Dynamic carbon factor modeling and adaptive threshold multi-system calibration method

Abstract

The invention provides a dynamic carbon factor modeling and adaptive threshold multi-system calibration method, and relates to the field of data processing. In the method, multi-modal data in a green park is acquired through a plurality of sensors, dynamic modeling is performed on carbon factors by using an LSTM model, and nonlinear features in the data are extracted. Modeling results are processed according to different working condition slices, and a self-adaptive threshold value is generated by adopting a deep reinforcement learning model. A carbon emission scheduling strategy of park equipment and regions is optimized through multi-agent reinforcement learning, the offset of a carbon factor and a threshold value is monitored in real time, an automatic calibration mechanism is triggered for adjustment, and finally real-time optimization and calibration of carbon emission are achieved. By implementing the technical scheme provided by the invention, the accuracy of dynamic carbon factor modeling and adaptive threshold calibration can be improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, specifically to a dynamic carbon factor modeling and adaptive threshold multi-system calibration method. Background Technology

[0002] With the increasing demand for green industrial park construction and intelligent management, effective monitoring and optimization of carbon emissions has become one of the important goals for achieving sustainable development.

[0003] In existing technologies, traditional carbon factor modeling methods and adaptive threshold generation methods usually rely on linear models or simplified statistical models. These methods have significant limitations when dealing with complex time-series data and nonlinear characteristics of carbon emissions, especially when dealing with the variable operating conditions in green parks, such as equipment load changes and climate change. They cannot fully consider the complex relationship between carbon emission factors and environmental factors under different operating conditions, which can easily lead to large errors in carbon factor prediction and make it difficult to meet the requirements of high-precision calibration.

[0004] Therefore, improving the accuracy of dynamic carbon factor modeling and adaptive threshold calibration has become a key technical challenge in carbon emission management in green industrial parks. Summary of the Invention

[0005] This application provides a dynamic carbon factor modeling and adaptive threshold multi-system calibration method, which facilitates the improvement of the accuracy of dynamic carbon factor modeling and adaptive threshold calibration.

[0006] A first aspect of this application provides a dynamic carbon factor modeling and adaptive threshold multi-mode calibration method. The method includes: acquiring multimodal data on carbon emissions collected by multiple sensors within a green park; dynamically modeling the carbon factor using an LSTM model based on the multimodal data, extracting nonlinear features from the multimodal data, and generating a dynamic carbon factor modeling result; slicing the dynamic carbon factor modeling result according to different operating conditions within the green park, and generating an adaptive threshold for each slice using a deep reinforcement learning model; optimizing the carbon emission scheduling of equipment and areas within the green park using a multi-agent reinforcement learning framework based on the dynamic carbon factor modeling result and the multimodal data, and generating an optimization strategy; monitoring the long-term offset between the dynamic carbon factor modeling result and the adaptive threshold in real time using a drift detection algorithm, and triggering an automatic calibration mechanism; and optimizing and calibrating the carbon emission situation of the green park in real time according to the optimization strategy and the automatic calibration mechanism.

[0007] A second aspect of this application provides a dynamic carbon factor modeling and adaptive threshold multi-mode calibration device. The device includes an acquisition module and a processing module. The acquisition module acquires multimodal data collected by multiple sensors within a green park regarding carbon emissions. The processing module dynamically models the carbon factor using an LSTM model based on the multimodal data, extracts nonlinear features from the multimodal data, and generates a dynamic carbon factor modeling result. The processing module further segments the dynamic carbon factor modeling result according to different operating conditions within the green park and generates an adaptive threshold for each segment using a deep reinforcement learning model. The processing module also optimizes the carbon emission scheduling of equipment and areas within the green park based on the dynamic carbon factor modeling result and the multimodal data using a multi-agent reinforcement learning framework, generating an optimization strategy. The processing module monitors the long-term offset between the dynamic carbon factor modeling result and the adaptive threshold in real time using a drift detection algorithm and triggers an automatic calibration mechanism. Finally, the processing module optimizes and calibrates the carbon emissions of the green park in real time according to the optimization strategy and the automatic calibration mechanism.

[0008] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, and both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described above.

[0009] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described above.

[0010] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By collecting various types of carbon emission data in real time from multiple sensors, the system can comprehensively and accurately monitor carbon emissions within green industrial parks. The multimodal data from different sensors provides rich environmental information, ensuring high accuracy and comprehensiveness in carbon emission monitoring. Employing an LSTM model for dynamic modeling of carbon factors effectively handles the temporal and nonlinear characteristics of carbon emission data. The LSTM model excels at capturing long-term dependencies and can cope with complex changes in carbon emission factors, improving the accuracy and predictive power of carbon factor modeling. By slicing the carbon factor modeling results according to different operating conditions and generating adaptive thresholds using a deep reinforcement learning model, the system can adjust carbon emission control strategies in real time to address different environmental changes. Adaptive thresholds provide high flexibility and adaptability to carbon emission management, avoiding the limitations imposed by fixed thresholds in traditional methods.

[0011] By employing a multi-agent reinforcement learning framework for carbon emission scheduling optimization, the carbon emission scheduling of different equipment and areas within the park can be optimized based on real-time data. Each equipment or area acts as an independent agent, coordinating and cooperating to optimize the overall carbon emission management of the park, improving scheduling efficiency and the accuracy of carbon emission control. A drift detection algorithm monitors the long-term deviation between the carbon factor modeling results and the adaptive threshold in real time. The system can quickly identify whether there are deviations in the carbon factor model and adjust them through an automatic calibration mechanism. This mechanism ensures the continuous accuracy of the carbon emission control strategy, especially when changes in the external environment or system deviations occur, enabling timely correction and preventing a decline in carbon emission control effectiveness. Combining the optimization strategy and the automatic calibration mechanism, the system can dynamically adjust the carbon emission management scheme based on real-time data and feedback. It ensures that the real-time carbon emission scheduling of equipment operation within the green park not only meets environmental protection requirements but also makes timely adjustments based on various factors, thereby achieving the best carbon emission control effect. Therefore, it facilitates improving the accuracy of dynamic carbon factor modeling and adaptive threshold calibration. Attached Figure Description

[0012] Figure 1 A flowchart illustrating a dynamic carbon factor modeling and adaptive threshold multi-system calibration method provided in this application embodiment; Figure 2 A schematic diagram of a module for a dynamic carbon factor modeling and adaptive threshold multi-system calibration device provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0013] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0015] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0016] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. 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 indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0017] To address the aforementioned technical problems, this application provides a dynamic carbon factor modeling and adaptive threshold multi-system calibration method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a dynamic carbon factor modeling and adaptive threshold multi-system calibration method provided in an embodiment of this application. The method is applied to a server and includes steps S110 to S160, as follows:

[0018] S110: Acquire multimodal data on carbon emissions collected by multiple sensors within the green park.

