, revealing the dependence relationship between different device nodes, especially their interaction in the time series, thereby providing a basis for the coordinated operation between devices. The association rule data is converted into a graph data structure, the time-sequence edge weights between nodes are constructed, and the connection relationship between nodes is sparsified, obtaining graph structure data. The sparsification process reduces unnecessary connections between nodes, reduces the computational complexity, makes the structure of the knowledge graph clearer, and thus improves the efficiency of graph analysis. Node embedding and edge embedding calculations are performed on the graph structure data to vectorize the time-sequence characteristics of the nodes and edges in the graph. Through node embedding and edge embedding, the characteristics of device operation are mapped to a high-dimensional vector space. These vector representations retain the relationship between devices and the time-sequence characteristics of energy transfer, and enable complex device interactions to be analyzed and modeled through simple mathematical operations. An equipment operation knowledge graph is formed. This knowledge graph contains the physical connection relationship of devices, the energy consumption transfer path, and the state transition link, and integrates multimodal operation feature data. Through the data representation of the graph structure, the time-sequence characteristics and association rules between devices are integrated. The equipment operation knowledge graph can intuitively reflect the energy transfer and coordinated operation relationship between devices in the cold chain system.

[0039] Step S400: Analyze the abnormal features of the device operation knowledge graph and the energy consumption pattern clustering data through the hybrid neural network model to obtain the device energy consumption abnormal feature data;

[0040] Specifically, for the equipment operation knowledge graph, graph convolutional layers at the node level and edge level are constructed respectively. The node-level graph convolutional layer adopts three-layer convolutional operations, and the sizes of the convolutional kernels are 32, 64, and 128 in sequence to gradually extract higher-level graph structure features. The edge-level graph convolutional layer contains two-layer convolutional operations, and the sizes of the convolutional kernels are 16 and 32 respectively to capture the edge information between nodes and obtain more comprehensive graph structure features. These graph structure features contain the representation of equipment nodes in the cold chain system and the complex relationships between them. The graph structure features are input into the relationship perception layer, which includes the equipment composition relationship sub-layer, the energy consumption transfer relationship sub-layer, and the state transition relationship sub-layer. Each sub-layer is set with 64 feature channels, and the information from different relationships is fused through the message passing mechanism to obtain a comprehensive graph feature vector. At the same time, for the energy consumption pattern clustering data, a multi-branch feature extraction network is constructed. The network includes a compressor unit feature branch, a condenser feature branch, and an evaporator feature branch. Each branch contains two fully connected layers, and the numbers of neurons are 128 and 64 respectively to extract important features from different characteristics of the equipment and obtain equipment feature vectors to describe the operating states of each equipment. The equipment feature vectors are input into the temporal feature extraction layer. This layer adopts a gated recurrent unit structure, including an input gate, a reset gate, and an update gate to effectively capture the temporal change characteristics of the equipment features. The hidden state dimension is set to 128, and the temporal window length is 24, indicating that the operating data within 24 time steps is modeled to obtain temporal feature vectors, revealing the time dependence and trend changes of the equipment during operation. The graph feature vector and the temporal feature vector are input into the cross-attention module. The module is set with 8 attention heads, and the dimension of each attention head is 32. Adaptive fusion is performed by calculating the correlation weights between features. The role of the cross-attention mechanism is to dynamically focus on the importance between the graph features and the temporal features to capture the interaction information between different modality data and obtain multi-modal fusion features, representing the complex states and energy consumption patterns of the equipment in the cold chain system. The multi-modal fusion features are input into the feature enhancement network for processing. The feature enhancement network contains three residual blocks, each of which consists of two convolutional layers and a shortcut connection, and batch normalization layers and ReLU activation functions are set between the residual blocks. The design of the residual block aims to solve the problem of gradient disappearance in deep networks while retaining the useful information of the input features. Through the operation of the residual blocks, enhanced feature vectors are obtained. The enhanced feature vectors are input into the abnormal pattern recognition layer, which includes an energy consumption anomaly sub-network, a temperature anomaly sub-network, and a pressure anomaly sub-network. Each sub-network consists of three fully connected layers, and the numbers of neurons are 256, 128, and 64 respectively. Each sub-network is specifically used to identify specific types of abnormal patterns. The energy consumption anomaly sub-network is used to detect whether the energy consumption of the equipment is abnormal, the temperature anomaly sub-network is used to identify abnormal temperature fluctuations, and the pressure anomaly sub-network focuses on the pressure state of the equipment.These sub-networks obtain abnormal pattern vectors by comprehensively analyzing enhanced feature vectors. Conduct a comprehensive analysis of the abnormal pattern vectors. Set the weight coefficients of energy consumption anomaly, temperature anomaly, and pressure anomaly to 0.4, 0.3, and 0.3 respectively. Through the method of weighted combination, synthesize the importance of different types of anomalies, and finally obtain the device energy consumption anomaly characteristic data.

[0041] Step S500: Generate an initial energy recovery plan based on the device energy consumption anomaly characteristic data;

[0042] Specifically, a recovery priority matrix is constructed for the device energy consumption anomaly characteristic data. By grading the anomaly degree based on the weight of energy consumption anomaly 0.4, the weight of temperature anomaly 0.3, and the weight of pressure anomaly 0.3, the anomaly priority data is obtained. By identifying the priorities corresponding to different types of anomalies, during the energy recovery process, the energy consumption anomalies with the greatest impact are resolved first, effectively improving the recovery efficiency. An energy recovery type mapping table is established according to the anomaly priority data, corresponding different anomaly characteristics to the corresponding energy recovery types. During this mapping process, a correspondence relationship is established between the energy consumption anomaly and three recovery modes, namely compressor waste heat recovery, condensation heat recovery, and system energy balance, to obtain the recovery type data. Feature augmentation is performed on the recovery type data. By introducing historical recovery efficiency indicators and device operating status indicators, a feature enhancement matrix is constructed to obtain the recovery feature data. Combining the historical performance of the device and the current operating status enhances the descriptive ability of the recovery type data, making the generation of the recovery strategy more in line with the actual operating conditions and historical efficiency performance of the device. The recovery feature data is input into the input layer of the Softmax regression classifier. After being processed by two hidden layers, with 128 neurons set in each layer and the BatchNorm and ReLU activation functions configured, the classification feature data is obtained. BatchNorm accelerates the training speed of the model and stabilizes the training process, while the ReLU activation function introduces non-linearity, enabling the classifier to model complex feature relationships and obtain the classification feature data. Based on the classification feature data, the probability distribution of the recovery strategy is generated. Probability scores are given to the compressor waste heat recovery parameters, condensation heat recovery parameters, and system energy balance parameters to obtain the strategy probability data. The generation process of the strategy probability data reflects the applicability and its probability distribution of different recovery strategies under the current energy consumption anomaly characteristics. Based on the strategy probability data, a parameter configuration matrix is constructed, which includes the value ranges of the compressor speed, condensation temperature, and evaporation temperature, to obtain the parameter configuration data. The configuration ranges of these parameters provide feasible operating parameters for the specific implementation of energy recovery to ensure that the recovery process is carried out under reasonable conditions. Operational constraint analysis is performed on the parameter configuration data. Based on the operating boundary conditions of the device and the stability requirements of the system, the corresponding parameter adjustment step sizes are set to obtain the operating parameter data. Setting appropriate parameter adjustment step sizes is to ensure the stable operation of the device during the energy recovery process and avoid system instability caused by drastic changes in parameters. On this basis, according to the time sequence of the system response and the energy transfer link, the execution of the operating parameters is organized in a reasonable order, and finally a control instruction sequence is constructed to obtain the initial energy recovery plan.

[0043] Step S600: Apply the initial energy recovery plan to the cold chain system, collect the operation efficiency data and load matching data, and perform online iterative optimization to obtain the target energy management plan.

