Intelligent carbon metering device and carbon emission calculation method thereof

Abstract

The application discloses an intelligent carbon metering device and a carbon emission calculation method thereof, and the method comprises the following steps: pre-processing multi-source data; using a deep learning model to identify the working conditions of each device, and simultaneously considering device aging correction to improve identification accuracy; establishing a bottom-up calculation model of carbon emission driven by energy consumption and a top-down calculation model based on emission concentration for concentration inversion; and obtaining the final carbon emission result through adaptive Kalman filtering fusion. The application identifies the device state in real time through AI, dynamically matches the emission factor, and adopts adaptive Kalman filtering to fuse the energy consumption and concentration double models, thereby significantly improving the accuracy and robustness of carbon metering. The method can effectively solve the problems of hysteresis, singleness and error accumulation in the traditional carbon metering method, realize real-time, accurate and traceable carbon emission calculation from the device level to the regional level, and realize the digitization, dynamicization and intelligentization of carbon metering.

Description

Technical Field

[0001] This invention belongs to the field of carbon emission monitoring and intelligent metering technology, and relates to an intelligent carbon metering device and its carbon emission calculation method, specifically an intelligent carbon metering device and its carbon emission calculation method based on the fusion of multi-source sensing and artificial intelligence. Background Technology

[0002] Current carbon emission calculations are mainly divided into two categories:

[0003] 1. Indirect calculation methods based on energy consumption statistics rely on fixed emission factors, cannot dynamically reflect changes in the environment and power grid structure, and have low accuracy and poor timeliness, and cannot distinguish differences in equipment efficiency.

[0004] 2. While the direct method based on gas concentration monitoring can reflect actual emissions, its measurement accuracy is unstable due to factors such as sensor layout and ventilation disturbances. This method is also costly, applicable only to large sources, and has a narrow coverage area.

[0005] In addition, traditional systems cannot integrate multi-source heterogeneous data (such as power, temperature and humidity, CO2 concentration, etc.) and lack the ability to identify operating conditions, resulting in systematic biases in emission calculations.

[0006] Based on this, the present invention proposes an intelligent carbon metering method that can integrate energy consumption and concentration dual models, adaptively correct emission factors, and has AI operating condition recognition and intelligent source tracing functions, so as to break the extensive mode of traditional methods and realize precise monitoring and tracking of fine-grained emission sources. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing an intelligent carbon metering device and its carbon emission calculation method. This method combines energy consumption monitoring, CO2 concentration detection, environmental compensation, dynamic emission factor matching, and multi-model fusion algorithms to achieve real-time accurate calculation and intelligent source traceability of equipment-level carbon emissions.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A carbon emission calculation method, which collaboratively processes data from multiple devices in a study area, includes the following steps:

[0010] S1: Preprocess the multi-source data to obtain synchronized device-level electrical parameter data and environmental data on a unified time axis, and generate regional-level concentration field data from the spatial data;

[0011] S2: Based on the equipment-level electrical parameter data obtained from S1, the power signals of each device are analyzed using a deep learning model to identify the operating conditions of each device, while also considering equipment aging correction to improve identification accuracy.

[0012] S3: Based on the three-element relationship of equipment type, identified operating conditions, and environment, match dynamic emission factors and establish a bottom-up calculation model for energy consumption-driven carbon emissions.

[0013] Simultaneously, based on the regional concentration field data in S1, a top-down calculation model for concentration inversion based on emission concentration is established;

[0014] S4: The bottom-up and top-down computational models are fused using adaptive Kalman filtering to obtain the final regional carbon emission results.

[0015] In the above technical solution, step S1 further comprises: uniformly calibrating the clocks of multiple data acquisition devices installed in the area using a time synchronization protocol to generate synchronized time-domain multi-source data, including high-frequency sampled electrical parameter data, including power, voltage, and current, and low-frequency sampled environmental data, including CO2 concentration, temperature, humidity, and air pressure data; interpolating the low-frequency data and mapping it uniformly to the main time axis; and using inverse distance weighting to generate a regional concentration field for the spatial CO2 concentration data.