[0019] Specifically, a server is a computer system specifically designed for storing, managing, and processing data. It provides computing resources and data storage space, serving multiple clients in a network environment. In this embodiment, the server's role is to receive and process data from sensors and perform further analysis and decision-making. For example, the server might undertake data collection, processing, and scheduling tasks within the energy management system of a green park. A green park refers to a park that meets sustainable development standards such as environmental protection, resource conservation, and energy efficiency. Such parks include green buildings, renewable energy systems, and low-carbon emission equipment. Parks based on renewable energy sources such as solar and wind power may use smart devices to monitor carbon emissions in real time and optimize energy use through data analysis. Green parks aim to reduce environmental footprint and promote green development.

[0020] Sensors are devices used to detect and measure environmental parameters such as physical quantities, chemical quantities, temperature, humidity, and gas concentrations. Sensors can collect environmental data in real time, converting it into electronic signals or digital information for processing by computer systems. In green parks, sensors are used to monitor temperature, humidity, gas concentrations, as well as carbon dioxide, nitrogen oxides, and energy consumption. For example, gas sensors are used to detect the concentration of carbon dioxide in the air to calculate the park's carbon emissions. Carbon emissions refer to greenhouse gases, particularly carbon dioxide. Carbon emission monitoring refers to the monitoring and recording of carbon emissions in green parks. This monitoring can acquire real-time data through sensors and help analyze carbon emission trends, thereby enabling appropriate emission reduction measures to be taken. For example, industrial equipment within the park may produce carbon dioxide emissions by burning fossil fuels; sensors are used to monitor these emissions in real time.

[0021] Multimodal data refers to data from multiple sources, formats, or types. These data sources can be different sensors, devices, or systems, including information such as temperature, humidity, gas concentration, and energy consumption. By integrating these different types of data, more comprehensive environmental analysis results can be obtained. In green parks, sensors may simultaneously provide multiple data sources, such as a combination of temperature and gas sensors. This multimodal data can help to understand carbon emissions more accurately.

[0022] Furthermore, firstly, multiple types of sensors are deployed within the green park to collect carbon emission-related data in real time. These sensors include gas sensors, temperature and humidity sensors, power meters, weather monitoring sensors, and light sensors. Each sensor is responsible for detecting specific environmental parameters. For example, gas sensors monitor the concentration of greenhouse gases such as carbon dioxide and nitrogen oxides in the air; temperature and humidity sensors measure the temperature and humidity within the park; power meters monitor the energy consumption of equipment; weather monitoring sensors provide meteorological data such as wind speed, precipitation, and temperature changes; and light sensors monitor the light intensity within the park. These sensors transmit data to servers or data processing centers in real time via wireless networks or other communication methods, ensuring that all carbon emission-related environmental data can be acquired promptly.

[0023] The collected multimodal data needs to be timestamped to ensure data consistency and synchronization. Since data from different sensors have different sampling frequencies and time periods, a unified timestamping process is essential. Each sensor data point is assigned a unique timestamp upon acquisition, indicating the specific point in time when the data was collected. This unified timestamp not only ensures the temporal consistency of various data sources but also provides a precise temporal basis for subsequent data analysis, modeling, and optimization. Timestamping ensures that data from different sensors can be aligned, thereby accurately reflecting the actual carbon emissions of the park during subsequent data processing and fusion.

[0024] Next, a data fusion algorithm is used to integrate and fuse data from different sensors to obtain complete multimodal data. This process uses algorithms such as Kalman filtering and weighted averaging to achieve data fusion, eliminating noise and inconsistencies between data from different sensors and providing a unified and accurate carbon emission dataset. Kalman filtering is a recursive algorithm used to estimate the state of a system from a series of noisy data, and is particularly suitable for dynamically changing systems. Weighted averaging calculates a weighted average by assigning different weights to data from different sources, thus giving higher weight to more reliable data during the fusion process. The formula for Kalman filtering is as follows:

[0025] in, It is the state estimate at the current moment. It is the state estimate from the previous moment. It is Kalman gain. This is the current observation value. It is the observation matrix.

[0026] Kalman gain It can be calculated using the following formula: in, It is the covariance of the estimation error at the previous time step. It is the covariance of the observation noise.

[0027] During the data fusion process, a weighted average algorithm is used to process the data weights from different sensors. Weighting coefficients are used to sum the data of different data types, resulting in fused multimodal data. This method not only effectively removes the noise influence from each sensor but also ensures a reasonable contribution from each data type. Through the aforementioned timestamp marking and data fusion steps, a complete multimodal dataset is finally obtained. This dataset contains real-time data collected by various sensors within the green park, and all data is synchronized uniformly according to time series. At this point, the obtained multimodal data has eliminated time misalignment and noise interference between different sensors and can be used as input for subsequent carbon emission modeling, optimization scheduling, and other algorithms. The fused data can more accurately reflect the changes and influencing factors of carbon emissions within the park, providing solid data support for subsequent carbon emission prediction and scheduling optimization.

[0028] S120. Based on the multimodal data, the carbon factor is dynamically modeled using the LSTM model, and the nonlinear features in the multimodal data are extracted to generate dynamic carbon factor modeling results.

