Park-level energy and carbon accounting cooperative scheduling method and system

Abstract

The invention discloses a park-level energy and carbon accounting collaborative scheduling method and system, and the method and system achieve the efficient integration and collaborative utilization of data through breaking the barrier between energy data and IT data, improve the green power consumption rate, effectively reduce the carbon emission, and remarkably improve the overall operation efficiency and economical efficiency of a park energy system. According to the method, energy and IT data are collected and processed in real time, a basis is provided for accurate prediction and optimal scheduling, an LSTM model and an XGBoost model are adopted for prediction, an NSGA-II multi-target optimization algorithm is combined, and two-channel optimal control over green power consumption and carbon emission is achieved under the condition that energy storage and server operation constraints are met; meanwhile, scheduling instructions are continuously optimized through a feedback mechanism, and continuous stability and high efficiency of operation of the whole system are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, in particular to a park-level energy and carbon accounting collaborative scheduling method and system. BACKGROUND

[0002] With the in-depth implementation of the "double carbon" strategy, the park-level energy system, especially the "source network load storage" integrated system containing photovoltaic, energy storage, heat pump and other units, is becoming a key scene for energy saving and emission reduction. At the same time, in order to support scientific research, data processing and other businesses, the scale of IT computing load such as high-performance computing clusters, cloud servers and other IT in the park is continuously growing, and its energy consumption and carbon emissions have become an important part of the total energy consumption of the park.

[0003] At present, in most parks, there are the following problems, 1. Data island problem, lack of collaborative decision basis, the data of IT system and energy system are not interconnected, and the basis for collaborative decision cannot be formed; 2. Green electricity consumption and IT load mismatch, photovoltaic and other renewable energy output has volatility, while IT load has periodical demand, and the energy system cannot predict the demand of IT load in traditional scheduling, and the IT system also does not preferentially use green electricity, resulting in a low level of green electricity consumption; 3. Carbon emission reduction target is not integrated into the scheduling logic, the existing scheduling only takes "lowest energy cost" or "fastest IT task response" as a single target, without combining the carbon emission factor of IT equipment and the carbon footprint of energy system, resulting in suboptimal overall carbon emission reduction effect of the park.

[0004] In the prior art, such as "a park energy scheduling system" with publication number CN115273456A only focuses on the cooperation of photovoltaic, energy storage and load, and does not involve IT system; "IT resource intelligent scheduling method" with publication number CN114861238A only optimizes server resources, and does not associate the carbon properties of energy supply; therefore, there is an urgent need for a collaborative scheduling scheme that integrates energy data and IT data and takes "maximum green electricity consumption + optimal carbon emission reduction" as the target. SUMMARY

[0005] In view of the above-mentioned technical problems that the data of park energy system and IT system are not interconnected, the basis for collaborative decision cannot be formed, and the green electricity consumption rate is low due to the inability to accurately predict the demand of IT load, the present application provides a park-level energy and carbon accounting collaborative scheduling method.

[0006] To achieve the above-mentioned purpose, the present application provides a park-level energy and carbon accounting collaborative scheduling method, comprising: S1: real-time collection of energy data and IT data, and storage of the energy data and IT data in a time series database after cleaning and standardization operation; S2: output green electricity output prediction value and carbon emission estimation value respectively based on the processed data in the time sequence database; S3: set green electricity consumption rate maximization and carbon emission minimization as core indexes, combine the green electricity output prediction value and the carbon emission estimation value, and generate a scheduling instruction set through a multi-objective optimization algorithm; S4: control energy equipment and IT resources according to the scheduling instruction, and collect actual green electricity consumption rate and carbon emission after execution by a feedback module, and if the deviation from the prediction is greater than 10%, return to S2 for re-optimization.

[0007] As an improved scheme of the present application, the method for obtaining the green electricity output prediction value comprises: adopting a long short-term memory network (LSTM) model, inputting historical green electricity output data and meteorological data, and outputting a green electricity generation power prediction value. The method for obtaining the carbon emission estimation value comprises: adopting a carbon accounting model based on a national unified emission factor, inputting energy scheduling instructions and historical carbon emission data, and outputting a carbon emission estimation value.

[0008] As an improved scheme of the present application, in S3, the multi-objective optimization algorithm is a non-dominated sorting genetic algorithm II (NSGA-II).

