Cold storage dynamic scheduling cooperative control method and system based on Internet of Things

Abstract

The invention belongs to the technical field of cold chain warehouse management, and particularly relates to a cold storage dynamic scheduling cooperative control method and system based on the Internet of Things, and the method comprises the steps: obtaining operation parameters and environment parameters; according to the operation parameters and the environment parameters, an E-LSTM model is used for analyzing the operation state of the refrigeration equipment, an analysis result is generated, and through a TFT neural network model, the environment parameters and the current time and production plan are combined, and the refrigeration demand change trend is predicted; based on the analysis result and the refrigeration demand change trend, a dynamic scheduling strategy of the refrigeration equipment is generated by combining a scheduling rule and an optimization target and utilizing a PPO algorithm; the scheduling strategy is sent to the PLC control end of the refrigeration equipment through the Internet of Things communication network, the PLC control end performs cooperative control on the refrigeration equipment according to the scheduling strategy, and meanwhile, the execution state of the equipment is fed back to the cloud server. Therefore, the problems of rigid scheduling mode, serious energy waste, insufficient emergency capability and the like in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of cold chain warehousing management technology, specifically involving a method and system for dynamic scheduling and collaborative control of cold storage based on the Internet of Things. Background Technology

[0002] In industrial production, refrigeration equipment is a crucial facility for ensuring the production environment and process requirements are met. Traditional industrial refrigeration equipment scheduling and management typically rely on manual experience or fixed control logic, with basic control performed by a PLC. This approach has significant limitations, such as the inability to flexibly adjust equipment operating combinations and parameters based on real-time cooling demand, equipment operating status, and changes in the external environment. In practical applications, on the one hand, when cooling demand is low, the equipment continues to operate in a fixed mode, leading to energy waste; on the other hand, when sudden high cooling demand or equipment failure occurs, timely and reasonable scheduling cannot be implemented, affecting the continuity and stability of production.

[0003] With the development of IoT technology, there is an urgent need to introduce it into the management of industrial refrigeration equipment. However, there is currently a lack of mature IoT-based dynamic scheduling and collaborative control technology solutions for refrigeration equipment. Existing models are difficult to cope with complex and ever-changing industrial production environments, and have shortcomings in energy consumption control, system stability, and intelligent management. There is an urgent need for a technical solution that can achieve dynamic and collaborative control of refrigeration equipment to improve the overall operating efficiency and intelligence level of industrial refrigeration systems. Summary of the Invention

[0004] This application provides a method and system for dynamic scheduling and collaborative control of cold storage based on the Internet of Things, in order to solve the problems of rigid scheduling methods, serious energy waste and insufficient emergency response capabilities in the existing technology.

[0005] The first aspect of this application provides a method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things (IoT), comprising the following steps: acquiring operating parameters of refrigeration equipment and environmental parameters of the refrigeration area; analyzing the operating status of the refrigeration equipment using an E-LSTM model based on the operating parameters and environmental parameters, generating analysis results, training a TFT neural network model using historical data, and predicting the trend of refrigeration demand changes by combining the environmental parameters with the current time and production plan; generating a dynamic scheduling strategy for the refrigeration equipment using a PPO algorithm based on the analysis results and the trend of refrigeration demand changes, combined with preset scheduling rules and optimization objectives, wherein the optimization objectives include minimizing energy consumption and improving refrigeration efficiency; sending the scheduling strategy to the PLC control terminal of the refrigeration equipment through an IoT communication network, wherein the PLC control terminal performs collaborative control of the refrigeration equipment according to the scheduling strategy, and simultaneously feeds back the execution status of the equipment to a cloud server.

[0006] Preferably, based on the operating parameters and environmental parameters, the operating status of the refrigeration equipment is analyzed using an E-LSTM model, and analysis results are generated. This includes: constructing an E-LSTM model; inputting the operating parameters and environmental parameters into the E-LSTM model, identifying the changing trends and abnormal fluctuations of the equipment operating parameters, and outputting an aggregated feature vector; based on a preset equipment health assessment standard, mapping the aggregated feature vector to equipment performance score, fault risk level, and energy efficiency level through a regression algorithm, and combining the attention mechanism weights of the E-LSTM model to analyze the input feature parameters that have the greatest impact on the equipment status; and generating analysis results based on the equipment performance score, fault risk level, energy efficiency level, and the most influential input feature parameters, combined with the equipment maintenance history.

[0007] Preferably, the formula for the E-LSTM model is:

[0008]

[0009]

[0010]

[0011]

[0012]

[0013]

[0014]

[0015] in, Forget gate vector; Input the device's operating parameters and environmental parameters at the current moment; The parameter change trend in the previous hour; For activation functions; Here is the forget gate weight matrix; This is the forget gate bias vector; Input gate vector; The input gate weight matrix; The input gate bias vector; New memory candidates are generated from the current input and historical state; Activation function; Historical fluctuation patterns; Overall assessment of equipment status; The candidate memory unit weight matrix; The bias vector for candidate memory cells; This is the output gate vector; This is the output gate weight matrix; This is the output gate bias vector; The current hidden state; Attention weights; Score for attention; The fusion of features at the current moment; for Attention weight matrix; This is the attention vector; The length of the monitoring cycle; It is an exponential function.

[0016] Preferably, the TFT neural network model trained with historical data, combined with the environmental parameters and the current time and production plan, predicts the changing trend of cooling demand, including: constructing a feature vector dataset; inputting the feature vector dataset into the TFT neural network model, using the VSN in the model to dynamically filter key features, and simultaneously capturing time-dynamic features through a multi-head attention mechanism; and using GRN to fuse the key features and the time-dynamic features to generate a joint feature set. Based on the joint feature set, the probability distribution of cooling demand is output using quantile regression of the TFT neural network model, and the changing trend of cooling demand is predicted.

[0017] Preferably, based on the analysis results and the cooling demand change trend, combined with preset scheduling rules and optimization objectives, a dynamic scheduling strategy for the cooling equipment is generated using the PPO algorithm. This includes: constructing a DRL state space containing analysis results, cooling demand change trends, and real-time operating parameters; defining an action space including equipment start / stop, load adjustment, and multi-machine collaborative mode based on the state space, combined with preset scheduling rules and optimization objectives, and designing a reward function using a reward function formula; and learning in the state space using the PPO algorithm based on the action space and the reward function to generate a scheduling strategy that adapts to dynamic demands and conforms to scheduling rules.

[0018] Preferably, the preset scheduling rules are as follows: when the cooling demand is less than or equal to a first preset value, the number of operating compressors is reduced, and the frequency or volume of the operating compressors is reduced; when the cooling demand is greater than the first preset value but less than a second preset value, the standby compressors are started, and the load rate of the operating compressors is balanced; when the cooling demand is greater than the second preset value, all standby compressors are started, and the load is dynamically adjusted to match the demand; when a compressor suddenly fails, the failed compressor is shut down, and the standby compressors are started to supplement the load.

