Microgrid intelligent economic regulation and control system and method based on carbon emission optimization

Abstract

The invention discloses a micro-grid intelligent economic regulation and control system and method based on carbon emission optimization, and relates to the technical field of economic regulation and control. The method comprises the following steps: deploying an Internet of Things sensor to collect photovoltaic generating capacity, energy storage SOC, load demand and power grid carbon intensity data in real time, and transmitting the data to an edge computing node through a 5G / optical fiber hybrid communication network; preprocessing the data by adopting wavelet transform, and establishing a time sequence prediction model of photovoltaic output / load demand; constructing a dynamic carbon flow tracking matrix; performing decision optimization according to the output dynamic carbon emission spectrum and the power carbon flow traceability data; and the edge node issues a control instruction through a Modbus-TCP protocol to carry out economic regulation and control. According to the method, multi-source data are collected in real time, a minute-level carbon emission equation is established based on a dynamic carbon flow tracking model, dynamically-changed energy carbon emission factors and line transmission loss are fused, and the micro-grid carbon footprint is accurately quantified.

Description

Technical Field

[0001] The present invention belongs to the technical field of economic regulation, and in particular relates to a microgrid intelligent economic regulation system and method based on carbon emission optimization. Background Art

[0002] Existing microgrid dispatch systems often use static carbon emission factors (such as annual averages), which fail to reflect minute-by-minute fluctuations in grid carbon intensity, leading to carbon flow tracking errors of 15-30%. The carbon emission characteristics of thermal power and renewable energy units are mixed in calculations, undermining the environmental value of green electricity.

[0003] Because mainstream optimization models often convert carbon emission costs into fixed electricity price add-ons, ignoring the dynamic pricing mechanisms of the carbon market, traditional linear weighted methods cannot achieve Pareto optimality. Furthermore, rigid scheduling models based on day-ahead planning struggle to adapt to second-by-second fluctuations in PV output, requiring reliance on frequent charging and discharging of energy storage to compensate, which accelerates battery degradation.

[0004] To solve the above problems, the present invention provides a microgrid intelligent economic control system and method based on carbon emission optimization. Summary of the Invention

[0005] The purpose of the present invention is to provide a microgrid intelligent economic regulation system and method based on carbon emission optimization. By real-time collection of multi-source data such as photovoltaic, energy storage, load and grid carbon intensity, and using wavelet noise reduction and LSTM models for data cleaning and prediction correction, a minute-level carbon emission equation is established based on a dynamic carbon flow tracking model, integrating dynamically changing energy carbon emission factors and line transmission losses, and accurately quantifying the carbon footprint of the microgrid, solving the existing problems of being unable to reflect minute-level fluctuations in grid carbon intensity and large carbon flow tracking errors.

[0006] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0007] The present invention is a microgrid intelligent economic control method based on carbon emission optimization, comprising the following steps:

[0008] Step S1: Deploy IoT sensors to collect real-time data on photovoltaic power generation, energy storage SOC, load demand, and grid carbon intensity, and transmit the data to edge computing nodes via a 5G / fiber hybrid communication network;

[0009] Step S2: Use wavelet transform to preprocess the data to eliminate voltage sag and harmonic interference noise, and use LSTM neural network to establish a time series prediction model for photovoltaic output / load demand;

[0010] Step S3: constructing a dynamic carbon flow tracking matrix;

[0011] Step S4: Optimize decision making based on the output dynamic carbon emission map and power carbon flow traceability data;

[0012] Step S5: The edge node issues control instructions through the Modbus-TCP protocol to perform economic regulation.

[0013] As a preferred technical solution, in step S1, a micro-inverter integrated voltage / current sensor (accuracy ±0.5%) is deployed in the photovoltaic array to achieve real-time acquisition of single-board power generation and support DC side fault location. Environmental parameter acquisition sensors can also be set around the photovoltaic array, such as: irradiance sensor (range 0-2000W / m 2 ) monitor light intensity fluctuations, backplane temperature sensor (PT100 platinum resistor) tracks component hot spot effects, and laser scattering dust sensor detects panel cleanliness degradation rate;

[0014] When monitoring energy storage SOC, the health status of the battery module is monitored through a voltage Hall sensor (0-1000V range), the instantaneous power of charge and discharge is tracked through a bidirectional current sensor (±300A accuracy), the charge and discharge capacity errors are accumulated through a coulomb meter, and the impedance spectrum analysis module evaluates the degree of electrochemical aging in real time.

[0015] As a preferred technical solution, in step S2, the specific process of preprocessing the data by wavelet transform is as follows:

[0016] Step S21: Use the DB4 wavelet to perform a four-layer multi-resolution decomposition (MRA) to obtain approximation coefficients (cA4) and detail coefficients (cD1-cD4); the detail coefficients include high-frequency detail coefficients (cD1-cD2) and low-frequency detail coefficients (cD3-cD4); the high-frequency detail coefficients contain harmonic components (2nd-50th harmonics), and the low-frequency detail coefficients contain voltage sag characteristics; the fourth-order vanishing moment can effectively capture the singular points of the voltage sag, the tight support (support length 8) balances the time-frequency resolution, and the approximate symmetry reduces the reconstruction phase distortion;

[0017] Step S22: Use unbiased risk estimation to determine the thresholds of each layer. The specific formula is as follows:

[0018]

[0019] Where, T j is the j-th layer threshold, σ j is the noise intensity of the high-frequency coefficient of the jth layer, N j is the number of wavelet coefficients in the jth layer;

[0020] Step S23: Apply segmentation processing to the wavelet coefficients:

[0021] When |w|≤T (noise-dominated interval), w′ j =sign(|w j |)(|w j |-aT j ); progressive compression is achieved by shrinkage factor a = 0.5, avoiding the breakpoint effect of hard threshold and excessive attenuation of traditional soft threshold;

[0022] When |w|>T, the original coefficient is retained to ensure that the effective signal components pass through without distortion;

[0023] Where w j is the wavelet coefficient of the jth layer, w′ j is the coefficient after threshold processing, a is the shrinkage factor, a=0.5, sign(|w j |) is to retain the phase information of the coefficient and output ±1 or 0;

[0024] Step S24: performing inverse wavelet transform on the processed coefficients;

[0025] Step S25: Using the Mallat reconstruction algorithm, the signals are combined through two-channel orthogonal mirror filters;

[0026] The noise separation mechanism is as follows: harmonic energy is concentrated in the cD1-cD2 layer (3-150kHz frequency band), voltage sag characteristics are reflected in the cD3 layer (0.1-3kHz), and the power frequency fundamental wave is retained in the cA4 layer (0-0.1kHz).

[0027] As a preferred technical solution, in step S2, the specific process of using the LSTM neural network to establish a time series prediction model for photovoltaic output / load demand is as follows:

[0028] Step LS01: Obtain the data after wavelet transform preprocessing and perform feature processing; the acquired data includes:

[0029] Photovoltaic output data: collects module-level current / voltage, irradiance, backplane temperature and other parameters, with a sampling interval of ≤5 minutes;

[0030] Load demand data: Synchronously obtain three-phase voltage / current, harmonic characteristics, and typical load start and stop events;

[0031] Associated meteorological data: Introducing external parameters such as temperature, humidity, and wind speed as auxiliary features for prediction;

[0032] Step LS02: Design the LSTM model architecture;

[0033] The LSTM model architecture includes:

[0034] The configuration process of the output layer is as follows:

[0035] Configure the time window: 24 hours of historical data, a total of 288 5-minute sampling points;

[0036] Configure the input dimensions: PV forecast input features are 12-dimensional (output + environmental parameters), and load forecast input features are 8-dimensional (power + harmonic characteristics).

[0037] The LSTM network structure is as follows:

[0038] Hidden layer: 3 layers of LSTM units (with 128 / 64 / 32 neurons respectively), with 20% dropout added between layers to prevent overfitting;

[0039] Attention mechanism: Add a temporal attention module after the second LSTM layer to dynamically weight key time period features;

[0040] Output layer: The fully connected layer is mapped to the predicted target (PV output / load demand), and the activation function uses ReLU.

[0041] Step LS03: Obtain historical data as the training set (accounting for 70%), the data of the last three months as the validation set (accounting for 15%), and the data of the latest month as the test set (accounting for 15%), and use time series cross-validation;

[0042] Step LS04: Customize the training strategy, including the loss function, optimizer, and early stopping mechanism. The loss function uses a customized weighted MAE, with a 1.5x penalty factor for peak hours. The optimizer uses the Nadam optimizer with an initial learning rate of 0.001, which decays by 20% every 10 epochs. The early stopping mechanism terminates training when the validation set loss shows no improvement after five consecutive epochs.

[0043] Step LS05: Correct and evaluate the prediction structure; use a dynamic correction mechanism to correct it, combine the latest collected data, and update the prediction sequence through Kalman filtering; when the prediction error exceeds 15%, trigger a reassessment of feature importance;

[0044] When conducting photovoltaic forecast evaluation, NRMSE ≤ 8.5% (sunny day) and ≤ 12.3% (cloudy day); when conducting load forecast evaluation, MAPE ≤ 5.2% (steady-state load) and ≤ 9.7% (impact load); at the same time, a forecast deviation report is generated every 24 hours, and abnormal samples are automatically marked and added to the training set.

[0045] As a preferred technical solution, in step LS01, the feature processing includes time series feature processing, spatial feature processing and dimensionality reduction processing; the time series feature processing extracts statistics such as the mean / variance / extreme value of the 24-hour activity window; the spatial feature processing constructs an adjacent component output correlation matrix for the distributed photovoltaic array; and the dimensionality reduction processing retains the principal components with a cumulative contribution rate ≥95% through PCA.

[0046] As a preferred technical solution, in step S3, the specific process of constructing a dynamic carbon flow tracking matrix is ​​as follows:

[0047] Step S31: Collecting power structure data of thermal power units and new energy stations connected to the dispatching center;

[0048] Step S32: Calculate the thermal power factor and the new energy factor, and adjust the dynamic weights;

[0049] Step S33: construct a 33-node impedance matrix, inject power to calculate branch power flow distribution, and mark key transmission paths;

[0050] Step S34: allocating line losses according to the power generation / load ratio;

[0051] Step S35: synchronize the timestamps of the multi-source data, establish the carbon flow correlation matrix, and perform real-time calculations. As a preferred technical solution, in step S32, the thermal power factor calculation formula is as follows:

[0052] EF coal =α×Q net ×η comb ×(1-η ccs ) / P out ;

[0053] The calculation formula of the new energy factor is as follows:

[0054] EF renew =β×(E manufacture / Lifetime) / P avg ;

[0055] Where, EF coal is the carbon emission coefficient per unit power generation of the thermal power unit, α is the correction coefficient of the carbon content of the fuel, Q net is the low calorific value of the fuel, η comb is the fuel combustion efficiency, η ccs is the carbon capture system efficiency, P out is the net output power of the unit; EF renew is the carbon emission factor of renewable energy throughout its life cycle, β is the expansion coefficient of its life cycle, E manufactureis the embodied carbon emissions during the equipment manufacturing phase, Lifetime is the average annual power generation during the entire life cycle of the equipment, and P avg It is the average annual power generation of the equipment throughout its life cycle.