[0009] As an improved scheme of the present application, while generating the collaborative scheduling instruction set, the estimated carbon emission corresponding to the scheduling instruction set is calculated in real time based on the national unified emission factor, and the estimated carbon emission is taken as a feedback signal to participate in the solving process of the optimization target.

[0010] As an improved scheme of the present application, in S1, the sampling frequency of the energy data is 1 min / time, and the sampling frequency of the carbon accounting data is 5 s / time, and the energy data and the carbon accounting data are transmitted through an MQTT protocol.

[0011] The present application also provides a park-level energy and carbon accounting collaborative scheduling system, comprising: An energy data acquisition module is configured to acquire energy data of photovoltaic, energy storage and heat pump. A carbon accounting data acquisition module is configured to acquire carbon emission factor data, carbon asset data and carbon quota data. A data preprocessing module is configured to perform AI-enhanced cleaning and intelligent standardization on the energy data and carbon accounting data. A prediction sub-module is configured to predict green electricity output prediction value and carbon emission estimation value. A carbon accounting sub-module is configured to calculate carbon emission of a scheduling scheme in real time based on a national unified emission factor, and input the result into the multi-objective decision sub-module. A multi-objective decision sub-module is configured to maximize green power consumption rate and minimize carbon emission based on the predicted curve to generate a set of coordinated scheduling instructions; An energy execution module is configured to control the operation state of the energy equipment according to the set of coordinated scheduling instructions; An IT execution module is configured to schedule the execution of the IT task and the allocation of the server resource according to the scheduling instruction; A feedback module is configured to collect the executed data and feed back to the prediction sub-module and the multi-objective decision sub-module.

[0012] As an improved scheme of the present application, the prediction sub-module inputs historical green power output data and meteorological data into a pre-trained LSTM model and a carbon accounting model to output green power generation power prediction values, and inputs energy scheduling instructions and historical carbon emission data into a carbon accounting model based on a national unified emission factor to output carbon emission estimation values.

[0013] As an improved scheme of the present application, the multi-objective decision sub-module adopts an NSGA-II algorithm to solve a multi-objective optimization problem under the constraints of energy storage SOC being greater than or equal to 20% and server CPU utilization being less than or equal to 85%.

[0014] As an improved scheme of the present application, the carbon accounting data collection module is connected to a network platform through an API interface to obtain carbon emission data and carbon quota information.

[0015] As an improved scheme of the present application, the carbon accounting sub-module converts energy consumption corresponding to the scheduling instruction into carbon emission based on an open-source carbon accounting tool and supports CCER development.

[0016] The present application has the following advantages: compared with the prior art, the AI-driven park-level energy and carbon accounting coordinated scheduling method and system provided by the present application breaks the barriers between energy data and carbon accounting data, realizes efficient integration and collaborative use of data, not only improves the green power consumption rate and effectively reduces the carbon emission, but also significantly improves the overall operation efficiency and economy of the park energy system; the present application collects and processes energy data and carbon accounting data in real time to provide a basis for accurate prediction and optimized scheduling, and uses an LSTM model and a carbon accounting model for prediction, combined with an AI-enhanced NSGA-II multi-objective optimization algorithm, to realize double-channel optimization control of green power consumption and carbon emission under the condition of meeting the energy storage and server operation constraints; at the same time, the carbon accounting sub-module calculates the carbon emission in real time to ensure the continuous stability and efficiency of the overall system operation and provides a data basis for park carbon asset development. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1A method step diagram for the present application; Figure 2 A module interaction diagram for the present application. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the application, the application will be further described below with reference to the accompanying drawings.

[0019] In the following description, example specific details are given to provide a more thorough understanding of the application. Obviously, the described embodiments are only a part of the embodiments of the application, and are not all the embodiments. It should be understood that the specific embodiments are only used to explain the application, and are not used to limit the application.

[0020] It should be understood that when the terms "include" and / or "contain" are used in the specification, it means that the features, integers, steps, operations, elements or components exist, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components or combinations thereof.