[0019] Preferably, the PPO algorithm formula is:

[0020] in, Parameters of the policy network; Time step; The ratio of the probability of the new strategy and the old strategy choosing the same action in the current state; Dominance function estimation; Clipping threshold; Expectation operator; It is a piecewise function; It is a minimum value function; This is the proxy objective function.

[0021] A second aspect of this application provides an IoT-based dynamic scheduling and collaborative control system for cold storage, comprising: an acquisition module for acquiring operating parameters of refrigeration equipment and environmental parameters of the refrigeration area; an analysis module for analyzing the operating status of the refrigeration equipment using an E-LSTM model based on the operating parameters and environmental parameters, generating analysis results, training a TFT neural network model using historical data, and predicting the trend of refrigeration demand changes by combining the environmental parameters with the current time and production plan; a generation module for generating a dynamic scheduling strategy for the refrigeration equipment using a PPO algorithm based on the analysis results and the trend of refrigeration demand changes, combined with preset scheduling rules and optimization objectives, wherein the optimization objectives include minimizing energy consumption and improving refrigeration efficiency; and an execution module for sending the scheduling strategy to the PLC control terminal of the refrigeration equipment via an IoT communication network, wherein the PLC control terminal performs scheduling and collaborative control of the refrigeration equipment according to the scheduling strategy, and simultaneously feeds back the execution status of the equipment to a cloud server.

[0022] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement a dynamic scheduling and collaborative control method for cold storage based on the Internet of Things as described in the above embodiments.

[0023] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement an Internet of Things-based dynamic scheduling and collaborative control method for cold storage as described in the above embodiments.

[0024] Therefore, this application has the following beneficial effects: This application's embodiments utilize IoT sensors to collect real-time equipment operating parameters and environmental data, and combine this with an E-LSTM model to evaluate the equipment's operating status. This breaks through the traditional scheduling mode that relies on manual experience or fixed logic, enhancing the ability to identify abnormal equipment fluctuations and providing a basis for dynamic scheduling. A TFT neural network model is used to predict the changing trend of cooling demand, and a PPO algorithm is combined to generate a dynamic scheduling strategy, improving the flexibility of response to dynamic demands. By preset optimization goals such as minimum energy consumption and maximum efficiency, energy utilization efficiency is improved. With the help of real-time IoT transmission and rapid cloud response, when equipment suddenly fails or cooling demand changes abruptly, the operating plan can be adjusted in a timely manner through the dynamic scheduling strategy, enhancing the equipment's cooling stability and reducing the risk of interruptions due to unforeseen circumstances. This provides support for the efficient, energy-saving, and stable operation of industrial refrigeration equipment. Therefore, it solves the problems of rigid scheduling methods, serious energy waste, and insufficient emergency response capabilities in existing technologies.

[0025] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0026] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a dynamic scheduling and collaborative control method for cold storage based on the Internet of Things, provided according to an embodiment of this application. Figure 2 This is an example diagram illustrating the predicted refrigeration demand of cold chain logistics cold storage according to an embodiment of this application; Figure 3 This is an example diagram illustrating the analysis of the operating status of a large-scale fruit and vegetable cold storage equipment according to an embodiment of this application; Figure 4 This is an example diagram illustrating a dynamic scheduling strategy for generating aquatic cold storage facilities according to an embodiment of this application; Figure 5 This is an example diagram illustrating the implementation of a dynamic scheduling strategy in a meat cold storage facility according to an embodiment of this application; Figure 6 This is an example diagram of an IoT-based dynamic scheduling and collaborative control method for cold storage provided according to an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an IoT-based dynamic scheduling and collaborative control system for cold storage provided according to an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0027] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0028] The following describes an embodiment of the IoT-based dynamic scheduling and collaborative control method and system for cold storage, with reference to the accompanying drawings. Addressing the issue of insufficient emergency response capabilities mentioned in the background section, this application provides an IoT-based dynamic scheduling and collaborative control method for cold storage. In this method, IoT sensors are deployed to collect real-time equipment operating parameters and environmental data. An E-LSTM model is used to evaluate the equipment's operating status, breaking through the traditional scheduling model that relies on manual experience or fixed logic. This enhances the ability to identify abnormal equipment fluctuations and provides a basis for dynamic scheduling. A TFT neural network model is used to predict the trend of refrigeration demand changes, and a PPO algorithm is combined to generate dynamic scheduling strategies, improving the flexibility of response to dynamic demands. Energy utilization efficiency is improved by pre-setting optimization goals such as minimum energy consumption and maximum efficiency. With the help of real-time transmission from the IoT and rapid cloud response, when equipment suddenly fails or refrigeration demand changes abruptly, the operating plan can be adjusted in a timely manner through dynamic scheduling strategies, enhancing equipment refrigeration stability and reducing the risk of interruptions due to unforeseen circumstances. This provides support for the efficient, energy-saving, and stable operation of industrial refrigeration equipment. Therefore, it solves the problems of rigid scheduling methods, serious energy waste, and insufficient emergency response capabilities in the prior art.

[0029] Specifically, Figure 1 This is a flowchart illustrating an IoT-based dynamic scheduling and collaborative control method for cold storage, as provided in an embodiment of this application.

[0030] like Figure 1 As shown, this IoT-based dynamic scheduling and collaborative control method for cold storage includes the following steps: In step S101, the operating parameters of the refrigeration equipment and the environmental parameters of the refrigeration area are obtained.

[0031] The operating parameters of the refrigeration equipment include the operating status and energy consumption indicators of the core components such as the compressor, condenser, and evaporator. The environmental parameters of the refrigeration area include temperature, humidity, gas concentration, and environmental factors related to the goods.

[0032] It is understood that the embodiments of this application lay the foundation for the generation of dynamic scheduling strategies by deploying IoT sensors to collect equipment operating parameters and environmental data in real time, thereby enhancing the accuracy of the analysis of the operating status of refrigeration equipment and the reliability of the prediction of the trend of refrigeration demand changes, and improving the overall collaborative control effect of cold storage.

[0033] In step S102, based on operating parameters and environmental parameters, the operating status of the refrigeration equipment is analyzed using an E-LSTM model to generate analysis results. The TFT neural network model is trained using historical data, and combined with environmental parameters, current time, and production plan, the trend of refrigeration demand changes is predicted.

[0034] The formula for the TFT neural network model is as follows:

[0035]

[0036]

[0037]

[0038]

[0039]

[0040]

[0041] in, For cold storage In time eigenvectors; A static covariate vector; It is a time-varying feature vector; This is a vector transpose operation; For time In the Attention weights for each feature; For gated residual networks; For the first Preprocessed representation of each feature; To encode the inherent attributes of cold storage; The weighted eigenvectors; For activation functions; For attention mechanisms; This is a cross-attention mechanism; Choose a network for the variables; For query matrix; The key matrix; It is a value matrix; The dimension of the key vector; These are the original static covariates; Input as a zero vector; The loss function; The number of samples; These are quantile values; For the set of quantiles; The quantile loss function; For the first The true value of each sample; For the model to the first The sample has a quantile value of The predicted value at that time; For the future The predicted value at any given time; It is a feedforward neural network; At any moment The hidden state; This is the output of the encoder.