[0029] Specifically, LSTM is a special type of recurrent neural network that excels at processing and predicting time-series data. Compared to traditional RNNs, LSTM can effectively remember long-term dependencies, avoiding the vanishing or exploding gradient problems found in traditional RNNs. Therefore, LSTM is well-suited for dynamic modeling of carbon factors, especially when dealing with time-series data such as the temporal variation of carbon emissions. LSTM uses its gating mechanism to control the flow of information, ensuring that important historical information is retained over longer periods, making it suitable for handling long-term dependency problems.

[0030] A carbon factor is a variable or indicator that measures or represents carbon emissions. In carbon emission management in green industrial parks, the carbon factor represents the amount of carbon emissions per unit of energy consumption, equipment operation, or under specific operating conditions. Changes in the carbon factor can reflect the dynamic fluctuations in carbon emissions within the park, and LSTM models are used to capture these dynamic changes. Dynamic modeling refers to using mathematical or computational models to describe a system or process that changes over time. In the embodiments of this application, dynamic modeling means building a model based on multimodal data to predict how the carbon factor changes over time. Through dynamic modeling, the system can understand the fluctuation patterns of the carbon emission factor and predict future carbon emission trends.

[0031] Nonlinear features refer to the complex, non-linear relationships within data. In carbon emission prediction, nonlinear features may include complex relationships between carbon factors and factors such as temperature, humidity, and energy use. Traditional linear models cannot effectively capture these complex interactions; therefore, deep learning models such as LSTM are needed to extract and model these nonlinear features. Dynamic carbon factor modeling results are obtained by dynamically modeling multimodal data and carbon factor variation patterns using an LSTM model. These results are presented as time series, reflecting the changing trends of carbon emissions within the industrial park at different time points. This result provides accurate data support for subsequent carbon emission scheduling and optimization.

[0032] Furthermore, firstly, the multimodal data collected by sensors within the green park needs to be input into an LSTM model for processing. This multimodal data may include data collected from devices such as temperature and humidity sensors, gas sensors, and power meters, and it exhibits both temporal and nonlinear characteristics. In this process, the LSTM model receives the time series data from these multimodal datasets as input, aiming to extract the temporal and nonlinear features of the carbon factor. The LSTM model is particularly well-suited for processing time series data because it can retain information over long periods and capture long-term dependencies.

[0033] LSTM models control information flow through input gates, forget gates, and output gates, enabling them to effectively respond to changes in carbon factors at each time step while learning the temporal and nonlinear dependencies in the data. For example, LSTM models can capture fluctuations in carbon factors over different time periods and identify the nonlinear effects of factors such as changes in temperature and humidity, and fluctuations in equipment load on carbon emissions.

[0034] During the training of the LSTM model, the model processes input data at multiple time steps, gradually extracting and learning temporal features. These temporal features reflect the changing trends and patterns of the carbon factor over time. Nonlinear features reflect the complex relationships between the carbon factor and other factors in the multimodal data, such as temperature, humidity, and gas concentration. In each layer of the LSTM, information flows through various gating mechanisms within the network. After multiple iterations, the model can extract the relationships between the carbon factor and these temporal and nonlinear features. For example, during training, the relationship between the carbon factor and changes in temperature and humidity is not linear. LSTM can effectively learn this complex relationship through its nonlinear activation functions, such as tanh or ReLU. The model automatically adjusts the weights to accurately fit the nonlinear relationships between the carbon factor and features such as temperature and humidity changes and energy consumption over time.

[0035] The LSTM model optimizes its network weights through feedforward and backpropagation processes. In the feedforward process, the model inputs multimodal data into each layer for processing. Each layer undergoes a nonlinear transformation using activation functions such as sigmoid and tanh, thereby fitting the input features. In the backpropagation process, the model calculates the error between the predicted and actual results, for example, using the mean squared error (MSE) as the loss function, and updates the weights of each layer according to the gradient descent algorithm. This allows the model to better fit the relationship between the carbon factor and temporal and nonlinear features.

[0036] Specifically, the carbon factor value at each time step is processed by each neuron in the network, and information is transmitted and adjusted under the action of the nonlinear activation function in each layer until the output modeling result can accurately reflect the changing trend of the carbon factor. In this process, the depth of the network layers and the use of nonlinear activation functions help to capture the complex nonlinear relationship between the carbon factor and other variables.

[0037] Ultimately, the LSTM model, through multiple training cycles, generates a dynamic carbon factor modeling result. This result is presented in time series form, reflecting the predicted values ​​of the carbon factor at different points in time. This dynamic modeling result can describe the fluctuation patterns of the carbon factor, helping to predict future carbon emission trends. For example, the system can predict future carbon emission levels based on historical carbon emission data, such as temperature changes, humidity changes, and equipment operation factors, providing important basis for subsequent carbon emission scheduling and optimization. Through the training and optimization of the LSTM model, the system can obtain high-precision carbon factor modeling results and adjust the carbon emission scheduling strategy within the park in real time, ensuring that carbon emissions within the green park remain within a controllable range.

[0038] S130. The dynamic carbon factor modeling results are sliced ​​according to different working conditions in the green park, and an adaptive threshold is generated for each slice through a deep reinforcement learning model.

[0039] Specifically, operating conditions refer to the state or environmental conditions under which a system or equipment operates. In carbon emission management, different operating conditions may include different equipment loads, weather conditions, energy usage patterns, etc., all of which affect the level of carbon emissions. For example, when the temperature rises, the load on the air conditioning system increases, and carbon emissions may rise accordingly. Based on these different environmental and operating conditions, the modeling and scheduling strategies for dynamic carbon factors may differ. Slicing refers to dividing the dynamic carbon factor modeling results into segments according to different operating conditions. Each slice represents the behavior and changes of carbon factors under a specific operating condition. For example, a park may have different seasons, equipment loads, or operating patterns, and the variation patterns of carbon factors are different under each operating condition. Therefore, by slicing the modeling results, carbon emissions can be analyzed and controlled more precisely. For example, in summer, when the air conditioning load is high, carbon factors may exhibit large fluctuations, while in winter, when the air conditioning load is low, carbon factor fluctuations are smaller.