[0021] Please refer to Figures 1-2 The park-level energy and carbon accounting collaborative scheduling method of the present application comprises: S1: Real-time acquisition of energy data and IT data, and storage of the energy data and IT data in a time series database after cleaning and standardization operation; S2: Based on the processed data in the time series database, output the green power output prediction value and the carbon emission estimation value of the preset period respectively; S3: Set the maximum green power consumption rate and the minimum carbon emission as the core indexes, combine the green power output prediction value and the carbon emission estimation value, and generate a set of scheduling instructions through a multi-objective optimization algorithm; S4: Control the energy equipment and IT resources according to the scheduling instructions, and collect the actual green power consumption rate and carbon emission after execution by the feedback module, if the deviation from the prediction is > 10%, return to S2 for re-optimization.

[0022] The application mainly adopts energy data and IT end data to analyze, realizes the collaborative optimization of green electricity consumption and carbon emission by constructing a multi-objective optimization model, and specifically, the application first stores the energy data and IT data collected in real time to a time sequence database after cleaning and standardization, then extracts data from the time sequence database, and takes the maximum green electricity consumption rate and the minimum carbon emission as core indexes, and predicts the green electricity output and IT load with high precision through a multi-objective optimization algorithm to provide a data basis for subsequent power scheduling, the green electricity output refers to the actual power generation capacity of the green energy power generation equipment in the park within a specific time period, which is affected by multiple factors such as weather conditions, equipment state and operation strategy, and also directly related to the improvement of green electricity consumption rate and the reduction of carbon emission; in the decision-making stage, the power of the energy system is accurately mobilized according to the scheduling instruction to be distributed to each device in the IT system, so that the contradiction between the green electricity consumption rate and the carbon emission can be effectively balanced, and the economy and environmental protection of the scheduling scheme are ensured; at the same time, the system dynamically evaluates the scheduling effect through a real-time feedback mechanism, and when the deviation between the actual execution result and the predicted value exceeds the threshold value, the re-optimization process is automatically triggered, thereby ensuring the adaptability and robustness of the scheduling strategy.

[0023] Among them, the cleaning method of the energy data and the carbon accounting data includes: An abnormality detection model based on LSTM is adopted to intelligently identify and process abnormal data of the energy data, and for the identified abnormal data points, the weighted average value of adjacent normal data points is used for interpolation repair to ensure data continuity; for the missing values in the carbon accounting data, the historical data and industry benchmark values are combined for dynamic filling, and the rationality of the filled values is verified through chi-square test.

[0024] A time series missing value filling method based on Transformer is adopted to realize accurate data recovery, the long-term dependence relationship in the time series data is captured through the self-attention mechanism, the continuous missing segment in the carbon accounting load data caused by equipment failure or communication interruption is modeled and filled as a whole, and the local deviation introduced by the traditional interpolation method is avoided; at the same time, combined with the sliding window statistical test, the stationarity of the filled data is ensured to meet the input requirements of the subsequent prediction model; A deep clustering algorithm is used to process repeated data, automatically identify and merge repeated data points with the same characteristics, define the similarity measurement standard between data points, group highly similar data into the same cluster, and only keep the cluster center point as the representative value to effectively reduce data redundancy; at the same time, combined with the time stamp verification mechanism, the time continuity of the original data is ensured to improve the input quality of the subsequent prediction model.

[0025] The cleaning and standardization method of the energy data and the carbon accounting data includes: An adaptive standardization method is adopted, automatically selecting a standardization strategy based on the data distribution characteristics. For normally distributed data, the Z-score standardization method is used, which calculates the mean and standard deviation of the data to transform it into a standard normal distribution with a mean of 0 and a standard deviation of 1, thereby eliminating the influence of dimensions and improving the model convergence speed. For data with significant skewness or long-tailed distribution, the Box-Cox transformation is applied for nonlinear standardization. The normalization adjustment of the data distribution is achieved through the optimized selection of parameter λ, ensuring that the data meets the statistical assumptions of the subsequent analysis model. At the same time, for multi-source heterogeneous data, the Min-Max standardization method is used to linearly scale indicators of different dimensions such as IT load power and energy consumption, uniformly mapping the data to the [0,1] interval, maintaining the relative proportion of data while improving the stability of model training. In addition, a dynamic threshold mechanism is introduced to perform secondary verification on the standardized data. When the data deviates from the historical mean by more than 3 times the standard deviation, an anomaly alarm is triggered. Combined with manual review, this ensures that the data quality meets the input accuracy requirements of the collaborative scheduling model.

