Carbon emission analysis method and device, electronic equipment and storage medium

Abstract

The invention provides a carbon emission analysis method and device, electronic equipment and a storage medium, and relates to the technical field of carbon emission data processing.The method comprises the steps that multi-source real-time data and energy consumption data of all energy consumption nodes in a park in the current time period are obtained, and multi-dimensional key features are extracted; determining initial energy structure data; according to the association relationship between the energy consumption node in the park and other energy consumption nodes, performing association correction on the initial energy structure data to obtain final energy structure data; then determining a dynamic emission factor matrix of the energy consumption nodes; performing carbon emission simulation on the energy consumption node through the hybrid model to obtain a carbon emission result of the energy consumption node; and performing dynamic analysis and explanation of carbon emission through an attention mechanism to obtain a carbon emission analysis result of the park in the current time period. According to the invention, high-precision dynamic analysis of park carbon emission is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carbon emission data processing, in particular to a carbon emission analysis method and device, electronic equipment and storage medium. BACKGROUND

[0002] At present, with the emphasis on environmental protection and sustainable development, as an important carrier of industrial agglomeration and economic development, the carbon emission analysis of the park is extremely important in energy management, environmental policy making and enterprise carbon emission reduction strategy. Among them, for the carbon emission analysis of the park, different emission factors need to be combined with different energy structures in the park to calculate the carbon emission, so as to obtain the analysis result of carbon emission.

[0003] In the related art, since the existing emission factor is based on fixed data and assumption, the emission factor mostly adopts fixed value, and the carbon emission in a certain period of time is analyzed statically by using fixed emission factor, so it is difficult to reflect the dynamic change of carbon emission, and thus it is unable to adapt to the dynamic change of energy structure of the park, resulting in large deviation of the analysis result of carbon emission. SUMMARY

[0004] The problem solved by the present application is how to improve the dynamic analysis accuracy of carbon emission of the park.

[0005] To solve the above problems, the present application provides a carbon emission analysis method, device, electronic equipment and storage medium.

[0006] In a first aspect, the present application provides a carbon emission analysis method, comprising: obtaining multi-source real-time data and energy consumption data of each energy consumption node in the park in a current time period; According to the carbon emission analysis requirement, the multi-dimensional key features of each energy consumption node are extracted from the multi-source real-time data of each energy consumption node; According to the multi-dimensional key features of each energy consumption node, the initial energy structure data of the energy consumption node is determined; According to the correlation relationship between the energy consumption node and other energy consumption nodes in the park, the initial energy structure data is corrected to obtain the final energy structure data; According to the final energy structure data, the dynamic emission factor matrix of the energy consumption node is obtained by using a hierarchical adaptive calculation model; Combining the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data, the carbon emission of the energy consumption node is simulated by using a hybrid model to obtain the carbon emission result of the energy consumption node; According to the carbon emission result of each energy consumption node, a dynamic analysis and interpretation of carbon emission is performed through an attention mechanism, and a carbon emission analysis result of the park in the current time period is obtained.

[0007] Optionally, the multi-dimensional key features of the energy consumption node are extracted from the multi-source real-time data of each energy consumption node according to the carbon emission analysis requirement, including: According to the management scene of the park, the type of the carbon emission analysis requirement is determined, and the type of the carbon emission analysis requirement includes short-term emission reduction monitoring requirement, long-term policy making requirement and equipment energy efficiency optimization requirement; For the type of the carbon emission analysis requirement, the target data type corresponding to the multi-source real-time data of each energy consumption node is matched; The original features of the target data type corresponding dimension are extracted from the target data type; wherein for the short-term emission reduction monitoring requirement, the energy consumption peak-valley fluctuation feature of time dimension and the real-time load feature of equipment dimension are extracted; for the long-term policy making requirement, the functional area energy consumption difference feature of space dimension and the clean energy proportion change feature of policy dimension are extracted; for the equipment energy efficiency optimization requirement, the aging loss feature of equipment dimension and the energy production-energy consumption correlation feature of working condition dimension are extracted; The original features of each extracted dimension are subjected to data preprocessing to obtain multi-dimensional key features of each energy consumption node.

[0008] Optionally, the initial energy structure data of the energy consumption node is determined according to the multi-dimensional key features of each energy consumption node, including: The multi-dimensional key features are analyzed and classified through an energy structure identification model to determine a plurality of energy structure types and scene correlation features of the energy consumption node; According to the energy structure type, the energy type proportion, energy conversion efficiency and energy consumption fluctuation data of the energy consumption node are determined; According to the scene correlation feature, the scene influence data strongly related to energy consumption is determined; The energy type proportion, energy conversion efficiency, energy consumption fluctuation data and scene influence data of the energy consumption node are taken as the initial energy structure data; The initial energy structure data is associated and corrected according to the correlation relationship between the energy consumption node and other energy consumption nodes in the park to obtain final energy structure data, including: According to the correlation relationship between each energy consumption node in the park, the influence coefficient of the energy consumption node is quantified; According to the affected coefficient, the energy type proportion, the energy conversion efficiency, the energy consumption fluctuation data and the scenario influence data are corrected to obtain final energy structure data.

[0009] Optionally, the hierarchical adaptive calculation model comprises a basic factor matching layer, an efficiency correction layer, a scenario dynamic adjustment layer and a bias calibration layer. According to the final energy structure data, the hierarchical adaptive calculation model is used to obtain a dynamic emission factor matrix of the energy consumption node, which comprises: According to the energy type proportion, the initial emission factor vector is obtained by combining a preset industry benchmark emission factor library through the basic factor matching layer. According to the energy conversion efficiency and the energy consumption fluctuation data, the initial emission factor vector is corrected to obtain an efficiency-corrected factor vector through the efficiency correction layer. According to the scenario influence data, the efficiency-corrected factor vector is dynamically adjusted in real time to obtain a scenario-adjusted factor matrix through the scenario dynamic adjustment layer. By combining historical carbon emission evaluation data, the bias value of the scenario-adjusted factor matrix and the historical actual monitoring data is determined, and the scenario-adjusted factor matrix is self-calibrated and optimized based on the bias value to obtain the dynamic emission factor matrix through the bias calibration layer.

[0010] Optionally, the hierarchical adaptive calculation model further comprises a real-time feedback layer. According to the final energy structure data, the hierarchical adaptive calculation model is used to obtain a dynamic emission factor matrix of the energy consumption node, which further comprises: Real-time carbon emission concentration data collected by real-time carbon emission monitoring equipment arranged in the park is obtained through the real-time feedback layer. According to the deviation between the real-time carbon emission concentration data and the estimated carbon emission concentration data generated by the dynamic emission factor matrix, a feedback compensation factor is constructed. The dynamic emission factor matrix is online iteratively optimized through the feedback compensation factor to update the emission factor values in the dynamic emission factor matrix.

[0011] Optionally, the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data are combined to simulate carbon emission of the energy consumption node through a hybrid model to obtain carbon emission results of the energy consumption node, which comprises: The energy consumption data is split according to energy types to obtain sub-consumption amounts of each energy type of the energy consumption node. multiply the sub-item consumption amount with an emission factor of a corresponding energy type in the dynamic emission factor matrix element by element to obtain a basic carbon emission amount of the energy consumption node; According to the multi-source real-time data, a bidirectional long short-term memory network model is used for fitting to obtain a nonlinear correction amount of the energy consumption node; According to the basic carbon emission amount and the nonlinear correction amount, a preliminary carbon emission result of the energy consumption node is obtained by weighted fusion; The preliminary carbon emission result is subjected to outlier detection and elimination, and is labeled in combination with the collection frequency of the energy consumption data to generate the carbon emission result containing a timestamp, an energy type sub-item carbon emission value, and a total carbon emission value.

[0012] Optionally, according to the carbon emission result of each energy consumption node, a dynamic analysis and interpretation of carbon emission are performed through an attention mechanism to obtain a carbon emission analysis result of the park in the current time period, including: The carbon emission result of the energy consumption node is input into an attention mechanism model; The energy type sub-item carbon emission value and the total carbon emission value in the carbon emission result are subjected to weight distribution through the attention mechanism model to determine a contribution degree of each energy consumption individual in the energy consumption node to carbon emission; According to the contribution degree, a dynamic analysis of the energy consumption individual is performed to obtain a carbon emission source and a key driving factor of the park; In combination with the carbon emission source and the key driving factor of the park, the carbon emission analysis result of the park in the current time period is generated.