[0016] Furthermore, in S2, the power sequence of a single device within the current sliding time window is normalized, and the probability of each working condition category is obtained based on a CNN-LSTM deep network. The CNN-LSTM deep network is configured with two one-dimensional convolutional layers, a pooling layer, a normalization layer, and a single-layer or double-layer LSTM network. The two one-dimensional convolutional layers extract the local waveform features and frequency domain features of the power sequence, the pooling layer and the normalization layer complete the feature dimensionality reduction processing, the single-layer or double-layer LSTM network captures the temporal change dependency of the working condition switching, and the probability of each working condition is output after passing through a fully connected layer and a Softmax activation function.

[0017] After first correcting the equipment power based on the equipment aging coefficient, the CNN-LSTM deep network is used to obtain the probability of the current identified working condition. The working condition label obtained at the previous identification time is matched to perform a smooth decision and output the current working condition label of the equipment.

[0018] Furthermore, the equipment aging coefficient is a function related to the cumulative operating time of the equipment. In addition to using the equipment aging coefficient to correct the original power characteristics of the equipment, the equipment aging correction also includes concatenating the cumulative operating time and maintenance frequency as additional features with the output of the LSTM and then feeding them into the fully connected layer. During the model training phase, labeled historical operating condition samples are used for supervised training, and the loss function is cross-entropy. The training data covers different equipment types, different load levels, and different aging stages. During the online phase, only forward inference is performed to obtain the current operating condition label.

[0019] Furthermore, the process of establishing the bottom-up calculation model for energy-driven carbon emissions is as follows:

[0020] Based on the standard emission factor, dynamic emission factors are formed by dynamically adjusting the standard emission factor according to equipment type, equipment operating conditions and environmental conditions. Specifically, the product of the environmental correction function, the real-time power grid emission correction function and the standard emission factor is used as the dynamic emission factor. The sum of the power data of each device in the study area and time period and the corresponding dynamic emission factor is the energy consumption-driven carbon emission.

[0021] Furthermore, the environmental correction function is a joint correction form formed by the product of temperature correction, humidity correction, and air pressure correction. The real-time grid emission correction function is: for the provincial or regional grid boundary where the studied area is located, the product of the proportion of the power generation of each energy type in the total power generation at time t within the same boundary and the unit power generation emission factor of that energy type is obtained, and the sum of the correction emission factors of all energy types within the boundary at time t is the real-time grid emission correction function.

[0022] Furthermore, the process of establishing the top-down calculation model for concentration inversion based on emission concentration is as follows:

[0023] The total regional emissions are calculated using the CO2 concentration change rate and spatial ventilation characteristics, based on a modified mass balance equation.

[0024]

[0025] in: V represents the carbon emission rate, expressed in kg / s; V represents the effective volume of the monitored spatial unit. When only a single independent space is monitored, V represents the volume of that single space. When the monitoring area contains multiple relatively independent subspaces, the emissions of each subspace are calculated separately, and then the total emissions of the area are summed. CO2 density; The rate of change of the volume-weighted average CO2 concentration C, calculated from the regional concentration field, within the current sliding time window; Ventilation rate; This represents the background concentration.

[0026] Furthermore, S4 specifically includes:

[0027] The actual carbon emissions x within the discrete time window k k As a state variable, the carbon emissions E obtained from the bottom-up calculation model energy (k) represents the prior state variables of the Kalman filter, and the carbon emissions E obtained from the top-down computation model. concentration (k) is used as the observation for Kalman filtering. The Kalman prediction and update steps are performed to obtain the fused carbon emissions E. fuse(k);

[0028] Real-time calculation of absolute error |E concentration (k)-E energy (k)|, when the absolute error exceeds a preset threshold, the process noise covariance Q in the Kalman filter is dynamically increased. k or R k And trigger an exception flag; otherwise, automatically decrease Q. k R k To improve the signal-to-noise ratio of the filter.

[0029] Furthermore, the method also includes intelligent source tracing and anomaly detection. When carbon emission data shows a sudden change or exceeds the standard, the operating conditions and environmental parameters are automatically traced back. If the regional concentration rises within the time window while the outdoor background concentration rises synchronously and the power of each device does not change significantly, it is determined to be environmental interference. If the power of a certain device suddenly increases and there is a significant linkage with the power of its adjacent exhaust device, it is determined to be an abnormal device coupling and an alarm is triggered.

[0030] A smart carbon metering device includes: an AI operating condition recognition module and a fusion computing module; wherein: the AI ​​operating condition recognition module is used to extract the time domain and frequency domain features of the equipment power signal based on a CNN-LSTM deep network, taking into account equipment aging, and simultaneously capture the operating condition change pattern in the time series, and output the equipment operating condition label.