[0026] For carbon accounting data, feature importance analysis based on graph neural networks (GNNs) is used to automatically determine the relative weights of each carbon emission factor. By constructing a carbon emission factor correlation graph, the message passing mechanism of the graph neural network is used to capture the complex interaction relationships between factors and identify key factors that have a significant impact on overall carbon emissions. For example, in IT load scenarios, the differentiated contributions of factors such as server model and task type to carbon emissions can be distinguished. At the same time, combined with the SHAP value interpretation method, the contribution of each factor in a specific data sample is quantified to ensure the transparency and interpretability of weight allocation. This allows for dynamic adjustments in the standardization strategy, such as using finer quantification granularity for high-weight factors and appropriately aggregating low-weight factors to reduce data dimensionality. Furthermore, to address the multi-timescale characteristics in carbon accounting data (such as the coupling between hourly equipment energy consumption and daily task planning), a hierarchical standardization strategy is adopted. First, low-frequency data (such as daily task volume) is linearly interpolated to a high-frequency time granularity, and then weighted and fused with the original high-frequency data (such as hourly energy consumption). By balancing the influence of different time scales through dynamic weight coefficients, the standardized data can reflect both instantaneous carbon emission characteristics and the cumulative effect of long-term task planning, ultimately providing multi-scale, highly consistent input data for the collaborative scheduling model.

[0027] A dimension conversion method based on deep learning is adopted to automatically convert carbon quota data from different sources into a unified dimension. A multi-layer perception (MLP) model is constructed, the numerical and unit information of the original carbon quota data are input, and the standardized dimensionless numerical value is output while preserving the proportional relationship of the original data. During model training, a dimension constraint loss function is used to ensure that the converted data is unified in dimension while not destroying its inherent numerical logic. For example, after converting the carbon quota (tons of CO2 equivalent / kilowatt-hour) of different regions into a standard unit, the linear correlation with the original energy consumption data is still maintained, thereby improving the compatibility of multi-region carbon accounting data in the collaborative scheduling model. In addition, for the time dynamic characteristics of carbon accounting data, a dynamic standardization window mechanism is adopted to automatically adjust the standardization parameters according to the data time span. For example, sliding window Z-score standardization is used for short-time high-frequency data (such as minute-level device emissions), and global Box-Cox transformation is used for long-time low-frequency data (such as monthly task plans), ensuring that the data of different time scales can maintain their original time series characteristics after standardization, and providing more accurate data support for the collaborative optimization of green electricity consumption and carbon emissions.

[0028] In this embodiment, the method for obtaining the green electricity output prediction value includes: inputting historical green electricity output data and meteorological data into a long short-term memory network (LSTM) model, and outputting a green electricity generation power prediction value. The LSTM model effectively deals with the complexity of green electricity output affected by multiple factors such as weather conditions and equipment states by capturing the time dependence in the historical data. The model first performs feature engineering processing on the input historical green electricity output data and meteorological data, extracts key features such as temperature, wind speed, and light intensity that are closely related to green electricity output, and then uses the memory characteristics of the LSTM unit. The model can learn the long-term and short-term dependence patterns in the data, thereby more accurately predicting future green electricity generation power.

[0029] The method for obtaining the carbon emission estimation value comprises: using a carbon accounting model based on a national unified emission factor, inputting energy scheduling instructions and historical carbon emission data, and outputting the carbon emission estimation value; in actual application, the carbon accounting model combines the emission factor data published by the state to ensure the accuracy and compliance of the carbon emission estimation; further, by dynamically correlating the energy scheduling instructions, the carbon accounting model can reflect the potential influence of different scheduling strategies on carbon emission in real time, providing a quantitative basis for optimized scheduling; at the same time, the introduction of historical carbon emission data enhances the prediction stability of the model, and is especially suitable for the scenario of a park energy system with greater volatility. Further, to improve the prediction accuracy, the model uses a sliding time window mechanism to dynamically weight the historical data by week / month / quarter, highlighting the indicative effect of recent data on future emission trends; for IT load-related indirect carbon emissions, the model analyzes the proportion of computationally intensive tasks in the task plan data, combined with the unit power consumption carbon emission coefficient of the server model, to achieve fine estimation in different scenarios and time periods; for example, during the night low valley period, the model will automatically reduce the carbon emission weight of non-critical tasks, while during the daytime peak period, it will strengthen the carbon emission contribution of real-time load; an abnormal emission identification module is also set in the carbon accounting model, when the estimation value deviates from the historical mean value by more than 2 times the standard deviation, triggering a secondary verification process, combining device-level energy consumption monitoring data to verify the applicability of the emission factor, to ensure that the estimation result always conforms to the actual operation characteristics of the park.

