Community demand response carbon emission reduction data monitoring method based on Internet of Things technology

Abstract

The invention relates to a community demand response carbon emission reduction data monitoring method based on the Internet of Things technology, and the method comprises the following steps: S1, collecting multi-source heterogeneous data of community carbon emission data, and carrying out the preprocessing, and obtaining a basic data set; s2, performing feature engineering and depth representation according to the basic data set, and constructing a high-dimensional feature vector set; s3, constructing a new energy output prediction model, and predicting a new energy output confidence interval based on the high-dimensional feature vector set; s4, constructing a load demand elastic prediction model, and predicting a rigid load and flexible load adjustment boundary based on the high-dimensional feature vector set; s5, based on the predicted new energy output confidence interval and the rigid load and flexible load adjustment boundary, adopting multi-objective optimization control to obtain an optimal equipment control instruction set; and S6, performing control based on the optimal equipment control instruction set, obtaining the verification emission reduction amount and a user response log, and generating a carbon perbenefit integral. According to the invention, the community can flexibly adjust the energy supply strategy according to the new energy volatility and the user demand.

Description

Technical Field

[0001] The present invention relates to the field of carbon emission reduction monitoring, and in particular to a community demand response carbon emission reduction data monitoring method based on Internet of Things technology. Background Art

[0002] As the fundamental unit of urban energy consumption, communities are increasingly attracting attention for their potential to reduce carbon emissions. Traditional community carbon reduction efforts primarily rely on static energy efficiency management and policy guidance, which presents two challenges: First, the fragmentation of carbon emission monitoring systems—community carbon emission data is heterogeneous and multi-sourced (e.g., equipment energy consumption, user behavior, environmental conditions, and renewable energy output), lacking multi-dimensional data integration and dynamic characterization; second, demand response accuracy is insufficient—the volatility of renewable energy output and the uncontrollable elasticity of load demand result in a lack of quantitative support for equipment regulation, and the incentive mechanism for users to engage in low-carbon behaviors is limited.

[0003] While a relatively comprehensive carbon credit system has been established for the existing power system, with recently released carbon credit emission reduction methodologies for distributed photovoltaic power generation, residential low-carbon electricity consumption, and pure electric passenger vehicles, there is still no systematic quantitative evaluation and incentive mechanism for the emission reduction behaviors brought about by community participation in electricity demand-side response. Furthermore, current carbon credit methodologies primarily focus on explicit low-carbon behaviors such as clean energy use and energy-saving renovations, while insufficient attention and quantification are given to the implicit carbon emission reduction effects generated by electricity demand response. Summary of the Invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a community demand response carbon emission reduction data monitoring method based on Internet of Things technology, so that the community can flexibly adjust the energy supply strategy according to the volatility of new energy and user demand, and at the same time achieve emission reduction-benefit positive feedback through carbon credit.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A community demand response carbon emission reduction data monitoring method based on Internet of Things technology includes the following steps:

[0007] S1: Collect multi-source heterogeneous data of community carbon emissions and preprocess them to obtain a basic data set;

[0008] S2: Based on the basic data set, perform feature engineering and deep characterization to construct a high-dimensional feature vector set;

[0009] S3: Build a new energy output prediction model and predict the confidence interval of new energy output based on a high-dimensional feature vector set;

[0010] S4: Construct a load demand elasticity prediction model and predict the rigid load and flexible load adjustment boundaries based on the high-dimensional feature vector set;

[0011] S5: Based on the predicted confidence interval of renewable energy output and the rigid and flexible load regulation boundaries, multi-objective optimization control is used to obtain the optimal equipment control instruction set;

[0012] S6: Control based on the optimal device control instruction set, obtain certified emission reductions and user response logs, and generate carbon credits.

[0013] Furthermore, we collected multi-source heterogeneous data of community carbon emission data and preprocessed them to obtain the basic data set, specifically

[0014] The multi-source heterogeneous data includes a list of public adjustable load equipment in the community D device , real-time equipment operation data collected by IoT sensors iot 、Environmental data D env and user behavior data D user ;

[0015] The data timestamps of each subsystem have an offset Δt;

[0016] First, the dynamic time warping algorithm is used to compensate for the time deviation:

[0017] C i，j =|t device，i -t user，j |+λ·SED(υ i ,υ j )

[0018] Among them, C i,j is the cost matrix element of dynamic time warping, which represents the alignment cost between the time series of community device data and the time series of user behavior data; SED is the standardized Euclidean distance, λ is the weight coefficient; i is the time index of device data; j is the time index of user behavior data;

[0019] t device,i is the device data timestamp, indicating the time of the i-th device data sampling point; t user,j is the timestamp of user behavior data, indicating the time of the jth user behavior data sampling point;

[0020] Then, the sliding window algorithm based on Grubbs test is used to detect outliers and remove them; finally, the missing values ​​are filled by the spatiotemporal correlation construction method; the preprocessed basic data set D is obtained. raw ={D device ,D iot ,D env ,D user}.