[0013] In a second aspect, a carbon emission analysis device is provided, including: A data acquisition unit is configured to acquire multi-source real-time data and energy consumption data of each energy consumption node in a park in a current time period; A feature extraction unit is configured to extract multi-dimensional key features of the energy consumption node from the multi-source real-time data of each energy consumption node according to carbon emission analysis requirements; An energy structure analysis unit is configured to determine initial energy structure data of the energy consumption node according to the multi-dimensional key features of each energy consumption node; A correction unit is configured to correct the initial energy structure data in association with other energy consumption nodes in the park to obtain final energy structure data; An emission factor matrix construction unit is configured to obtain a dynamic emission factor matrix of the energy consumption node according to the final energy structure data by a hierarchical adaptive calculation model; An analog unit is configured to simulate carbon emission of the energy consumption node by a hybrid model to obtain a carbon emission result of the energy consumption node by combining the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data. An analysis unit is configured to perform dynamic analysis and interpretation of carbon emission by an attention mechanism according to the carbon emission result of each energy consumption node to obtain a carbon emission analysis result of the park in the current time period.

[0014] In a third aspect, an electronic device of the present application comprises a processor and a memory, wherein the memory is configured to store a computer program. The computer program, when loaded by the processor, enables the processor to perform the carbon emission analysis method as described above.

[0015] In a fourth aspect, a computer readable storage medium of the present application stores a computer program, wherein the computer program, when executed by a processor, implements the carbon emission analysis method as described above.

[0016] The carbon emission analysis method, device, electronic device and storage medium of the present application can obtain multi-source real-time data and energy consumption data of each energy consumption node in the park, determine initial energy structure data from multi-dimensional key features, and correct the data in combination with the correlation between nodes, so that the energy structure data can dynamically reflect the actual state of energy consumption in the park. Furthermore, a hierarchical adaptive calculation model is used to generate a dynamic emission factor matrix, which can automatically adjust according to the real-time changes of the energy structure, so as to accurately capture the dynamic changes of carbon emission in the carbon emission simulation process and achieve high-precision dynamic analysis of carbon emission in the park. The introduction of the attention mechanism can dynamically analyze and interpret the carbon emission result of each energy consumption node, which can deeply mine the key information and dynamic change law behind the carbon emission data. Not only can it help to more clearly understand the dynamic changes of carbon emission in the park, but also can provide in-depth and practical guiding significance analysis results for formulating targeted carbon emission reduction strategies.

[0017] For the problem of dynamic change of energy structure in the park, the application captures the dynamic change information of the energy structure in time by acquiring multi-source real-time data and extracting multi-dimensional key features. At the same time, a hierarchical adaptive calculation model is used to generate a dynamic emission factor matrix, which can automatically adjust the calculation of the emission factor according to the real-time change of the energy structure, so as to accurately reflect the carbon emission under the dynamic change of the energy structure, and solve the problem of large deviation of the analysis result caused by the fixed emission factor in the traditional method. The previous carbon emission analysis may only determine the energy structure according to a single data source or limited data dimensions, resulting in insufficient and inaccurate data. The application acquires multi-source real-time data of each energy consumption node in the park, and extracts multi-dimensional key features therefrom, fully integrates data of different types and different sources, and makes the determination of the energy structure data more comprehensive and accurate. In addition, the correlation between the energy consumption nodes is considered to correct the initial energy structure data, further improving the accuracy and reliability of the data, providing a solid data foundation for subsequent carbon emission analysis, and solving the problem of deviation of the analysis result caused by insufficient and inaccurate data integration in the past. At the same time, the previous carbon emission analysis result is often general, lacking in-depth analysis and explanation of the carbon emission of each energy consumption node, and it is difficult to effectively guide the carbon emission reduction work of the park. The application dynamically analyzes and explains the carbon emission result of each node through the attention mechanism, deeply mines the information behind the carbon emission data, and clarifies the position and role of each node in the park carbon emission, as well as its dynamic change trend and influencing factors, further providing strong support for the park managers to formulate specific and effective carbon emission reduction strategies, and solving the problem of lack of depth and targeted guidance of the previous carbon emission analysis result. In summary, the application effectively solves the problems of dynamic deficiency and precision deviation caused by fixed emission factor in the traditional park carbon emission analysis by combining multi-source real-time data, dynamic emission factor matrix and attention mechanism, and realizes high-precision dynamic analysis of the park carbon emission. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of a carbon emission analysis method in an embodiment of the application is shown. Figure 2 A structural diagram of a carbon emission analysis system in an embodiment of the application is shown. Figure 3 A structural diagram of an electronic device in an embodiment of the application is shown. DETAILED DESCRIPTION

[0019] In order to make the above objectives, characteristics and advantages of the present application more apparent, concrete embodiments of the present application will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein, but rather, these embodiments are provided in order to more thoroughly and completely understand the present application. It should be understood that the drawings and embodiments of the present application are merely for illustrative purposes, and are not intended to limit the scope of protection of the present application.

[0020] It should be understood that each of the steps recited in the method embodiments of the present application can be performed in different orders, and / or in parallel. In addition, the method embodiments can include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0021] The term "comprising" and variations thereof as used herein are open-ended, that is, "comprising but not limited to"; the term "based on" is "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optional" means "optional in at least some embodiments". Related definitions will be given in the description below. It should be noted that the concepts "first", "second", etc. mentioned in the present application are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0022] It should be noted that the modification of "one" or "multiple" mentioned in the present application is illustrative rather than limiting, and those skilled in the art should understand that, unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0023] The names of the messages or information exchanged between the devices in the embodiments of the present application are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0024] In combination Figure 1 The carbon emission analysis method provided by the embodiments of the present application includes: Obtaining multi-source real-time data and energy consumption data of each energy consumption node in the park within a current time period.

[0025] Specifically, when acquiring multi-source real-time data and energy consumption data of each energy consumption node in the park within the current time period, first, the energy consumption nodes in the park are determined, such as production workshops, office buildings, charging piles, boiler rooms, etc. Through the deployment of Internet of Things sensors such as power sensors, gas flow meters, water temperature sensors, etc. at each node, real-time collection of equipment operating parameters such as voltage, current, flow, pressure, environmental data such as temperature, humidity, illumination is performed, and the park energy management system, equipment operation and maintenance log system, etc. are connected to acquire the energy consumption raw data of each node such as hourly power consumption, daily natural gas consumption, monthly coal consumption, etc. and set the current time period such as the past 24 hours, the first week of the month. The collected data is time-stamped and aligned, outliers such as jump data caused by sensor failure are removed, missing values are completed by interpolation with adjacent period means, and finally structured multi-source real-time data sets and energy consumption data sets are formed.

[0026] According to the carbon emission analysis requirements, multi-dimensional key features of each energy consumption node are extracted from the multi-source real-time data of each energy consumption node.

[0027] Specifically, according to the carbon emission analysis requirements, such as identifying key emission nodes, analyzing the impact of energy types on emissions, and evaluating periodical emission differences, when extracting multi-dimensional key features from the multi-source real-time data of each energy consumption node, first, the feature dimensions are determined based on the requirements. For example, if the requirement is to analyze the impact of device operating status on emissions, the dimensions can include device operating parameters such as motor power peak, boiler operating time, load characteristics such as production load rate, power load fluctuation coefficient; if the requirement is environmental factor influence analysis, the dimensions can include environmental parameters such as outdoor temperature mean, humidity change rate; then through statistical methods such as calculating the mean, variance, maximum value of each hour data to extract basic features, through time series analysis such as calculating the change amount, trend slope of adjacent period data to extract dynamic features, and then combining feature selection algorithms such as mutual information-based screening, variance threshold method to remove redundant features irrelevant to the analysis requirements, such as removing illumination data which has no significant impact on boiler house emission analysis, and finally obtaining the multi-dimensional key feature set of each node.

[0028] According to the multi-dimensional key features of each energy consumption node, the initial energy structure data of the energy consumption node is determined.