[0042] It is understood that, after training with historical data, this application's embodiments combine environmental parameters, current time, and production plans to predict the changing trend of refrigeration demand. It utilizes VSN to dynamically screen key features and a multi-head attention mechanism to capture dynamic features over time. After GRN fusion to generate a joint feature set, it outputs the probability distribution of future refrigeration demand through quantile regression. This provides a basis for generating a dynamic scheduling strategy for suitable refrigeration equipment, helping to achieve the optimization goals of minimizing energy consumption and improving refrigeration efficiency. It also enhances the accuracy of grasping changes in refrigeration demand and strengthens the scientific and effective collaborative control of cold storage.

[0043] For example, such as Figure 2 As shown, in the daily operation of a cold chain logistics cold storage facility, there are 3 frozen zones and 2 refrigerated zones, storing seafood, frozen drinks, and other goods. The operation of the refrigeration equipment needs to be adjusted according to the actual situation. Staff collected relevant data from the past 12 months to construct a feature vector dataset, including hourly temperatures (frozen zone -22℃ to -18℃, refrigerated zone 2℃ to 8℃), humidity (70% to 90%), external ambient temperature (-10℃ to 38℃), and the volume of goods entering and leaving the warehouse at different times (0 to 80 tons / day), as well as information such as the storage cycle of goods in the production plan. After inputting this data into a TFT neural network model, the model dynamically selects frozen zone temperature, the volume of goods entering and leaving the warehouse, and the external ambient temperature as key features through VSN. Using a multi-head attention mechanism, it captures the peak patterns of refrigeration demand caused by frequent goods entering and leaving the warehouse daily from 9:00-11:00 and 14:00-16:00. Then, it uses GRN to fuse features to generate a joint feature set, and finally outputs the trend of refrigeration demand changes through quantile regression. For example, the predicted cooling demand for a certain Tuesday was 1650 kWh, while the actual demand was 1632 kWh, with an error rate of 1.1%; the predicted cooling demand for a certain Friday (peak outbound period) was 2100 kWh, while the actual demand was 2085 kWh, with an error rate of 0.7%. These prediction results provide an accurate basis for subsequent generation of dynamic scheduling strategies using the PPO algorithm, helping cold storage facilities to optimize energy consumption while meeting cooling demands.

[0044] In this embodiment, the operating status of the refrigeration equipment is analyzed using an E-LSTM model based on operating parameters and environmental parameters, and analysis results are generated. This includes: constructing an E-LSTM model; inputting operating parameters and environmental parameters into the E-LSTM model to identify trends and abnormal fluctuations in the equipment's operating parameters, and outputting an aggregated feature vector; based on preset equipment health assessment standards, mapping the aggregated feature vector to equipment performance scores, fault risk levels, and energy efficiency levels using a regression algorithm, and combining the attention mechanism weights of the E-LSTM model to analyze the input feature parameters that have the greatest impact on the equipment status; and generating analysis results based on the equipment performance score, fault risk level, energy efficiency level, and the most influential input feature parameters, combined with the equipment's maintenance history.

[0045] Among them, the aggregated feature vector integrates, refines and compresses the effective features in the original input data to form a set of values ​​that can represent the essential characteristics of the device's operating status.

[0046] It should be noted that the regression algorithm formula is as follows:

[0047]

[0048]

[0049] in, To score the predicted equipment performance; This is the weight matrix; For aggregated feature vectors; For bias terms; This is the transpose of the matrix; For activation functions; The equipment belongs to the first The probability of each energy efficiency level; For the first Weight matrix corresponding to each energy efficiency level; For summation index; It is an exponential function.

[0050] Equipment performance rating refers to an indicator that quantifies and scores the performance of equipment. Failure risk level refers to the risk level classified by the probability of equipment failure, including: Very low risk: All parameters of the equipment are stable, there are 0 historical failure records, the lifespan of core components is more than 80% remaining, and there is no possibility of failure in the short term; Low risk: The equipment is operating normally, with only very minor wear and tear on non-core components, and the probability of equipment failure each month is less than 1%; Medium risk: The equipment has occasional minor abnormalities, and the probability of equipment failure each month is between 1% and 5%; High risk: The equipment has obvious abnormalities, and the probability of equipment failure each month is between 5% and 30%; Very high risk: The equipment has shown signs of critical failure, and the probability of equipment failure each month is more than 30%.

[0051] Energy efficiency ratings refer to the classification of equipment energy utilization efficiency, including: Highest energy efficiency: The equipment's energy utilization rate reaches the industry's top level, with energy consumption significantly lower than the average level, and optimal energy saving performance, such as a power consumption-to-output ratio lower than the industry average by more than 40%; High energy efficiency: The energy utilization rate is relatively high, better than the industry average, such as a power consumption-to-output ratio lower than the industry average by 20% to 40%; Medium energy efficiency: The energy utilization rate is at the industry average level, with energy consumption deviating from the industry average by within ±10%, meeting basic energy-saving requirements; Low energy efficiency: The energy utilization rate is lower than the industry average, with a power consumption-to-output ratio higher than the industry average by 10% to 30%, belonging to the energy efficiency level that needs improvement; Extremely low energy efficiency: The energy utilization rate is far below the industry standard, with a power consumption-to-output ratio higher than the industry average by more than 30%, and may be classified as obsolete or restricted in use.

[0052] It is understood that the embodiments of this application, by mapping aggregated feature vectors to equipment performance scores, fault risk levels, and energy efficiency levels using regression algorithms, can transform abstract feature information into intuitive and quantifiable evaluation indicators, facilitating the assessment of the current operating status of the equipment; providing clear direction for equipment maintenance and reducing blind maintenance. This improves the accuracy and reliability of cold storage equipment status assessment and enhances the rationality and effectiveness of dynamic scheduling and collaborative control.

[0053] For example, in the condition assessment of the refrigeration equipment in a meat cold storage facility, the refrigeration units include compressors, condensers, and other equipment. Operating parameters include compressor discharge temperature (70-95℃), suction pressure (0.3-0.6MPa), and motor current (15-30A). Environmental parameters include internal temperature (-18 to -15℃) and humidity (80%-85%). The E-LSTM model processes these parameters and outputs an aggregated feature vector. This vector integrates information such as the compressor's continuous 8-hour temperature fluctuation trend, abnormal suction pressure fluctuations (e.g., a sudden pressure drop of 0.1MPa within 10 minutes), and changes in internal humidity. After mapping using a regression algorithm, the equipment performance score is 82 points (out of 100), the fault risk level is "medium" (risk levels are divided into high, medium, and low), and the energy efficiency level is level two (out of five levels, with level one being the best). This provides accurate equipment status information for subsequent scheduling strategies.