[0031] The fusion computing module is used to run two models in parallel: a bottom-up calculation of carbon emissions driven by energy consumption and a top-down calculation of concentration inversion based on emission concentration. By dynamically adjusting the error weights through adaptive Kalman filtering, the optimal fusion of the results of the two models is achieved, and the carbon emission results are output.

[0032] The beneficial effects of this invention are:

[0033] This invention proposes a high-precision, real-time, traceable carbon emission calculation method through multi-source synchronous acquisition, AI operating condition identification, dynamic emission factor correction, and multi-model fusion algorithm. This method can effectively solve the problems of lag, uniformity, and error accumulation in traditional carbon measurement methods, and realize real-time, accurate, and traceable carbon emission calculation from the equipment level to the regional level, achieving the digitalization, dynamism, and intelligence of carbon measurement. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of one implementation process of the method of the present invention;

[0035] Figure 2 This is a schematic diagram of a specific embodiment of the system of the present invention. Detailed Implementation

[0036] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Example 1

[0038] A flowchart of a carbon emission calculation method according to the present invention is shown below. Figure 1 As shown, this method relies on multi-source sensor data acquisition and achieves real-time quantification, dynamic correction, and intelligent source tracing of carbon emissions through the fusion of energy consumption parameters, CO2 concentration, environmental data, and AI operating condition recognition results. This method describes a multi-device-single-region collaborative processing flow: multiple intelligent carbon metering devices are installed on the monitored equipment or regional nodes. Each device collects its own power, local CO2, and environmental data in real time. A data processing center, such as a cloud platform, simultaneously processes device-level time-series data and regional-level concentration field data on a unified time axis. This invention can output real-time device-level emission results in the steps, or ultimately output the total regional emissions within a set time window. In a specific embodiment of this invention, device-level results are updated at the second or minute level, and regional-level results are updated at 1-minute intervals, with the concentration change rate calculated using a 5-minute sliding window. The method includes the following steps:

[0039] S1: Multi-source data preprocessing

[0040] Multiple data acquisition devices installed in the area are clock-calibrated uniformly using a time synchronization protocol to generate synchronized time-domain multi-source data, including high-frequency sampled electrical parameter data (power, voltage, current) and low-frequency sampled environmental data (CO2 concentration, temperature, humidity, air pressure). The low-frequency data is interpolated and uniformly mapped to the main time axis. The spatial CO2 concentration data is then weighted by inverse distance to generate a regional concentration field. According to a specific embodiment of the present invention, the process can be as follows:

[0041] The cloud platform receives synchronized time-series data uploaded by multiple smart carbon metering devices, achieving ±1ms synchronization via the IEEE 1588 PTP protocol, synchronizing the timestamps of each device to within milliseconds. High-frequency data preferably refers to electrical parameters such as power, voltage, and current, with a sampling frequency of 10Hz to 1kHz; low-frequency data preferably refers to CO2 concentration, temperature, humidity, and air pressure data, with a sampling frequency of 0.2Hz to 1Hz. Low-frequency data is uniformly mapped to the main time axis after cubic spline interpolation; spatial data, such as CO2 concentration, undergoes spatial representativeness correction, i.e., based on the location of each CO2 sensor, an inverse distance weighted (IDW) algorithm is used to generate a regional concentration field, ensuring that the concentration data reflects the overall region rather than single-point anomalies.

[0042]

[0043] in, Let (x, y) be the concentration at the spatial point (x, y). Let i be the concentration detected by the i-th sensor. For spatial points The distance to the i-th sensor, where n is the number of sensors.

[0044] After generating the regional concentration field, the volume-weighted average concentration C is then calculated. avg (t) is used for subsequent concentration inversion.

[0045] S2: AI-powered fine-grained identification of equipment operating conditions

[0046] Equipment operating conditions directly affect emission factors and energy efficiency, thus requiring precise identification and correction of operating conditions. This invention utilizes a CNN-LSTM deep learning model to analyze power signals and identify equipment operating conditions such as standby, loading, no-load, and startup; combined with aging coefficients... Modify features to improve recognition accuracy.