[0029] Specifically, when determining the initial energy structure data according to the multi-dimensional key features of each energy consumption node, a mapping relationship between the key features and the energy structure is first established. For example, the device type feature, such as the identification of an electric forklift corresponding to its consumed energy of electricity, and the operating parameter feature, such as the gas flow sensor data of a gas boiler directly correlating to the natural gas consumption. For a composite node such as a workshop using both electricity and natural gas, the proportion of the consumption of each type of energy is calculated through the energy-related parameters in the key features, such as the cumulative value of the power sensor and the cumulative value of the gas flow meter. The average electricity / gas consumption ratio of the same type of workshop is combined with the historical energy structure template of the same type of node to calibrate the initial consumption of each type of energy. If the characteristics of a node show that its production load has increased by 20% compared to the historical template, the initial consumption of each type of energy is adjusted in proportion. Finally, the initial energy structure data containing the types of energy consumed by each node, such as electricity, natural gas, coal, and biomass energy, and the corresponding consumption amounts are formed.

[0030] According to the association relationship between the energy consumption node and other energy consumption nodes in the park, the initial energy structure data is associated and corrected to obtain the final energy structure data.

[0031] Specifically, when the initial energy structure data is associated and corrected to obtain the final energy structure data according to the association relationship between the energy consumption node and other nodes in the park, a node association graph is first constructed through the park energy network topology graph and the device connection relationship table to clearly define the association types, such as power supply association: the power station supplies power to workshops A and B; and energy sharing association: multiple offices share a central air conditioning system. Then, the initial data is verified based on the association rules. For example, the total power supply amount of the power station should be equal to the sum of the initial data of the power consumption of each downstream node such as workshops A and B. If there is a deviation, such as the power station records a power supply of 1000 kWh, while the total of the downstream nodes is 900 kWh, the correction amount is allocated according to the association strength, such as the historical electricity consumption proportion of workshop A is 60% and that of workshop B is 40%, to complete 100 kWh. Workshop A is supplemented with 60 kWh and workshop B is supplemented with 40 kWh. For associations with energy conversion, such as a boiler room converting coal into steam to supply to a workshop, the matching of the steam consumption and the coal consumption is verified according to the conversion efficiency, such as the conversion rate of coal to steam is 80%, to correct unreasonable initial data. Finally, the final energy structure data that meets the association logic is obtained.

[0032] Through a hierarchical self-adaptive calculation model, the dynamic emission factor matrix of the energy consumption node is obtained according to the final energy structure data.

[0033] Specifically, when obtaining the dynamic emission factor matrix of the energy consumption node according to the final energy structure data by the hierarchical adaptive calculation model, the hierarchical model includes three layers: the bottom layer is a basic emission factor layer, which stores the benchmark emission factors of various energies, such as the emission factor of each kWh of electricity and the emission factor of each cubic meter of natural gas published by the state; the middle layer is a node type correction layer, which sets a correction coefficient according to the node attributes, such as industrial nodes and civil nodes, for example, the emission factor of industrial electricity is 5% higher than that of civil electricity because the grid load of industrial electricity is high; the top layer is a time dynamic layer, which adjusts the emission factor by combining external factors of the current time period, such as the real-time energy composition of the power grid and seasonal factors, for example, the emission factor of electricity in summer is 10% lower than that in winter because the photovoltaic power generation capacity is high in summer; the adaptive mechanism automatically increases the correction coefficient weight of the corresponding node type in the middle layer when the change rate of the energy type proportion in the final energy structure data is suddenly increased, such as a 30% sudden increase in the natural gas proportion of a node, and at the same time triggers the time dynamic layer of the top layer to recalculate, and finally outputs a dynamic matrix containing the emission factors of each node, each energy type and each time period, such as the emission factor of electricity of node A in the t1 period and the emission factor of natural gas in the t2 period.

[0034] In combination with the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data, the carbon emission of the energy consumption node is simulated by a hybrid model to obtain the carbon emission result of the energy consumption node.

[0035] Specifically, in combination with the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data, the carbon emission of the energy consumption node is simulated by a hybrid model to obtain the carbon emission result, and the hybrid model fuses a physical model and a data-driven model: the physical model part calculates the basic emission based on the formula certain energy carbon emission = corresponding dynamic emission factor x energy consumption amount, for example, the basic emission of node B consuming natural gas 500m 3 , and the corresponding dynamic emission factor is 2.0 kgCO2 / m 3 , is 1000 kgCO2; the data-driven model part, such as the LSTM neural network, uses multi-source real-time data, such as device operating temperature and load rate, to correct the basic value, for example, when the device load rate exceeds 80%, the emission efficiency decreases, and the basic value is adjusted to 1050 kgCO2 by the model output correction coefficient 1.05; the hybrid model dynamically adjusts the weights of the two parts according to the historical simulation error, for example, the weight of the physical model is increased to 0.7 when the error is small, and finally outputs the hourly carbon emission data and the cumulative carbon emission result of each node in the current time period.

[0036] According to the carbon emission result of each energy consumption node, the carbon emission is dynamically analyzed and explained by an attention mechanism to obtain the carbon emission analysis result of the park in the current time period.

[0037] Specifically, according to the carbon emission results of each energy consumption node, the dynamic analysis and interpretation of carbon emissions by the attention mechanism can obtain the carbon emission analysis results of the park in the current time period. First, the carbon emission results of each node are input into the attention model, and the attention weight of each node is calculated. For example, the carbon emission of node C accounts for 30% of the total emissions of the park, and the Pearson correlation coefficient with the total emissions is 0.9. Therefore, the weight is 0.3, and the top 20% of the key nodes are screened out. Then, the key nodes are analyzed for their emission driving factors, and the attention mechanism is used to locate specific features such as the carbon emission of node C. 80% of the carbon emission comes from coal consumption, and the emission increases sharply during the daily 8-10 pm peak load period. Dynamic analysis results are generated, including the total carbon emissions of the park, the emission proportion pie chart of each node, the key emission source list containing the node name and contribution, the emission curve with time, the main driving factor explanation such as coal consumption being the primary source of park emissions accounting for 65%, etc. A complete carbon emission analysis report is formed.

[0038] The carbon emission analysis method of the present embodiment obtains multi-source real-time data and energy consumption data of each energy consumption node in the park, determines the initial energy structure data from multiple dimensions of key features, and corrects it in combination with the correlation between nodes, so that the energy structure data can dynamically reflect the actual state of energy consumption in the park. Then, a dynamic emission factor matrix is generated by using a hierarchical adaptive calculation model. This matrix can automatically adjust according to the real-time changes of the energy structure, so as to accurately capture the dynamic changes of carbon emissions during the carbon emission simulation process, and realize high-precision dynamic analysis of carbon emissions in the park. The introduction of the attention mechanism can dynamically analyze and interpret the carbon emission results of each energy consumption node, which can deeply mine the key information and dynamic change rules behind the carbon emission data. Not only can it help to better understand the dynamic changes of carbon emissions in the park, but also can provide in-depth and practical guidance for the development of targeted carbon emission reduction strategies.

[0039] In view of the dynamic change of the energy structure of the park, the embodiment captures the dynamic change information of the energy structure in time by acquiring multi-source real-time data and extracting multi-dimensional key features. Meanwhile, a hierarchical adaptive calculation model is used to generate a dynamic emission factor matrix. The model can automatically adjust the calculation of the emission factor according to the real-time change of the energy structure, so as to accurately reflect the carbon emission under the dynamic change of the energy structure, and solve the problem of large deviation of the analysis result caused by the fixed emission factor in the traditional method. The previous carbon emission analysis may only determine the energy structure according to a single data source or limited data dimensions, resulting in insufficient and inaccurate data. The embodiment acquires multi-source real-time data of each energy consumption node in the park, and extracts multi-dimensional key features therefrom, fully integrates data of different types and different sources, and makes the determination of the energy structure data more comprehensive and accurate. In addition, the correlation between the energy consumption nodes is considered to correct the initial energy structure data, further improving the accuracy and reliability of the data, providing a solid data foundation for subsequent carbon emission analysis, and solving the problem of deviation of the analysis result caused by insufficient and inaccurate data integration in the past. At the same time, the previous carbon emission analysis result is often general, lacking in-depth analysis and explanation of the carbon emission of each energy consumption node, and it is difficult to effectively guide the carbon emission reduction work of the park. The embodiment dynamically analyzes and explains the carbon emission result of each node through the attention mechanism, deeply mines the information behind the carbon emission data, and clarifies the position and role of each node in the carbon emission of the park, as well as the dynamic change trend and influencing factors, further providing strong support for the park managers to formulate specific and effective carbon emission reduction strategies, and solving the problem of lack of depth and targeted guidance of the previous carbon emission analysis result. In summary, the embodiment effectively solves the problems of dynamic deficiency and precision deviation caused by the fixed emission factor in the traditional park carbon emission analysis by combining multi-source real-time data, a dynamic emission factor matrix and an attention mechanism, and realizes high-precision dynamic analysis of the carbon emission of the park.