[0054] In this embodiment of the application, the formula for the E-LSTM model is:

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062] in, Forget gate vector; Input the device's operating parameters and environmental parameters at the current moment; The parameter change trend in the previous hour; For activation functions; Here is the forget gate weight matrix; This is the forget gate bias vector; Input gate vector; The input gate weight matrix; The input gate bias vector; New memory candidates are generated from the current input and historical state; Activation function; Historical fluctuation patterns; Overall assessment of equipment status; The candidate memory unit weight matrix; The bias vector for candidate memory cells; This is the output gate vector; This is the output gate weight matrix; This is the output gate bias vector; The current hidden state; Attention weights; Score for attention; The fusion of features at the current moment; for Attention weight matrix; This is the attention vector; The length of the monitoring cycle; It is an exponential function.

[0063] It is understood that, in this application embodiment, by constructing an E-LSTM model, the operating parameters and environmental parameters of the refrigeration equipment are input into the model. The model identifies the changing trends and abnormal fluctuations of the equipment operating parameters and outputs aggregated feature vectors. These vectors are then mapped to indicators such as equipment performance scores using preset standards. Simultaneously, an attention mechanism is used to analyze the input feature parameters that have the greatest impact on the equipment status, ultimately generating analysis results. This process captures the dynamic changes and potential anomalies in the equipment's operating status, providing a basis for equipment status assessment. It helps to identify fault risks in advance, understand the equipment's energy efficiency level, lay a reliable foundation for the formulation of subsequent dynamic scheduling strategies, improve the accuracy and foresight of cold storage equipment status analysis, and enhance the rationality and efficiency of collaborative control of refrigeration equipment.

[0064] For example, such as Figure 3 As shown, in the status monitoring of refrigeration equipment in a large fruit and vegetable cold storage facility, the operating parameters to be monitored include compressor suction temperature (-30℃~-10℃), discharge pressure (1.2MPa~1.8MPa), operating frequency (30Hz~60Hz), and lubricating oil temperature (40℃~60℃). Environmental parameters include indoor temperature (0℃~10℃), indoor humidity (75%~95%), and outdoor ambient temperature (-5℃~38℃). Data acquisition interval is 15 minutes, and a total of 10,800 sets of historical data have been input into the model. After inputting the above parameters into the constructed E-LSTM model, the model processes the time-series information through its internal structure. For example, at t=90 minutes, the forgetting gate vector f... t =0.68, retaining 68% of the previous moment's intake temperature change trend; input gate vector i t=0.71, incorporating the abnormal fluctuation of a sudden increase of 0.15 MPa in the current exhaust pressure. After model calculation, the output is an aggregated feature vector with a dimension of 48, which integrates key information such as changes in operating frequency (weight 27%), fluctuations in intake temperature (weight 23%), and the influence of humidity in the refrigeration unit (weight 18%). Based on the preset equipment health assessment standards, the aggregated feature vector is mapped to: equipment performance score of 79 points (out of 100), fault risk level "medium" (risk level classification: high ≥60 points, medium 30~59 points, low ≤29 points), and energy efficiency level 2 (level 1 energy efficiency ratio ≥4.0, level 2 3.5~3.9). Combining the model's attention mechanism weights, operating frequency (weight 31%) and exhaust pressure (weight 26%) are determined to be the parameters with the greatest impact on equipment status. Based on the equipment performance score, fault risk level, energy efficiency level, and the most influential input feature parameters, combined with the equipment maintenance history, analysis results are generated, providing accurate status basis for the subsequent dynamic scheduling of refrigeration equipment.

[0065] In this embodiment, a TFT neural network model trained with historical data is used to predict the changing trend of cooling demand by combining environmental parameters with the current time and production plan. This includes: constructing a feature vector dataset; inputting the feature vector dataset into the TFT neural network model, using the VSN in the model to dynamically filter key features, and simultaneously capturing time-dynamic features through a multi-head attention mechanism; using GRN to fuse key features and time-dynamic features to generate a joint feature set; and using quantile regression of the TFT neural network model based on the joint feature set to output the probability distribution of cooling demand and predict the changing trend of cooling demand.

[0066] VSN is a variable selection network that filters key information by dynamically assigning weights to features.

[0067] It should be noted that multi-head attention is a deep learning method that captures complex relationships between information by decomposing the input vector into multiple subspaces, computing attention in parallel, and merging the results.

[0068] GRN stands for Gated Residual Network, which optimizes information transmission by using gating and residual connections.

[0069] It is understood that the embodiments of this application use VSN to dynamically filter key features from the feature vector dataset, eliminate redundant features, reduce the computational complexity of the model, and provide accurate basic feature support for cooling demand prediction; capture time-dynamic features through multi-head attention mechanism to provide time-dimensional basis for predicting the changing trend of cooling demand; and fuse key features and time-dynamic features through GRN to generate a joint feature set, optimize the transmission and integration of feature information, ensure that effective features are fully utilized, and jointly improve the accuracy and timeliness of cooling demand prediction.

[0070] For example, in predicting refrigeration demand for a cold chain logistics cold storage facility, a feature vector dataset is first constructed, containing environmental parameters (outdoor temperature -10~40℃, indoor humidity 65%~95%), production plans (daily cargo throughput 8~30 tons), and historical refrigeration data (13,140 sets over 18 months). This dataset is then input into a TFT neural network model, which dynamically filters out key features such as outdoor temperature (weight 35%), cargo throughput (weight 30%), and indoor humidity (weight 20%) using VSN. Simultaneously, a multi-head attention mechanism captures temporal dynamics: refrigeration demand increases by 25%~40% during peak unloading hours from 14:00-16:00 on weekdays, decreases by 30% at night, and quarterly demand shows cyclical growth with the peak season for fresh produce. A joint feature set is generated through GRN fusion, and quantile regression using the TFT model outputs the probability distribution of refrigeration demand for the next 48 hours. This provides a basis for generating dynamic scheduling strategies for refrigeration equipment.

[0071] In step S103, based on the analysis results and the changing trend of cooling demand, combined with the preset scheduling rules and optimization objectives, the PPO algorithm is used to generate a dynamic scheduling strategy for the cooling equipment. The optimization objectives include minimizing energy consumption and improving cooling efficiency.

[0072] Among them, the dynamic scheduling strategy is a collaborative control scheme that dynamically adjusts the operation of refrigeration equipment according to the actual situation in order to achieve the goals of minimizing energy consumption and improving refrigeration efficiency.

[0073] It is understood that the embodiments of this application, through the analysis results of equipment operating status and the changing trend of refrigeration demand, combined with preset scheduling rules and the optimization goals of minimizing energy consumption and improving refrigeration efficiency, utilize the PPO algorithm to generate a dynamic scheduling strategy for refrigeration equipment; flexibly adjust the equipment operating mode according to the actual equipment status and real-time demand, reasonably start and stop the compressor, and adjust the load rate, so that the operation of the refrigeration equipment is more in line with actual needs, ensuring the refrigeration effect while minimizing energy consumption and improving refrigeration efficiency; at the same time, the dynamic scheduling strategy can cope with demand fluctuations and equipment status changes, ensure the stability and coordination of multi-equipment collaborative operation, reduce resource waste, provide support for the efficient operation of cold storage, and enhance the overall economy and reliability of refrigeration equipment.