[0040] Deep reinforcement learning is an algorithmic framework that combines deep learning and reinforcement learning. It enables agents, such as devices, regions, or systems, to optimize their decisions through trial and error as they interact with their environment. In carbon emission management, deep reinforcement learning models are used to generate adaptive thresholds, determining how to optimize carbon emission control strategies under specific operating conditions. By training on historical data, deep reinforcement learning can learn optimal control behavior under various environmental conditions through reward and penalty mechanisms. For example, a system might use deep reinforcement learning to determine how to adjust the temperature settings of an air conditioning system to minimize carbon emissions.

[0041] An adaptive threshold refers to a threshold that is dynamically adjusted based on different operating conditions and real-time data. In carbon emission control, thresholds are used to limit the range of carbon emission factors or other related parameters to ensure that carbon emissions do not exceed predetermined limits. Adaptive thresholds automatically adjust according to environmental changes, such as temperature variations and load fluctuations, allowing carbon emission control strategies to adapt to different operating conditions. For example, during the high temperatures of summer, air conditioning equipment experiences higher loads, leading to increased carbon emissions. In this case, the adaptive threshold might be adjusted to a higher level to accommodate the increased energy consumption; conversely, during the lower loads of winter, the threshold can be adjusted to a lower level.

[0042] Furthermore, when executing operating condition slicing based on the dynamic carbon factor modeling results, an operating condition feature vector is first constructed for each time point to uniformly represent meteorological conditions, equipment load status, and seasonal variations. Based on this, time indices are assigned to sets of different operating condition slices, ensuring that the data distribution within each subsequent slice is consistent with the operating conditions. To this end, an operating condition spatial partitioning domain is defined, and the mapping from time points to operating condition slices is completed using an indicator method, as shown in the following formula:

[0043] in, Indicates the first The set of time indices corresponding to each working condition slice; Indicates at time The operating condition feature vector is a concatenated representation of meteorological condition sub-vectors, equipment load status sub-vectors, and seasonal variation sub-vectors. Indicates the first The feature domain of each working condition is derived from clustering of historical working condition distributions or rule-based boundary sets. This mapping ensures the homogeneity of samples within a slice by adhering to the principle that "if the working condition feature vector is within the working condition domain, it is included in the corresponding slice," thereby improving the discriminability and robustness of policy learning and threshold generation on each slice.

[0044] After completing the work condition slices, a Markov decision process description of a deep reinforcement learning model is constructed separately for each work condition slice. The carbon factor time series within the slice and the work condition features at the same time point are uniformly used as the state, and adaptive threshold action candidates are used as the action set. A policy network outputs threshold suggestions, and the policy is updated by environmental feedback. Therefore, the state and actions at the slice level are defined as follows:

[0045] in, Indicates the first In each working condition slice, time point The state vector contains the current dynamic carbon factor modeling results. Recent historical fragments of carbon factors With the current operating condition feature vector ; This represents the adaptive threshold output by the deep reinforcement learning model in this state. This formulation, based on the principle that "states characterize foreseeable risks and contexts, and actions provide threshold decisions," transforms the threshold generation problem into a policy optimization problem, making it easier to obtain stable and reliable threshold policies through centralized training and distributed execution within slices.

[0046] During training, a reward function is constructed with bias suppression and out-of-bounds penalty as the core objectives. Smoothing constraints are applied to prevent unnecessary sharp fluctuations in the threshold, enabling the deep reinforcement learning model to automatically adjust the adaptive threshold within each scenario slice, thus balancing prediction accuracy, out-of-bounds risk, and policy stability. The reward function can be formalized as:

[0047] in, Indicates the first Each working condition slice at time Instant rewards; This represents the results of dynamic carbon factor modeling; Represents the reference carbon factor observation or the calibrated labeled value at the same time. and These represent the adaptive thresholds at the current and previous time steps, respectively. , , These are non-negative weighting coefficients, used to balance the prediction bias term, the out-of-bounds penalty term, and the threshold smoothing term, respectively. This design, based on the principle that "maximizing reward is equivalent to minimizing bias, minimizing out-of-bounds, and ensuring smooth threshold changes," drives the strategy to converge within a slice to a threshold trajectory that balances safety and operability.

[0048] In multi-round iterative optimization, parameter updates using a policy network or value network are employed to gradually improve the quality of threshold decisions. At each work condition slice, a soft update rule stably converges to an equilibrium point with an adaptive threshold, thus forming a threshold library that can be switched according to changes in work conditions. The iterative update of the threshold can be represented by the following form:

[0049] in, and They represent the first Wheel and the first The threshold obtained from rounds of iteration; Indicates being between and The soft update coefficient between them; Indicates parameters Threshold generation mapping for deep reinforcement learning models of representation; This represents the current state. The update uses the principle of "slowly approximating the policy output from the current threshold" to suppress training oscillations and improve generalization stability under non-stationary conditions. Ultimately, it automatically converges within each condition slice to obtain an adaptive threshold that can be directly used for online control.

[0050] S140. Based on the dynamic carbon factor modeling results and multimodal data, a multi-agent reinforcement learning framework is used to optimize the carbon emission scheduling of equipment and areas within the green park and generate optimization strategies.

[0051] Specifically, the multi-agent reinforcement learning framework is a reinforcement learning method in which multiple agents participate in the decision-making process. Each agent independently perceives the environment and adjusts its behavioral strategy based on environmental feedback. In carbon emission scheduling optimization, each device or area within a park can be considered an agent, making decisions based on its own state and surrounding environment to collectively optimize the overall carbon emissions of the park. Through this framework, each agent not only considers its own goals but can also collaborate or compete with other agents to ultimately achieve the optimal carbon emission control strategy. For example, each component of the park, such as the air conditioning system, lighting system, and production equipment, can be considered an agent, working together to optimize overall carbon emissions.