[0056] As a preferred technical solution, in step S35, a carbon flow correlation matrix is ​​constructed. And calculate the total carbon emissions of the system in real time Where a ij represents the carbon flow distribution coefficient from node i to j, P it represents the active power injection of node i at time t, EF it represents the emission factor of node i at time t, T loss represents the carbon emissions from network losses, is the Hadamard product; the specific construction process is as follows:

[0057] Step S351: constructing a node-branch matrix based on Kirchhoff's law and calculating the power transmission path;

[0058] Step S352: Derivation of carbon flow distribution coefficient a through branch power flow distribution matrix PB ij =PB ij / ∑PB jk ;

[0059] Step S353: Update matrix A every 15 minutes and set a mark for the new energy node to reflect the network reconstruction and the change of the power flow direction;

[0060] Step S354: Synchronously obtain the system's P it , and pull EF from the system it ;

[0061] Step S355: Generate node carbon emission intensity vector, Realize the spatial distribution of carbon flow and output N×1 column vector; sum the matrix multiplication results and add T loss ;

[0062] In order to ensure the accuracy of the data, carbon conservation verification can be performed after calculating the total carbon emissions, that is, satisfying: |C t -∑(P it ×EF it )|<5%×T loss ; When the calculation results do not meet the carbon conservation verification, timely exception handling is required to trigger the manual review process.

[0063] As a preferred technical solution, in step S351, the specific distance of the power transmission path is calculated as follows:

[0064] Step L1: Starting from the power source node (such as a power plant or a new energy station), the grid nodes are hierarchically sorted along the power transmission direction to form a directed acyclic graph.

[0065] Step L2: Mark all power outflow paths (including main lines and backup lines) for each node;

[0066] Step L3: The total power received by each node (including local power generation and upstream transmission power) is distributed to each outflow branch according to the equivalent admittance ratio of the downstream branch. For example, if there are two lines downstream of a node with an admittance ratio of 3:2, the power will be distributed according to this ratio.

[0067] Step L4: Based on the electricity demand of the load node, trace back from the load end to the power source end and calculate the proportion of power generation capacity occupied by each electricity demand. For example, a household's electricity consumption may be 30% from thermal power and 70% from photovoltaic power.

[0068] Step L5: Superimpose the distribution ratios of all power transmission paths in the network (including direct paths and multi-level transit paths) to form a complete power transmission path matrix. If a ring network structure exists (such as a city power grid), a virtual decomposition method is used to equate the ring network to multiple groups of radial paths for calculation. This ensures the uniqueness of power distribution and transforms the previously invisible power transmission process into a quantifiable and traceable path network, providing a physically clear mathematical mapping foundation for carbon flow tracking.

[0069] When wind and solar power generation fluctuates, the injected power weight of the new energy station node is dynamically adjusted, and its instantaneous adjustable capacity is reflected in the downstream branch allocation; when wind and solar power are abandoned, the equivalent admittance value of the new energy node in the transmission path matrix is ​​corrected to suppress the carbon flow distribution of the invalid path; when the power grid switches the operating mode (such as switching operation), the node hierarchy relationship and admittance ratio are automatically reconstructed. For example, after a backup line is put into use, the power allocation weight of the node where it is located needs to be recalculated; when a branch is disconnected due to a fault, the allocation ratio of the path is automatically eliminated, and the originally allocated power is redistributed to the remaining valid paths.

[0070] The present invention is a microgrid intelligent economic control system based on carbon emission optimization, including a multi-source data perception layer, a data transmission architecture, and edge nodes;

[0071] The multi-source data perception layer includes a voltage / current sensor, a BMS integrated sensor, a three-phase electricity meter, a harmonic analysis module and a carbon flow sensor; the voltage / current sensor is used to collect the DC output parameters of each photovoltaic panel in real time; the BMS integrated sensor includes a coulomb meter and an impedance spectrum analysis module; the BMS integrated sensor includes a coulomb meter for accumulating the charge and discharge of the battery; the impedance spectrum analysis module is used to detect the electrochemical state of the battery; the three-phase smart meter is used to monitor power consumption; the harmonic analysis module decomposes the voltage / current signal into the fundamental wave and each harmonic component through the Fourier transform algorithm, and accurately calculates the total harmonic distortion rate and the 3-50 harmonic content rate; the carbon flow sensor includes a power electronic carbon flow metering terminal, an energy type identification module and a blockchain gateway; the power electronic carbon flow metering terminal is used to calculate the carbon emissions corresponding to each kilowatt-hour of electricity in combination with the dynamically updated electric carbon factor; the energy type identification module is used to analyze the characteristics of thermal power / new energy and identify the energy type; the blockchain gateway is used to receive power grid carbon quota data;

[0072] The data transmission structure includes a front-end access layer and a backbone transmission layer; the front-end access layer includes a LoRaWAN wireless mesh network, an RS485 bus, and a high-speed power line carrier; the backbone transmission layer includes a 5G private network and a fiber ring network; the 5G private network is deployed with uRLLC slices;

[0073] The edge node includes a data processing unit and a communication interface; the data processing unit includes an industrial-grade ARM processor and local storage; the communication interface includes a multi-protocol gateway and a time-sensitive network;

[0074] The industrial-grade ARM processor is deployed with a preprocessing module, a time series prediction model, a dynamic carbon flow tracking matrix, a transmission path calculation module, and a carbon footprint modeling module; the preprocessing module is used to preprocess the data collected through wavelet transform; the time series prediction model is used to establish a time series prediction model for photovoltaic output / load demand using an LSTM neural network; the dynamic carbon flow tracking matrix is ​​used to construct a carbon flow tracking matrix based on the power structure data of thermal power units and new energy stations; the transmission path calculation module is used to superimpose the distribution ratios of all power transmission paths in the network to form a complete power transmission path matrix; the carbon footprint modeling module is used to dynamically track the carbon footprint of the microgrid.