[0074] For example, after a meat cold storage facility generates a dynamic scheduling strategy, it coordinates the control of six compressors. When the TFT model predicts that the cooling demand will reach 150kW (second preset value 120kW) from 9:00 to 11:00 the next day, and the E-LSTM model analyzes that the failure risk level of compressor No. 3 is "low" and its performance score is 90, the strategy instructs to start all two standby compressors and adjust the load rate of the operating compressors to 80%. In actual operation, the cooling demand during this period is 148kW, and the equipment load matching degree reaches 98.7%, a significant improvement compared to 75% before the strategy was not adopted. When the demand drops to 80kW in the afternoon (first preset value 90kW), the strategy automatically reduces the number of operating compressors by two and reduces the frequency of the remaining four compressors from 50Hz to 35Hz. During this period, the energy consumption is reduced by 22.3kW·h compared to the same demand. During this period, compressor No. 1 suddenly fails. The strategy immediately shuts down the equipment and starts the standby compressor, restoring the cooling capacity within 30 seconds. The temperature fluctuation in the storage facility is controlled within ±0.5℃, without affecting the quality of meat storage. The overall daily energy consumption was reduced by 18.6%, and the cooling efficiency was improved by 23%, which verified the effectiveness of the dynamic scheduling strategy.

[0075] In this embodiment, based on the analysis results and the changing trend of cooling demand, combined with preset scheduling rules and optimization objectives, a dynamic scheduling strategy for cooling equipment is generated using the PPO algorithm. This includes: constructing a DRL state space containing analysis results, the changing trend of cooling demand, and real-time operating parameters; defining an action space including equipment start-up and shutdown, load adjustment, and multi-machine collaborative mode based on the state space, combined with preset scheduling rules and optimization objectives, and designing a reward function using a reward function formula; and learning in the state space using the PPO algorithm based on the action space and the reward function to generate a scheduling strategy that adapts to dynamic demands and conforms to scheduling rules.

[0076] The reward function is used to quantify the quality of the algorithm's decision-making module's behavior, and guides the algorithm's decision-making module to learn better strategies by outputting numerical rewards.

[0077] The reward function formula is:

[0078] in, Total reward value; This is an energy consumption incentive item; This is a reward item for temperature control. This is a reward item for equipment operation; This is an electricity price incentive item.

[0079] It is understood that the embodiments of this application are based on energy consumption indicators, refrigeration efficiency parameters and equipment operating status, and are converted into quantifiable reward and penalty values. Positive rewards are given when the scheduling strategy reduces energy consumption, improves efficiency and keeps the equipment running stably, and negative penalties are given when energy consumption exceeds the standard, efficiency decreases and equipment risk increases. This guides the PPO algorithm to iteratively update in a direction that meets the optimization goal, reduces the blindness of the algorithm, enhances the pertinence of strategy optimization, and provides a strong guarantee for the scientific nature of collaborative control of refrigeration equipment and the economic efficiency of cold storage operation.

[0080] For example, a cold chain cold storage facility designed a multi-dimensional reward function when using the PPO algorithm to generate a dynamic scheduling strategy. This function uses hourly energy consumption (target ≤ 120 kWh), refrigeration efficiency (target ≥ 3.8), and equipment failure risk (target ≤ 29 points) as core indicators, setting a base reward value of 100 points. When the strategy reduces hourly energy consumption to 105 kWh (12.5% ​​below the target), an additional 20 points are awarded; when refrigeration efficiency reaches 4.0 (5.3% above the target), an additional 15 points are awarded; when equipment failure risk remains at 25 points (low risk), an additional 10 points are awarded, resulting in a positive reward of 145 points per strategy iteration. If energy consumption rises to 130 kWh (8.3% above the target), 30 points are deducted; if efficiency drops to 3.4 (10.5% below the target), 25 points are deducted; if risk rises to 40 points (medium risk), 20 points are deducted, resulting in a negative penalty of 25 points for the strategy. After 1000 iterations, the strategy achieved an optimization rate of 18% in energy consumption control, with cooling efficiency remaining stable above 3.9 and equipment failure risk reduced by 40%. This verified the effectiveness of the reward function in guiding algorithm iteration through quantitative rewards and penalties, ensuring that the final strategy aligns with the goals of minimizing energy consumption and improving efficiency.

[0081] In this embodiment, the preset scheduling rules are as follows: when the cooling demand is less than or equal to a first preset value, the number of operating compressors is reduced, and the frequency or volume of the operating compressors is reduced; when the cooling demand is greater than the first preset value but less than a second preset value, the standby compressors are started, and the load rate of the operating compressors is balanced; when the cooling demand is greater than the second preset value, all standby compressors are started, and the load is dynamically adjusted to match the demand value; when a compressor suddenly fails, the failed compressor is shut down, and the standby compressors are started to supplement the load.

[0082] The first preset value for refrigeration demand is 30% of the total refrigeration capacity of the cold storage, and the second preset value is 70% of the total refrigeration capacity of the cold storage.

[0083] It is understood that the embodiments of this application improve the stability of cold storage operation by standardizing the basic logic and behavioral boundaries of equipment scheduling, avoid the generation of scheduling actions by the algorithm that do not conform to the equipment safety operation specifications or actual scenario requirements, reduce the complexity of the algorithm to generate the optimal strategy, improve the reliability of decision-making, and ensure that the cold storage refrigeration can still maintain efficient and coordinated operation under dynamically changing environments and loads.

[0084] In this embodiment of the application, the PPO algorithm formula is as follows:

[0085] in, Parameters of the policy network; Time step; The ratio of the probability of the new strategy and the old strategy choosing the same action in the current state; Dominance function estimation; Clipping threshold; Expectation operator; It is a piecewise function; It is a minimum value function; This is the proxy objective function.

[0086] It is understood that the embodiments of this application use the PPO algorithm to learn in the constructed DRL state space based on the analysis results of equipment operation status and the changing trend of cooling demand, combined with preset scheduling rules and the optimization goals of minimizing energy consumption and improving cooling efficiency. This provides a scientific decision-making basis for the dynamic scheduling of cooling equipment, balances the stability of strategy updates, and improves the efficiency and economy of collaborative control of cooling equipment.

[0087] For example, such as Figure 4As shown, the aquatic product cold storage is equipped with 3 compressors (2 operating, 1 standby). E-LSTM model analysis shows that compressor 1 has a performance score of 85 and a failure risk score of 22 (low), while compressor 2 has a score of 82 and a risk score of 26 (medium-low). The TFT model predicts that the cooling demand will increase from 150kW to 170kW (+13.3%) in the next 2 hours. The state space constructed by the PPO algorithm includes the above equipment status, demand trends, and real-time energy consumption (currently 280kW·h / h). The action space is defined as "compressor frequency adjustment (40-60Hz), standby compressor start / stop". Combining the scheduling rules (the second preset value is 80% of the rated load, i.e., 180kW) and the reward function (2 points for every 1% decrease in energy consumption, 3 points for every 0.1 increase in COP), the algorithm iteratively generates the following strategy: compressor 1's frequency is adjusted from 50Hz to 53Hz, compressor 2 is maintained at 48Hz, and the standby compressor is not started (current demand does not exceed 180kW). After the strategy was executed, the PLC control terminal reported the following data: the temperature in the freezing area stabilized at -20.1℃±0.2℃ within one hour; the current of unit 1 was 23A (rated 25A), and that of unit 2 was 21A; total energy consumption decreased to 265kW・h / h (a 5.4% reduction); and the COP increased from 3.6 to 3.8. The reward function calculation score was 100 + 5.4 × 2 + (3.8 - 3.6) / 0.1 × 3 = 116 points. The algorithm was further optimized based on this score to ensure that subsequent scheduling always adapts to dynamic needs and meets the goal of minimum energy consumption.