[0052] Carbon emission scheduling refers to developing reasonable operational plans and schedules based on the carbon emission needs of different equipment and areas to reduce overall carbon emissions. In the management of green industrial parks, carbon emission scheduling refers to rationally arranging the start-up and shutdown times and load distribution of equipment based on different equipment loads, environmental changes, and energy consumption demands to reduce carbon emissions during peak periods. The purpose of carbon emission scheduling is to optimize the park's carbon emission structure and ensure that all equipment can meet the park's actual needs while maximizing energy conservation and reducing emissions.

[0053] Optimization strategies are operational solutions that minimize or optimize carbon emissions by addressing the carbon emissions scheduling problem. These strategies are automatically generated based on reinforcement learning models, combined with real-time environmental data and carbon emissions predictions from the industrial park. In implementation, optimization strategies go beyond controlling individual devices; they also consider the synergistic effects between devices across the entire park. Strategies may include adjusting equipment load, start-up and shutdown timings, and operating modes to minimize carbon emissions.

[0054] Furthermore, firstly, carbon emission scheduling tasks need to be defined for each device and area within the green park. Each task should include key information such as the energy demand, load changes, carbon emission factors, and operating status of the device or area. These tasks provide the necessary environmental context for the multi-agent reinforcement learning framework, ensuring that each agent can make decisions under specific environmental conditions. For example, a device's carbon emission task might include predicting changes in the carbon emission factor based on current load and energy consumption, while an area's carbon emission task might include adjusting energy allocation among different devices to ensure overall carbon emissions are minimized. The definition of a carbon emission scheduling task can be formalized as follows:

[0055] in, Indicates the first Carbon emission scheduling tasks for individual equipment or areas; Is it equipment or area? Energy demand; Is it equipment or area? Load changes; It is a carbon emission factor, calculated based on the operating status of the equipment; Is it equipment or area? The running status.

[0056] Each agent determines its current state by sensing the environment, such as temperature, humidity, equipment load, and carbon emission factors, and makes decisions based on these states. Each agent not only considers its own carbon emission scheduling tasks but also needs to collaborate or compete with other agents to optimize the overall carbon emissions of the park. The interaction between agents drives the carbon emission scheduling within the park towards global optimization. In this way, each device or area participates in the overall carbon emission optimization process through independent learning. The state is represented as:

[0057] in, It is an intelligent agent The state vector includes energy demand, load change, carbon emission factor, operating state, and the action selected in the last operation. .

[0058] During the intensive training phase, multiple agents share global state information and collaboratively optimize based on local feedback. At this stage, all agents can learn and optimize their behavior through shared environmental information. In reinforcement learning, agents continuously interact with the environment and adjust their behavioral strategies based on the feedback received. During this process, the deep reinforcement learning model learns how to select the optimal carbon emission scheduling scheme to optimize the overall carbon emission target of the park. Each agent makes decisions based on global state information; although each agent performs tasks locally, they cooperate to maximize the overall carbon emission optimization of the park. During training, the strategies of all agents can be optimized through a shared objective function. The objective function may include multiple factors such as carbon emissions, energy consumption, and equipment load balancing.

[0059] During multiple rounds of iterative training, the agent continuously optimizes its decision-making strategy through reinforcement learning algorithms. The goal of reinforcement learning is to enable each agent to learn from environmental feedback and gradually select the optimal action through trial and error. Each agent selects the action most relevant to its current state based on dynamic carbon factor modeling results and multimodal data, thereby influencing its carbon emission scheduling. The optimal action is obtained through reinforcement learning algorithms such as Q-learning and Deep Q-Networks (DQN), which maximize the reward function and ensure that carbon emissions are minimized while meeting the scheduling objectives. The selection of the optimal action can be expressed as:

[0060] in, In time step The optimal action to choose; It is an action In state The Q-value represents the expected reward of the action; Is the intelligent agent at time step The state at that time.

[0061] Once the optimal carbon emission scheduling actions are obtained through multiple rounds of training, the agent will execute the optimization strategy in practical applications. At this stage, equipment and areas within the green park will be dynamically scheduled according to the optimization strategy to ensure overall carbon emissions are controlled within the optimal range. When executing the optimization strategy, the agent selects the most appropriate action based on its current state and environmental information, continuously optimizing carbon emissions within the park. The execution of the optimization strategy relies not only on the local scheduling tasks of each device and area but also on the coordination and cooperation between agents to ensure optimal overall carbon emission management for the park. At this stage, the agent's behavior may be affected by real-time environmental changes within the park, such as seasonal variations and fluctuations in energy demand. Through the execution of the optimization strategy, the park's carbon emissions will be dynamically optimized to meet sustainable development goals.

[0062] By employing a multi-agent reinforcement learning framework, each device and area within the green park is treated as an independent agent. Through sharing global state information, optimizing strategies based on local feedback, iterative training to select the optimal action, and finally executing the optimization strategy, dynamic optimization of carbon emission scheduling is ensured. This process achieves fine-grained control and intelligent optimization of carbon emission scheduling tasks through reinforcement learning.

[0063] S150 uses a drift detection algorithm to monitor the long-term offset between the dynamic carbon factor modeling results and the adaptive threshold in real time, and triggers an automatic calibration mechanism.

[0064] Specifically, drift detection algorithms are used to monitor and identify deviations or anomalies in data trends in real time. In carbon emission management systems, the goal of drift detection is to promptly detect long-term deviations between predicted and actual carbon factor values, ensuring the system can automatically identify and adapt to these changes. For example, when the difference between the modeled carbon factor results and actual carbon emission values ​​exceeds a preset tolerance range, the drift detection algorithm will trigger an alarm, indicating that the model may need adjustment or recalibration. For instance, suppose a carbon emission prediction model outputs expected carbon dioxide concentration values ​​for a certain time period based on sensor data. If the actual measured carbon dioxide concentration deviates continuously from the predicted value, the drift detection algorithm will determine if a deviation exists and initiate the corresponding adjustment mechanism.