[0021] Furthermore, the high-dimensional feature vector set vfinal Including statistical features v statistical , frequency domain features v freq , device cluster characteristics v cluster And the deeply fused multimodal features h fused , v final =[v statistical ;v freq ;v cluster ;h fused ], as follows:

[0022] Statistical characteristics include mean u, standard deviation σ and trend term β:

[0023]

[0024] Where T is the total length of the time series; x t is the power value of the time series at the tth moment; is the time center point of the time series;

[0025] The frequency domain feature is the main frequency component extracted by Fourier transform (FFT) of the power time series data:

[0026]

[0027] Spatial aggregation of devices in the same area (such as buildings and charging stations) is performed to obtain device cluster characteristics:

[0028]

[0029] The deeply fused multimodal features are obtained based on the fusion of device, environment and user features.

[0030] Furthermore, the deeply integrated multimodal features are obtained based on the fusion of device, environment, and user features, as follows:

[0031] Get device characteristics in the community device 、Environmental characteristics env , user characteristics h use , and construct cross-attention weights for device, environment, and user features:

[0032]

[0033] Where Q = W Q h device For query, K = W K h user is a key, V=W V h env is the value, the superscript T indicates transposition, d k is the dimension of the key; W Q ,WK ,W V are the weight coefficients of query, key and value respectively; Attention(Q,K,V) is the attention output; h device is the device characteristic vector, including the power and load rate of the device; h user is the user feature vector, including user behavior preference and elasticity index; h env is the environmental feature vector, including light intensity and ambient temperature;

[0034] The final multimodal features are obtained by dynamically weighting the multimodal features through the gating mechanism:

[0035] g=σ(W g [h device ;h user ;h env ])

[0036] h fused =g☉h device +(1-g)☉h user ;

[0037] Where g is the gating weight vector; Wg is the learnable gating parameter matrix; σ is the Sigmoid activation function; and ⊙ is element-wise multiplication.

[0038] Furthermore, S3 is specifically

[0039] S31: From the high-dimensional feature vector set v final Extract multimodal key features;

[0040] S32: Construct a new energy output prediction model based on the Transformer-base model, wherein the Transformer-base model includes a Transformer and a regression head; the Transformer embeds the high-dimensional feature vector v final After input, it is embedded into a fixed-dimensional space d model Then, the temporal and spatial dependencies between the input features are captured by the multi-head self-attention output, and the regression head regresses the Transformer output to calculate the output of the next h time steps.

[0041] S33: Use uncertainty quantification methods to evaluate model output intervals.

[0042] Furthermore, the uncertainty quantification method is used to evaluate the model output range, as follows:

[0043] The prediction confidence interval CI(t+h) reflects the model's confidence in the prediction of renewable energy output. and the prediction uncertainty σ(t+h) is expressed as:

[0044]

[0045] Where k′ is the confidence factor;

[0046] Monte Carlo sampling is used for uncertainty estimation. In the inference phase, M forward propagations are performed to calculate the predicted mean and variance:

[0047]

[0048] Furthermore, the load demand elasticity prediction model uses a dual-branch deep network to simultaneously predict rigid loads and flexible boundaries, as follows:

[0049] The dual-branch deep network includes a shared encoder, a rigid load prediction branch, and a flexible boundary prediction branch;

[0050] The shared encoder, the high-dimensional feature vector set v final Extract common features:

[0051] h shared =MLP(v final );

[0052] Among them, MLP is a multi-layer perceptron; h shared is the common feature extracted by MLP;

[0053] The rigid load prediction branch and the flexible boundary prediction branch are based on the common characteristics h shared Make a prediction:

[0054]