[0088] In step S104, the scheduling strategy is sent to the PLC control terminal of the refrigeration equipment through the Internet of Things communication network. The PLC control terminal performs coordinated control of the refrigeration equipment according to the scheduling strategy, and at the same time, feeds back the execution status of the equipment to the cloud server.

[0089] Among them, the PLC control terminal refers to the terminal module that performs real-time control and command execution on the operating status of the cold storage refrigeration equipment through a programmable logic controller.

[0090] It is understood that the embodiments of this application use a PLC control terminal to coordinate the control of refrigeration equipment according to a strategy, and feed back the equipment execution status to the cloud server, so that the scheduling strategy can be transmitted, issued and executed, ensuring the implementation of the scheduling strategy, improving the timeliness of coordinated control, enhancing the controllability of cold storage operation, providing data support for the continuous optimization of models and algorithms, and providing a strong guarantee for the efficient and stable operation of cold storage.

[0091] For example, such as Figure 5As shown, in a meat cold storage facility, the scheduling strategy generated by the cloud server indicates that due to a 15% increase in current refrigeration demand compared to one hour ago (TFT model prediction result), it is necessary to start the No. 2 standby compressor, adjust the frequency of the No. 1 main compressor from 50Hz to 55Hz, maintain the No. 3 compressor at 45Hz, and simultaneously increase the evaporative air cooler speed to 1400r / min. The IoT communication network (using a 5G industrial module) sends this strategy to the PLC control terminal of the refrigeration equipment in the freezing area within 0.8 seconds. The PLC control terminal immediately executes the strategy: the No. 2 compressor starts within 15 seconds, the frequency of the No. 1 compressor gradually increases to 55Hz (fluctuation ≤0.3Hz), the No. 3 compressor maintains a stable operation at 45Hz, and the evaporative air cooler speed reaches 1400r / min within 20 seconds. During execution, the PLC control terminal fed back status data to the cloud server every 20 seconds: the real-time temperature of the freezing zone dropped from -21.5℃ to -22.1℃ (fluctuation ±0.2℃); the current of compressor 1 was 18A (rated 20A), the current of compressor 2 stabilized at 16A after startup, and the current of compressor 3 was 14A; the total energy consumption dropped from 65kW·h / h before the strategy execution to 62kW·h / h; equipment vibration, pressure, and other parameters were all within safe ranges (vibration ≤0.06g, condensing pressure ≤1.8MPa). After receiving the data, the cloud server combined the analysis of the equipment status using the E-LSTM model (no abnormal fluctuations) to confirm the effectiveness of the strategy execution, laying the foundation for the next round of scheduling adjustments.

[0092] This application proposes an IoT-based dynamic scheduling and collaborative control method for cold storage. By deploying IoT sensors to collect real-time equipment operating parameters and environmental data, and combining this with an E-LSTM model to evaluate equipment operating status, it breaks through the traditional scheduling mode that relies on manual experience or fixed logic, enhancing the ability to identify abnormal equipment fluctuations and providing a basis for dynamic scheduling. A TFT neural network model is used to predict the changing trend of refrigeration demand, and a PPO algorithm is combined to generate a dynamic scheduling strategy, improving the flexibility of response to dynamic demands. By setting optimization goals such as minimum energy consumption and maximum efficiency, energy utilization efficiency is improved. With the help of real-time transmission from the IoT and rapid cloud response, when equipment suddenly fails or refrigeration demand changes abruptly, the operating plan can be adjusted in a timely manner through the dynamic scheduling strategy, enhancing the stability of equipment refrigeration and reducing the risk of interruption due to unforeseen circumstances. This provides support for the efficient, energy-saving, and stable operation of industrial refrigeration equipment. Therefore, it solves the problems of rigid scheduling methods, serious energy waste, and insufficient emergency response capabilities in existing technologies.

[0093] The following will illustrate a dynamic scheduling and collaborative control method for cold storage based on the Internet of Things through a specific embodiment, such as... Figure 6 As shown, it includes: In the fresh food cold chain storage facility, at 7:00 AM on Monday, IoT sensors distributed across various devices and areas simultaneously went into operation, comprehensively capturing operational details and environmental data related to refrigeration. At the refrigeration equipment level, compressor No. 1 in the refrigerated section operated at 14.5A current (rated 18A) and 42Hz frequency, with an exhaust temperature of 73℃ and a suction pressure of 0.32MPa; compressor No. 2 operated at 13.8A current and 38Hz frequency, with an exhaust temperature of 71℃ and a suction pressure of 0.3MPa, both operating within their stable ranges. In the frozen section, compressor No. 1, responsible for lower temperatures, operated at 21A current (rated 25A) and 50Hz frequency, with an exhaust temperature of 82℃ and a suction pressure of 0.16MPa. Compressor No. 2, a standby compressor, was preheated to 52℃ and ready to be put into operation at any time. Regarding environmental parameters, the temperatures in cold storage zones A and B were 5.2℃ and 4.8℃, with humidity levels of 88% and 85%, respectively; the temperature in frozen zone C was -17.5℃, with humidity at 68%, all meeting the temperature and humidity standards for fresh food storage. The outdoor temperature was 23℃, and the humidity was 62%, providing a reference for assessing the impact of the external environment. This data, aggregated via a LoRa gateway, was transmitted completely to the cloud server within 3 minutes, with a transmission latency of only 1.8 seconds and 100% data integrity, laying a precise data foundation for subsequent intelligent analysis.

[0094] After receiving the data, the cloud server immediately activates the E-LSTM model and the TFT model for deep analysis. The E-LSTM model focuses on device status assessment, taking 21-dimensional data containing operational and environmental parameters as input, and passing it through a forget gate (f... t =0.83) Filter valid historical information, input gate (i t =0.77) Integrating new features, a 128-dimensional aggregated feature vector is generated. After regression algorithm: , , The performance of the equipment is mapped as follows: Compressor No. 1 in the refrigeration section scores 89 points, has a failure risk of 21 points (low risk), and is rated Energy Efficiency Level 1; Compressor No. 2 scores 84 points, has a risk of 29 points (low to medium risk), and is rated Energy Efficiency Level 2; Compressor No. 1 in the freezer section scores 87 points, has a risk of 23 points, and is rated Energy Efficiency Level 1. This is combined with the model's attention mechanism weights (α). t =0.86), clearly identifying "refrigeration compressor frequency stability" and "freezing compressor suction pressure" as key parameters affecting performance. Meanwhile, based on nearly three months of historical data (daily refrigeration capacity and inbound volume from 7:00 to 12:00), combined with the current environment, time (Monday 7:00), and production plan (500kg of fruits and vegetables and 800kg of frozen meat inbound at 10:00), the TFT model accurately predicts demand changes from 7:00 to 10:00: refrigeration demand increases from 95kW to 120kW (+26.3%), and freezing demand increases from 140kW to 160kW (+14.3%), providing a forward-looking basis for scheduling strategy formulation.