[0047] According to a specific embodiment of the present invention, based on a CNN-LSTM fusion structure: the CNN layer extracts the time-domain and frequency-domain features of the power signal; the LSTM layer captures the time-series dependencies; and outputs device condition labels (such as standby, stable, idle, loaded, fault, etc.); the specific process of condition prediction is as follows:

[0048] The CNN-LSTM operating condition recognition model takes the power sequence of a single device within a sliding time window of length L as input, where L can range from 60s to 300s. The network structure may include two one-dimensional convolutional layers to extract local waveform and frequency domain features, pooling and normalization layers for dimensionality reduction, and one or two LSTM layers to extract the timing dependency of operating condition switching. Finally, it outputs the probabilities of operating conditions such as standby, idle, stable, loading, startup, and fault through fully connected layers and Softmax.

[0049] Equipment aging factor:

[0050]

[0051] Power characteristic correction:

[0052]

[0053] Equipment aging correction is preferably directly embedded in the identification model: on the one hand, it is used to correct the original power characteristics; on the other hand, it can also incorporate the cumulative operating time t. run Features such as maintenance frequency or efficiency decay coefficient are added as supplementary features and concatenated with the LSTM output before being fed into the fully connected layer. A smooth decision is then made by combining the aging-corrected features with the previous time-stamped condition label to output the current condition. This correction reduces the condition identification error from 15% to 3%.

[0054] During the model training phase, labeled historical operating condition samples are used for supervised training. The preferred loss function is cross-entropy. The training data covers different equipment types, different load levels, and different aging stages. During the online phase, only forward inference is performed to obtain the current operating condition label.

[0055] S3: Dynamic Emission Factor Matching

[0056] The system matches dynamic emission factors based on the identified three-element relationship between equipment type, operating condition, and environment. :

[0057]

[0058] And update the emission factor database in real time; among which The standard emission factor is determined according to national or international standards (such as the IPCC guidelines); The environmental correction function in this application preferably adopts a joint correction form of temperature T, humidity H, and air pressure P, i.e., f(T,H,P)=f(T)×f(H)×f(P). According to a specific embodiment of the present invention, the temperature correction term is... The air pressure correction term is Humidity correction item is (H represents relative humidity %RH, with a reference humidity of 50%RH);

[0059] Real-time grid emission correction function:

[0060]

[0061] This is a correction function for the real-time emission structure of the regional power grid corresponding to the power consumption boundary of the device. It is preferably obtained according to the provincial or regional power grid boundary where the enterprise or factory is located, and is not constructed individually for a single device, nor is it used indiscriminately for the entire city. In the formula, m represents the number of energy types in the regional power grid. Let be the power generation of the i-th energy source (thermal power, hydropower, wind power, photovoltaic power, nuclear power, etc.) at time t. E is the emission factor per unit of electricity generated from this energy source. totalgrid (t)=ΣE i (t) represents the total power generation within the same boundary.

[0062] S4: Dual-model carbon emission calculation and fusion

[0063] This method integrates an indirect calculation model based on energy consumption (i.e., a bottom-up calculation model of carbon emissions driven by energy consumption) and a direct calculation model based on emission concentration (i.e., a top-down calculation model of concentration inversion based on emission concentration), performs parallel calculations, and achieves result consistency and error self-correction through a dynamic fusion algorithm.

[0064] The bottom-up carbon emission calculation model driven by energy consumption calculates the energy consumption-driven carbon emissions by summing the products of the power data of each device within the study area and time period and their corresponding dynamic emission factors. For a single device, when the system collects the device's power data... At that time, it can be based on the corresponding dynamic emission factor. Calculate carbon emissions:

[0065]

[0066] Total energy consumption drives carbon emissions as follows:

[0067]

[0068] The top-down calculation model for concentration inversion based on emission concentration is as follows:

[0069] The total regional emissions are calculated using the CO2 concentration change rate and spatial ventilation characteristics, based on a modified mass balance equation.

[0070]

[0071] in: V represents the carbon emission rate, expressed in kg / s; V is the effective volume of the monitored spatial unit. When monitoring only a single independent space, V represents the volume of that single space. When the monitoring area contains multiple relatively independent subspaces, the emissions of each subspace are calculated separately, and then summed to obtain the total emissions of the area. V is preferably adjusted for temperature and pressure, i.e. ; The density of CO2 is 1.977 kg / m³. 3 ; C is the rate of change of the volume-weighted average CO2 concentration C, calculated from the regional concentration field, within the current sliding time window; Q is the ventilation volume of the monitored space unit, which can be calculated from the rated air volume of the fresh air unit or exhaust fan, the opening degree of the air valve, and the differential pressure sensor; in natural ventilation scenarios, it can also be estimated from the opening area of ​​doors and windows, the differential pressure ΔP, and the empirical coefficient k, for example, it can be... , where k is the spatial coefficient. For the background concentration, the average CO2 concentration of outdoor or upwind background nodes within the same time window is preferred.