[0044] Specifically, the operation parameter data in the initial energy recovery scheme are sent to the control systems of the compressor unit, condenser, and evaporator. In this way, the key equipment of the cold chain system is controlled in real time, and the compressor waste heat recovery data, condensation heat recovery data, and system energy balance data are collected to obtain the first operation data. The first operation data are collected in segments to facilitate the evaluation of the equipment's operation under different time scales. The collection is segmented according to three time scales of 5 minutes, 15 minutes, and 30 minutes, recording the dynamic change process of the equipment operation efficiency parameters and load matching parameters to obtain the second operation data. The segmented collection method can reflect the operation status of the cold chain system in the short term, medium term, and long term. Based on the second operation data, the operation efficiency indicators for the short term, medium term, and long term are calculated respectively, including the compressor energy efficiency ratio, condenser heat transfer efficiency, and evaporator refrigeration coefficient, thereby obtaining the efficiency evaluation data. These indicators can measure the energy efficiency performance of the equipment under different time scales and provide detailed information about the equipment's energy utilization rate. The compressor energy efficiency ratio reflects the relationship between the energy consumption and refrigeration output of the compressor, the condenser heat transfer efficiency reflects its heat transfer performance, and the evaporator refrigeration coefficient is used to evaluate the effectiveness of its refrigeration capacity. Multi-scale load matching analysis is performed on the efficiency evaluation data. By calculating the deviations of the real-time load, hourly load, and daily load from the set values and performing dynamic balance calculations of the load, the load matching data are obtained to ensure that the load of the cold chain system always remains consistent with the set value, thereby maximizing the energy efficiency of the system under various operating conditions. Through load matching analysis, the deviations between the load and the actual operating requirements are identified, and corresponding adjustment measures are taken to maintain the stable operation and efficient energy utilization of the system. The efficiency evaluation data and load matching data are input into the adaptive optimization iterator, and the energy recovery parameters are dynamically adjusted by the stochastic gradient descent method with momentum to obtain the first optimized parameter data. The use of the stochastic gradient descent method can effectively reduce the local minimum problem encountered in the optimization process, while the momentum term accelerates the convergence process, thus finding the global optimal solution faster. Through the adaptive optimization iterator, the energy recovery parameters are continuously adjusted to make the equipment operation more tend to the optimal state. Hierarchical operation constraint verification is performed on the first optimized parameter data. According to the safe operation boundary of the equipment, system stability requirements, and energy balance constraints, the optimized parameters are corrected to ensure that the optimized parameters operate within the safe range of the equipment and at the same time meet the stability requirements of the system. Through multiple verifications, the second optimized parameter data are obtained, thus ensuring that the optimization process will not have a negative impact on the safety of the equipment and the stability of the system. According to the second optimized parameter data, a hierarchical control instruction sequence is constructed. According to the different response requirements of the cold chain system, the execution strategy is reconstructed according to the timing characteristics of the fast response layer, medium-term adjustment layer, and slow optimization layer to obtain the hierarchical control data.The hierarchical control method enables the system to make corresponding adjustments according to different time-scale requirements. The fast response layer is used to handle instantaneous changes, the medium-term adjustment layer is used to adapt to fluctuations in a longer time range, and the slow optimization layer focuses on improving the long-term energy efficiency of the system. The hierarchical control data and the initial energy recovery scheme are optimized through multi-objective fusion, and a closed-loop control process with a prediction-feedback-compensation mechanism is constructed to finally obtain the target energy management scheme. Through closed-loop control, the energy utilization rate is continuously optimized during the continuous prediction, feedback, and compensation processes to ensure that the system can maintain the best energy efficiency under different operating conditions.

[0045] In the embodiment of the present invention, through a multi-level data acquisition and preprocessing mechanism, a comprehensive perception of the operating parameters and environmental parameters of the cold chain system is realized, providing a high-quality data basis for energy management; a clustering analysis method based on energy consumption distance is adopted to accurately identify the energy consumption pattern characteristics of the system, providing data support for the formulation of energy recovery strategies; combined with knowledge graph technology, a time-series association model between devices is constructed to deeply explore the energy transfer law and device state evolution characteristics in the system; through the feature fusion mechanism of the hybrid neural network model, an accurate identification of the abnormal energy consumption state of the device is realized, providing a reliable basis for the generation of the energy recovery scheme; based on the multi-time-scale operating efficiency evaluation and load matching analysis, a hierarchical energy management optimization mechanism is established to ensure the stability and efficiency of the system operation; through a closed-loop online iterative optimization strategy, the dynamic adjustment and continuous optimization of the energy recovery scheme are realized, significantly improving the energy utilization efficiency of the system.

[0046] In a specific embodiment, the process of executing step S100 may specifically include the following steps:

[0047] Data collection and normalization processing are performed on the refrigerant pressure, refrigerant temperature, motor current, and vibration frequency of the compressor unit to obtain the operating characteristic data of the compressor unit; data collection and standardization processing are performed on the heat exchange pressure, condensation temperature, fan speed, and load power of the condenser to obtain the operating characteristic data of the condenser; data collection and dimension unification processing are performed on the evaporation pressure, evaporation temperature, defrosting state, and refrigeration capacity of the evaporator to obtain the operating characteristic data of the evaporator;

[0048] Time-series alignment processing is performed on the operating characteristic data of the compressor unit, the operating characteristic data of the condenser, and the operating characteristic data of the evaporator to obtain the device operating characteristic matrix;

[0049] Data collection is performed on the temperature inside the cold storage, the humidity inside the cold storage, and the door state of the cold storage, and data smoothing processing is performed through the sliding window method to obtain the environmental parameter characteristic data;

[0050] Perform local linear regression correction on the outliers in the device operation feature matrix to obtain the first operation feature data, and perform time series interpolation to complete the missing values in the first operation feature data to obtain the second operation feature data;

[0051] Perform feature fusion and dimensionality reduction processing on the second operation feature data and the environmental parameter feature data to obtain multi-dimensional feature data.

[0052] Specifically, collect and preprocess the operation parameters of the key devices in the cold chain system, including the operation data of the compressor unit, condenser, and evaporator. During the data collection process of the compressor unit, collect key parameters such as refrigerant pressure, refrigerant temperature, motor current, and vibration frequency. To ensure data consistency and model effectiveness, perform normalization processing on these data to convert the operation parameters of the compressor unit to a unified range, usually [0,1]. Through normalization processing, obtain the operation feature data of the compressor unit. Collect the operation data of the condenser, including heat exchange pressure, condensation temperature, fan speed, and load power. Perform standardization processing on these data to convert data with different dimensions to a standard normal distribution to maintain the importance and comparability of each feature in subsequent analysis. The calculation formula for standardization is:

[0053] ;

[0054] Among them, represents the original data, is the mean of the data, is the standard deviation, is the value after standardization. Through standardization processing, obtain the operation feature data of the condenser, reflecting the operation status of the condenser at different time periods. At the same time, detect the operation status of the evaporator, and the collected parameters include evaporation pressure, evaporation temperature, defrosting status, and refrigeration capacity. For the data of the evaporator, since the dimensions of different features are inconsistent, for example, pressure is in pascals, temperature is in degrees Celsius, and refrigeration capacity is in watts, perform dimensionality unification processing on these data to ensure effective comparison and modeling between features. Through the process of dimensionality unification, obtain the operation feature data of the evaporator. Perform time series alignment processing on the operation feature data of the compressor unit, condenser, and evaporator to construct a device operation feature matrix, including the time series operation data of the device. Collect the temperature, humidity, and door status in the cold storage, and perform data smoothing processing through the sliding window method to obtain environmental parameter feature data. The calculation of the sliding window method is expressed as:

[0055]

[0056] Among them, is the smoothed value at time point and is the original data, is the size of the sliding window. The random fluctuations in the environmental data are eliminated by the sliding window method, making the data smoother and more stable, and better reflecting the true change trend of the environment. The outliers in the device operation feature matrix are corrected. The method of local linear regression is used to correct the outliers. Local linear regression corrects the abnormal data by fitting a linear regression line using adjacent data points to obtain the first operation feature data. The missing values in the first operation feature data are interpolated in time series to ensure the continuity and integrity of the data. The interpolation method uses linear interpolation or spline interpolation to obtain the second operation feature data. The second operation feature data and the environmental parameter feature data are subjected to feature fusion and dimensionality reduction processing to obtain multi-dimensional feature data. The device operation features and environmental features are combined to comprehensively reflect the operation status of the cold chain system and environmental impacts. The dimensionality reduction processing uses the principal component analysis method to project the high-dimensional data onto a lower-dimensional subspace to reduce the computational complexity while maintaining the main information of the data, obtaining multi-dimensional feature data.