[0095] Based on the analysis results and predicted trends, the PPO algorithm: The strategy is generated by combining scheduling rules (first preset value 60% of rated load, second preset value 80%) and optimization objectives (lowest energy consumption, highest efficiency). The state space of the strategy formulation covers equipment performance score (84~89 points), failure risk (21~29 points), predicted demand (95~160kW), and real-time total energy consumption (260kW・h / h); the action space includes compressor start / stop and frequency adjustment (35~60Hz); the reward function is based on 100 points, adding 2 points for every 1% decrease in energy consumption and 3 points for every 0.1 increase in COP, guiding strategy optimization through an incentive mechanism. The final strategy is as follows: 7:10-10:00, in the refrigeration area, standby unit No. 3 (36Hz) is started, unit No. 1 is maintained at 42Hz, and unit No. 2 is adjusted to 45Hz, with a total load rate of 72%; in the freezer area, unit No. 1 is adjusted to 53Hz, and the standby unit is not started (load 77.8%). After 10:00, the refrigeration unit No. 3 will be reduced to 32Hz, and if the demand in the freezing unit exceeds 160kW, the standby unit No. 2 (48Hz) will be started to ensure a dynamic balance between supply and demand.

[0096] After the strategy is generated, it is transmitted to the PLC control terminal via a 5G industrial network with encryption, with a delay of only 0.7 seconds, ensuring efficient delivery of instructions. The PLC then executes the operation precisely: Unit 3 in the refrigeration zone starts up within 15 seconds, Unit 2 smoothly rises to 45Hz within 5 minutes (frequency fluctuation ≤0.2Hz); Unit 1 in the freezing zone rises to 53Hz within 8 minutes (current 22.5A≤25A), and operates stably. During execution, the PLC provides data feedback every 20 seconds: At 8:00, the temperature in the refrigerated area is 3.5℃ (±0.1℃), and the temperature in the frozen area is -18.2℃ (±0.2℃). Total energy consumption drops to 245kW・h / h (a decrease of 5.8%), and COP increases from 3.5 to 3.8, demonstrating a significant improvement in energy efficiency. At 10:00, when goods are put into storage, the refrigerated demand reaches 120kW (72% load), the frozen demand is 158kW, and the total energy consumption is 252kW・h / h (a decrease of 1.2%). Equipment vibration is 0.05~0.07g (≤0.1g), and all parameters are within the normal range. The feedback data is transmitted back to the cloud in real time to verify the effectiveness of the strategy and optimize the next round of control, forming a closed-loop intelligent management system of "collection-analysis-decision-execution-feedback," providing continuous assurance for the efficient and safe operation of the fresh food cold chain cold storage.

[0097] In summary, this invention acquires operating parameters of refrigeration equipment and environmental parameters of the refrigeration area, analyzes the equipment's operating status using an E-LSTM model, and generates analysis results including performance scores, fault risk levels, and energy efficiency levels. Simultaneously, it uses a TFT neural network model combined with environmental parameters, time, and production plans to predict trends in refrigeration demand. Based on these results, and combining preset scheduling rules with optimization goals of minimizing energy consumption and improving refrigeration efficiency, a dynamic scheduling strategy is generated using the PPO algorithm. This strategy is then transmitted to the PLC control terminal via the Internet of Things (IoT) for collaborative control, while the execution status is fed back to the cloud. This enables intelligent dynamic scheduling of the refrigeration equipment. Through multi-model collaboration and algorithm optimization, the economic efficiency and effectiveness of refrigeration equipment operation are improved, the stability and safety of equipment operation are ensured, energy waste is reduced, and the flexibility of cold storage in responding to demand changes is enhanced, providing support for the efficient and stable operation of cold storage facilities.

[0098] Next, referring to the accompanying drawings, a dynamic scheduling and collaborative control system for cold storage based on the Internet of Things is described according to an embodiment of this application.

[0099] Figure 7 This is a block diagram of an IoT-based dynamic scheduling and collaborative control system for cold storage, according to an embodiment of this application.

[0100] like Figure 7 As shown, the IoT-based dynamic scheduling and collaborative control system 10 for cold storage includes: an acquisition module 100, an analysis module 200, a generation module 300, and an execution module 400.

[0101] The system comprises the following modules: Acquisition module 100 acquires the operating parameters of the refrigeration equipment and the environmental parameters of the refrigeration area; Analysis module 200 analyzes the operating status of the refrigeration equipment using an E-LSTM model based on the operating and environmental parameters, generates analysis results, trains a TFT neural network model using historical data, and predicts the trend of refrigeration demand changes by combining environmental parameters with the current time and production plan; Generation module 300 generates a dynamic scheduling strategy for the refrigeration equipment based on the analysis results and the trend of refrigeration demand changes, combined with preset scheduling rules and optimization objectives, using the PPO algorithm, where optimization objectives include minimizing energy consumption and improving refrigeration efficiency; Execution module 400 sends the scheduling strategy to the PLC control terminal of the refrigeration equipment via an IoT communication network. The PLC control terminal performs coordinated control of the refrigeration equipment according to the scheduling strategy, and simultaneously feeds back the execution status of the equipment to the cloud server.

[0102] It should be noted that the foregoing explanation of an embodiment of a dynamic scheduling and collaborative control method for cold storage based on the Internet of Things also applies to this embodiment of a dynamic scheduling and collaborative control system for cold storage based on the Internet of Things, and will not be repeated here.

[0103] This application proposes an IoT-based dynamic scheduling and collaborative control system for cold storage. By deploying IoT sensors to collect real-time equipment operating parameters and environmental data, and combining this with an E-LSTM model to evaluate equipment operating status, it breaks through the traditional scheduling mode that relies on manual experience or fixed logic, enhancing the ability to identify abnormal equipment fluctuations and providing a basis for dynamic scheduling. It utilizes a TFT neural network model to predict the changing trend of refrigeration demand and combines it with a PPO algorithm to generate dynamic scheduling strategies, improving the flexibility of response to dynamic demands. By setting optimization goals such as minimum energy consumption and maximum efficiency, it improves energy utilization efficiency. With the help of real-time transmission from the IoT and rapid cloud response, when equipment suddenly fails or refrigeration demand changes abruptly, the dynamic scheduling strategy can adjust the operating plan in a timely manner, enhancing the stability of equipment refrigeration and reducing the risk of interruption due to unforeseen circumstances. This provides support for the efficient, energy-saving, and stable operation of industrial refrigeration equipment. Therefore, it solves the problems of rigid scheduling methods, serious energy waste, and insufficient emergency response capabilities in existing technologies.