[0040] Optionally, the multi-dimensional key features of each energy consumption node are extracted from the multi-source real-time data of the energy consumption node according to the carbon emission analysis requirement, including: According to the management scene of the park, the type of the carbon emission analysis requirement is determined, and the type of the carbon emission analysis requirement includes short-term emission reduction monitoring requirement, long-term policy making requirement and equipment energy efficiency optimization requirement; According to the type of the carbon emission analysis requirement, the target data type corresponding to the multi-source real-time data of each energy consumption node is matched; extracting original features of a corresponding dimension of the target data type from the target data type; wherein, for the short-term emission reduction monitoring requirement, extracting energy consumption peak-valley fluctuation features of time dimension and real-time load features of equipment dimension; for the long-term policy making requirement, extracting functional area energy consumption difference features of space dimension and clean energy proportion change features of policy dimension; for the equipment energy efficiency optimization requirement, extracting aging loss features of equipment dimension and production capacity-energy consumption correlation features of working condition dimension; performing data preprocessing on the original features of each of the extracted dimensions to obtain multi-dimensional key features of each of the energy consumption nodes.

[0041] Specifically, when determining the type of carbon emission analysis requirement according to the management scene of the park, the core management goal of the park is first determined. For example, if the management scene is daily operation monitoring such as real-time supervision of the implementation of emission reduction measures by the dispatching center, the short-term emission reduction monitoring requirement is corresponded, and the emission fluctuation and immediate emission reduction effect in the near 1-7 days are focused on. If the management scene is annual / five-year planning making such as the formulation of carbon peak path of the park by the environmental protection department, the long-term policy making requirement is corresponded, and the emission trend and structural optimization space for more than 1 year are concerned. If the management scene is equipment operation and maintenance upgrade such as energy efficiency reconstruction of high energy consumption equipment in the workshop, the equipment energy efficiency optimization requirement is corresponded, and the energy consumption and output matching of a single / class of equipment are focused on.

[0042] For type matching target data types according to carbon emission analysis demand, data sources are filtered based on time scale of demand and objects of interest. For short-term emission reduction monitoring demand, high-frequency real-time data such as device current / power data every 5 minutes, hourly energy consumption statistical data and device state data such as running / offline state and load switching records are matched. For long-term policy making demand, periodic cumulative data such as monthly energy consumption of each functional area, quarterly clean energy photovoltaic / wind power generation and policy-related data such as new energy device installed capacity and carbon quota execution records are matched. For device energy efficiency optimization demand, device full life cycle data such as running time, maintenance records and factory parameters and production data such as unit time capacity and product qualification rate are matched. Specifically, when extracting original features from target data types, key dimensions are focused according to demand types. For short-term emission reduction monitoring demand, energy consumption peak-valley fluctuation features in time dimension are extracted by calculating deviation rate of hourly energy consumption value from daily average energy consumption, duration of peak-valley period such as 9:00-11:00 for peak and 0:00-6:00 for valley and energy consumption difference, and real-time load features in device dimension are extracted by current load rate actual power / rated power and load change rate adjacent 10-minute load difference / base load. For long-term policy making demand, functional area energy consumption difference features in space dimension are extracted by calculating standard deviation of unit area energy consumption of each functional area such as production area / office area / living area and energy consumption density total energy consumption / functional area area, and clean energy proportion change features in policy dimension are extracted by ratio of monthly clean energy consumption to total energy consumption and year-on-year growth rate of the ratio. For device energy efficiency optimization demand, aging loss features in device dimension are extracted by linear regression coefficient of running time and energy consumption growth such as energy consumption increase proportion per 1000 hours of running and Pearson correlation coefficient of monthly fault frequency and energy consumption, and production capacity-energy consumption correlation features in working condition dimension are extracted by unit product energy consumption total energy consumption / total production and synchronous rate of production capacity fluctuation amplitude daily production capacity standard deviation and energy consumption fluctuation amplitude.

[0043] When performing data preprocessing on original features of each dimension extracted, cleaning, standardization and dimension reduction operations are sequentially performed. In the data cleaning stage, 3σ rule is used to remove outliers such as energy consumption fluctuation values exceeding mean value ± 3 times standard deviation, and missing values such as 1-2 data point missing caused by temporary device offline are completed by linear interpolation. In the standardization stage, min-max normalization is used to convert different dimension features such as energy consumption value unit kWh and load rate percentage to [0, 1] interval, eliminating the influence of dimension difference. In the dimension reduction stage, principal component analysis PCA is used to retain principal components with cumulative contribution rate ≥ 90%, and redundant features such as highly correlated load rate and power fluctuation features are removed.

[0044] In the embodiment of the present application, based on the park management scenario, three types of demands are divided, including short-term emission reduction monitoring, long-term policy making and equipment energy efficiency optimization, the target boundary of feature extraction is clearly defined from the source, the information redundancy problem caused by indiscriminate feature collection in traditional methods is avoided, the subsequent analysis is always carried out around the actual business demand, and the foundation for the pertinence of features is laid. Secondly, different target data types are matched for different demand types, such as matching high-frequency real-time data for short-term demand and matching periodic cumulative data for long-term demand. Through accurate alignment of data and demand, the interference of irrelevant data is reduced, the calculation cost of data processing is reduced, and the efficiency of feature extraction is improved. Thirdly, based on demand-oriented extraction of original features, such as short-term demand focusing on peak-valley fluctuation and device load in time dimension, long-term demand focusing on spatial functional area difference and clean energy proportion change, the extracted features are directly related to the core analysis target, and the relevance of features and carbon emission influencing factors is strengthened, providing high information density input variables for subsequent models. Finally, through preprocessing operations such as cleaning, standardization and dimension reduction, the interference caused by abnormal values, dimension differences and feature redundancy is eliminated, the features more accurately reflect the essential properties of energy consumption nodes, and the disturbance of noise data on subsequent energy structure analysis and carbon emission simulation is avoided.

[0045] Optionally, the initial energy structure data of the energy consumption node is determined according to the multi-dimensional key features of each energy consumption node, including: The multi-dimensional key features are analyzed and classified by an energy structure identification model to determine a plurality of energy structure types and scene correlation features of the energy consumption node; According to the energy structure type, the energy type proportion, energy conversion efficiency and energy consumption fluctuation data of the energy consumption node are determined; According to the scene correlation feature, the energy consumption strongly related scene influence data is determined; The energy type proportion, energy conversion efficiency, energy consumption fluctuation data and scene influence data of the energy consumption node are taken as the initial energy structure data; The initial energy structure data is associated and corrected according to the correlation relationship between the energy consumption node and other energy consumption nodes in the park to obtain the final energy structure data, including: According to the correlation relationship between each energy consumption node in the park, the influence coefficient of the energy consumption node is quantified; According to the influence coefficient, the energy type proportion, energy conversion efficiency, energy consumption fluctuation data and scene influence data are corrected to obtain the final energy structure data.

[0046] Optionally, the layered adaptive calculation model comprises a basic factor matching layer, an efficiency correction layer, a scenario dynamic adjustment layer and a bias calibration layer. According to the final energy structure data, the layered adaptive calculation model is used to obtain a dynamic emission factor matrix of the energy consumption node, comprising: Through the basic factor matching layer, an initial emission factor vector is obtained according to the energy type proportion and in combination with a preset industry benchmark emission factor library. Through the efficiency correction layer, the initial emission factor vector is corrected according to the energy conversion efficiency and the energy consumption fluctuation data to obtain an efficiency-corrected factor vector. Through the scenario dynamic adjustment layer, the efficiency-corrected factor vector is dynamically adjusted in real time according to the scenario influence data to obtain a scenario-adjusted factor matrix. Through the bias calibration layer, a bias value of the scenario-adjusted factor matrix and historical actual monitoring data is determined in combination with historical carbon emission evaluation data, the scenario-adjusted factor matrix is self-calibrated and optimized based on the bias value, and the dynamic emission factor matrix is obtained.