[0057] In a specific embodiment, the process of executing step S200 may specifically include the following steps:

[0058] The multi-dimensional feature data is grouped according to device operation features, environmental features, and energy consumption features, and the mean and standard deviation are calculated for each group of features to obtain feature statistical parameters;

[0059] An energy consumption feature normalization matrix is constructed based on the feature statistical parameters, the multi-dimensional feature data is subjected to standardization transformation to obtain standardized feature data, and based on the standardized feature data, a feature weighted matrix is constructed through the device operation feature weight coefficient, environmental feature weight coefficient, and energy consumption feature weight coefficient to obtain weighted feature data;

[0060] Using the Euclidean distance between any two data samples in the weighted feature data and the feature weighted matrix, weighted distance calculation is performed to obtain a sample distance matrix, and a search radius threshold is set for the sample distance matrix, and the number of neighboring samples within the search radius threshold of each sample point is counted to obtain local density distribution data;

[0061] According to the local density distribution data, for each sample point, a set of sample points with a larger density value is found, and the distance to the nearest sample point in the set is calculated to obtain density distance data;

[0062] The local density distribution data and the density distance data are jointly analyzed, and the density aggregation points are determined by the product of the local density value and the density distance to obtain clustering center data, and based on the clustering center data and the sample distance matrix, the remaining sample points are assigned to the nearest clustering center to obtain energy consumption pattern clustering data.

[0063] Specifically, the multi-dimensional feature data is grouped according to device operation characteristics, environmental characteristics, and energy consumption characteristics. After dividing the device operation characteristics (such as compressor speed, evaporator pressure, etc.), environmental characteristics (such as cold storage temperature, humidity, etc.), and energy consumption characteristics (such as power consumption), the mean and standard deviation are calculated for each group of characteristics respectively to obtain characteristic statistical parameters. An energy consumption characteristic normalization matrix is constructed based on the characteristic statistical parameters to perform a standardization transformation on the multi-dimensional feature data and obtain standardized feature data. The feature data is linearly transformed so that its mean is 0 and its standard deviation is 1, which helps to reduce the influence between different feature dimensions and ensure the stability of subsequent algorithms. Through the standardization transformation, all feature data is mapped to the same scale, eliminating the influence caused by the dimensional difference between different features and obtaining standardized feature data. Based on the standardized feature data, feature weighting processing is performed. By setting the weight coefficients of device operation characteristics, environmental characteristics, and energy consumption characteristics, a feature weighting matrix is constructed to obtain weighted feature data, highlighting the importance of different features in energy consumption management. The feature weighting matrix is expressed as:

[0064] ;

[0065] where, is the weighted feature data, is the standardized feature data, is the weight coefficient of the th group of features. Using the Euclidean distance between any two data samples in the weighted feature data and combining with the feature weighting matrix, weighted distance calculation is performed to obtain a sample distance matrix. The calculation formula of the Euclidean distance is:

[0066] ;

[0067] where, represents the weighted Euclidean distance between sample and sample , and are the weighted values of sample and sample on the th feature respectively, is the number of features. By calculating the sample distances, a sample distance matrix is obtained to characterize the similarity between samples. After obtaining the sample distance matrix, a search radius threshold is set to count the number of neighboring samples within the range of the search radius for each sample point, thereby obtaining the local density distribution data. The magnitude of the local density reflects the degree of aggregation of samples in space. The smaller the search radius, the higher the degree of aggregation of samples that can be identified. Based on the local density distribution data, for each sample point, a set of sample points with larger density values is found, and the distance to the nearest sample point in this set is calculated to obtain the density distance data. The local density distribution data and the density distance data are jointly analyzed, and the density aggregation points are determined by the product of the local density value and the density distance, thereby obtaining the clustering center data. The formula is expressed as:

[0068] ;

[0069] wherein, represents the aggregation value of the sample point , is the local density value of the sample point , is the distance between the sample point and the sample point with a larger density value. By calculating the aggregation value of each sample point, the point with the highest degree of aggregation is found as the center point of the clustering. Based on the clustering center data and the sample distance matrix, the remaining sample points are assigned to the nearest clustering center to obtain the energy consumption pattern clustering data, which reflects the internal relationship between the equipment operation status, environmental conditions, and energy consumption patterns in the cold chain system.

[0070] In a specific embodiment, the process of executing step S300 may specifically include the following steps:

[0071] Perform type classification and coding processing on the equipment information, energy consumption data, and operation parameters in the energy consumption pattern clustering data to construct a node attribute table, thereby obtaining the basic node data;

[0072] Construct an equipment composition relationship diagram based on the basic node data, and convert the composition relationship and operation parameters of the compressor unit, condenser, and evaporator into physical connection relationships between nodes to obtain the equipment relationship data;

[0073] Use the equipment relationship data to establish an energy consumption transfer path, and convert the energy flow process in the cold chain system into a directed edge relationship to obtain the energy consumption relationship data;

[0074] Establish an equipment state transition link based on the energy consumption relationship data, and convert the dynamic changes of the equipment operation state into a time-sequence state sequence to obtain the state migration data;

[0075] Perform multimodal information integration on the state transition data, map temperature, power, and pressure into the dynamic attributes of the corresponding nodes to obtain node feature data, and perform association rule mining based on the node feature data to construct a temporal dependence relationship matrix between nodes to obtain association rule data;

[0076] Convert the association rule data into a graph data structure, construct the temporal edge weights between nodes, and perform sparsification processing on the connection relationships between nodes to obtain graph structure data. Perform node embedding and edge embedding calculations on the graph structure data to vectorize the temporal features of the nodes and edges in the graph to obtain the device operation knowledge graph.