[0030] In the present embodiment, in S3, the multi-objective optimization algorithm is a non-dominated sorting genetic algorithm NSGA-II; The NSGA-II algorithm efficiently approximates the Pareto front while maintaining the diversity of the solution set through fast non-dominated sorting and congestion calculation; in specific implementation, the algorithm maximizes green power consumption and minimizes carbon emissions as dual objectives, under the constraint conditions that the energy storage SOC is not less than 20% and the server CPU utilization is not higher than 85%, dynamically generates energy scheduling instructions including photovoltaic inverter output, energy storage charging and discharging power, and heat pump operating state, as well as IT resource scheduling instructions for server cluster task allocation and node start-stop; through the elite reservation strategy and cross variation operation, it is ensured that each generation solution set is better than the last generation, and the finally output scheduling instruction set can simultaneously optimize energy utilization efficiency and carbon emission reduction effect.

[0031] In this embodiment, while generating the coordinated scheduling instruction set, the estimated carbon emissions corresponding to the scheduling instruction set are calculated in real time based on the national unified emission factor, and the estimated carbon emissions are taken as a feedback signal to participate in the solving process of the optimization target. Specifically, the carbon accounting sub-module accesses an open-source carbon accounting tool, calculates the estimated carbon emissions under the current scheduling scheme in real time according to the photovoltaic power generation amount, energy storage charging and discharging power and heat pump energy consumption data in the energy scheduling instruction, and combines the national unified emission factor. The value is input into the multi-objective decision-making sub-module through the feedback interface as an important basis for dynamically adjusting the green power consumption rate and the carbon emission weight. When the estimated carbon emissions exceed the preset threshold, the NSGA-II algorithm automatically increases the priority of the carbon emission minimization target, reduces the proportion of fossil energy power generation or optimizes the IT task allocation to reduce the overall emissions, and ensures that the scheduling strategy evolves in the optimal direction of carbon emission reduction while meeting the real-time operation constraints through the closed-loop feedback mechanism, and maintains the steady operation of the entire energy and IT coordinated scheduling system.

[0032] In this embodiment, in S1, the sampling frequency of the energy data is 1 min / time, and the sampling frequency of the IT data is 5 s / time.

[0033] The application also provides a park-level energy and carbon accounting coordinated scheduling system, comprising: An energy data acquisition module is configured to acquire energy data of photovoltaic, energy storage and heat pump. A carbon accounting data acquisition module is configured to acquire carbon emission factor, carbon asset data and carbon quota data. A data preprocessing module is configured to perform AI-enhanced cleaning and intelligent standardization on the energy data and carbon accounting data. A prediction sub-module is configured to predict green power output prediction values and carbon emission estimation values. A carbon accounting sub-module is configured to calculate the carbon emissions of the scheduling scheme in real time based on the national unified emission factor, and input the result into the multi-objective decision-making sub-module. A multi-objective decision-making sub-module is configured to maximize the green power consumption rate and minimize the carbon emissions as the target, and generate a coordinated scheduling instruction set based on the prediction curve. An energy execution module is configured to control the operating state of the energy equipment according to the coordinated scheduling instruction set. An IT execution module is configured to schedule the execution of IT tasks and the allocation of server resources according to the scheduling instruction. A feedback module is configured to acquire the data after execution and feed back to the prediction sub-module and the multi-objective decision-making sub-module.

[0034] In this embodiment, the prediction submodule inputs historical green power output data and meteorological data through a pre-trained LSTM model and a carbon accounting model, and outputs green power generation power prediction values and carbon emission estimates by inputting energy scheduling instructions and historical carbon emission data into a carbon accounting model based on a national unified emission factor.

[0035] In this embodiment, the multi-objective decision-making submodule uses the NSGA-II algorithm to solve the multi-objective optimization problem under the constraints of energy storage SOC>20% and server CPU utilization<85%.