[0047] Specifically, first, an energy structure identification model is constructed. The model can adopt a classification model based on random forest, take multi-dimensional key features (such as energy type related features, equipment operation features, environmental features, etc.) as input, and train the model through training samples (containing node feature data of known energy structure types) to make the model have the ability to identify the energy structure type. Then, the multi-dimensional key features of the target node are input into the trained model, and the energy structure type of the node is output, such as electricity-natural gas mixed type, coal dominant type, and renewable energy auxiliary type. At the same time, through the feature importance analysis module of the model, the scene features (such as production shift features, seasonal features, park activity arrangement features, etc.) strongly related to the energy structure are selected as scene correlation features, for example, the energy structure of a certain production workshop is strongly related to the three-shift production scene, and the energy structure of a certain office building is strongly related to the weekday / holiday scene. For the energy type proportion, combined with the consumption data of each type of energy in the multi-dimensional key features of the node (such as electric power sensor cumulative value, gas flowmeter data, etc.), the proportion of the consumption of each type of energy in the total energy consumption is calculated, for example, in the electricity-natural gas mixed type node, if the electricity consumption is 600 kWh and the natural gas consumption is about 400 kWh (converted according to the heat value), the electricity proportion is 60% and the natural gas proportion is 40%. For energy conversion efficiency, reference is made to the equipment parameters (such as the rated conversion efficiency of the boiler) and operation features (such as the load rate) corresponding to the energy structure type, and the actual conversion efficiency is calculated by the formula actual conversion efficiency = rated efficiency x load rate correction coefficient, for example, the rated efficiency of the gas boiler is 90%, the current load rate is 80%, and the correction coefficient is 0.95, so the actual conversion efficiency is 90% x 0.95 = 85.5%. For energy consumption fluctuation data, the standard deviation and peak-valley difference of energy consumption in a certain period (such as 1 day) are calculated to quantify the fluctuation degree, for example, the hourly energy consumption of a certain node is 100, 120, 90, and 110 kWh, the standard deviation is 12.9, and the peak-valley difference is 30 kWh.

[0048] According to the scene correlation feature determination and energy consumption strong correlation scene influence data, first, the mapping relationship between scene correlation feature and energy consumption data is established, the correlation between scene feature and energy consumption is calculated through Pearson correlation coefficient, and the scene features with absolute value of correlation coefficient greater than or equal to 0.6 are screened out; then, for these scene features, the influence degree of the scene features on energy consumption is quantified, for example, in the three-shift production scene, the energy consumption of the middle shift (16:00-24:00) is increased by 20% compared with that of the early shift (8:00-16:00), and the 20% increment is the scene influence data; in the winter low temperature scene, when the outdoor temperature is lower than 5℃, the heating energy consumption is increased by 50kWh / day compared with that in normal temperature, and the 50kWh / day increment also belongs to the scene influence data. Finally, the above calculated data is integrated to form a structured data set, for example, the initial energy structure data of a node can be represented as: energy type proportion (electricity 70%, natural gas 30%), energy conversion efficiency (electric power equipment 92%, natural gas equipment 88%), energy consumption fluctuation data (daily standard deviation 15kWh, peak valley difference 40kWh), scene influence data (energy consumption on weekdays is 30% higher than that on weekends, cooling energy consumption in high temperature weather is increased by 25kWh / day). The correlation relationship type between nodes (such as power supply correlation: node A supplies power to nodes B and C; heating correlation: node D supplies heat to node E) is determined through the park energy network topology diagram; then, the correlation strength is calculated based on the historical data, for example, for power supply correlation, the ratio of the change amount of the power supply of node A to the change amount of the energy consumption of node B is calculated to obtain the dependence coefficient of node B to node A, if the energy consumption of node B is increased by 80kWh when the power supply of node A is increased by 100kWh, then the dependence coefficient is 0.8; the dependence coefficient and the correlation type weight (such as power supply correlation weight 0.6, heating correlation weight 0.4) are combined, and the influence coefficient is quantified through the formula influence coefficient = dependence coefficient × correlation type weight, for example, the influence coefficient of node B is 0.8×0.6=0.48. For energy type proportion, if node B is affected by upstream power supply node A (influence coefficient 0.48), and the electricity supply proportion of node A is decreased by 5% due to grid adjustment, then the electricity proportion of node B is simultaneously corrected to original proportion × (1-5%×0.48); for energy conversion efficiency, if node E is affected by heating node D (influence coefficient 0.3), and the heat conversion efficiency of node D is increased by 3%, then the heating related conversion efficiency of node E is corrected to original efficiency × (1+3%×0.3); for energy consumption fluctuation data, if node C is strongly correlated with node A (influence coefficient 0.5), and the energy consumption fluctuation standard deviation of node A is increased by 10kWh, then the fluctuation standard deviation of node C is corrected to original standard deviation +10×0.5; for scene influence data, if node E is affected by the maintenance scene of node D (influence coefficient 0.4), and the heating scene influence data of node D is increased by 20kWh during maintenance, then the corresponding scene influence data of node E is corrected to original data +20×0.4, and finally the final energy structure data conforming to the correlation logic is formed.

[0049] In the embodiments of the present application, the analysis and classification of multi-dimensional key features by means of the energy structure identification model breaks through the limitations of traditional reliance on artificial experience to judge the energy structure, automatically identifies the energy structure type by the model and extracts the scene correlation features, so that the determination of the initial energy structure data is more in line with the actual energy consumption characteristics of the node, avoiding subjective bias, and the introduction of the scenario influence data makes up for the neglect of external scene factors such as environment and production by pure energy consumption data, enhancing the comprehensiveness of the data; secondly, the affected coefficient is quantified based on the node correlation and the initial data is corrected, solving the problem that single node data independent calculation is difficult to reflect the interaction of power supply, heat supply and other related interactions in the park energy network, for example, the energy ratio, conversion efficiency and other factors are corrected by the influence coefficient of the upstream node on the downstream node, so that the final energy structure data can accurately reflect the interaction between nodes, eliminating the data island bias caused by neglecting the correlation; finally, the final energy structure data obtained through this process not only retains the details of the single node energy consumption characteristics, but also conforms to the operation logic of the overall energy network of the park, providing a high-quality data basis for subsequent calculation of dynamic emission factor matrix and carbon emission simulation, improving the reliability and practicality of carbon emission analysis results from the source, and ensuring that the analysis conclusion can truly reflect the internal correlation between energy consumption and carbon emission in the park.

[0050] Optionally, the hierarchical adaptive calculation model further comprises a real-time feedback layer; The hierarchical adaptive calculation model is used to obtain the dynamic emission factor matrix of the energy consumption node according to the final energy structure data, and further comprises: Through the real-time feedback layer, real-time carbon emission concentration data collected by the real-time carbon emission monitoring device arranged in the park is obtained; According to the deviation between the real-time carbon emission concentration data and the estimated carbon emission concentration data generated by the dynamic emission factor matrix, a feedback compensation factor is constructed; The dynamic emission factor matrix is iteratively optimized online by the feedback compensation factor, and the emission factor value in the dynamic emission factor matrix is updated.

[0051] Specifically, high-precision carbon emission monitoring devices (such as non-dispersive infrared gas analyzers, laser spectrum detectors, etc.) are deployed around the key energy consumption nodes in the park (such as boiler room exhaust ports, high-energy consumption workshop ventilation ports, park total discharge ports, etc.), the data collection frequency is set to once every 10 minutes, real-time CO2 concentration, temperature, air pressure and other data are collected, the original data is sent to the data receiving end of the real-time feedback layer through a wireless transmission module (such as LoRa, 5G), the receiving end performs preliminary verification on the data (such as eliminating abnormal values exceeding the device range and converting the concentration data to standard state values through air pressure calibration), and finally a time-stamped real-time carbon emission concentration data set is formed.