[0055] in, is the predicted value of rigid load; h shared The common features generated by the shared encoder serve as the input for rigid load prediction; W r is the weight matrix of the fully connected layer, which maps the implicit features after nonlinear processing to the predicted value of the rigid load; Ur is the transformation matrix; b r is a bias term that compensates for the linear deviation in the feature mapping process; ReLU(·) is the activation function; is the lower boundary of the predicted flexible load; is the upper boundary of the predicted flexible load; W min , W max is the weight matrix; output boundary After being constrained by Sigmoid mapping, it is normalized to the interval [0,1];

[0056] The error function L is obtained by jointly optimizing the rigidity error L rigid , flexible boundary tightness Lflex and physical constraints L constraint :

[0057] L=L rigid +λ1L flex +λ2L constraint ;

[0058]

[0059] Where N′ is the number of samples; λ1, λ2 are weight hyperparameters; is the predicted value of the rigid load of the i′th sample; is the actual rigid load value of the i′th sample; is the actual flexible load value of the i′th sample; The lower bound of the flexible load predicted for the i′th sample; is the upper bound of the flexible load predicted for the i-th sample; τ is the confidence level; () + is the positive part function, which is used to measure the uncovered part of the boundary;

[0060] Monte Carlo Dropout is used to evaluate the uncertainty of boundary predictions. M forward propagations are performed during the inference phase with Dropout enabled to statistically analyze the distribution of flexible boundaries:

[0061]

[0062] Final output, rigid load prediction value Flexible load adjustment boundary

[0063] Furthermore, based on the predicted confidence interval of renewable energy output and the rigid and flexible load regulation boundaries, multi-objective optimization control is used to obtain the optimal equipment control instruction set, as follows:

[0064] Randomly initialize the initial control strategy of each device in the system (such as photovoltaic power, wind power, thermal power, and flexible load regulation power) to form a control instruction set;

[0065] For each solution in the population, multi-objective values ​​are calculated based on the objective function and constraints: the objective function includes minimizing the total energy cost f1, minimizing the carbon footprint f2, and minimizing the net power fluctuation f3 of the grid, as follows:

[0066]

[0067]

[0068] in, is the power value purchased from the grid at time t; is the power value sold to the grid at time t; is the charging power of device d at time t; is the discharge power of device d at time t; It is the time-of-use electricity purchase price; is the time-of-use electricity price; γ d is the efficiency loss coefficient of device d, which is related to the charge and discharge characteristics of the device itself; δ grid,t is the carbon emission factor of the power grid; δ d is the fossil fuel emission rate of device d;

[0069] Constraints include power balance constraints, uncertainty in renewable energy output, unadjustable rigid load constraints, and flexible load adjustment boundary constraints:

[0070]

[0071] Among them, P new，t The actual output of new energy at time t; The power sold to the grid; t is the standard deviation of the prediction error;

[0072] For solutions that exceed the upper and lower boundaries of flexible load regulation or the confidence interval of renewable energy output, a correction mechanism is used to limit them to the valid range;

[0073] The fitness of each solution is calculated using a target trade-off strategy based on neighborhood decomposition;

[0074] Apply crossover and mutation operations to improve the diversity of solutions and explore better solutions;

[0075] After a preset round of iterations, the Pareto optimal solution set that meets the constraints is obtained.

[0076] Furthermore, based on the optimal device control instruction set, control is performed to obtain certified emission reductions and user response logs, and generate carbon credits, as follows:

[0077] Define the baseline carbon emissions when users do not participate in regulation:

[0078]

[0079] in, is the historical average power purchase amount from the power grid in period t when the user does not participate in the optimization regulation; grid，t is the carbon emission factor of the regional power grid in the corresponding period;

[0080] After optimizing the control based on the optimal device control instruction set, the actual power purchase amount of the optimized user in the t period is collected in real time through the smart meter. Calculate actual carbon emissions:

[0081]

[0082] Then the emission reduction ΔE reduce =E baseline -E actual ;

[0083] Obtain records of user flexible load adjustment response logs, including time, adjustment amount, duration, and response speed, and generate carbon credits:

[0084]

[0085] in, is the emission reduction of user u; is the response speed of the g-th action of user u; is the number of user participations; α, β, and γ are the basic integral rate of emission reduction, the response speed gain coefficient, and the fixed reward for single participation in regulation, respectively; is the average response speed.

[0086] A community demand response carbon emission reduction data monitoring system based on Internet of Things technology includes a processor, a memory, and a computer program stored on the memory. When the processor executes the computer program, it specifically performs the steps of the community demand response carbon emission reduction data monitoring method based on Internet of Things technology as described above.