[0036] In this embodiment, the carbon accounting data acquisition module is connected to the network platform through an API interface to obtain carbon emission data and carbon quota information.

[0037] In this embodiment, the carbon accounting submodule converts the energy consumption corresponding to the scheduling instructions into carbon emissions based on an open-source carbon accounting tool, and supports CCER development.

[0038] The working principle of the present application is: real-time acquisition of energy data and IT data, receiving the green power output curve and IT load demand curve output by the prediction submodule through the multi-objective decision-making submodule, immediately starting the NSGA-II algorithm for multi-objective optimization solution, the NSGA-II algorithm maximizes the green power consumption rate and minimizes the carbon emissions, under the constraint conditions that the energy storage SOC state is not less than 20% and the server CPU utilization is not higher than 85%, through non-dominated sorting and congestion calculation to generate a Pareto optimal solution set, the carbon accounting submodule synchronously accesses the open-source carbon accounting tool, converts the energy consumption corresponding to the scheduling instruction set into real-time carbon emissions based on the national unified emission factor, this value is fed back to the optimization process as a dynamic constraint condition, after optimization, the generated collaborative scheduling instruction set includes photovoltaic inverter output adjustment value, energy storage charging and discharging power, heat pump operating state and other energy control parameters, as well as server cluster task migration instructions, node start and stop signals and other IT resource allocation parameters; the energy execution module accurately issues energy control instructions to photovoltaic inverters, energy storage devices and heat pump equipment through PLC controllers, the IT execution module completes the dynamic allocation of server resources through the Kubernetes scheduler API interface; the feedback module continuously collects actual running data, when the actual green power consumption rate or carbon emissions deviates more than 10% from the predicted value, the system automatically triggers the re-optimization process, and executes steps S2 to S4 again until convergence. The cooperative processing time delay of energy data and IT data is controlled within 500ms during the entire scheduling period, ensuring the real-time and effectiveness of the scheduling instructions.

[0039] Application Example One: Campus Application Take a university park (200,000 square meters, including 100kW photovoltaic, 50kWh multi-element energy storage, 100kW reclaimed water source heat pump, 200 scientific research servers) in Shenzhen Xili University Town as an example, the implementation process of the application is described in detail: Data acquisition layer: install current voltage sensor (collect output) on photovoltaic array, install energy consumption meter (collect kWh) on heat pump, install SOC detector on energy storage system, and transmit data through Huawei IoTEdge gateway; deploy Crane carbon accounting module (interface server cluster) on IT side, collect load data through K8smetrics-server, and store data in InfluxDB after data preprocessing.

[0040] Collaborative decision-making layer: deployed on local server (CPU: Intel Xeon Gold 6330, memory 64GB) in the park, running LSTM green electricity prediction model (training data is photovoltaic output + meteorological data in the past year), XGBoost IT load prediction model (training data is IT load data in the past six months), and NSGA-II decision-making algorithm (iteration number 100 times, population size 50).

[0041] Execution control layer: energy side connects photovoltaic inverter, energy storage charge and discharge switch, and heat pump controller through PLC controller (Siemens S7-1200); IT side connects server cluster through K8sAPI to realize task scheduling.

[0042] Period 1 (10:00-14:00, peak of green electricity output): Green electricity prediction value P_photo(t)=80-100kW, IT load prediction value P_IT(t)=40-50kW; Decision instruction: 1. Energy storage system stops charging, maintaining SOC=80%; 2. IT system schedules high-priority scientific research tasks (such as AI model training), increases IT load to 70-80kW, and gives priority to green electricity; 3. Heat pump maintains low load operation (energy consumption ≤20kW) to avoid occupying green electricity.

[0043] Execution effect: green electricity consumption rate 92%, IT task completion efficiency improved by 30%.

[0044] Period 2 (20:00-23:00, IT load peak): Green electricity prediction value P_photo(t)=0kW (no photovoltaic at night), IT load prediction value P_IT(t)=90-100kW; Decision instruction: 1. Energy storage system discharges (discharge power 50kW) to supplement power supply; 2. Turn off 10 idle servers to reduce IT load to 80kW; 3. Power grid purchases 30kW of thermal power to meet IT demand combined with energy storage discharge.