[0072] To reduce measurement errors and model uncertainties, the system integrates bottom-up (energy consumption model) and top-down (concentration inversion model) results. The multi-model fusion uses the actual emissions x within a discrete time window k. k As a state variable, prior estimates are first obtained from the bottom-up energy consumption model. This prior reflects the emission results calculated based on power data within the current time window; then, the observations are obtained from the top-down concentration inversion model. This observation reflects the emission results obtained by inverting CO2 concentration changes, ventilation volume, and background concentration within the same time window.

[0073] State equations (based on energy consumption):

[0074]

[0075] Observation equation (based on concentration inversion):

[0076]

[0077] Among them: dynamic adjustment based on equipment operating condition stability; dynamic adjustment based on ventilation stability.

[0078] The Kalman update process can be represented as:

[0079]

[0080]

[0081]

[0082] in, The process noise covariance is determined based on the power fluctuation variance, operating condition switching frequency, and equipment state stability. To observe the noise covariance, it was determined based on CO2 sensor noise, ventilation fluctuations, and background concentration stability; assuming the concentration model can directly characterize the total emissions, H... k Option 1 is acceptable.

[0083] The aforementioned integrated carbon emissions can be equivalently expressed as:

[0084]

[0085] in, , The weights are dynamically adaptive based on model confidence, and change dynamically according to the inverse variance normalization rule, i.e.:

[0086] ,

[0087] in, The confidence variance of the bottom-up model. The confidence variance of the top-down model essentially corresponds to the variance in adaptive Kalman filtering. , .

[0088] Real-time calculation of absolute error |E concentration (k)-E energy (k)|, when the absolute error exceeds a preset threshold, the process noise covariance Q in the Kalman filter is dynamically increased. k or R k And trigger an exception flag; otherwise, automatically decrease Q. k R k To improve the filtering signal-to-noise ratio. When the system is running stably, , Automatically reduced to improve the filter signal-to-noise ratio.

[0089] S5: Intelligent Traceability and Anomaly Detection

[0090] If carbon emission data shows sudden changes or exceeds the standard, the system will automatically backtrack the operating conditions and environmental parameters:

[0091] If the concentration in the area increases within the time window while the outdoor background concentration increases synchronously and the power of each device does not change significantly, it is determined to be environmental interference; if the power of a certain device suddenly increases and there is a significant linkage with the power of its adjacent exhaust device, it is determined to be an abnormal device coupling and an alarm is triggered.

[0092] S6: Cloud Visualization

[0093] The integrated data is uploaded to the cloud platform to enable trend analysis, anomaly tracing, and optimization decisions.

[0094] Example 2

[0095] The following uses an industrial fixed scenario as an example to illustrate the implementation process of this invention:

[0096] (1) Scenario and Hardware Deployment: Three monitored devices were set up in an electronic assembly workshop: an air compressor (A), a reflow oven (B), and an air conditioning unit (C). Each device was equipped with an energy acquisition unit. Four CO2 monitoring nodes and one outdoor background node were arranged in the workshop, and temperature, humidity, air pressure, and differential pressure or air volume detection modules were configured. Each terminal uploaded data via RS485 or Ethernet, and the platform was synchronized according to the IEEE 1588PTP unified clock.

[0097] (2) Multi-source data preprocessing: Electrical parameter data were sampled at 100Hz, and CO2 and environmental parameters were sampled at 1Hz; the platform interpolated the low-frequency data to the 1Hz main time axis and used the IDW algorithm to construct the workshop CO2 concentration field, and further obtained the volume-weighted average concentration C. avg(t). The effective volume V of the workshop is corrected based on the measured geometric volume V0 and in combination with temperature and air pressure.

[0098] (3) Operating condition identification: The system analyzes the power sequence of each device using a 60s sliding time window, and identifies that device A is in a loading condition, device B is in a stable operating condition, and device C is in an intermittent start-stop condition; at the same time, it calculates the aging correction coefficient of each device based on the cumulative running time and corrects the power characteristics.