[0087] The present invention has the following beneficial effects:

[0088] 1. This invention enables communities to flexibly adjust their energy supply strategies based on the volatility of new energy and user needs. It also achieves a positive feedback loop of emission reduction and benefits through carbon credits, precisely supporting the implementation of dual carbon goals.

[0089] 2. The present invention constructs an elastic prediction model for rigid loads and flexible loads, accurately predicts the rigid load demand curve and the flexible load adjustment boundary. The rigid load prediction ensures that the basic needs of users are met, and the flexible load adjustment boundary provides optimization space for demand response, accurately distinguishes rigid loads from flexible loads, improves control flexibility, and ensures a balance between user electricity demand and system optimization goals.

[0090] 3. This invention combines the confidence interval of renewable energy output forecast, rigid load forecast and flexible load adjustment boundary, and adopts a multi-objective optimization method to make the scheduling strategy take into account the supply and demand balance, operating cost and renewable energy utilization rate. At the same time, the optimal solution set is solved through genetic algorithm, and the control instructions for each device are finally generated. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0092] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0093] refer to Figure 1 In this embodiment, a community demand response carbon emission reduction data monitoring method based on Internet of Things technology is provided, including the following steps:

[0094] S1: Collect multi-source heterogeneous data of community carbon emissions and preprocess them to obtain a basic data set;

[0095] S2: Based on the basic data set, perform feature engineering and deep characterization to construct a high-dimensional feature vector set;

[0096] S3: Build a new energy output prediction model and predict the confidence interval of new energy output based on a high-dimensional feature vector set;

[0097] S4: Construct a load demand elasticity prediction model and predict the rigid load and flexible load adjustment boundaries based on the high-dimensional feature vector set;

[0098] S5: Based on the predicted confidence interval of renewable energy output and the rigid and flexible load regulation boundaries, multi-objective optimization control is used to obtain the optimal equipment control instruction set;

[0099] S6: Control based on the optimal device control instruction set, obtain certified emission reductions and user response logs, and generate carbon credits.

[0100] In this embodiment, multi-source heterogeneous data of community carbon emission data is collected and preprocessed to obtain a basic data set, specifically

[0101] The multi-source heterogeneous data includes a list of public adjustable load equipment in the community D device (charging piles, public lighting, central air conditioning, etc.), real-time equipment operation data collected by IoT sensors iot (power, current, voltage, start / stop status, etc.), environmental data D env (light intensity, temperature, humidity, wind speed, etc.) and user behavior data D user (power consumption patterns, energy usage habits, and user adjustment participation records);

[0102] The data timestamps of each subsystem have an offset Δt;

[0103] First, the dynamic time warping algorithm is used to compensate for the time deviation:

[0104] C i，j =|t device，i -t user，j|+λ·SED(υ i ,υ j )

[0105] Among them, C i,j is the cost matrix element of dynamic time warping, which represents the alignment cost between the time series of community device data (such as device operation data) and the time series of user behavior data; SED is the standardized Euclidean distance, λ is the weight coefficient; i is the time index of device data; j is the time index of user behavior data;

[0106] t device,i is the device data timestamp, indicating the time of the i-th device data sampling point; t user,j is the timestamp of user behavior data, indicating the time of the jth user behavior data sampling point;

[0107] Then, the sliding window algorithm based on Grubbs test is used to detect outliers and remove them; finally, the missing values ​​are filled by the spatiotemporal correlation construction method; the preprocessed basic data set D is obtained. raw ={D device ,D iot ,D env ,D user}.

[0108] In this embodiment, the high-dimensional feature vector set v final Including statistical features v statistical , frequency domain features v freq , device cluster characteristics v cluster And the deeply fused multimodal features h fused , v final =[v statistical ;v freq ;v cluster ;h fused ], as follows:

[0109] Statistical characteristics include mean u, standard deviation σ and trend term β:

[0110]

[0111] Where T is the total length of the time series; x t is the power value of the time series at the tth moment; is the time center point of the time series;

[0112] The frequency domain feature is the main frequency component extracted by Fourier transform (FFT) of the power time series data:

[0113]

[0114] Spatial aggregation of devices in the same area (such as buildings and charging stations) is performed to obtain device cluster characteristics:

[0115]

[0116] The deeply fused multimodal features are obtained based on the fusion of device, environment and user features.