[0077] Specifically, perform type division and coding processing on the device information, energy consumption data, and operation parameters in the energy consumption mode clustering data. Classify various types of information. For example, independently divide device information such as compressor units, condensers, and evaporators, classify the energy consumption data of each device (such as energy consumption rate, electrical energy utilization rate) into one group, and classify operation parameters (such as temperature, pressure, wind speed, etc.) into another group. Perform standardized coding on each type of feature so that it can be uniformly managed and processed during the subsequent construction of the graph, construct a node attribute table to obtain basic node data. Construct a device composition relationship graph based on the basic node data, and convert the physical composition relationship and operation parameters of devices such as compressor units, condensers, and evaporators into physical connection relationships between nodes. Each device is regarded as a node, and the connection relationship between nodes is represented by an edge. For example, the connection between a compressor and a condenser represents the heat exchange process of the refrigerant of the compressor passing through the condenser. These physical connection relationships are represented in the form of an adjacency matrix. The definition of the adjacency matrix A is as follows:

[0078] ;

[0079] where, represents the connection relationship between device and device . If there is a physical connection between and , then It is 1; otherwise it is 0. In this way, device relationship data is obtained, and a complete device composition relationship diagram is constructed to reflect the physical connections and energy transfer paths between devices. The energy transfer path in the cold chain system is established using the device relationship data. The energy transfer between devices is regarded as a directed relationship, such as the energy flow process from the compressor to the condenser and then to the evaporator. By establishing a directed edge relationship, the energy consumption transfer process is mapped into a graph structure to obtain energy consumption relationship data. The edges in the directed graph are represented by a directed adjacency matrix, and the elements in the matrix represent whether there is a connection between nodes and indicate the direction of energy flow. Based on the energy consumption relationship data, a state transition link of the device is established, and the dynamic changes in the device operation state are transformed into a time series state sequence to obtain state migration data. The state transition link describes the changes in the state during the device operation process, such as the process from the start-up, load change to the stop of the compressor, and these changes are represented by a state sequence. The state transition matrix represents that the device at time transfers its state to the device and is defined by the following formula:

[0080] ;

[0081] where, represents the probability that the device transfers to the device at time at time , and represent the states of the devices and at their respective times. These state transition information constitutes the temporal mutual influence between devices. Multimodal information integration is performed on the state migration data, and the operation parameters such as the temperature, power, and pressure of the device are mapped into the dynamic attributes of the corresponding nodes to obtain node feature data. Through multimodal information integration, the state and characteristics of the nodes are described. Association rule mining is performed based on the node feature data to construct a temporal dependence relationship matrix between nodes. The temporal association relationship between devices is found, such as the operation state of a certain device affecting the energy consumption performance of another device. By mining this association, association rule data is obtained to reflect the dependence relationship between devices. The association rule data is transformed into a graph data structure to construct the temporal edge weights between nodes. For example, the temporal dependence relationship between nodes is described by the weight matrix and the formula is as follows:

[0082] ;

[0083] where, represents the node The edge weight between node i and node j represents node and The association strength between them is a function that maps the association strength to the edge weight. To reduce the computational complexity and improve the representation ability of the graph structure, the connection relationship between nodes is sparsified, retaining the most important connections and ignoring the less influential connections, resulting in a more concise graph structure data. Node embedding and edge embedding calculations are performed on the graph structure data. The nodes and edges in the graph are represented as vectors to facilitate subsequent processing by machine learning models. Node embedding uses an embedding method based on graph convolution, such as calculating the embedding vector of node features through a graph convolutional neural network. Edge embedding is comprehensively calculated through the attributes of the edge and the features of the connected nodes to obtain the embedding vector of the edge. Through node embedding and edge embedding, the temporal features in the graph are vectorized to obtain the knowledge graph of device operation.

[0084] In a specific embodiment, the process of executing step S400 may specifically include the following steps:

[0085] Construct node-level and edge-level graph convolutional layers for the device operation knowledge graph respectively. The node-level graph convolutional layer contains three convolutional operations, and the kernel sizes are 32, 64, and 128 respectively. The edge-level graph convolutional layer contains two convolutional operations, and the kernel sizes are 16 and 32 respectively, to obtain the graph structure features;

[0086] Input the graph structure features into the relationship perception layer. The relationship perception layer includes a device composition relationship sub-layer, an energy consumption transfer relationship sub-layer, and a state transition relationship sub-layer. Each sub-layer is set with 64 feature channels, and multiple relationship information is fused through a message passing mechanism to obtain the graph feature vector;

[0087] Construct a multi-branch feature extraction network for the energy consumption pattern clustering data. The multi-branch feature extraction network includes a compressor unit feature branch, a condenser feature branch, and an evaporator feature branch. Each branch contains two fully connected layers, and the number of neurons is 128 and 64 respectively, to obtain the device feature vector;

[0088] Input the device feature vector into the temporal feature extraction layer. The temporal feature extraction layer adopts a gated recurrent unit structure, including an input gate, a reset gate, and an update gate. The hidden state dimension is set to 128, and the temporal window length is 24, to obtain the temporal feature vector;

[0089] Input the graph feature vector and the temporal feature vector into the cross-attention module. The cross-attention module is set with 8 attention heads, and the dimension of each attention head is 32. Adaptive fusion is performed by calculating the correlation weights between features to obtain the multi-modal fusion feature;

[0090] For the multi-modal fusion feature input to the feature enhancement network, the feature enhancement network includes three residual blocks. Each residual block consists of two convolutional layers and a shortcut connection, and batch normalization layers and ReLU activation functions are set between the residual blocks to obtain an enhanced feature vector;

[0091] Input the enhanced feature vector into the abnormal pattern recognition layer. The abnormal pattern recognition layer includes an energy consumption abnormal sub-network, a temperature abnormal sub-network, and a pressure abnormal sub-network. Each sub-network consists of three fully connected layers, and the number of neurons is 256, 128, and 64 respectively to obtain an abnormal pattern vector;

[0092] Conduct comprehensive analysis on the abnormal pattern vector. Set the weight coefficients of energy consumption abnormality, temperature abnormality, and pressure abnormality to 0.4, 0.3, and 0.3 respectively, and obtain the device energy consumption abnormal feature data through weighted combination.

[0093] Specifically, construct node-level and edge-level graph convolutional layers for the device operation knowledge graph. The node-level graph convolutional layer includes three convolutional operations, and the convolutional kernel sizes are 32, 64, and 128 respectively to gradually extract the deep features of the nodes in the graph. The edge-level graph convolutional layer includes two convolutional operations, and the convolutional kernel sizes are 16 and 32 respectively, which are used to capture the relationship information of the edges between nodes. The settings of the node-level and edge-level graph convolutional layers help to extract multi-level graph structure features from the device operation knowledge graph. Node-level convolution is used to extract the features of device nodes (such as compressors, condensers, evaporators, etc.) themselves and their direct interaction features with surrounding nodes; while edge-level convolution focuses on describing the physical and energy consumption transfer relationships between nodes. These operations can effectively extract the complex relationship information contained in the graph. Input the graph structure features after convolutional operations into the relationship perception layer. The relationship perception layer includes a device composition relationship sub-layer, an energy consumption transfer relationship sub-layer, and a state transition relationship sub-layer. Each sub-layer is set with 64 feature channels to process the graph structure features from multiple dimensions. In the relationship perception layer, multiple relationship information is fused through the message passing mechanism. Each node not only considers its own features when updating features, but also considers the information of neighbor nodes, so as to fully integrate the complex associations between devices in the composition, energy consumption transfer, and state change processes to obtain the graph feature vector. The process of message passing is represented by the following formula:

[0094] ;

[0095] Among them, is the feature vector of node at the layer, represents the set of neighbor nodes of node , is the weight matrix at the layer, is the bias term, is the activation function. Through this mechanism, the composition relationship, energy consumption transfer, and state transfer information of the fusion device are integrated to form a comprehensive graph feature vector. A multi-branch feature extraction network is constructed for the energy consumption pattern clustering data. The multi-branch feature extraction network includes a compressor unit feature branch, a condenser feature branch, and an evaporator feature branch. Each branch contains two fully connected layers with 128 and 64 neurons respectively, which are used to extract the feature vectors of the devices. Through the design of the multi-branch network, the features of different devices are processed separately, so as to retain the uniqueness of the devices when extracting features, and at the same time ensure the consistency between the features of different devices. These feature vectors describe the performance of each device in the energy consumption pattern and are important representations of the device state. The device feature vectors are input into the temporal feature extraction layer. The temporal feature extraction layer adopts a gated recurrent unit structure, which includes an input gate, a reset gate, and an update gate. The hidden state dimension is set to 128, and the temporal window length is 24. The gated recurrent unit can effectively capture the time dependence of the device features. Especially when the device operating state has strong temporal characteristics, the gated recurrent unit can remember important information and forget irrelevant historical states through the gating mechanism, obtain the temporal representation of the device features, and finally output the temporal feature vector. The graph feature vector and the temporal feature vector are input into the cross-attention module together. The cross-attention module sets 8 attention heads, and the dimension of each attention head is 32. Adaptive fusion is achieved by calculating the correlation weights between the features. The purpose of the cross-attention mechanism is to establish connections between different modal data (such as graph features and temporal features), so that the mutual correlation between the features can be better expressed. By learning the attention weights, it is adaptively judged which features are more important for the energy consumption anomaly recognition of the device, and multi-modal fusion features are obtained. The calculation of the attention mechanism is expressed by the following formula:

[0096] ;

[0097] where, 、 、 represent the query matrix, the key matrix, and the value matrix respectively, Indicates the dimension of the key vector. Through the attention mechanism, weighted summation is performed on each feature vector, making the multi-modal fusion feature more representative. The multi-modal fusion feature is input into the feature enhancement network, which contains three residual blocks. Each residual block consists of two convolutional layers and a shortcut connection, and batch normalization layers and ReLU activation functions are set between the residual blocks. The design of the residual block can solve the problem of gradient disappearance in deep networks. Through the shortcut connection, the input features are directly passed to the output layer, thus retaining the original feature information. The batch normalization layer accelerates the convergence process of the network, while the ReLU activation function introduces non-linearity, making the feature representation ability more abundant. After being processed by the feature enhancement network, an enhanced feature vector is finally obtained. The enhanced feature vector is input into the abnormal pattern recognition layer, which contains an energy consumption abnormal sub-network, a temperature abnormal sub-network, and a pressure abnormal sub-network. Each sub-network consists of three fully connected layers, and the number of neurons is 256, 128, and 64 respectively. These sub-networks are used to identify different types of abnormal patterns. For example, the energy consumption abnormal sub-network identifies the abnormality in the energy consumption of the device, the temperature abnormal sub-network is used to identify whether the fluctuation of the device temperature is abnormal, and the pressure abnormal sub-network focuses on the pressure change of the device. Each sub-network outputs an abnormal pattern vector, and these vectors reflect the abnormal states of the device in different aspects. Comprehensive analysis is performed on the abnormal pattern vectors. By setting the weight coefficients of energy consumption abnormality, temperature abnormality, and pressure abnormality to 0.4, 0.3, and 0.3 respectively, different types of abnormalities are weighted and combined to obtain the energy consumption abnormal feature data of the device. The formula for weighted combination is:

[0098] ;

[0099] where, represents the final abnormal feature data, 、 、 represent the feature data of energy consumption abnormality, temperature abnormality, and pressure abnormality respectively. Through this weighted method, comprehensive analysis is performed according to the relative importance of different abnormalities to obtain a more accurate abnormal recognition result.

[0100] In a specific embodiment, the process of executing step S500 may specifically include the following steps:

[0101] Construct a recovery priority matrix for the device energy consumption abnormal feature data, and perform abnormal degree classification based on the energy consumption abnormal weight of 0.4, the temperature abnormal weight of 0.3, and the pressure abnormal weight of 0.3 to obtain abnormal priority data;

[0102] Establish an energy recovery type mapping table according to the abnormal priority data, and establish a corresponding relationship between the abnormal features and three recovery modes of compressor waste heat recovery, condensation heat recovery, and system energy balance to obtain recovery type data;

[0103] Augment the features of the recycling type data. By introducing historical recycling efficiency indicators and equipment operating status indicators, construct a feature enhancement matrix to obtain recycling feature data;

[0104] Input the recycling feature data into the input layer of the Softmax regression classifier. After being processed by two hidden layers, with 128 neurons set in each layer and the BatchNorm and ReLU activation functions configured, obtain classification feature data;

[0105] Generate a recycling strategy probability distribution based on the classification feature data. Perform probability scoring on the compressor waste heat recovery parameters, condensation heat recovery parameters, and system energy balance parameters to obtain strategy probability data, and construct a parameter configuration matrix based on the strategy probability data, including the value ranges of the compressor speed, condensation temperature, and evaporation temperature, to obtain parameter configuration data;

[0106] Conduct an analysis of the operating constraints on the parameter configuration data. Set the parameter adjustment step size according to the equipment operating boundary conditions and system stability requirements to obtain operating parameter data, and organize the execution order of the operating parameter data according to the system response time sequence and energy transfer link to construct a control instruction sequence to obtain an initial energy recovery plan.

[0107] Specifically, construct a recycling priority matrix for the equipment energy consumption abnormal feature data to determine the priority order of different equipment in energy recovery. When constructing the recycling priority matrix, weight coefficients are respectively set for the different influence degrees of energy consumption abnormality, temperature abnormality, and pressure abnormality, where the weight of energy consumption abnormality is 0.4, the weight of temperature abnormality is 0.3, and the weight of pressure abnormality is 0.3. Through these weights, the equipment energy consumption abnormal feature data is weighted and summed according to the degree of each abnormality to obtain the abnormal priority data of each equipment. The specific calculation formula is as follows:

[0108] ;

[0109] Among them, represents the abnormal priority of equipment , , , respectively represent equipment Abnormal eigenvalue in terms of energy consumption, temperature and pressure. The anomalies of each device are weighted by this formula to obtain a priority data matrix. According to the abnormal priority data, a mapping table of energy recovery types is established. The abnormal characteristics are correlated with three recovery modes: compressor waste heat recovery, condensation heat recovery, and system energy balance. For example, if the abnormal priority of the device's energy consumption is high, the compressor waste heat recovery method is preferably selected to utilize the excess heat; if the abnormal priority of the temperature is high, energy recovery is considered through the condensation heat recovery method; for pressure anomalies, the system energy balance method is used for processing. Through this mapping method, each abnormal characteristic is associated with the most suitable energy recovery mode to obtain recovery type data. Feature augmentation is performed on the recovery type data. By introducing historical recovery efficiency indicators and device operating status indicators, a feature enhancement matrix is constructed to obtain recovery feature data. The historical recovery efficiency indicator reflects the actual effect of energy recovery under similar operating conditions, while the device operating status indicator describes the current operating state of the device, such as temperature, pressure, etc. The introduction of these features enables the recovery feature data to better combine historical experience with the current state, providing more comprehensive information for energy recovery decision-making. The recovery feature data is input into the input layer of the Softmax regression classifier and processed through two hidden layers, with 128 neurons set in each layer, and the batch normalization (BatchNorm) and ReLU activation function are configured at the same time. The purpose of batch normalization is to accelerate the training process and prevent gradient disappearance, while the ReLU activation function can introduce non-linearity and enhance the expression ability of the model. After processing through these two hidden layers, classification feature data is obtained. Based on the classification feature data, the Softmax regression classifier outputs the probability distribution of each energy recovery mode, including the probability scores of compressor waste heat recovery parameters, condensation heat recovery parameters, and system energy balance parameters. Assume , , respectively represent the probability scores of these three energy recovery modes, which are expressed by the following formula:

[0110] ;

[0111] where, represents the score of the energy recovery mode, Represents the probability of the corresponding mode. These probability scores reflect the applicability of different recovery modes in the current situation. Based on the policy probability data, a parameter configuration matrix is constructed, which includes the value ranges of the compressor speed, condensation temperature, and evaporation temperature, to obtain specific parameter configuration data. For example, at what speed the compressor should operate and how the temperature of the condenser should be adjusted, etc. Perform an operating constraint analysis on the parameter configuration data to ensure that all parameters are within the operating boundary conditions of the equipment and meet the stability requirements of the system. By setting a reasonable parameter adjustment step size, ensure that the equipment remains stable during the adjustment process and avoid fluctuations caused by too rapid adjustment. For example, the speed of the compressor needs to be adjusted in steps of 100 revolutions per minute to ensure a smooth transition of the system. After the constraint analysis, the final operating parameter data is obtained. According to the operating parameter data, organize the execution order according to the time sequence of the system response and the energy transfer link, construct a control instruction sequence, and form an initial energy recovery plan. In the control instruction sequence, consider the energy transfer relationship between devices. For example, the waste heat of the compressor is sent to the condenser for condensation treatment, so it is necessary to coordinate the operating order of each device to achieve the maximum utilization of energy. The generation of control instructions needs to consider the current state of the device and predict the response of the system, so as to achieve optimal control at each step of adjustment.