[0075] As a preferred technical solution, the carbon footprint modeling module includes a dynamic carbon emission map and power carbon flow traceability data; the dynamic carbon emission map includes node-level real-time carbon trajectory and line loss compensation; the node-level real-time carbon trajectory is used to output the minute-level carbon emission factor evolution curve of each power grid node based on the IEEE 33-node model; the line loss compensation generates a T_loss correction coefficient matrix through power flow calculation to quantify the additional carbon emissions caused by transmission network losses; the power carbon flow traceability data is a power-carbon flow coupling mapping table, which is used to record the carbon emission transmission path of each kWh of electricity from the power generation end to the power consumption end.

[0076] The present invention has the following beneficial effects:

[0077] (1) This paper collects multi-source data such as photovoltaic, energy storage, load, and grid carbon intensity in real time, and uses wavelet noise reduction and LSTM models for data cleaning and prediction correction. Based on a dynamic carbon flow tracking model, a minute-level carbon emission equation is established, integrating the dynamically changing energy carbon emission factors and line transmission losses to accurately quantify the carbon footprint of the microgrid.

[0078] (2) The present invention performs carbon emission deviation detection every 6 hours. If the threshold is exceeded, it triggers cloud-based model retraining to achieve dynamic strategy correction, and ultimately forms a closed-loop control system of "perception-optimization-execution-verification", taking into account both low-carbon and economic goals.

[0079] (3) This invention combines LSTM time series modeling with spatial feature extraction of the attention mechanism, uses online learning mechanisms to cope with equipment aging and environmental changes, and customizes the loss function to enhance the prediction accuracy during peak hours;

[0080] (4) The present invention breaks through the limitations of static factors by dynamically binding the traditional power flow matrix with the carbon emission factor, and uses the network loss carbon emission calculation of the node model to improve the accuracy of cross-border power transmission carbon accounting. It avoids the dimensional expansion problem of traditional matrix multiplication through element-level multiplication, and unifies the physical power flow and carbon flow analysis through matrix operations to achieve minute-level dynamic tracking of the carbon footprint of the microgrid.

[0081] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0083] Figure 1This is a flow chart of a microgrid intelligent economic control method based on carbon emission optimization according to the present invention;

[0084] Figure 2 This is a structural schematic diagram of a microgrid intelligent economic control system based on carbon emission optimization according to the present invention. DETAILED DESCRIPTION

[0085] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0086] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0087] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-2 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0088] Example 1

[0089] See also Figure 1 As shown, the present invention is a microgrid intelligent economic control method based on carbon emission optimization, comprising the following steps:

[0090] Step S1: Deploy IoT sensors to collect real-time data on photovoltaic power generation, energy storage SOC, load demand, and grid carbon intensity, and transmit the data to edge computing nodes via a 5G / fiber hybrid communication network;

[0091] Step S2: Use wavelet transform to preprocess the data to eliminate voltage sag and harmonic interference noise, and use LSTM neural network to establish a time series prediction model for photovoltaic output / load demand;

[0092] Step S3: constructing a dynamic carbon flow tracking matrix;

[0093] Step S4: Optimize decision making based on the output dynamic carbon emission map and power carbon flow traceability data;

[0094] Step S5: The edge node issues control instructions through the Modbus-TCP protocol to perform economic regulation.

[0095] In step S1, a micro-inverter integrated voltage / current sensor (accuracy ±0.5%) is deployed in the photovoltaic array to achieve real-time acquisition of single-board power generation and support DC side fault location. Environmental parameter acquisition sensors can also be set around the photovoltaic array, such as irradiance sensor (range 0-2000W / m 2 ) monitor light intensity fluctuations, backplane temperature sensor (PT100 platinum resistor) tracks component hot spot effects, and laser scattering dust sensor detects panel cleanliness degradation rate;

[0096] When monitoring energy storage SOC, the health status of the battery module is monitored through a voltage Hall sensor (0-1000V range), the instantaneous power of charge and discharge is tracked through a bidirectional current sensor (±300A accuracy), the charge and discharge capacity errors are accumulated through a coulomb meter, and the impedance spectrum analysis module evaluates the degree of electrochemical aging in real time.