[0052] Based on the dynamic emission factor matrix and the final energy structure data, the estimated data (diffusion coefficient is adjusted in real time according to meteorological data) can be generated by the formula: estimated carbon emission concentration = (Σ energy consumption × corresponding emission factor) / park space volume × diffusion coefficient; then the deviation value is calculated, using the relative deviation formula: deviation = (real-time concentration - estimated concentration) / estimated concentration × 100%, if the real-time concentration is 800 ppm, and the estimated concentration is 750 ppm, then the deviation is 6.67%; then a compensation factor is constructed according to the size and continuous trend of the deviation, and a deviation threshold (such as ± 5%) is set, when the absolute value of the deviation ≤ 5%, the compensation factor is 1 (no adjustment is needed); when the deviation > 5% (real-time is higher than the estimated value), the compensation factor = 1 + deviation × 0.3 (emission factor is amplified); when the deviation < -5% (real-time is lower than the estimated value), the compensation factor = 1 + deviation × 0.3 (emission factor is reduced), for example, the compensation factor corresponding to the above 6.67% deviation is 1 + 6.67% × 0.3 ≈ 1.02, and if the deviation direction is consistent for 3 times in a row, the adjustment coefficient of the compensation factor is increased from 0.3 to 0.5 to enhance the correction strength. Then the energy consumption node and energy type corresponding to the deviation are located (such as the deviation is mainly caused by the coal consumption of the boiler room), and the emission factor of the energy type corresponding to the node is extracted from the dynamic emission factor matrix (such as the current coal emission factor is 2.6 kgCO2 / kg); then the updated value is obtained by multiplying the original emission factor by the feedback compensation factor (such as 2.6 × 1.02 ≈ 2.652 kgCO2 / kg); at the same time, an iteration period (such as updated every hour) is set, and the change amount of the emission factor and the corresponding deviation data are recorded after each update, and the compensation effect is counted through a sliding window (such as the past 24 hours), if the deviation of a node still exceeds the threshold after multiple compensations, the baseline value of the underlying basic emission factor layer is triggered to recheck, to ensure that the dynamic emission factor matrix continuously fits the actual emission situation, and finally the optimized dynamic emission factor matrix is output.

[0053] In the embodiment of the present application, the real carbon emission concentration data collected by the high-precision monitoring equipment in the park is introduced through real-time feedback layer, breaking the limitation of relying only on final energy structure data and theoretical model calculation to estimate the concentration, providing an objective and real reality benchmark for the optimization of emission factors, and avoiding the problem of disconnection between estimation and reality caused by model assumptions (such as fixed diffusion coefficient and theoretical value deviation of energy conversion efficiency); secondly, by calculating the relative deviation of real-time concentration and estimated concentration and constructing a differentiated feedback compensation factor (such as dynamically adjusting the compensation coefficient according to the deviation size, and enhancing the correction intensity combined with the deviation trend), the fine and targeted adjustment of the emission factor is realized, which not only avoids the factor fluctuation caused by excessive correction under small deviation, but also quickly responds when there is a significant deviation, ensuring that the correction direction and amplitude are consistent with the actual emission change; thirdly, based on the online iterative optimization mechanism of the feedback compensation factor, the dynamic emission factor matrix can be updated in real time according to the park working conditions (such as equipment load change, weather condition fluctuation, energy supply structure adjustment), eliminating the lag of traditional static emission factor or fixed period update factor, so that the emission factor is always adapted to the current node running state and the overall environment of the park; finally, the data precision of the dynamic emission factor matrix optimized by real-time feedback is greatly improved, which directly provides more actual core calculation parameters for subsequent carbon emission simulation, thereby reducing the simulation error of carbon emission results, making the park carbon emission analysis results more truly reflect the actual emission situation, providing more reliable data support for real-time decision-making such as short-term emission reduction regulation and equipment operation optimization, and also enhancing the adaptation ability of the hierarchical adaptive calculation model to complex and dynamic park scenarios.

[0054] Optionally, the carbon emission simulation of the energy consumption node is performed by a hybrid model based on the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data, to obtain the carbon emission result of the energy consumption node, including: The energy consumption data is split according to energy types to obtain the sub-consumption amount of each energy type of the energy consumption node; The sub-consumption amount is multiplied by the emission factor of the corresponding energy type in the dynamic emission factor matrix element by element to obtain the basic carbon emission amount of the energy consumption node; According to the multi-source real-time data, a bidirectional long short-term memory network model is used for fitting to obtain a nonlinear correction amount of the energy consumption node; The basic carbon emission amount and the nonlinear correction amount are weighted and fused to obtain a preliminary carbon emission result of the energy consumption node; The preliminary carbon emission result is subjected to outlier detection and elimination, and is labeled in combination with the collection frequency of the energy consumption data to generate the carbon emission result containing time stamp, energy type sub-carbon emission value and total carbon emission value.

[0055] Specifically, the energy types (such as electricity, natural gas, coal, biomass energy, etc.) contained in the first selected data are classified and identified by energy metering account or sensor data (such as electricity data marked as electricity-10kV, natural gas data marked as gas-municipal), and the total energy consumption data of the node is divided into the consumption of each single energy type, for example, the monthly total energy consumption of a workshop is 10000 kWh equivalent, after splitting, the electricity consumption is 6000 kWh, the natural gas consumption is about 4000 kWh (converted according to the calorific value), and the sub-consumption list of each energy type is formed. Multiply the sub-consumption amount by the corresponding emission factor of each energy type in the dynamic emission factor matrix element by element, first match the real-time emission factor of each energy type in the dynamic emission factor matrix (such as the electricity emission factor is 0.5 kgCO2 / kWh, and the natural gas is 2.0 kgCO2 / m 3 ), and then perform the calculation of consumption amount x emission factor for each energy type, for example, the electricity sub-consumption of 6000 kWh multiplied by 0.5 kgCO2 / kWh gets 3000 kgCO2, and the natural gas sub-consumption of 2000 m 3 multiplied by 2.0 kgCO2 / m 3 gets 4000 kgCO2, and the calculation results of all energy types are accumulated to get the basic carbon emission of the node (such as 7000 kgCO2). Based on multi-source real-time data, a bidirectional long short-term memory network model is used to fit to obtain a nonlinear correction amount, first, the multi-source real-time data (such as device real-time load rate, running temperature, environmental humidity, production rhythm, etc.) are standardized (converted to [0, 1] interval numerical value), and the processed data are divided into training samples (containing historical data and corresponding actual carbon emission deviation) according to time sequence, and the model is trained to learn the nonlinear correlation in the data (such as the carbon emission nonlinearly increases when the load rate exceeds 80%); the real-time data of the current node are input into the trained model, and the model output value is the nonlinear correction amount of the basic carbon emission (such as the basic value of 7000 kgCO2, the model outputs a correction amount of +350 kgCO2, reflecting the additional emission caused by high load). The basic carbon emission and the nonlinear correction amount are weighted and fused to obtain the preliminary carbon emission result, and the fusion weights of the two are set (such as the basic carbon emission weight 0.8, the nonlinear correction amount weight 0.2, the weights can be dynamically adjusted according to the historical simulation error, and the basic value weight is increased when the error is small), and the formula preliminary result = basic value x basic weight + correction amount x correction weight is calculated, for example, the basic value of 7000 kgCO2 x 0.8 + 350 kgCO2 x 0.2 = 5600 + 70 = 5670 kgCO2, and the preliminary carbon emission result of the node is obtained.

[0056] The preliminary carbon emission results are subjected to outlier detection and elimination, and combined with the collection frequency of energy consumption data to complete the labeling. The interquartile range (IQR) method is used to detect outliers. The upper quartile (Q3) and lower quartile (Q1) of the preliminary result sequence are calculated. Values outside the range of [Q1-1.5×IQR, Q3+1.5×IQR] are determined as outliers and eliminated (such as abnormally high values caused by transient equipment failure). According to the collection frequency of energy consumption data (such as every 15 minutes), time stamps (such as 2025-10-24 08:00:00, 08:15:00) are added to the retained valid data, and the sub-item carbon emission values (such as electricity 3000 kg, natural gas 4000 kg) and total carbon emission value (5670 kg) of each energy type are summarized. Finally, a structured carbon emission result containing time stamp, sub-item value, and total value is generated.