[0045] Results: Carbon emissions from IT equipment were reduced by 18% compared to traditional dispatching, and the SOC of energy storage was maintained at ≥25%.

[0046] After 30 days of coordinated scheduling using the method and system of this invention, compared with traditional independent scheduling: Table 1 Indicator Traditional independent scheduling Inventive coordinated scheduling Lift / lower amplitude Green power accommodation rate 58% 91% +33% Park carbon emissions 32 tons 27 tons -15.6% Park energy cost 1.2 million yuan / year 0.98 million yuan / year -18.3% Carbon asset value 0 5.2 million yuan / year +5.2 million yuan Abnormality detection accuracy 78% 95% +17% Data integrity 82% 98% +16% Carbon accounting accuracy 85% 97% +12% Scheduling optimization efficiency Benchmark Benchmark + 40% +40% Prediction error Benchmark Benchmark - 25% -25% Application Example 2: Industrial Park Application Taking a certain electronics industrial park (500,000 square meters, including 500kW photovoltaic, 500kWh energy storage, 500kW ground source heat pump, and 500 production servers) as an example: Data acquisition layer: Deploy smart meters, photovoltaic inverter data acquisition devices, energy storage BMS systems, and heat pump controllers to transmit data via industrial Ethernet; Carbon accounting side connects to the Guangdong Provincial Carbon Trading Platform API to obtain enterprise carbon quotas and carbon trading data.

[0047] AI-enhanced collaborative decision-making layer: Deployed on the park cloud platform (using Alibaba Cloud ECS instance, 8 cores and 32GB memory), running a distributed LSTM model and a carbon accounting model, and an AI-enhanced NSGA-II algorithm.

[0048] Results: Green electricity consumption increased from 62% to 94%, carbon emissions decreased by 22%, annual energy cost savings reached 3.2 million yuan, and carbon asset value reached 180,000 yuan per year. AI-enhanced data processing improved anomaly detection accuracy to 96%, data integrity to 99%, and carbon accounting accuracy to 98%.

[0049] Application Example 3: Business Park Application Taking a commercial complex park (300,000 square meters, including 200kW photovoltaic, 200kWh energy storage, 300kW air source heat pump, and 100 office servers) as an example: Data acquisition layer: Deploy smart sockets, energy management systems, and HVAC control systems to transmit data via LoRa gateways; Carbon accounting side connects to the Shenzhen Carbon Inclusive Platform to realize carbon credit redemption.

[0050] AI-enhanced collaborative decision-making layer: deployed on an edge server (4 cores, 16GB memory), running the AI-enhanced NSGA-II algorithm.

[0051] Results: Green electricity consumption increased from 55% to 89%, carbon emissions decreased by 18%, annual energy cost savings reached 850,000 yuan, and carbon asset value reached 68,000 yuan per year. AI-enhanced data processing significantly improved the overall system performance.

[0052] The advantages of the present application are: 1、The present application realizes the dual optimization of green electricity consumption rate and carbon emission through the deep integration and real-time collaborative processing of energy data and IT data, compared with the traditional independent scheduling mode, the present application is improved in green electricity consumption rate, effectively reduces the phenomenon of light abandonment, at the same time, reduces the annual carbon emission of IT equipment, significantly reduces the influence of park operation on the environment, reduces the energy cost of the park, through accurate energy scheduling and IT resource allocation, avoids energy waste and excessive purchase.

[0053] 2、The present application also ensures the adaptability and robustness of the scheduling strategy through the closed-loop feedback mechanism and multi-objective optimization algorithm, which can automatically adjust in different time periods and working conditions to achieve the best balance between economy and environmental protection.

[0054] 3、The present application innovatively applies AI technology to each link of carbon accounting and energy scheduling, through AI enhanced data preprocessing, intelligent prediction model and reinforcement learning optimization algorithm, realizes the intelligentization and self-adaptation of scheduling decision, makes the system can dynamically optimize the decision strategy according to the actual running state.

[0055] The above disclosure is only a few specific embodiments of the present application, but the present application is not limited to this, any changes that can be thought of by those skilled in the art should fall within the protection scope of the present application.