[0112] In a specific embodiment, the process of executing step S600 may specifically include the following steps:

[0113] Send the operating parameter data in the initial energy recovery plan to the control systems of the compressor unit, condenser, and evaporator, collect the compressor waste heat recovery data, condensation heat recovery data, and system energy balance data, and obtain the first operating data;

[0114] Segment the first operating data according to three time scales of 5 minutes, 15 minutes, and 30 minutes, record the dynamic change process of the equipment operating efficiency parameters and load matching parameters, and obtain the second operating data;

[0115] Calculate the short-term operating efficiency index, medium-term operating efficiency index, and long-term operating efficiency index based on the second operating data, including the compressor energy efficiency ratio, condenser heat transfer efficiency, and evaporator refrigeration coefficient, and obtain the efficiency evaluation data;

[0116] Perform a multi-scale load matching analysis on the efficiency evaluation data, calculate the deviations of the real-time load, hourly load, and daily load from the set values, and perform load dynamic balance calculations to obtain the load matching data;

[0117] Input the efficiency evaluation data and load matching data into the adaptive optimization iterator, and dynamically adjust the energy recovery parameters by the stochastic gradient descent method with momentum term to obtain the first optimized parameter data;

[0118] Perform hierarchical operation constraint verification on the first optimized parameter data, and perform parameter correction according to the equipment safe operation boundary, system stability requirements, and energy balance constraints to obtain the second optimized parameter data;

[0119] Construct a hierarchical control instruction sequence based on the second optimized parameter data, and reconstruct the execution strategy according to the timing characteristics of the fast response layer, medium-term regulation layer, and slow optimization layer to obtain hierarchical control data;

[0120] Perform multi-objective fusion optimization on the hierarchical control data and the initial energy recovery scheme, and construct a closed-loop control process with a prediction-feedback-compensation mechanism to obtain the target energy management scheme.

[0121] Specifically, the operation parameter data in the initial energy recovery scheme is sent to each device, including the control systems of the compressor unit, condenser, and evaporator. Through the control system, the operation parameters such as the rotational speed of the compressor, the condensation temperature of the condenser, and the refrigeration temperature of the evaporator are set, so that each device enters the desired operation state. During the operation of the device, the waste heat recovery data of the compressor, the condensation heat recovery data of the condenser, and the energy balance data of the entire system are collected in real time to obtain the first operation data. The first operation data reflects the energy utilization status of each device in actual operation, and also includes the mutual influence and coupling relationship between devices. The first operation data is segmented and collected according to three time scales of 5 minutes, 15 minutes, and 30 minutes. Through the multi-time scale collection method, the dynamic changes of the operation efficiency parameters and load matching parameters of the device in different time periods are recorded to obtain the second operation data, which reflects the performance of the device under different time scales of short term, medium term, and long term. Based on the second operation data, the operation efficiency indicators of short term, medium term, and long term are calculated respectively, including the coefficient of performance (COP) of the compressor, the heat transfer efficiency of the condenser, and the refrigeration coefficient of the evaporator. The calculation of these indicators is to quantify the performance of the device under different time scales and evaluate its energy efficiency status. The calculation formula for the compressor coefficient of performance is:

[0122] ;

[0123] where, represents the coefficient of performance of the compressor, is the refrigeration capacity provided by the compressor, is the input power of the compressor. By calculating this indicator, the energy efficiency performance of the compressor under different load conditions is reflected. For the condenser, the calculation of its heat transfer efficiency is expressed by the following formula:

[0124] ;

[0125] where, represents the heat transfer efficiency of the condenser, is the heat transfer amount of the condenser, is the heat transfer area, is the average value of the temperature difference during the heat transfer process. The higher the heat transfer efficiency, the better the performance of the condenser. For the refrigeration coefficient of the evaporator, it is calculated using the following formula:

[0126] ;

[0127] where, represents the refrigeration coefficient of the evaporator, is the refrigeration capacity provided by the evaporator, is the energy input of the evaporator. These efficiency indicators constitute the efficiency evaluation data, reflecting the operating performance of each device in the cold chain system. Perform multi-scale load matching analysis on the efficiency evaluation data. The load matching analysis evaluates whether the device is operating within the appropriate load range by calculating the deviations between the real-time load, hourly load, and daily load of the device at different time scales and the set values, and obtains the load matching data. For example, for the compressor, calculate its load matching degree in each time period to determine whether it is operating near the optimal energy efficiency point. The load matching analysis also involves the calculation of dynamic balance of the load to ensure that the loads of each device can be smoothly adjusted at different time scales. Input the efficiency evaluation data and the load matching data into the adaptive optimization iterator, and dynamically adjust the energy recovery parameters by the stochastic gradient descent method with momentum term to obtain the first optimized parameter data. The optimization process of the stochastic gradient descent method is expressed as:

[0128] ;

[0129] where, represents the parameter in the th iteration, is the learning rate, is the gradient of the loss function with respect to the parameter, is the weight of the momentum term, Represents the cumulative momentum term. By introducing momentum, the convergence speed is accelerated, and the oscillations occurring in the optimization process are reduced, enabling the optimal value of the energy recovery parameter to be found more quickly. Hierarchical operation constraint verification is performed on the first optimization parameter data to ensure that the optimized parameters are within the safe operation boundary of the device, while satisfying the constraints of system stability and energy balance. By verifying whether the rotational speed change of the compressor is within its allowed safe range and whether the temperature regulation of the condenser will not have too much impact on other parts of the system, the second optimization parameter data is obtained. If the parameters do not meet these constraint conditions, corresponding corrections need to be made to ensure the safety of device operation. According to the second optimization parameter data, a hierarchical control instruction sequence is constructed. The hierarchical control instructions include a fast response layer, an intermediate regulation layer, and a slow optimization layer. These different control levels are used to achieve the optimal regulation of the system at different time scales. For example, the fast response layer is used to handle the instantaneous load change of the compressor to ensure that the device adapts to load fluctuations in a short time; the intermediate regulation layer is used to regulate the temperature of the condenser to cope with changes in the ambient temperature; the slow optimization layer is used for the long-term energy efficiency optimization of the system to ensure that the entire system reaches the optimal operating state on a longer time scale. Through hierarchical control, hierarchical control data is obtained. The hierarchical control data is multi-objective fusion optimized with the initial energy recovery scheme to construct a closed-loop control process with prediction, feedback, and compensation mechanisms, and an objective energy management scheme is obtained. The closed-loop control process continuously predicts the energy consumption and state changes of the device, real-time feedbacks the operating conditions of the device, and adjusts the parameters of the device through the compensation mechanism to achieve the continuous optimization of the system. For example, when the load of the compressor changes, the rotational speed is adjusted in a timely manner based on the energy efficiency change information obtained through feedback, thereby avoiding excessive energy consumption; when the ambient temperature rises, the heat exchange efficiency of the condenser is adjusted through the prediction model to maintain the energy balance of the system. The introduction of closed-loop control ensures that the system can maintain the optimal energy efficiency under different operating conditions.