[0097] In step S2, the specific process of wavelet transform preprocessing the data is as follows:

[0098] Step S21: Use the DB4 wavelet to perform a four-layer multi-resolution decomposition (MRA) to obtain approximation coefficients (cA4) and detail coefficients (cD1-cD4); the detail coefficients include high-frequency detail coefficients (cD1-cD2) and low-frequency detail coefficients (cD3-cD4); the high-frequency detail coefficients contain harmonic components (2nd-50th harmonics), and the low-frequency detail coefficients contain voltage sag characteristics; the fourth-order vanishing moment can effectively capture the singular points of the voltage sag, the tight support (support length 8) balances the time-frequency resolution, and the approximate symmetry reduces the reconstruction phase distortion;

[0099] Step S22: Use unbiased risk estimation to determine the thresholds of each layer. The specific formula is as follows:

[0100]

[0101] Where, T j is the j-th layer threshold, σ j is the noise intensity of the high-frequency coefficient of the jth layer, N j is the number of wavelet coefficients in the jth layer;

[0102] Step S23: Apply segmentation processing to the wavelet coefficients:

[0103] When |w|≤T (noise-dominated interval), w′ j =sign(|w j |)(|w j |-aT j ); progressive compression is achieved by shrinkage factor a = 0.5, avoiding the breakpoint effect of hard threshold and excessive attenuation of traditional soft threshold;

[0104] When |w|>T, the original coefficient is retained to ensure that the effective signal components pass through without distortion;

[0105] Where w j is the wavelet coefficient of the jth layer, w′ j is the coefficient after threshold processing, a is the shrinkage factor, a=0.5, sign(|w j |) is to retain the phase information of the coefficient and output ±1 or 0;

[0106] Step S24: performing inverse wavelet transform on the processed coefficients;

[0107] Step S25: Using the Mallat reconstruction algorithm, the signals are combined through two-channel orthogonal mirror filters;

[0108] The noise separation mechanism is as follows: harmonic energy is concentrated in the cD1-cD2 layer (3-150kHz frequency band), voltage sag characteristics are reflected in the cD3 layer (0.1-3kHz), and the power frequency fundamental wave is retained in the cA4 layer (0-0.1kHz).

[0109] In step S2, the specific process of using the LSTM neural network to establish a time series prediction model for photovoltaic output / load demand is as follows:

[0110] Step LS01: Obtain the data after wavelet transform preprocessing and perform feature processing; the acquired data includes:

[0111] Photovoltaic output data: collects module-level current / voltage, irradiance, backplane temperature and other parameters, with a sampling interval of ≤5 minutes;

[0112] Load demand data: Synchronously obtain three-phase voltage / current, harmonic characteristics, and typical load start and stop events;

[0113] Associated meteorological data: Introducing external parameters such as temperature, humidity, and wind speed as auxiliary features for prediction;

[0114] Step LS02: Design the LSTM model architecture;

[0115] The LSTM model architecture includes:

[0116] The configuration process of the output layer is as follows:

[0117] Configure the time window: 24 hours of historical data, a total of 288 5-minute sampling points;

[0118] Configure the input dimensions: PV forecast input features are 12-dimensional (output + environmental parameters), and load forecast input features are 8-dimensional (power + harmonic characteristics).

[0119] The LSTM network structure is as follows:

[0120] Hidden layer: 3 layers of LSTM units (with 128 / 64 / 32 neurons respectively), with 20% dropout added between layers to prevent overfitting;

[0121] Attention mechanism: Add a temporal attention module after the second LSTM layer to dynamically weight key time period features;

[0122] Output layer: The fully connected layer is mapped to the predicted target (PV output / load demand), and the activation function uses ReLU.

[0123] Step LS03: Obtain historical data as the training set (accounting for 70%), the data of the last three months as the validation set (accounting for 15%), and the data of the latest month as the test set (accounting for 15%), and use time series cross-validation;

[0124] Step LS04: Customize the training strategy, including the loss function, optimizer, and early stopping mechanism. The loss function uses a customized weighted MAE, with a 1.5x penalty factor for peak hours. The optimizer uses the Nadam optimizer with an initial learning rate of 0.001, which decays by 20% every 10 epochs. The early stopping mechanism terminates training when the validation set loss shows no improvement after five consecutive epochs.

[0125] Step LS05: Correct and evaluate the prediction structure; use a dynamic correction mechanism to correct it, combine the latest collected data, and update the prediction sequence through Kalman filtering; when the prediction error exceeds 15%, trigger a reassessment of feature importance;

[0126] When conducting photovoltaic forecast evaluation, NRMSE ≤ 8.5% (sunny day) and ≤ 12.3% (cloudy day); when conducting load forecast evaluation, MAPE ≤ 5.2% (steady-state load) and ≤ 9.7% (impact load); at the same time, a forecast deviation report is generated every 24 hours, and abnormal samples are automatically marked and added to the training set.

[0127] In step LS01, feature processing includes time series feature processing, spatial feature processing, and dimensionality reduction processing. Time series feature processing extracts statistics such as the mean / variance / extreme value of the 24-hour activity window. Spatial feature processing constructs an output correlation matrix of adjacent components for the distributed photovoltaic array. Dimensionality reduction processing uses PCA to retain principal components with a cumulative contribution rate ≥ 95%.

[0128] In step S3, the specific process of constructing a dynamic carbon flow tracking matrix is ​​as follows:

[0129] Step S31: Collecting power structure data of thermal power units and new energy stations connected to the dispatching center;

[0130] Step S32: Calculate the thermal power factor and the new energy factor, and adjust the dynamic weights;

[0131] Step S33: construct a 33-node impedance matrix, inject power to calculate branch power flow distribution, and mark key transmission paths;

[0132] Step S34: allocating line losses according to the power generation / load ratio;

[0133] Step S35: synchronize the timestamps of multi-source data, establish a carbon flow correlation matrix, and perform real-time calculations.