[0057] In the embodiments of the present application, first, the energy consumption data is split according to the energy type, and the basic carbon emission is calculated by matching the dynamic emission factor, breaking the traditional rough mode of multiplying total energy consumption by a fixed average factor. The emission factors of different energies (such as electricity and natural gas) differ significantly, and after splitting, the actual emission contribution of each energy can be accurately matched by calculating each type, avoiding the basic error caused by the total energy consumption average calculation covering up the differences between items, and laying a precise underlying data foundation for subsequent simulation results. Secondly, the Bi-LSTM (Bidirectional Long Short-Term Memory) is used to fit the nonlinear correction quantity of multi-source real-time data, and the nonlinear correlation problem between carbon emission and factors such as device load and environmental temperature in the actual scene (such as when the load rate exceeds 80%, the emission will increase nonlinearly due to the decrease of device efficiency) is solved. Compared with the traditional linear correction model, Bi-LSTM can more accurately capture the complex dynamic correlation in the time series, making the correction quantity more in line with the actual working condition, and greatly reducing the deviation caused by the disconnection between linear assumption and actual scene. Thirdly, the basic emission and nonlinear correction quantity are fused through dynamic weight, rather than simply superimposed, and the contribution ratio of the two can be adjusted in real time according to the historical simulation error (such as increasing the weight of the basic value when the error is small), which not only retains the stability of the basic calculation, but also plays the dynamic adaptability of the nonlinear correction, avoiding the result imbalance caused by the dominance of a single factor. Finally, the abnormal value detection and elimination (such as abnormal high value caused by device failure) and timestamp labeling not only eliminate the interference of noise data on the result, but also give the carbon emission result a time dimension and itemized details (such as the proportion of carbon emission of each energy type), making the result not only accurately reflect the actual emission level of the node, but also support subsequent periodical emission trend analysis, key energy emission tracing and other refined management needs. In summary, the whole process forms a precise conversion link from raw data to high-quality carbon emission results, and compared with the traditional method, the simulation error is significantly reduced, and the result is more in line with the actual energy consumption and emission rules of the park, which can not only provide accurate data support for short-term emission reduction regulation (such as real-time adjustment of high-emission device load), but also provide detailed basis for long-term energy efficiency optimization (such as identifying high-emission energy types).

[0058] Optionally, according to the carbon emission results of each energy consumption node, a dynamic analysis and explanation of carbon emission is performed through an attention mechanism to obtain a carbon emission analysis result of the park in the current time period, including: inputting the carbon emission results of the energy consumption nodes into an attention mechanism model; performing weight distribution on the energy type itemized carbon emission values and total carbon emission values in the carbon emission results through the attention mechanism model to determine the contribution degree of each energy consumption individual in the energy consumption node to carbon emission; performing dynamic analysis on the energy consumption individuals according to the contribution degree to obtain the carbon emission sources and key driving factors of the park; The carbon emission analysis result of the park in the current time period is generated in combination with the carbon emission sources and the key driving factors of the park.

[0059] Specifically, the carbon emission result of the energy consumption node is input into an attention mechanism model, the structured data in the carbon emission result is sorted, including the energy type sub-item carbon emission value (such as the emission amount corresponding to electricity and natural gas) at each time stamp, the total carbon emission value, and the associated node attribute (such as the equipment number and the function area type), the data is converted into a vector form recognizable by the model (such as splicing the sub-item value and the total emission value of each hour into a feature vector), and the model is input in the order of time sequence to provide structured input for subsequent weight calculation. The attention mechanism model allocates weights to the sub-item values and the total emission value in the carbon emission result to determine the contribution degree of each energy consumption individual: the model calculates the correlation strength between the emission data of each energy consumption individual (such as the power consumption of a certain gas-fired boiler or a certain production line) and the total carbon emission, for example, the cosine similarity between the natural gas sub-item emission value and the total emission value in a certain period is 0.92, which is significantly higher than that of other energy types, and the model allocates an attention weight of 0.45 to it; combined with the weight and the actual emission amount of the individual, the specific contribution proportion of the individual to carbon emission (such as 38%) is obtained through calculation by contribution degree = weight x sub-item emission value / total emission value x 100%, which clearly shows the emission influence degree of different individuals. According to the contribution degree, the energy consumption individual is dynamically analyzed to obtain the carbon emission sources and the key driving factors of the park, the energy consumption individuals are ranked in descending order of contribution degree, and the top 30% individuals in terms of contribution degree are selected as the main carbon emission sources (such as coal consumption of the boiler room and power consumption of the refrigeration station); for these main sources, the driving factors are analyzed in combination with their operation data (such as the daily operation time and the load rate of the coal consumption individual), for example, it is found that the contribution degree of the boiler room increases significantly from 8 to 12 o'clock on weekdays, and the load rate in the corresponding period reaches 90%, so the high load operation is determined as the key driving factor of the source; at the same time, the time sequence change of the contribution degree is tracked to identify the time nodes of emission sudden increase / sudden decrease and the corresponding individuals, and the driving factors are supplemented (such as equipment maintenance leading to a sharp drop in the emission of a production line). Finally, in combination with the carbon emission sources and the key driving factors of the park, the carbon emission analysis result of the current time period is generated, the core indicators are integrated, including the total carbon emission of the park, the contribution proportion of each main source (such as 45% for coal and 30% for electricity), the emission peak period and the value; the interpretation content is written to clearly show the specific influence of the key driving factors (such as coal consumption being the primary source, mainly due to the full-load operation of the boiler room during the production peak period, which contributes 42% of the emission); the visual charts (such as the contribution degree pie chart, the emission time sequence curve and the driving factor annotation) are attached, and finally the analysis result that can reflect the overall emission level of the park and accurately locate the key emission sources and causes is formed, which provides a clear direction for emission reduction decision-making.

[0060] In the embodiment of the present application, the attention mechanism model is used to allocate weights to the sub-item values and the total emission value in the carbon emission result, breaking the rough mode of only looking at the total emission and ignoring individual differences in traditional analysis. By quantifying the correlation strength between each type of energy consumption individual (such as a single device or a single energy type) and the total carbon emission and calculating the contribution degree, the high-contribution emission individual (such as a natural gas consumption device contributing 38%) can be accurately located, avoiding the masking of key emission sources caused by average processing, allowing the park to clearly understand which individual is the core object of emission reduction, and providing clear targets for subsequent regulation. Secondly, the dynamic analysis based on the contribution degree breaks through the limitations of static analysis. By tracking the time sequence change of the contribution degree (such as the sudden increase in the contribution degree of the boiler room at 8-12 am on weekdays), and combining with the operation data to mine the key driving factors (such as high-load operation), both the normal emission rule (such as high emission in a fixed period) and the temporary impact (such as emission sudden drop caused by equipment maintenance) can be captured. Compared with the traditional static report of post-mortem summary, it can better adapt to the dynamic changes of carbon emission in the park in real time, helping managers respond to sudden emission fluctuations in time. Thirdly, the final analysis result combines data indicators and cause interpretation, solving the interpretability problem of traditional black-box output model. The report not only contains hard indicators such as total emission and contribution ratio, but also clearly marks the specific impact of key driving factors (such as coal consumption contributing 42% due to full-load operation during peak period), allowing managers to not only know how much emission is, but also understand why it is emitted, avoiding the problem of blindly developing emission reduction measures based on data values. Finally, from the actual management value of the park, this technology upgrades carbon emission analysis from general statistics to precise guidance. The park can develop targeted emission reduction schemes (such as adjusting production shifts and optimizing equipment load) for high-contribution equipment (such as boiler room) and key driving factors (such as peak load), rather than investing resources indiscriminately, greatly improving the emission reduction efficiency and resource utilization rate. At the same time, the dynamic tracking capability can also allow the park to evaluate the effect of emission reduction measures in real time (such as whether the contribution degree of a certain device decreases after adjustment), forming a closed-loop management of analysis-decision-evaluation, and effectively converting carbon emission data into actionable management strategies, promoting the park's carbon management from passive statistics to active regulation.

[0061] In combination Figure 2 As shown in the figure, the carbon emission analysis device of the present application comprises: A data acquisition unit for acquiring multi-source real-time data and energy consumption data of each energy consumption node in the park within a current time period; A feature extraction unit for extracting multi-dimensional key features of each energy consumption node from the multi-source real-time data of the energy consumption node according to the carbon emission analysis requirement; An energy structure analysis unit for determining the initial energy structure data of each energy consumption node according to the multi-dimensional key features of the energy consumption node; A correction unit is configured to correct the initial energy structure data according to an association relationship between the energy consumption node and other energy consumption nodes in the park, and obtain final energy structure data; An emission factor matrix construction unit is configured to obtain a dynamic emission factor matrix of the energy consumption node according to the final energy structure data through a hierarchical adaptive calculation model. An simulation unit is configured to simulate carbon emission of the energy consumption node through a hybrid model by combining the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data, and obtain a carbon emission result of the energy consumption node. An analysis unit is configured to perform dynamic analysis and interpretation of carbon emission through an attention mechanism according to the carbon emission result of each energy consumption node, and obtain a carbon emission analysis result of the park in the current time period.

[0062] The carbon emission analysis device has the same advantages as the carbon emission analysis method described above, and details are not repeated here.