[0065] Long-term offset refers to a gradually increasing discrepancy between predicted and actual values ​​over a certain period. This offset occurs when data trends or system operating conditions change, and may be due to inaccurate system modeling or changes in the external environment. Long-term offset can lead to increased errors between predictions and actual values, affecting subsequent decision-making and optimization. For example, if a park's carbon emission prediction model does not take into account factors such as rising energy prices or weather changes in a timely manner, then a long-term deviation may occur between predicted and actual carbon emissions, causing carbon emission control strategies to fail.

[0066] Automatic calibration refers to the process by which a system automatically adjusts and optimizes itself when data offsets or model errors are detected. In carbon emission management, automatic calibration is used to adjust model parameters, update carbon factor modeling results, or adaptive thresholds to reduce the impact of long-term offsets on decision-making. The calibration process is triggered by a drift detection algorithm and dynamically adjusts based on current carbon emission data and environmental feedback to ensure the model remains accurate at all times.

[0067] Furthermore, the deviation between the dynamic carbon factor modeling results and the adaptive threshold is calculated and monitored in real time: First, the system needs to calculate the deviation between the dynamic carbon factor modeling results and the adaptive threshold. The deviation represents the difference between the predicted value and the actual emission value of the current carbon factor. This deviation calculation is the core of the entire drift detection process and is used to determine whether the carbon factor has experienced a long-term shift. The formula for calculating the deviation can be expressed as:

[0068] in, Indicates at time deviation, This is the result of dynamic carbon factor modeling. It is an adaptive threshold.

[0069] During real-time monitoring, the system continuously updates the deviation and analyzes it using a drift detection algorithm. If the deviation exceeds a set tolerance range, such as a preset threshold, the system proceeds to the next step. The drift detection algorithm calculates the deviation value at each moment, continuously monitoring whether any offset exceeds the set threshold to ensure timely detection of any abnormal changes.

[0070] When the deviation exceeds a preset threshold, the system will immediately trigger an alarm signal. The purpose of the alarm signal is to notify system administrators or automatic control mechanisms that the carbon factor has deviated from the control range of the adaptive threshold. This process can be triggered by setting a dynamic threshold, which is triggered when the currently calculated deviation... If the deviation exceeds the set allowable range, such as the maximum tolerable deviation, the system will automatically generate an alarm signal. This alarm signal will help the administrator or the system automatically adjust parameters to prevent large errors in carbon emission control. For example, the alarm trigger condition may be set to the deviation exceeding a certain tolerance threshold. ,Right now: If deviation If the value continuously exceeds this threshold for a certain period of time, an alarm will be triggered, indicating that the change in carbon factor has exceeded the normal control range and measures need to be taken.

[0071] An automatic calibration mechanism is triggered if a continuous alarm signal persists for more than a preset duration: If the alarm signal persists for a set time and the deviation remains greater than a threshold, the automatic calibration mechanism is activated. The purpose of this mechanism is to adjust the parameters or model weights used in the carbon factor modeling process so that the modeling results can be restored to the expected control range. Specifically, the automatic calibration mechanism automatically analyzes potential errors in the carbon factor modeling process, adjusts the model parameters, and ensures more accurate prediction results, thereby bringing the dynamic carbon factor modeling results back to the control range of the adaptive threshold.

[0072] Automatic calibration mechanisms involve adjusting model parameters, such as weights in an LSTM model, or control strategies. By analyzing real-time data, the system updates the parameters according to the calibration algorithm, correcting the model's predictions. For example, it may be necessary to retrain the LSTM model or modify its internal weights and bias terms using some adaptive adjustment method to ensure that the carbon factor predictions more closely match the actual data. The automatic calibration process can be represented as follows:

[0073] in, It is the result of dynamic carbon factor modeling after automatic calibration; This is the original dynamic carbon factor modeling result; It is the calculated deviation; These are model parameters, such as the weights and thresholds of the LSTM network, which will be adjusted during the calibration process.

[0074] By combining drift detection algorithms, alarm mechanisms, and automatic calibration mechanisms, the entire system can promptly adjust and optimize when anomalies occur in carbon factor prediction. The drift detection algorithm first monitors the deviation between the carbon factor modeling results and the adaptive threshold in real time, and triggers an alarm when the deviation exceeds a preset threshold. If the alarm persists for more than a preset duration, the automatic calibration mechanism will adjust the model, thereby ensuring the accuracy and stability of the carbon emission management system.

[0075] S160. Based on the optimization strategy and automatic calibration mechanism, the carbon emissions of the green park are optimized and calibrated in real time.

[0076] Specifically, real-time optimization and calibration refer to the process of immediately adjusting and optimizing data upon detecting changes or deviations. In carbon emission management, real-time optimization involves dynamically adjusting carbon emission control strategies based on real-time data from the park, such as energy use, equipment load, and weather conditions, to ensure the minimization of carbon emissions. Real-time calibration, on the other hand, involves promptly adjusting the parameters of the model or strategy when a deviation is found between the carbon factor modeling results and actual carbon emission data, ensuring accurate carbon emission predictions. For example, in a green park, if weather conditions change, such as rising temperatures leading to increased air conditioning system load and increased carbon emissions, the system can optimize the carbon emission scheduling scheme in real time, adjusting the air conditioning load allocation. Simultaneously, if there is a difference between the model's predicted values ​​and actual carbon emission data, the system will trigger an automatic calibration mechanism to adjust the parameters of the carbon factor prediction model, making future carbon emission predictions more accurate.

[0077] Furthermore, firstly, energy dispatching and carbon emission control in the green park are managed in real time through optimization strategies. These strategies consider not only the carbon emissions of individual devices but also the synergistic effects between multiple devices and areas within the park. For example, when multiple air conditioning units operate simultaneously, the optimization strategy might adjust the operating load of each unit to ensure effective control of overall carbon emissions while meeting comfort requirements. The optimization strategy uses reinforcement learning algorithms to dynamically adjust based on current equipment load, environmental changes, and energy consumption to determine the optimal operating mode. The ultimate goal is to minimize carbon emissions within the park and maximize energy efficiency.