[0134] In step S32, the thermal power factor calculation formula is as follows:

[0135] EF coal =α×Q net ×η comb ×(1-η ccs ) / P out ;

[0136] The calculation formula of the new energy factor is as follows:

[0137] EF renew =β×(E manufacture / Lifetime) / P avg ;

[0138] Where, EF coal is the carbon emission coefficient per unit power generation of the thermal power unit, α is the correction coefficient of the carbon content of the fuel, Q net is the low calorific value of the fuel, η comb is the fuel combustion efficiency, η ccs is the carbon capture system efficiency, P out is the net output power of the unit; EF renew is the carbon emission factor of renewable energy throughout its life cycle, β is the expansion coefficient of its life cycle, E manufacture is the embodied carbon emissions during the equipment manufacturing phase, Lifetime is the average annual power generation during the entire life cycle of the equipment, and P avg It is the average annual power generation of the equipment throughout its life cycle.

[0139] In step S35, a carbon flow correlation matrix is ​​constructed And calculate the total carbon emissions of the system in real time Where a ij represents the carbon flow distribution coefficient from node i to j, P it represents the active power injection of node i at time t, EF it represents the emission factor of node i at time t, T loss represents the carbon emissions from network losses, is the Hadamard product; the specific construction process is as follows:

[0140] Step S351: constructing a node-branch matrix based on Kirchhoff's law and calculating the power transmission path;

[0141] Step S352: Derivation of carbon flow distribution coefficient a through branch power flow distribution matrix PB ij =PB ij / ∑PB jk ;

[0142] Step S353: Update matrix A every 15 minutes and set a mark for the new energy node to reflect the network reconstruction and the change of the power flow direction;

[0143] Step S354: Synchronously obtain the system's P it , and pull EF from the system it ;

[0144] Step S355: Generate node carbon emission intensity vector, Realize the spatial distribution of carbon flow and output N×1 column vector; sum the matrix multiplication results and add T loss ;

[0145] In order to ensure the accuracy of the data, carbon conservation verification can be performed after calculating the total carbon emissions, that is, satisfying: |C t -∑(P it ×EF it )|<5%×T loss ; When the calculation results do not meet the carbon conservation verification, timely exception handling is required to trigger the manual review process.

[0146] In step S351, the specific distance of the power transmission path is calculated as follows:

[0147] Step L1: Starting from the power source node (such as a power plant or a new energy station), the grid nodes are hierarchically sorted along the power transmission direction to form a directed acyclic graph.

[0148] Step L2: Mark all power outflow paths (including main lines and backup lines) for each node;

[0149] Step L3: The total power received by each node (including local power generation and upstream transmission power) is distributed to each outflow branch according to the equivalent admittance ratio of the downstream branch. For example, if there are two lines downstream of a node with an admittance ratio of 3:2, the power will be distributed according to this ratio.

[0150] Step L4: Based on the electricity demand of the load node, trace back from the load end to the power source end and calculate the proportion of power generation capacity occupied by each electricity demand. For example, a household's electricity consumption may be 30% from thermal power and 70% from photovoltaic power.

[0151] Step L5: Superimpose the distribution ratios of all power transmission paths in the network (including direct paths and multi-level transit paths) to form a complete power transmission path matrix. If a ring network structure exists (such as a city power grid), a virtual decomposition method is used to equate the ring network to multiple groups of radial paths for calculation. This ensures the uniqueness of power distribution and transforms the previously invisible power transmission process into a quantifiable and traceable path network, providing a physically clear mathematical mapping foundation for carbon flow tracking.

[0152] When wind and solar power generation fluctuates, the injected power weight of the new energy station node is dynamically adjusted, and its instantaneous adjustable capacity is reflected in the downstream branch allocation; when wind and solar power are abandoned, the equivalent admittance value of the new energy node in the transmission path matrix is ​​corrected to suppress the carbon flow distribution of the invalid path; when the power grid switches the operating mode (such as switching operation), the node hierarchy relationship and admittance ratio are automatically reconstructed. For example, after a backup line is put into use, the power allocation weight of the node where it is located needs to be recalculated; when a branch is disconnected due to a fault, the allocation ratio of the path is automatically eliminated, and the originally allocated power is redistributed to the remaining valid paths.

[0153] Example 2

[0154] See Figure 2 As shown, the present invention is a microgrid intelligent economic control system based on carbon emission optimization, which can be used to execute the method content of Example 1 of the present invention, including: a multi-source data perception layer, a data transmission architecture, and an edge node;

[0155] The multi-source data perception layer includes voltage / current sensors, BMS integrated sensors, three-phase electricity meters, harmonic analysis modules and carbon flow sensors; the voltage / current sensors are used to collect the DC output parameters of each photovoltaic panel in real time; the BMS integrated sensors include coulomb meters and impedance spectrum analysis modules; the BMS integrated sensors include coulomb meters for accumulating the charge and discharge of the battery; the impedance spectrum analysis module is used to detect the electrochemical state of the battery; the three-phase smart meter is used to monitor electricity consumption; the harmonic analysis module decomposes the voltage / current signal into fundamental and harmonic components through the Fourier transform algorithm, and accurately calculates the total harmonic distortion rate and the 3rd-50th harmonic content rate; the carbon flow sensor includes a power electronic carbon flow metering terminal, an energy type identification module and a blockchain gateway; the power electronic carbon flow metering terminal is used to combine the dynamically updated electric carbon factor to calculate the carbon emissions corresponding to each kilowatt-hour of electricity; the energy type identification module is used to analyze the characteristics of thermal power / new energy and identify the energy type; the blockchain gateway is used to receive grid carbon quota data;

[0156] The data transmission structure includes a front-end access layer and a backbone transmission layer. The front-end access layer includes a LoRaWAN wireless mesh network, RS485 bus, and high-speed power line carrier. The backbone transmission layer includes a 5G private network and a fiber ring network. The 5G private network is deployed with uRLLC slices.