[0063] In combination with Figure 3 As shown in the drawings, the electronic device of the present application comprises a processor and a memory for storing a computer program. The computer program, when loaded by the processor, enables the processor to execute the carbon emission analysis method described above.

[0064] The electronic device of the present application has the same advantages as the carbon emission analysis method described above, and details are not repeated here.

[0065] The computer readable storage medium of the present application has the same advantages as the carbon emission analysis method described above, and details are not repeated here.

[0066] The computer readable storage medium of the present application has the same advantages as the carbon emission analysis method described above, and details are not repeated here.

[0067] Although the present application is disclosed as above, the protection scope of the present application is not limited to this. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application, and these changes and modifications will fall within the protection scope of the present application.

Claims

1. A carbon emission analysis method characterized by, The method comprises the following steps: acquiring multi-source real-time data and energy consumption data of each energy consumption node in the park within a current time period; extracting multi-dimensional key features of each energy consumption node from the multi-source real-time data of each energy consumption node according to carbon emission analysis requirements; determining initial energy structure data of each energy consumption node according to the multi-dimensional key features of each energy consumption node; correcting the initial energy structure data according to the correlation between each energy consumption node and other energy consumption nodes in the park to obtain final energy structure data; obtaining a dynamic emission factor matrix of each energy consumption node according to the final energy structure data through a hierarchical self-adaptive calculation model; combining the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data, simulating carbon emissions of each energy consumption node through a hybrid model to obtain carbon emission results of each energy consumption node; performing dynamic analysis and interpretation of carbon emissions through an attention mechanism according to the carbon emission results of each energy consumption node to obtain carbon emission analysis results of the park within the current time period.

2. The carbon emission analysis method according to claim 1, characterized by, The method comprises the following steps: determining the type of the carbon emission analysis requirements according to the management scenarios of the park, wherein the type of the carbon emission analysis requirements comprises short-term emission reduction monitoring requirements, long-term policy making requirements and equipment energy efficiency optimization requirements; matching target data types corresponding to the multi-source real-time data of each energy consumption node according to the type of the carbon emission analysis requirements; extracting original features of corresponding dimensions from the target data types, wherein for the short-term emission reduction monitoring requirements, time dimension energy consumption peak-valley fluctuation features and device dimension real-time load features are extracted; for the long-term policy making requirements, space dimension functional area energy consumption difference features and policy dimension clean energy proportion change features are extracted; for the equipment energy efficiency optimization requirements, device dimension aging loss features and working condition dimension energy production-energy consumption correlation features are extracted; performing data preprocessing on the extracted original features of each dimension to obtain multi-dimensional key features of each energy consumption node.

3. The carbon emission analysis method according to claim 1, characterized by, The method comprises the following steps: analyzing and classifying the multi-dimensional key features through an energy structure identification model to determine multiple energy structure types and scenario correlation features of each energy consumption node; determining energy type proportions, energy conversion efficiencies and energy consumption fluctuation data of each energy consumption node according to the energy structure types; determining scenario influence data strongly related to energy consumption according to the scenario correlation features; taking the energy type proportions, the energy conversion efficiencies, the energy consumption fluctuation data and the scenario influence data of each energy consumption node as the initial energy structure data. The initial energy structure data is associated and corrected according to the association relationship of the energy consumption nodes in the park, and final energy structure data is obtained, comprising: According to the association relationship of each energy consumption node in the park, the influence coefficient of the energy consumption node is quantified; According to the influence coefficient, the energy type proportion, the energy conversion efficiency, the energy consumption fluctuation data and the scenario influence data are corrected to obtain the final energy structure data.

4. The carbon emission analysis method according to claim 3, characterized by, The hierarchical adaptive calculation model includes a basic factor matching layer, an efficiency correction layer, a scene dynamic adjustment layer and a bias calibration layer; According to the final energy structure data, the hierarchical adaptive calculation model obtains the dynamic emission factor matrix of the energy consumption node, comprising: Through the basic factor matching layer, according to the energy type proportion, combined with the preset industry benchmark emission factor library, the initial emission factor vector is obtained; Through the efficiency correction layer, according to the energy conversion efficiency and the energy consumption fluctuation data, the initial emission factor vector is corrected to obtain the efficiency corrected factor vector; Through the scene dynamic adjustment layer, the efficiency corrected factor vector is adjusted in real time according to the scenario influence data to obtain the scene adjusted factor matrix; Through the bias calibration layer, combined with the historical carbon emission evaluation data, the bias value of the scene adjusted factor matrix and the historical actual monitoring data is determined, and the scene adjusted factor matrix is self-calibrated and optimized based on the bias value to obtain the dynamic emission factor matrix.

5. The carbon emission analysis method according to claim 4, characterized by, The hierarchical adaptive calculation model further includes a real-time feedback layer; According to the final energy structure data, the hierarchical adaptive calculation model obtains the dynamic emission factor matrix of the energy consumption node, further comprising: Through the real-time feedback layer, the real-time carbon emission concentration data collected by the real-time carbon emission monitoring device arranged in the park is obtained; According to the deviation between the real-time carbon emission concentration data and the estimated carbon emission concentration data generated by the dynamic emission factor matrix, a feedback compensation factor is constructed; Through the feedback compensation factor, the dynamic emission factor matrix is iteratively optimized online to update the emission factor value in the dynamic emission factor matrix.

6. The carbon emission analysis method of claim 1, wherein, The combination of the dynamic emission factor matrix, the multi-source real-time data and the energy consumption data is simulated by a hybrid model to obtain the carbon emission result of the energy consumption node, comprising: The energy consumption data is split according to the energy type to obtain the sub-consumption of each energy type of the energy consumption node; The sub-consumption is multiplied by the emission factor of the corresponding energy type in the dynamic emission factor matrix to obtain the basic carbon emission of the energy consumption node; According to the multi-source real-time data, a bidirectional long short-term memory network model is used for fitting to obtain the nonlinear correction amount of the energy consumption node; The basic carbon emission and the nonlinear correction amount are weighted and fused to obtain the preliminary carbon emission result of the energy consumption node; The preliminary carbon emission results are subjected to outlier detection and removal, and labeled in conjunction with the collection frequency of the energy consumption data to generate the carbon emission results including timestamps, energy type-specific carbon emission values, and total carbon emission values.

7. The carbon emission analysis method according to claim 6, characterized by, The step of dynamically analyzing and interpreting carbon emissions based on the carbon emission results of each energy consumption node, using an attention mechanism, to obtain the carbon emission analysis results of the park within the current time period, includes: The carbon emission results of the energy consumption nodes are input into the attention mechanism model; The attention mechanism model is used to assign weights to the carbon emission values ​​of the energy type and the total carbon emission value in the carbon emission results, thereby determining the contribution of each energy consumer in the energy consumption node to carbon emissions. Based on the contribution level, a dynamic analysis is performed on the energy-consuming individuals to obtain the carbon emission sources and key driving factors of the park; By combining the carbon emission sources and key driving factors of the park, the carbon emission analysis results of the park during the current time period are generated.

8. A carbon emission analysis device, characterized by, include: The data acquisition unit is used to acquire multi-source real-time data and energy consumption data of each energy consumption node in the park within the current time period. The feature extraction unit is used to extract multi-dimensional key features of the energy consumption node from the multi-source real-time data of each energy consumption node according to the carbon emission analysis requirements. An energy structure analysis unit is used to determine the initial energy structure data of each energy consumption node based on the multi-dimensional key features of each energy consumption node. The correction unit is used to perform correlation correction on the initial energy structure data based on the correlation between the energy consumption nodes and other energy consumption nodes in the park, so as to obtain the final energy structure data. The emission factor matrix construction unit is used to obtain the dynamic emission factor matrix of the energy consumption node based on the final energy structure data through a hierarchical adaptive calculation model. The simulation unit is used to combine the dynamic emission factor matrix, the multi-source real-time data, and the energy consumption data to simulate the carbon emissions of the energy consumption node through a hybrid model, and obtain the carbon emission results of the energy consumption node. The analysis unit is used to perform dynamic analysis and interpretation of carbon emissions based on the carbon emission results of each energy consumption node through an attention mechanism, so as to obtain the carbon emission analysis results of the park in the current time period.

9. An electronic device, comprising: include: Processor and memory, the memory being used to store computer programs; When the computer program is loaded by the processor, it causes the processor to execute the carbon emission analysis method as described in any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the carbon emission analysis method as described in any one of claims 1-7.

Images