[0078] During this process, the optimization strategy can adjust the on / off status, load configuration, and other operating parameters of equipment in real time to achieve coordinated operation between different devices and areas, thereby optimizing the overall carbon emissions of the park. For example, if an air conditioner in the park has an excessive load, the optimization strategy may balance carbon emissions by reducing the load of that air conditioner or activating other low-load devices. The optimization strategy controls emissions in the following ways:

[0079] in, Indicates time Carbon emission scheduling and control strategies; It is the first Each device at time Energy demand; It is the first Each device at time Load changes.

[0080] The automatic calibration mechanism retrains or adjusts the parameters of the dynamic carbon factor model and adaptive threshold: After implementing the optimization strategy, the system will adjust the dynamic carbon factor model and adaptive threshold according to the actual operation of the park through the automatic calibration mechanism. Specifically, the automatic calibration mechanism will retrain or adjust the parameters of the carbon factor model based on real-time carbon emission data and environmental changes. If the difference between the carbon factor predicted by the model and the actual carbon emission value is large, the system will adjust the model parameters or update the adaptive threshold to make the carbon factor modeling results more accurate. For example, when the automatic calibration mechanism finds that the model prediction deviates significantly from the actual carbon emission value, it can fine-tune the weights of the LSTM network through the optimization algorithm to ensure that the model can adapt to the current carbon emission situation in the park. This process adjusts the parameters of the carbon factor modeling through feedback to ensure that the carbon emission prediction is always consistent with the actual situation.

[0081] During automatic calibration, the system periodically analyzes the feedback after implementing the optimization strategy to ensure the stability and long-term effectiveness of the optimization results. Optimization feedback refers to the changes in carbon emissions within the park after implementing the optimization strategy and its impact on energy dispatch. Based on this feedback, the system dynamically adjusts the strategy weights in the deep reinforcement learning model. Adjusting the strategy weights helps the model better cope with environmental changes and equipment load fluctuations, ensuring the optimization strategy remains effective under changing park conditions. By analyzing carbon emission changes and optimization effects in each cycle, the system can optimize the deep reinforcement learning model's strategy, making each adjustment more accurately adapt to the park's energy management needs. For example, if energy consumption increases and carbon emissions are not effectively controlled during a certain period, the system will adjust the reward function or strategy weights in the reinforcement learning model, enabling the model to make more appropriate dispatch decisions in the next cycle.

[0082] By combining optimization strategies, automatic calibration mechanisms, and reinforcement learning models, the system enables real-time optimization and calibration of carbon emissions within green industrial parks. The optimization strategy dynamically adjusts energy dispatch based on the carbon emissions of equipment and the region, while the automatic calibration mechanism periodically analyzes optimization feedback and real-time data to adjust model parameters and strategy weights, ensuring that the park's carbon emissions remain at an optimal level. Through multiple iterations and adjustments, the system can adapt to changes both inside and outside the park, providing an efficient and flexible carbon emission control solution.

[0083] This application also provides a dynamic carbon factor modeling and adaptive threshold multi-system calibration device, referring to... Figure 2 , Figure 2This is a schematic diagram of a multi-mode calibration device for dynamic carbon factor modeling and adaptive thresholding provided in an embodiment of this application. The device is a server, comprising an acquisition module 21 and a processing module 22. The acquisition module 21 acquires multi-modal data on carbon emissions collected by multiple sensors within a green park. The processing module 22 dynamically models the carbon factor using an LSTM model based on the multi-modal data, extracts nonlinear features from the multi-modal data, and generates dynamic carbon factor modeling results. The processing module 22 also slices the dynamic carbon factor modeling results according to different operating conditions within the green park and generates adaptive thresholds for each slice using a deep reinforcement learning model. Furthermore, based on the dynamic carbon factor modeling results and multi-modal data, the processing module 22 optimizes the carbon emission scheduling of equipment and areas within the green park using a multi-agent reinforcement learning framework, generating optimization strategies. The processing module 22 also monitors the long-term offset between the dynamic carbon factor modeling results and the adaptive thresholds in real time using a drift detection algorithm and triggers an automatic calibration mechanism. Finally, the processing module 22 performs real-time optimization and calibration of the carbon emissions in the green park according to the optimization strategies and the automatic calibration mechanism.

[0084] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0085] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0086] The communication bus 32 is used to enable communication between these components.

[0087] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0088] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0089] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0090] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a dynamic carbon factor modeling and adaptive threshold multi-system calibration method.

[0091] exist Figure 3In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call the application program stored in the memory 35, which is a dynamic carbon factor modeling and adaptive threshold multi-system calibration method. When executed by one or more processors, the electronic device performs one or more methods as described in the above embodiments.

[0092] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0093] This application also provides a computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0094] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0095] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0096] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0097] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0098] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this 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. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0099] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A dynamic carbon factor modeling and adaptive threshold multi-system calibration method, characterized in that, The method includes: Acquire multimodal data on carbon emissions collected by multiple sensors within the green park; Based on the multimodal data, an LSTM model is used to dynamically model the carbon factor, and nonlinear features in the multimodal data are extracted to generate dynamic carbon factor modeling results. The dynamic carbon factor modeling results are sliced ​​according to different working conditions in the green park, and an adaptive threshold is generated for each slice through a deep reinforcement learning model. Based on the dynamic carbon factor modeling results and the multimodal data, a multi-agent reinforcement learning framework is used to optimize the carbon emission scheduling of equipment and areas within the green park and generate optimization strategies. The drift detection algorithm monitors the long-term offset between the dynamic carbon factor modeling result and the adaptive threshold in real time and triggers an automatic calibration mechanism. The carbon emissions of the green park are optimized and calibrated in real time according to the optimization strategy and the automatic calibration mechanism.