[0157] The edge node includes a data processing unit and a communication interface; the data processing unit includes an industrial-grade ARM processor and local storage; the communication interface includes a multi-protocol gateway and a time-sensitive network;

[0158] The industrial-grade ARM processor is deployed with a preprocessing module, a time series prediction model, a dynamic carbon flow tracking matrix, a transmission path calculation module, and a carbon footprint modeling module; the preprocessing module is used to preprocess the data collected through wavelet transform; the time series prediction model is used to use the LSTM neural network to establish a time series prediction model for photovoltaic output / load demand; the dynamic carbon flow tracking matrix is ​​used to construct a carbon flow tracking matrix based on the power structure data of thermal power units and new energy stations; the transmission path calculation module is used to superimpose the distribution ratios of all power transmission paths in the network to form a complete power transmission path matrix; the carbon footprint modeling module is used to dynamically track the carbon footprint of the microgrid.

[0159] The carbon footprint modeling module includes a dynamic carbon emission map and power carbon flow traceability data; the dynamic carbon emission map includes node-level real-time carbon trajectories and line loss compensation; the node-level real-time carbon trajectory is used to output the minute-level carbon emission factor evolution curve of each grid node based on the IEEE 33-node model; the line loss compensation is generated by power flow calculation to generate a T_loss correction coefficient matrix to quantify the additional carbon emissions caused by transmission network losses; the power carbon flow traceability data is a power-carbon flow coupling mapping table, which is used to record the carbon emission transmission path of each kWh of electricity from the power generation end to the power consumption end.

[0160] It is worth noting that in the above system embodiment, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0161] In addition, those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiments can be accomplished by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0162] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A microgrid intelligent economic control method based on carbon emission optimization, characterized in that: The steps include: Step S1: Deploy IoT sensors to collect real-time data on photovoltaic power generation, energy storage SOC, load demand, and grid carbon intensity, and transmit the data to edge computing nodes via a 5G / fiber hybrid communication network; Step S2: Use wavelet transform to preprocess the data to eliminate voltage sag and harmonic interference noise, and use LSTM neural network to establish a time series prediction model for photovoltaic output / load demand; Step S3: constructing a dynamic carbon flow tracking matrix; Step S4: Optimize decision making based on the output dynamic carbon emission map and power carbon flow traceability data; Step S5: The edge node issues control instructions through the Modbus-TCP protocol to perform economic regulation.

2. The microgrid intelligent economic control method based on carbon emission optimization according to claim 1 is characterized in that: In step S2, the specific process of preprocessing the data by wavelet transform is as follows: Step S21: using DB4 wavelet to perform four-layer multi-resolution decomposition to obtain approximate coefficients and detail coefficients; the detail coefficients include high-frequency detail coefficients and low-frequency detail coefficients; the high-frequency detail coefficients contain harmonic components, and the low-frequency detail coefficients contain voltage sag characteristics; Step S22: Use unbiased risk estimation to determine the thresholds of each layer. The specific formula is as follows: Where, T j is the j-th layer threshold, σ j is the noise intensity of the high-frequency coefficient of the jth layer, N j is the number of wavelet coefficients in the jth layer; Step S23: Apply segmentation processing to the wavelet coefficients: When |w|≤T, w′ j =sign(|w j |)(|w j |-aT j ); When |w|>T, the original coefficient is retained; Where w j is the wavelet coefficient of the jth layer, w′ j is the coefficient after threshold processing, a is the shrinkage factor, a=0.5, sign(|w j |) is to retain the phase information of the coefficient; Step S24: performing inverse wavelet transform on the processed coefficients; Step S25: Using the Mallat reconstruction algorithm, a signal is combined through two-channel orthogonal mirror filters.

3. The microgrid intelligent economic control method based on carbon emission optimization according to claim 1 is characterized in that: In step S2, the specific process of using the LSTM neural network to establish a time series prediction model for photovoltaic output / load demand is as follows: Step LS01: Obtain the data after wavelet transform preprocessing and perform feature processing; Step LS02: Design the LSTM model architecture; Step LS03: Obtain historical data as the training set, the data of the last three months as the validation set, and the data of the last month as the test set; Step LS04: Customize the training strategy, including loss function, optimizer, and early stopping mechanism; Step LS05: Modify and evaluate the predicted structure.

4. The microgrid intelligent economic control method based on carbon emission optimization according to claim 1 is characterized in that: In step LS01, feature processing includes time series feature processing, spatial feature processing, and dimensionality reduction processing; the time series feature processing extracts mean / variance / extreme value statistics of a 24-hour activity window; the spatial feature processing constructs an adjacent component output correlation matrix for a distributed photovoltaic array; and the dimensionality reduction processing retains principal components with a cumulative contribution rate ≥ 95% through PCA.

5. The microgrid intelligent economic control method based on carbon emission optimization according to claim 1 is characterized in that: In step S3, the specific process of constructing the dynamic carbon flow tracking matrix is ​​as follows: Step S31: Collecting power structure data of thermal power units and new energy stations connected to the dispatching center; Step S32: Calculate the thermal power factor and the new energy factor, and adjust the dynamic weights; Step S33: construct a 33-node impedance matrix, inject power to calculate branch power flow distribution, and mark key transmission paths; Step S34: allocating line losses according to the power generation / load ratio; Step S35: synchronize the timestamps of multi-source data, establish a carbon flow correlation matrix, and perform real-time calculations.