[0130] Please refer to Figure 2 , Figure 2 which is a schematic block diagram of the structure of the energy management and recovery system 200 for the cold chain supply chain provided by the embodiment of the present application, as Figure 2 shown. The energy management and recovery system 200 for the cold chain supply chain includes:

[0131] An acquisition module 210, configured to collect and preprocess the operating parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data;

[0132] A calculation module 220, configured to perform feature extraction and local density distribution calculation on the multi-dimensional feature data to obtain energy consumption pattern clustering data;

[0133] A construction module 230, configured to construct a device operation knowledge graph according to the energy consumption pattern clustering data;

[0134] An analysis module 240, configured to perform anomaly feature analysis on the equipment operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model, so as to obtain equipment energy consumption anomaly feature data;

[0135] A generation module 250, configured to generate an initial energy recovery plan according to the equipment energy consumption anomaly feature data;

[0136] An optimization module 260, configured to apply the initial energy recovery plan to the cold chain system, collect operation efficiency data and load matching data, and perform online iterative optimization to obtain a target energy management plan.

[0137] Through the collaborative cooperation of the above-mentioned various components, through a multi-level data collection and preprocessing mechanism, the comprehensive perception of the operation parameters and environmental parameters of the cold chain system is realized, providing a high-quality data basis for energy management; adopting a clustering analysis method based on energy consumption distance, the energy consumption pattern characteristics of the system are accurately identified, providing data support for the formulation of energy recovery strategies; combining knowledge graph technology, a time-series correlation model between devices is constructed, deeply mining the energy transfer law and device state evolution characteristics in the system; through the feature fusion mechanism of the hybrid neural network model, the accurate identification of the abnormal state of equipment energy consumption is realized, providing a reliable basis for the generation of energy recovery plans; based on the operation efficiency evaluation and load matching analysis of multiple time scales, a hierarchical energy management optimization mechanism is established to ensure the stability and efficiency of the system operation; through a closed-loop online iterative optimization strategy, the dynamic adjustment and continuous optimization of the energy recovery plan are realized, significantly improving the energy utilization efficiency of the system.

[0138] Please refer to Figure 3 , Figure 3 FIG. is a schematic structural diagram of an energy management and recovery device 300 for a cold chain supply chain provided by an embodiment of the present application. The energy management and recovery device 300 for the cold chain supply chain includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected through a system bus 303. Among them, the memory 302 may include a non-volatile storage medium and an internal memory.

[0139] The non-volatile storage medium may store a computer program. The computer program includes program instructions. When the program instructions are executed by the processor 301, the processor 301 may be enabled to execute any one of the energy management and recovery methods for the cold chain supply chain described above.

[0140] The processor 301 is configured to provide computing and control capabilities to support the operation of the entire energy management and recovery device 300 for the cold chain supply chain.

[0141] The internal memory provides an environment for the operation of a computer program in a non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can be caused to execute any of the above energy management and recovery methods for the cold chain supply chain.

[0142] Those skilled in the art can understand that Figure 3 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the energy management and recovery device 300 of the cold chain supply chain involved in the solution of this application. Specifically, the energy management and recovery device 300 of the cold chain supply chain may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0143] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0144] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described energy management and recovery device 300 of the cold chain supply chain can refer to the corresponding process of the foregoing energy management and recovery method for the cold chain supply chain, and will not be elaborated herein.

[0145] This application embodiment also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by one or more processors, the one or more processors are caused to implement the energy management and recovery method for the cold chain supply chain provided by this application embodiment.

[0146] Among them, the computer-readable storage medium may be the internal storage unit of the energy management and recovery device 300 of the cold chain supply chain in the foregoing embodiments, such as the hard disk or memory of the energy management and recovery device 300 of the cold chain supply chain. The computer-readable storage medium may also be an external storage device of the energy management and recovery device 300 of the cold chain supply chain, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped with the energy management and recovery device 300 of the cold chain supply chain.

[0147] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0148] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0149] The above is the case. The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. An energy management and recovery method for a cold chain supply chain, characterized in that, Including: Collect and preprocess the operating parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data; Extract features from the multi-dimensional feature data and calculate the local density distribution to obtain energy consumption pattern clustering data; Construct a device operation knowledge graph based on the energy consumption pattern clustering data; specifically including: classify and encode the device information, energy consumption data, and operating parameters in the energy consumption pattern clustering data to construct a node attribute table and obtain basic node data; construct a device composition relationship graph based on the basic node data, and convert the composition relationship and operating parameters of the compressor unit, condenser, and evaporator into physical connection relationships between nodes to obtain device relationship data; use the device relationship data to establish an energy consumption transfer path, and convert the energy flow process in the cold chain system into a directed edge relationship to obtain energy consumption relationship data; establish a device state transition link based on the energy consumption relationship data, and convert the dynamic changes in the device operating state into a time series state sequence to obtain state migration data; perform multi-modal information integration on the state migration data, map temperature, electricity, and pressure to the dynamic attributes of corresponding nodes to obtain node feature data, and perform association rule mining based on the node feature data to construct a time series dependence relationship matrix between nodes to obtain association rule data; convert the association rule data into a graph data structure, construct time series edge weights between nodes, and perform sparsification processing on the connection relationships between nodes to obtain graph structure data, perform node embedding and edge embedding calculations on the graph structure data, vectorize the time series features of nodes and edges in the graph to obtain a device operation knowledge graph; Perform abnormal feature analysis on the device operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model to obtain device energy consumption abnormal feature data; Generate an initial energy recovery plan based on the device energy consumption abnormal feature data; Apply the initial energy recovery plan to the cold chain system, collect operation efficiency data and load matching data, and perform online iterative optimization to obtain a target energy management plan.

2. The energy management and recovery method for the cold chain supply chain according to claim 1, wherein The collection and preprocessing of the operating parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data includes: Collect and normalize the refrigerant pressure, refrigerant temperature, motor current, and vibration frequency of the compressor unit to obtain compressor unit operation feature data; collect and standardize the heat exchange pressure, condensation temperature, fan speed, and load power of the condenser to obtain condenser operation feature data; collect and unify the dimension of the evaporation pressure, evaporation temperature, defrosting state, and refrigeration capacity of the evaporator to obtain evaporator operation feature data; Perform time series alignment processing on the compressor unit operation feature data, the condenser operation feature data, and the evaporator operation feature data to obtain a device operation feature matrix; Collect the temperature inside the cold storage, the humidity inside the cold storage, and the state of the cold storage door, and perform data smoothing processing through the sliding window method to obtain environmental parameter feature data; Perform local linear regression correction on the outliers in the device operation feature matrix to obtain the first operation feature data, and perform time series interpolation to complete the missing values in the first operation feature data to obtain the second operation feature data; Perform feature fusion and dimensionality reduction processing on the second operation feature data and the environmental parameter feature data to obtain multi-dimensional feature data.

3. The energy management and recovery method for the cold chain supply chain according to claim 2, wherein Performing feature extraction and local density distribution calculation on the multi-dimensional feature data to obtain energy consumption pattern clustering data, including: Group the multi-dimensional feature data according to device operation features, environmental features, and energy consumption features, and calculate the mean and standard deviation for each group of features to obtain feature statistical parameters; Construct an energy consumption feature normalization matrix according to the feature statistical parameters, perform standardization transformation on the multi-dimensional feature data to obtain standardized feature data, and based on the standardized feature data, construct a feature weighted matrix through device operation feature weight coefficients, environmental feature weight coefficients, and energy consumption feature weight coefficients to obtain weighted feature data; Use the Euclidean distance between any two data samples in the weighted feature data and the feature weighted matrix to perform weighted distance calculation to obtain a sample distance matrix, and set a search radius threshold for the sample distance matrix, and count the number of neighboring samples of each sample point within the search radius threshold to obtain local density distribution data; According to the local density distribution data, find a set of sample points with a larger density value for each sample point, and calculate the distance to the nearest sample point in the set to obtain density distance data; Perform joint analysis on the local density distribution data and the density distance data, determine the density aggregation points through the product of the local density value and the density distance to obtain clustering center data, and based on the clustering center data and the sample distance matrix, assign the remaining sample points to the nearest clustering center to obtain energy consumption pattern clustering data.