Claims

1. A park-level energy and carbon accounting coordinated dispatching method, characterized in that, The method comprises the following steps: S1: Real-time collection of energy data and carbon accounting data, and storage of the energy data and carbon accounting data in a time series database after cleaning and standardization; S2: Generation of green power output prediction values and carbon emission estimates based on the processed data in the time series database; S3: Setting the maximum green power consumption rate and the minimum carbon emission as core indicators, combining the green power output prediction values and carbon emission estimates, and generating a scheduling instruction set through an AI-enhanced multi-objective optimization algorithm; S4: Controlling energy equipment and IT resources according to the scheduling instructions, collecting the actual green power consumption rate and carbon emission after execution by the feedback module, setting a prediction deviation, and returning to S2 for re-optimization if the actual green power consumption rate and carbon emission are greater than the prediction deviation.

2. The park-level energy and carbon accounting collaborative scheduling method according to claim 1, wherein the method for obtaining the green power output prediction values comprises: using a long short-term memory (LSTM) model, inputting historical green power output data and meteorological data, and outputting green power generation power prediction values; The method for obtaining the carbon emission estimate comprises: using a carbon accounting model based on a national unified emission factor, inputting energy scheduling instructions and historical carbon emission data, and outputting carbon emission estimates. In S3, the multi-objective optimization algorithm is a non-dominated sorting genetic algorithm II (NSGA-II).

3. The park-level energy and carbon accounting coordinated scheduling method according to claim 1, characterized in that, While generating the collaborative scheduling instruction set, the estimated carbon emission corresponding to the scheduling instruction set is calculated in real time based on the national unified emission factor, and the estimated carbon emission is used as a feedback signal to participate in the solution process of the optimization target.

4. The park-level energy and carbon accounting coordinated scheduling method according to claim 1, characterized in that, In S1, the sampling frequency of the energy data is 1 min / time, and the sampling frequency of the carbon accounting data is 5 s / time. The energy data and the carbon accounting data are transmitted through the MQTT protocol.

5. The park-level energy and carbon accounting coordinated scheduling method according to claim 1, characterized in that, The method comprises the following steps:

6. A park-level energy and carbon accounting coordinated dispatch system, characterized in that, An energy data collection module is configured to collect energy data of photovoltaic, energy storage, and heat pump; A carbon accounting data collection module is configured to collect carbon emission factor data, carbon asset data, and carbon quota data; A data preprocessing module is configured to perform AI-enhanced cleaning and intelligent standardization on the energy data and carbon accounting data; A prediction submodule is configured to predict green power output prediction values and carbon emission estimates; A carbon accounting submodule is configured to calculate the carbon emission of a scheduling scheme in real time based on a national unified emission factor, and input the result into the multi-objective decision submodule; A multi-objective decision submodule is configured to generate a collaborative scheduling instruction set based on the prediction curve, with the maximum green power consumption rate and the minimum carbon emission as the target; An energy execution module is configured to control the operating state of energy equipment according to the collaborative scheduling instruction set; An IT execution module is configured to schedule the execution of IT tasks and the allocation of server resources according to the scheduling instructions; A feedback module is configured to collect the data after execution and feed it back to the prediction submodule and the multi-objective decision submodule. ​ 7. The park-level energy and carbon accounting coordinated scheduling system according to claim 6, characterized in that, The prediction submodule inputs historical green electricity output data and meteorological data through a pre-trained LSTM model and a carbon accounting model, and outputs green electricity generation power prediction values and carbon emission estimates using a carbon accounting model based on national unified emission factors, inputs energy dispatch instructions and historical carbon emission data, and outputs carbon emission estimates.

8. The park-level energy and carbon accounting coordinated scheduling system according to claim 6, characterized in that, The multi-objective decision-making submodule uses the NSGA-II algorithm to solve the multi-objective optimization problem under the constraints of energy storage SOC ≥ 20% and server CPU utilization ≤ 85%.

9. The park-level energy and carbon accounting coordinated scheduling system according to claim 6, characterized in that, The carbon accounting data acquisition module is connected to a network platform through an API interface to obtain carbon emission data and carbon quota information.

10. The park-level energy and carbon accounting coordinated scheduling system according to claim 9, characterized in that, The carbon accounting submodule converts the energy consumption corresponding to the dispatch instructions into carbon emissions based on an open-source carbon accounting tool and supports CCER development.

Images