2. The dynamic carbon factor modeling and adaptive threshold multi-system calibration method according to claim 1, characterized in that, The acquisition of multimodal data on carbon emissions collected by multiple sensors within the green park specifically includes: Multiple different types of sensors are deployed in the green park to collect real-time data on carbon emissions. These sensors include gas sensors, temperature and humidity sensors, power meters, weather monitoring sensors, and light sensors. The carbon emission data is timestamped using a unified timestamp, and the carbon emission data is integrated using a data fusion algorithm to obtain the multimodal data. The data fusion algorithm includes Kalman filtering and weighted averaging.

3. The dynamic carbon factor modeling and adaptive threshold multi-system calibration method according to claim 1, characterized in that, The step of dynamically modeling the carbon factor using an LSTM model based on the multimodal data and extracting nonlinear features from the multimodal data to generate dynamic carbon factor modeling results specifically includes: The LSTM model is used to process the multimodal data to extract the temporal and nonlinear features from the multimodal data. Based on the feedforward and backpropagation process of the LSTM model, the nonlinear relationship between the carbon factor and the temporal features and the nonlinear features is fitted through multiple network layers and nonlinear activation functions, and the dynamic carbon factor modeling result is generated.

4. The dynamic carbon factor modeling and adaptive threshold multi-system calibration method according to claim 1, characterized in that, The step of slicing the dynamic carbon factor modeling results according to different working conditions within the green park, and generating adaptive thresholds for each slice using a deep reinforcement learning model, specifically includes: Based on the dynamic carbon factor modeling results, the carbon factor data is sliced ​​according to different operating conditions, including meteorological conditions, equipment load status and seasonal changes. Each slice represents the carbon factor change under a specific operating condition. After slicing is completed, the deep reinforcement learning model is trained for each slice. The input of the deep reinforcement learning model is the carbon factor data of the slice. The model generates the corresponding adaptive threshold based on the feedback information through interaction with the environment. During training, the deep reinforcement learning model is controlled to adjust the adaptive threshold according to the carbon factor fluctuation of each slice based on the set reward mechanism. The reward mechanism includes evaluating the deviation between the predicted and actual carbon factor values ​​and rewarding or penalizing according to the magnitude of the deviation. Through a multi-round iterative optimization process of the deep reinforcement learning model, the adaptive threshold of each slice is automatically updated.

5. The dynamic carbon factor modeling and adaptive threshold multi-system calibration method according to claim 1, characterized in that, Based on the dynamic carbon factor modeling results and the multimodal data, a multi-agent reinforcement learning framework is used to optimize the carbon emission scheduling of equipment and areas within the green park, generating optimization strategies, specifically including: Within the green park, a carbon emission scheduling task is defined for each device and area. The carbon emission scheduling task includes the energy demand, load changes, carbon emission factors, and operating status of the device and area. Through the multi-agent reinforcement learning framework, each device and area within the green park is identified as an independent agent; During the intensive training phase, based on the carbon emission scheduling task, multiple agents are controlled to share global state information and perform collaborative optimization based on local feedback information to obtain optimized actions. Through multiple rounds of iterative training, the agent is controlled to select the optimal action from the optimized actions based on the dynamic carbon factor modeling results and the multimodal data, so as to determine the optimization strategy based on the optimal action, wherein the optimal action is obtained by optimization through a reinforcement learning algorithm; By implementing the optimization strategy, the carbon emission scheduling of equipment and areas within the green park is dynamically optimized.

6. The dynamic carbon factor modeling and adaptive threshold multi-system calibration method according to claim 1, characterized in that, The method of monitoring the long-term shift between the dynamic carbon factor modeling result and the adaptive threshold in real time through a drift detection algorithm and triggering an automatic calibration mechanism specifically includes: The deviation between the dynamic carbon factor modeling result and the adaptive threshold is calculated, and the drift detection algorithm is used to monitor the deviation in real time. If the deviation is determined to exceed a preset threshold, an alarm signal is triggered, indicating that the carbon factor has deviated from the control range of the adaptive threshold. If it is determined that the duration of the alarm signal exceeds the preset duration, the automatic calibration mechanism is triggered. The automatic calibration mechanism recalibrates the dynamic carbon factor modeling result by adjusting the parameters or model weights in the carbon factor modeling process, so that the dynamic carbon factor modeling result returns to the control range of the adaptive threshold.

7. The dynamic carbon factor modeling and adaptive threshold multi-system calibration method according to claim 1, characterized in that, The process of optimizing and calibrating the carbon emissions of the green park in real time according to the optimization strategy and the automatic calibration mechanism specifically includes: The optimization strategy is used to control the energy scheduling and carbon emissions of each device and area in the green park in real time. The optimization strategy is used to optimize the carbon emissions of a single device and also to optimize the synergistic effect between multiple devices and areas. The automatic calibration mechanism is used to retrain or adjust the parameters of dynamic carbon factor modeling and adaptive threshold. The automatic calibration mechanism is used to periodically analyze the optimization feedback after the optimization strategy is implemented, and dynamically adjust the strategy weights in the deep reinforcement learning model to maintain the optimization and control of carbon emissions in the green park.

8. A dynamic carbon factor modeling and adaptive threshold multi-system calibration device, characterized in that, The apparatus is used to perform the method as described in any one of claims 1 to 7, the apparatus comprising an acquisition module and a processing module, wherein... The acquisition module is used to acquire multimodal data on carbon emissions collected by multiple sensors within the green park. The processing module is used to dynamically model the carbon factor using an LSTM model based on the multimodal data, extract the nonlinear features in the multimodal data, and generate dynamic carbon factor modeling results. The processing module is also used to slice the dynamic carbon factor modeling results according to different working conditions in the green park, and generate an adaptive threshold for each slice through a deep reinforcement learning model. The processing module is also used to optimize the carbon emission scheduling of equipment and areas in the green park based on the dynamic carbon factor modeling results and the multimodal data, using a multi-agent reinforcement learning framework, and generate optimization strategies. The processing module is also used to monitor the long-term offset between the dynamic carbon factor modeling result and the adaptive threshold in real time through a drift detection algorithm, and trigger an automatic calibration mechanism. The processing module is also used to optimize and calibrate the carbon emissions of the green park in real time according to the optimization strategy and the automatic calibration mechanism.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images