4. The energy management and recovery method for the cold chain supply chain according to claim 1, characterized in that, Performing abnormal feature analysis on the device operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model to obtain device energy consumption abnormal feature data, including: Construct node-level and edge-level graph convolutional layers for the device operation knowledge graph respectively. The node-level graph convolutional layer contains three convolutional operations with kernel sizes of 32, 64, and 128 respectively, and the edge-level graph convolutional layer contains two convolutional operations with kernel sizes of 16 and 32 respectively to obtain graph structure features; Input the graph structure features into the relationship perception layer. The relationship perception layer includes a device composition relationship sub-layer, an energy consumption transfer relationship sub-layer, and a state transition relationship sub-layer. Each sub-layer is set with 64 feature channels, and multiple relationship information is fused through a message passing mechanism to obtain a graph feature vector; Construct a multi-branch feature extraction network for the energy consumption pattern clustering data. The multi-branch feature extraction network includes a compressor unit feature branch, a condenser feature branch, and an evaporator feature branch. Each branch contains two fully connected layers with 128 and 64 neurons respectively to obtain device feature vectors; Input the device feature vector into the temporal feature extraction layer. The temporal feature extraction layer adopts a gated recurrent unit structure, including an input gate, a reset gate, and an update gate. The hidden state dimension is set to 128, and the temporal window length is 24, to obtain a temporal feature vector; Input the graph feature vector and the temporal feature vector into the cross-attention module. The cross-attention module sets 8 attention heads, and the dimension of each attention head is 32. Adaptive fusion is performed by calculating the correlation weights between features to obtain a multi-modal fusion feature; Input the multi-modal fusion feature into the feature enhancement network. The feature enhancement network includes three residual blocks. Each residual block consists of two convolutional layers and a shortcut connection, and a batch normalization layer and a ReLU activation function are set between the residual blocks to obtain an enhanced feature vector; Input the enhanced feature vector into the abnormal pattern recognition layer. The abnormal pattern recognition layer includes an energy consumption abnormal sub-network, a temperature abnormal sub-network, and a pressure abnormal sub-network. Each sub-network consists of three fully connected layers, and the number of neurons is 256, 128, and 64 respectively, to obtain an abnormal pattern vector; Conduct a comprehensive analysis of the abnormal pattern vector. Set the weight coefficients of energy consumption abnormality, temperature abnormality, and pressure abnormality to 0.4, 0.3, and 0.3 respectively, and obtain the device energy consumption abnormal feature data through weighted combination.

5. The energy management and recovery method for the cold chain supply chain according to claim 4, characterized in that Generate an initial energy recovery plan according to the device energy consumption abnormal feature data, including: Construct a recovery priority matrix for the device energy consumption abnormal feature data, and perform abnormal degree grading based on the energy consumption abnormal weight of 0.4, the temperature abnormal weight of 0.3, and the pressure abnormal weight of 0.3 to obtain abnormal priority data; Establish an energy recovery type mapping table according to the abnormal priority data, and establish a corresponding relationship between the abnormal features and three recovery modes: compressor waste heat recovery, condensation heat recovery, and system energy balance, to obtain recovery type data; Expand the features of the recovery type data. Construct a feature enhancement matrix by introducing historical recovery efficiency indicators and device operating status indicators to obtain recovery feature data; Input the recovery feature data into the input layer of the Softmax regression classifier. After being processed by two hidden layers, 128 neurons are set in each layer and the BatchNorm and ReLU activation functions are configured to obtain classification feature data; Generate a recovery strategy probability distribution according to the classification feature data. Perform probability scoring on the compressor waste heat recovery parameters, condensation heat recovery parameters, and system energy balance parameters to obtain strategy probability data, and construct a parameter configuration matrix based on the strategy probability data, including the value ranges of the compressor speed, condensation temperature, and evaporation temperature, to obtain parameter configuration data; Conduct an analysis of the operating constraints for the parameter configuration data. Set the parameter adjustment step size according to the device operating boundary conditions and system stability requirements to obtain operating parameter data, and organize the execution order of the operating parameter data according to the system response time series and energy transfer link to construct a control instruction sequence to obtain an initial energy recovery plan.

6. The energy management and recovery method for the cold chain supply chain according to claim 5, wherein Applying the initial energy recovery scheme to the cold chain system, collecting operation efficiency data and load matching data, and performing online iterative optimization to obtain a target energy management scheme, including: Issuing the operation parameter data in the initial energy recovery scheme to the control systems of the compressor unit, condenser, and evaporator, collecting compressor waste heat recovery data, condensation heat recovery data, and system energy balance data, and obtaining first operation data; Segmenting the first operation data according to three time scales of 5 minutes, 15 minutes, and 30 minutes, recording the dynamic change process of equipment operation efficiency parameters and load matching parameters, and obtaining second operation data; Calculating short-term operation efficiency indicators, medium-term operation efficiency indicators, and long-term operation efficiency indicators respectively according to the second operation data, including compressor energy efficiency ratio, condenser heat transfer efficiency, and evaporator refrigeration coefficient, and obtaining efficiency evaluation data; Performing multi-scale load matching analysis on the efficiency evaluation data, calculating the deviations of real-time load, hourly load, and daily load from the set values, and performing load dynamic balance calculation to obtain load matching data; Inputting the efficiency evaluation data and the load matching data into an adaptive optimization iterator, and dynamically adjusting the energy recovery parameters by the stochastic gradient descent method with momentum term to obtain first optimized parameter data; Performing hierarchical operation constraint verification on the first optimized parameter data, and correcting the parameters according to the equipment safe operation boundary, system stability requirements, and energy balance constraints to obtain second optimized parameter data; Constructing a hierarchical control instruction sequence according to the second optimized parameter data, and reconstructing the execution strategy according to the timing characteristics of the fast response layer, medium-term regulation layer, and slow optimization layer to obtain hierarchical control data; Performing multi-objective fusion optimization on the hierarchical control data and the initial energy recovery scheme, and constructing a closed-loop control process with a prediction-feedback-compensation mechanism to obtain a target energy management scheme.

7. An energy management and recovery system for a cold chain supply chain, characterized in that, Used to execute the energy management recovery method of the cold chain supply chain as described in any one of claims 1-6, including: A collection module, used to collect and preprocess operation parameters and environmental parameters of the cold chain system to obtain multi-dimensional feature data; A calculation module, used to perform feature extraction and local density distribution calculation on the multi-dimensional feature data to obtain energy consumption pattern clustering data; A construction module, used to construct an equipment operation knowledge graph according to the energy consumption pattern clustering data; An analysis module, used to perform abnormal feature analysis on the equipment operation knowledge graph and the energy consumption pattern clustering data through a hybrid neural network model to obtain equipment energy consumption abnormal feature data; A generation module, used to generate an initial energy recovery scheme according to the equipment energy consumption abnormal feature data; An optimization module, used to apply the initial energy recovery scheme to the cold chain system, collect operation efficiency data and load matching data, and perform online iterative optimization to obtain a target energy management scheme.

8. An energy management and recovery device for a cold chain supply chain, characterized in that, The energy management recovery device of the cold chain supply chain includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor invokes the instructions in the memory to cause the energy management recovery device of the cold chain supply chain to execute the energy management recovery method of the cold chain supply chain according to any one of claims 1-6.

9. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by the processor, the energy management recovery method of the cold chain supply chain according to any one of claims 1-6 is implemented.

Images