[0104] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 801, the processor 802, and the computer program stored on the memory 801 and capable of running on the processor 802.

[0105] When the processor 802 executes the program, it implements the IoT-based dynamic scheduling and collaborative control method for cold storage provided in the above embodiments.

[0106] Furthermore, electronic devices also include: Communication interface 803 is used for communication between memory 801 and processor 802.

[0107] The memory 801 is used to store computer programs that can run on the processor 802.

[0108] The memory 801 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0109] If the memory 801, processor 802, and communication interface 803 are implemented independently, then the communication interface 803, memory 801, and processor 802 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0110] Optionally, in a specific implementation, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, then the memory 801, processor 802, and communication interface 803 can communicate with each other through an internal interface.

[0111] The processor 802 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0112] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described IoT-based dynamic scheduling and collaborative control method for cold storage.

[0113] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0114] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0115] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0116] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0117] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0118] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things, characterized in that, include: Obtain the operating parameters of the refrigeration equipment and the environmental parameters of the refrigeration area; Based on the operating parameters and environmental parameters, the operating status of the refrigeration equipment is analyzed using an E-LSTM model to generate analysis results. A TFT neural network model is trained using historical data, and combined with the environmental parameters, the current time, and the production plan, the trend of refrigeration demand changes is predicted. Based on the analysis results and the trend of cooling demand changes, combined with the preset scheduling rules and optimization objectives, the PPO algorithm is used to generate a dynamic scheduling strategy for the cooling equipment. The optimization objectives include minimizing energy consumption and improving cooling efficiency. The scheduling strategy is sent to the PLC control terminal of the refrigeration equipment via the Internet of Things (IoT) communication network. The PLC control terminal performs coordinated control of the refrigeration equipment according to the scheduling strategy, and at the same time, feeds back the execution status of the equipment to the cloud server.

2. The method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things according to claim 1, characterized in that, Based on the aforementioned operating parameters and environmental parameters, the operating status of the refrigeration equipment is analyzed using an E-LSTM model, and analysis results are generated, including: Construct an E-LSTM model; The operating parameters and environmental parameters are input into the E-LSTM model to identify the changing trends and abnormal fluctuations of the equipment operating parameters, and output an aggregated feature vector. Based on the preset equipment health assessment standards, the aggregated feature vector is mapped to equipment performance score, fault risk level and energy efficiency level through regression algorithm. Combined with the attention mechanism weight of the E-LSTM model, the input feature parameters that have the greatest impact on equipment status are analyzed. Based on the equipment performance score, fault risk level, energy efficiency level, and the most influential input feature parameters, combined with the equipment maintenance history, analysis results are generated.

3. The method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things according to claim 1, characterized in that, The formula for the E-LSTM model is as follows: ； ； ； ； ； ； ； ； in, Forget gate vector; Input the device's operating parameters and environmental parameters at the current moment; This shows the parameter change trend for the previous hour; For activation functions; Here is the forget gate weight matrix; This is the forget gate bias vector; The input gate vector; The input gate weight matrix; The input gate bias vector; New memory candidates generated based on the current input and historical state; For activation functions; This reflects historical fluctuation patterns. For an overall assessment of the equipment status; The candidate memory unit weight matrix; The bias vector for candidate memory cells; This is the output gate vector; This is the output gate weight matrix; This is the output gate bias vector; The current hidden state; Attention weights; Score for attention; The fusion of features at the current moment; for Attention weight matrix; This is the attention vector; The length of the monitoring cycle; It is an exponential function.

4. The method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things according to claim 1, characterized in that, A TFT neural network model trained on historical data, combined with the aforementioned environmental parameters and current time and production plans, predicts the changing trends in cooling demand, including: Construct a feature vector dataset; The feature vector dataset is input into the TFT neural network model, and key features are dynamically filtered using the VSN in the model. At the same time, the time dynamic features are captured through the multi-head attention mechanism. The key features and the time-dynamic features are fused using GRN to generate a joint feature set; Based on the joint feature set, the probability distribution of cooling demand is output using quantile regression of the TFT neural network model, and the changing trend of cooling demand is predicted.

5. The method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things according to claim 1, characterized in that, Based on the analysis results and the changing trend of cooling demand, combined with preset scheduling rules and optimization objectives, a dynamic scheduling strategy for cooling equipment is generated using the PPO algorithm, including: Construct a DRL state space that includes analysis results, cooling demand change trends, and real-time operating parameters; Based on the state space, combined with the preset scheduling rules and optimization objectives, an action space including device start-up and shutdown, load adjustment, and multi-machine collaborative mode is defined, and a reward function is designed through the reward function formula; Based on the action space and the reward function, the PPO algorithm is used to learn in the state space to generate a scheduling strategy that adapts to dynamic requirements and conforms to scheduling rules.

6. The method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things according to claim 1, characterized in that, The preset scheduling rules are as follows: When the cooling demand is less than or equal to the first preset value, reduce the number of operating compressors and reduce the frequency or volume of the operating compressors. When the cooling demand is greater than the first preset value but less than the second preset value, the standby compressor is started and the compressor load rate is balanced. When the cooling demand exceeds the second preset value, all standby compressors are started, and the load is dynamically adjusted to match the demand value. When a compressor suddenly fails, shut down the faulty compressor and start the backup compressor to supplement the load.

7. The method for dynamic scheduling and collaborative control of cold storage based on the Internet of Things according to claim 4, characterized in that, The PPO algorithm formula is as follows: ； in, These are the parameters of the policy network; For time steps; This represents the ratio of the probability of the new strategy and the old strategy choosing the same action in the current state. For the estimation of the advantage function; This is the clipping threshold; For expectation operators; It is a piecewise function; It is a minimum value function; This is the proxy objective function.

8. A dynamic scheduling and collaborative control system for cold storage based on the Internet of Things, characterized in that, include: The acquisition module is used to acquire the operating parameters of the refrigeration equipment and the environmental parameters of the refrigeration area; The analysis module is used to analyze the operating status of the refrigeration equipment using an E-LSTM model based on the operating parameters and environmental parameters, generate analysis results, train a TFT neural network model using historical data, and predict the trend of refrigeration demand changes by combining the environmental parameters with the current time and production plan. The generation module, based on the analysis results and the changing trend of cooling demand, combined with preset scheduling rules and optimization objectives, uses the PPO algorithm to generate a dynamic scheduling strategy for the cooling equipment, wherein the optimization objectives include minimizing energy consumption and improving cooling efficiency. The execution module is used to send the scheduling strategy to the PLC control terminal of the refrigeration equipment through the Internet of Things communication network. The PLC control terminal performs scheduling and coordinated control of the refrigeration equipment according to the scheduling strategy, and at the same time, feeds back the execution status of the equipment to the cloud server.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the Internet of Things-based dynamic scheduling and collaborative control method for cold storage as claimed in claims 1-7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the IoT-based dynamic scheduling and collaborative control method for cold storage as claimed in claims 1-7.

Images