[0099] (4) Dynamic emission factor matching: The platform matches the EF of each device from the dynamic emission factor library according to the three-dimensional relationship of equipment type, operating condition and environment. dyn A, EF dyn B and EF dyn C, and combined with the real-time emission correction function f of the power grid in the region. grid (t) Update the current emission factor.

[0100] (5) Bottom-up calculation: During the time window from 10:00 to 10:01, the average power of equipment A, B, and C is 18kW, 42kW, and 12kW, respectively; the energy consumption model emission E during this time window is calculated by combining their respective dynamic emission factors. energy (10:01) = 0.86 kg CO2 / min.

[0101] (6) Calculated from top to bottom: Within the same time window, the regional average CO2 concentration increased from 612 ppm to 645 ppm, and the background concentration C bg =422ppm, ventilation volume Q=0.82m3 / s, E is calculated using the mass balance equation. concentration (10:01) = 0.91 kg CO2 / min.

[0102] (7) Fusion computing: based on E energy As a priori, with E concentration For the measurement, Q is set by combining the stability of the operating conditions and the stability of the ventilation. k R k After performing the Kalman update, the merged emissions E are obtained. fuse (10:01) = 0.89 kg CO2 / min. The platform also retains the single-equipment energy consumption model results as an equipment-level emission reference, and uses E... fuse As part of the region's total emissions.

[0103] (8) Anomaly determination: If the regional concentration rises within a certain time window and the outdoor background concentration rises synchronously, and the power of each device does not change significantly, the system determines that it is an environmental interference; if the power of device B suddenly increases and there is a significant linkage with its neighboring exhaust device in the power correlation matrix, it is determined that the device coupling is abnormal and an alarm is triggered.

[0104] (9) Cloud display: The platform refreshes the carbon emission curves of equipment level and region level every 1 minute, and generates cumulative emission reports by shift, hour and day, providing a basis for process optimization, energy-saving transformation and anomaly tracing.

[0105] Example 3

[0106] A smart carbon metering device includes a multi-parameter acquisition module, an edge processing and synchronization module, and a cloud platform, such as... Figure 2 As shown, specifically:

[0107] 1. The multi-parameter acquisition module includes the following:

[0108] Power Acquisition Unit: Utilizing a Hall effect sensor as the core sensing element, and paired with the ADE7755 professional metering chip, it achieves synchronous sampling of voltage and current. The voltage channel uses a voltage divider resistor network to step down the mains voltage to the chip's operating range, while the current channel acquires signals through a manganese-copper shunt or a miniature current transformer, ensuring sampling stability under wide load conditions. The main control unit reads 32-bit high-precision power data via an SPI interface and integrates an RS485 communication module supporting the MODBUS protocol for data upload.

[0109] The CO2 acquisition unit, used as a direct monitoring method, captures changes in CO2 concentration around the equipment or at the exhaust port. Its core employs dual-path NDIR (non-dispersive infrared) technology. Compared to traditional sensors, NDIR technology offers advantages such as longer lifespan, stronger anti-interference capabilities, and less temperature drift. A single light source is split into a measurement path (4.26μm, corresponding to the characteristic absorption wavelength of CO2) and a reference path (3.9μm, with no CO2 absorption) via a beam splitter. The differential signal compensates for errors caused by ambient temperature fluctuations. The module incorporates a rapid preheating circuit, enabling rapid tracking of concentration changes.

[0110] Environmental Compensation Module: Integrates high-precision temperature, humidity, and atmospheric pressure sensors. Because CO2 sensor readings are significantly affected by ambient temperature and humidity, this module performs real-time compensation and calibration of CO2 concentration data to ensure accuracy and reliability. Based on the Continuous Ambient Compensation principle, it integrates an SHT31 temperature and humidity sensor and an MS5611 barometer, via I... 2 C interacts with the main control MCU in real time. For the CO2 module, a three-dimensional compensation model for temperature, air pressure, and humidity is established; for the power module, the effect of ambient temperature on the manganese copper resistor is compensated.

[0111] 2. The edge processing and synchronization module includes the following:

[0112] Edge computing core: It has a built-in embedded processor and memory, and has functions such as sensor driving, data packaging and protocol conversion. It can run lightweight AI models to complete local data preprocessing and preliminary identification of working conditions, reducing the load on the cloud platform. At the same time, it supports data storage and breakpoint resume in the event of network interruption to avoid data loss.