[0117] In this embodiment, the deeply fused multimodal features are obtained based on the fusion of device, environment, and user features, as follows:

[0118] Get device characteristics in the community device 、Environmental characteristics env , user characteristics h use , and construct cross-attention weights for device, environment, and user features:

[0119]

[0120] Where Q = W Q h device For query, K = W K h user is a key, V=W V h env is the value, the superscript T indicates transposition, d k is the dimension of the key; W Q ,W K ,W V are the weight coefficients of query, key and value respectively; Attention(Q,K,V) is the attention output; h device is the device characteristic vector, including the power and load rate of the device; h user is the user feature vector, including user behavior preference and elasticity index; h env is the environmental feature vector, including light intensity and ambient temperature;

[0121] The final multimodal features are obtained by dynamically weighting the multimodal features through the gating mechanism:

[0122] g=σ(W g [h device ;h user ;h env ])

[0123] h fused =g☉h device +(1-g)☉h user ;

[0124] Where g is the gating weight vector; Wg is the learnable gating parameter matrix; σ is the Sigmoid activation function; and ⊙ is element-wise multiplication.

[0125] In this embodiment, S3 is specifically

[0126] S31: From the high-dimensional feature vector set v final Extract multimodal key features;

[0127] S32: Construct a new energy output prediction model based on the Transformer-base model, wherein the Transformer-base model includes a Transformer and a regression head; the Transformer embeds the high-dimensional feature vector v final After input, it is embedded into a fixed-dimensional space d model Then, the temporal and spatial dependencies between the input features are captured by the multi-head self-attention output, and the regression head regresses the Transformer output to calculate the output of the next h time steps.

[0128] S33: Use uncertainty quantification methods to evaluate model output intervals.

[0129] In this embodiment, the uncertainty quantification method is used to evaluate the model output interval, as follows:

[0130] The prediction confidence interval CI(t+h) reflects the model's confidence in the prediction of renewable energy output. and the prediction uncertainty σ(t+h) is expressed as:

[0131]

[0132] Where k′ is the confidence factor;

[0133] Monte Carlo sampling is used for uncertainty estimation. In the inference phase, M forward propagations (repeated sampling) are performed to calculate the predicted mean and variance:

[0134]

[0135] In this embodiment, the load demand elasticity prediction model uses a dual-branch deep network to simultaneously predict rigid load and flexible boundary, as follows:

[0136] The dual-branch deep network includes a shared encoder, a rigid load prediction branch, and a flexible boundary prediction branch;

[0137] The shared encoder, the high-dimensional feature vector set v final Extract common features:

[0138] h shared =MLP(v final );

[0139] Among them, MLP is a multi-layer perceptron; h shared is the common feature extracted by MLP;

[0140] The rigid load prediction branch and the flexible boundary prediction branch are based on the common characteristics h shared Make a prediction:

[0141]

[0142] in, is the predicted value of rigid load; h shared The common features generated by the shared encoder serve as the input for rigid load prediction; W r is the weight matrix of the fully connected layer, which maps the implicit features after nonlinear processing to the predicted value of the rigid load; Ur is the transformation matrix; b r is a bias term that compensates for the linear deviation in the feature mapping process; ReLU(·) is the activation function; is the lower boundary of the predicted flexible load; is the upper boundary of the predicted flexible load; W min , W max is the weight matrix; output boundary After being constrained by Sigmoid mapping, it is normalized to the interval [0,1];

[0143] The error function L is obtained by jointly optimizing the rigidity error L rigid , flexible boundary tightness L flex and physical constraints L constraint :

[0144] L=L rigid +λ1L flex +λ2L constraint ;

[0145]

[0146] Where N′ is the number of samples; λ1, λ2 are weight hyperparameters; is the predicted value of the rigid load of the i′th sample; is the actual rigid load value of the i′th sample; is the actual flexible load value of the i′th sample; The lower bound of the flexible load predicted for the i′th sample; is the upper bound of the flexible load predicted for the i-th sample; τ is the confidence level; () + is the positive part function, which is used to measure the uncovered part of the boundary;

[0147] Monte Carlo Dropout is used to evaluate the uncertainty of boundary predictions. M forward propagations are performed during the inference phase with Dropout enabled to statistically analyze the distribution of flexible boundaries:

[0148]

[0149] Final output, rigid load prediction value Flexible load adjustment boundary