6. The microgrid intelligent economic control method based on carbon emission optimization according to claim 5 is characterized in that: In step S32, the thermal power factor calculation formula is as follows: EF coal =α×Q net ×η comb ×(1-η ccs ) / P out ; The calculation formula of the new energy factor is as follows: EF renew =β×(E manufacture / Lifetime) / P avg ; Where, EF coal is the carbon emission coefficient per unit power generation of the thermal power unit, α is the correction coefficient of the carbon content of the fuel, Q net is the low calorific value of the fuel, η comb is the fuel combustion efficiency, η ccs is the carbon capture system efficiency, P out is the net output power of the unit; EF renew is the carbon emission factor of renewable energy throughout its life cycle, β is the expansion coefficient of its life cycle, E manufacture is the embodied carbon emissions during the equipment manufacturing phase, Lifetime is the average annual power generation during the entire life cycle of the equipment, and P avg It is the average annual power generation of the equipment throughout its life cycle.

7. The microgrid intelligent economic control method based on carbon emission optimization according to claim 1 is characterized in that: In step S35, a carbon flow correlation matrix is ​​constructed. And calculate the total carbon emissions of the system in real time Where a ij represents the carbon flow distribution coefficient from node i to j, P it represents the active power injection of node i at time t, EF it represents the emission factor of node i at time t, T loss represents the carbon emissions from network losses, is the Hadamard product; the specific construction process is as follows: Step S351: constructing a node-branch matrix and calculating the power transmission path; Step S352: Derivation of carbon flow distribution coefficient a through branch power flow distribution matrix PB ij =PB ij / ∑PB jk ; Step S353: Update matrix A every 15 minutes and set a mark for the new energy node; Step S354: Synchronously obtain the system's P it , and pull EF from the system it ; Step S355: Generate node carbon emission intensity vector, Realize the spatial distribution of carbon flow and output N×1 column vector; sum the matrix multiplication results and add T loss .

8. The microgrid intelligent economic control method based on carbon emission optimization according to claim 7 is characterized in that: In step S351, the specific distance of the power transmission path is calculated as follows: Step L1: Starting from the power source node, the grid nodes are hierarchically sorted along the power transmission direction to form a directed acyclic graph; Step L2: Mark all power outflow paths for each node; Step L3: The total power received by each node is distributed to each outflow branch according to the equivalent admittance ratio of the downstream branch; Step L4: Based on the power demand of the load node, trace back from the load end to the power supply end and calculate the proportion of power generation capacity occupied by each power demand; Step L5: Superimpose the allocation ratios of all power transmission paths in the network to form a complete power transmission path matrix.

9. A microgrid intelligent economic control system based on carbon emission optimization, including a multi-source data perception layer, a data transmission architecture, and edge nodes, characterized by: The multi-source data perception layer includes a voltage / current sensor, a BMS integrated sensor, a three-phase electricity meter, a harmonic analysis module and a carbon flow sensor; the voltage / current sensor is used to collect the DC output parameters of each photovoltaic panel in real time; the BMS integrated sensor includes a coulomb meter and an impedance spectrum analysis module; the BMS integrated sensor includes a coulomb meter for accumulating the charge and discharge of the battery; the impedance spectrum analysis module is used to detect the electrochemical state of the battery; the three-phase smart meter is used to monitor power consumption; the harmonic analysis module decomposes the voltage / current signal into the fundamental wave and each harmonic component through the Fourier transform algorithm, and accurately calculates the total harmonic distortion rate and the 3-50 harmonic content rate; the carbon flow sensor includes a power electronic carbon flow metering terminal, an energy type identification module and a blockchain gateway; the power electronic carbon flow metering terminal is used to calculate the carbon emissions corresponding to each kilowatt-hour of electricity in combination with the dynamically updated electric carbon factor; the energy type identification module is used to analyze the characteristics of thermal power / new energy and identify the energy type; the blockchain gateway is used to receive power grid carbon quota data; The data transmission structure includes a front-end access layer and a backbone transmission layer; the front-end access layer includes a LoRaWAN wireless mesh network, an RS485 bus, and a high-speed power line carrier; the backbone transmission layer includes a 5G private network and a fiber ring network; the 5G private network is deployed with uRLLC slices; The edge node includes a data processing unit and a communication interface; the data processing unit includes an industrial-grade ARM processor and local storage; the communication interface includes a multi-protocol gateway and a time-sensitive network; The industrial-grade ARM processor is equipped with a preprocessing module, a time series prediction model, a dynamic carbon flow tracking matrix, a transmission path calculation module, and a carbon footprint modeling module. The preprocessing module is used to preprocess the collected data through wavelet transform. The time series prediction model is used to establish a time series prediction model for photovoltaic output / load demand using an LSTM neural network. The dynamic carbon flow tracking matrix is ​​used to construct a carbon flow tracking matrix based on the power structure data of thermal power units and new energy stations. The transmission path calculation module is used to form a complete power transmission path matrix based on the distribution ratio of all power transmission paths in the network; the carbon footprint modeling module is used to dynamically track the carbon footprint of the microgrid.

10. A microgrid intelligent economic control system based on carbon emission optimization according to claim 9, characterized in that: The carbon footprint modeling module includes a dynamic carbon emission map and power carbon flow traceability data; the dynamic carbon emission map includes node-level real-time carbon trajectory and line loss compensation; the node-level real-time carbon trajectory is used to output a minute-level carbon emission factor evolution curve for each grid node based on the IEEE 33-node model; The line loss compensation amount is calculated through power flow to generate a T_loss correction coefficient matrix to quantify the additional carbon emissions caused by transmission network losses; The electricity carbon flow traceability data is a power-carbon flow coupling mapping table, which is used to record the carbon emission transmission path of each kWh of electricity from the power generation end to the power consumption end.

Images