[0113] Sensor synchronization mechanism: To avoid fusion errors caused by time differences in data from different sensors, data acquisition is triggered by a unified internal clock. At the same time, the timestamps of electrical energy, CO2, and environmental parameters are synchronized to within milliseconds using the IEEE 1588 PTP precise time protocol. This ensures that each set of acquired data corresponds to the device status at the same time node, providing a time-consistent data foundation for subsequent multi-model fusion.

[0114] 3. The cloud platform includes the following:

[0115] AI operating condition recognition module: It is used to extract the time domain and frequency domain features of the equipment power signal based on the CNN-LSTM deep network, taking into account the aging of the equipment, while capturing the operating condition change pattern in the time series and outputting the equipment operating condition label.

[0116] Fusion Computing Module: This module is used to run two parallel models: a bottom-up calculation of carbon emissions driven by energy consumption and a top-down calculation of concentration inversion based on emission concentration. It dynamically adjusts the error weights through adaptive Kalman filtering to achieve optimal fusion of the results from the two models and outputs the carbon emission results.

[0117] Dynamic emission factor library: Stores emission factors corresponding to different equipment types (such as motors, air conditioners, charging piles), operating conditions, and environmental conditions (such as high temperature, high pressure, etc.). The basic library is established based on IPCC guidelines, national standards, etc., and dynamic factors are matched in real time according to the three-element relationship of equipment, operating conditions, and environment.

[0118] Cloud-based visualization: Enables emission data uploading, trend analysis, anomaly tracing, and optimization decision-making.

[0119] This invention innovatively constructs a four-layer data fusion architecture encompassing energy characteristics, operating condition identification, dynamic factors, and concentration inversion. Through AI, it identifies equipment status in real time and dynamically matches emission factors. Furthermore, it employs an adaptive Kalman filter to fuse energy consumption and concentration models, significantly improving the accuracy and robustness of carbon metering. Ultimately, the system achieves closed-loop management capabilities, from precise carbon emission measurement to anomaly tracing.

[0120] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0121] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0122] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0123] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0124] The embodiments described above are merely some preferred embodiments of the present invention, and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for calculating carbon emissions, characterized in that, Collaborative processing of multiple devices in the study area includes the following steps: S1: Preprocess the multi-source data to obtain synchronized device-level electrical parameter data and environmental data on a unified time axis, and generate regional-level concentration field data from the spatial data; S2: Based on the equipment-level electrical parameter data obtained from S1, the power signals of each device are analyzed using a deep learning model to identify the operating conditions of each device, while also considering equipment aging correction to improve identification accuracy. S3: Based on the three-element relationship of equipment type, identified operating conditions, and environment, match dynamic emission factors and establish a bottom-up calculation model for energy consumption-driven carbon emissions. Simultaneously, based on the regional concentration field data in S1, a top-down calculation model for concentration inversion based on emission concentration is established; S4: The bottom-up and top-down computational models are fused using adaptive Kalman filtering to obtain the final regional carbon emission results.

2. The carbon emission calculation method according to claim 1, characterized in that, The specific steps of step S1 are as follows: the multiple data acquisition devices installed in the area are clocked uniformly through a time synchronization protocol to generate synchronized time-domain multi-source data, including high-frequency sampled electrical parameter data, including power, voltage, and current, and low-frequency sampled environmental data, including CO2 concentration, temperature, humidity, and air pressure data. Low-frequency data are interpolated and uniformly mapped to the main time axis, and the spatial CO2 concentration data is generated by inverse distance weighting to form a regional concentration field.

3. The carbon emission calculation method according to claim 1, characterized in that, In step S2, the power sequence of a single device within the current sliding time window is normalized, and the probability of each working condition category is obtained based on a CNN-LSTM deep network. The CNN-LSTM deep network is configured with two one-dimensional convolutional layers, a pooling layer, a normalization layer, and a single or double LSTM network. The two one-dimensional convolutional layers extract local waveform features and frequency domain features of the power sequence, the pooling layer and the normalization layer complete the feature dimensionality reduction processing, the single or double LSTM network captures the temporal change dependency of the working condition switching, and the probability of each working condition is output after passing through a fully connected layer and a Softmax activation function. After first correcting the equipment power based on the equipment aging coefficient, the CNN-LSTM deep network is used to obtain the probability of the current identified working condition. The working condition label obtained at the previous identification time is matched to perform a smooth decision and output the current working condition label of the equipment.

4. The carbon emission calculation method according to claim 3, characterized in that, The equipment aging coefficient is a function related to the cumulative operating time of the equipment. In addition to using the equipment aging coefficient to correct the original power characteristics of the equipment, the equipment aging correction also includes concatenating the cumulative operating time and maintenance frequency as additional features with the output of the LSTM and then feeding them into the fully connected layer. During the model training phase, labeled historical operating condition samples are used for supervised training. The loss function is cross-entropy. The training data covers different equipment types, different load levels and different aging stages. During the online phase, only forward inference is performed to obtain the current operating condition label.

5. The carbon emission calculation method according to claim 1, characterized in that, The process of establishing the bottom-up calculation model for energy-driven carbon emissions is as follows: Based on the standard emission factor, dynamic emission factors are formed by dynamically adjusting the standard emission factor according to equipment type, equipment operating conditions and environmental conditions. Specifically, the product of the environmental correction function, the real-time power grid emission correction function and the standard emission factor is used as the dynamic emission factor. The sum of the power data of each device in the study area and time period and the corresponding dynamic emission factor is the energy consumption-driven carbon emission.

6. The carbon emission calculation method according to claim 5, characterized in that, The environmental correction function is a joint correction form formed by the product of temperature correction, humidity correction, and air pressure correction. The real-time grid emission correction function is: for the provincial or regional grid boundary of the studied area, the product of the proportion of the power generation of each energy type in the total power generation at time t within the same boundary and the unit power generation emission factor of that energy type is obtained, and the sum of the correction emission factors of all energy types within the boundary at time t is the real-time grid emission correction function.

7. The carbon emission calculation method according to claim 1, characterized in that, The process of establishing the top-down calculation model for concentration inversion based on emission concentration is as follows: The total regional emissions are calculated using the CO2 concentration change rate and spatial ventilation characteristics, based on a modified mass balance equation. ， in: V represents the carbon emission rate, expressed in kg / s; V represents the effective volume of the monitored spatial unit. When only a single independent space is monitored, V represents the volume of that single space. When the monitoring area contains multiple relatively independent subspaces, the emissions of each subspace are calculated separately, and then the total emissions of the area are summed. CO2 density; This represents the rate of change of the volume-weighted average CO2 concentration C, calculated from the regional concentration field, within the current sliding time window. Ventilation rate; This represents the background concentration.

8. The carbon emission calculation method according to claim 1, characterized in that, Specifically, S4 is: The actual carbon emissions x within the discrete time window k k As a state variable, the carbon emissions E obtained from the bottom-up calculation model energy (k) represents the prior state variables of the Kalman filter, and the carbon emissions E obtained from the top-down computation model. concentration (k) is used as the observation for Kalman filtering. The Kalman prediction and update steps are performed to obtain the fused carbon emissions E. fuse (k); Real-time calculation of absolute error |E concentration (k)-E energy (k)|, when the absolute error exceeds a preset threshold, the process noise covariance Q in the Kalman filter is dynamically increased. k Or R k And trigger an exception flag; otherwise, automatically decrease Q. k R k To improve the signal-to-noise ratio of the filter.

9. The carbon emission calculation method according to claim 1, characterized in that, The method also includes intelligent source tracing and anomaly detection. When carbon emission data shows a sudden change or exceeds the standard, the operating conditions and environmental parameters are automatically traced back. If the regional concentration rises within the time window while the outdoor background concentration rises synchronously and the power of each device does not change significantly, it is determined to be environmental interference. If the power of a certain device suddenly increases and there is a significant linkage with the power of its adjacent exhaust device, it is determined to be an abnormal device coupling and an alarm is triggered.

10. An intelligent carbon metering device, characterized in that, include: The AI ​​operating condition recognition module and the fusion computing module are used to extract the time-domain and frequency-domain features of the equipment power signal based on the CNN-LSTM deep network, taking into account equipment aging, and capture the operating condition change pattern in the time series, and output the equipment operating condition label. The fusion computing module is used to run two models in parallel: a bottom-up calculation of carbon emissions driven by energy consumption and a top-down calculation of concentration inversion based on emission concentration. By dynamically adjusting the error weights through adaptive Kalman filtering, the optimal fusion of the results of the two models is achieved, and the carbon emission results are output.

Images