[0150] In this embodiment, based on the predicted confidence interval of renewable energy output and the rigid load and flexible load regulation boundaries, multi-objective optimization control is adopted to obtain the optimal device control instruction set, as follows:

[0151] Randomly initialize the initial control strategy of each device in the system (such as photovoltaic power, wind power, thermal power, and flexible load regulation power) to form a control instruction set;

[0152] For each solution in the population, multi-objective values ​​are calculated based on the objective function and constraints: the objective function includes minimizing the total energy cost f1, minimizing the carbon footprint f2, and minimizing the net power fluctuation f3 of the grid, as follows:

[0153]

[0154]

[0155] in, is the power value purchased from the grid at time t; is the power value sold to the grid at time t; is the charging power of device d at time t; is the discharge power of device d at time t; It is the time-of-use electricity purchase price; is the time-of-use electricity price; γ d is the efficiency loss coefficient of device d, which is related to the charge and discharge characteristics of the device itself; δ grid,t is the carbon emission factor of the power grid; δ d is the fossil fuel emission rate of device d;

[0156] Constraints include power balance constraints, uncertainty in renewable energy output, unadjustable rigid load constraints, and flexible load adjustment boundary constraints:

[0157]

[0158] Among them, P new，t is the actual output of new energy (such as photovoltaic and wind power) at time t; The power sold to the grid; t is the standard deviation of the prediction error;

[0159] For solutions that exceed the upper and lower boundaries of flexible load regulation or the confidence interval of renewable energy output, a correction mechanism is used to limit them to the valid range;

[0160] The fitness of each solution is calculated using a target trade-off strategy based on neighborhood decomposition;

[0161] Apply crossover and mutation operations to improve the diversity of solutions and explore better solutions;

[0162] After a preset round of iterations, the Pareto optimal solution set that meets the constraints is obtained.

[0163] In this embodiment, control is performed based on the optimal device control instruction set, certified emission reductions and user response logs are obtained, and carbon credits are generated, as follows:

[0164] Define the baseline carbon emissions when users do not participate in regulation:

[0165]

[0166] in, is the historical average power purchase amount from the power grid in period t when the user does not participate in the optimization regulation; grid，t is the carbon emission factor of the regional power grid in the corresponding period;

[0167] After optimizing the control based on the optimal device control instruction set, the actual power purchase amount of the optimized user in the t period is collected in real time through the smart meter. Calculate actual carbon emissions:

[0168]

[0169] Then the emission reduction ΔE reduce =E baseline -E actual ;

[0170] Obtain records of user flexible load adjustment response logs, including time, adjustment amount, duration, and response speed, and generate carbon credits:

[0171]

[0172] in, is the emission reduction of user u; is the response speed of the g-th action of user u; is the number of user participations; α, β, and γ are the basic integral rate of emission reduction, the response speed gain coefficient, and the fixed reward for single participation in regulation, respectively; is the average response speed.

[0173] In this embodiment, a community demand response carbon emission reduction data monitoring system based on the Internet of Things technology is also provided, including a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps of the community demand response carbon emission reduction data monitoring method based on the Internet of Things technology.

[0174] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0176] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0177] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0178] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

Claims

1. A community demand response carbon emission reduction data monitoring method based on Internet of Things technology, characterized by: The following steps are involved: S1: Collect multi-source heterogeneous data of community carbon emissions and preprocess them to obtain a basic data set; S2: Based on the basic data set, perform feature engineering and deep characterization to construct a high-dimensional feature vector set; S3: Build a new energy output prediction model and predict the confidence interval of new energy output based on a high-dimensional feature vector set; S4: Construct a load demand elasticity prediction model and predict the rigid load and flexible load adjustment boundaries based on the high-dimensional feature vector set; S5: Based on the predicted confidence interval of renewable energy output and the rigid and flexible load regulation boundaries, multi-objective optimization control is used to obtain the optimal equipment control instruction set; S6: Control based on the optimal device control instruction set, obtain certified emission reductions and user response logs, and generate carbon credits.

2. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 1 is characterized in that: The multi-source heterogeneous data of community carbon emission data is collected and pre-processed to obtain a basic data set, specifically The multi-source heterogeneous data includes a list of public adjustable load equipment in the community D device , real-time equipment operation data collected by IoT sensors iot 、Environmental data D env and user behavior data D user ; The data timestamps of each subsystem have an offset Δt; First, the dynamic time warping algorithm is used to compensate for the time deviation: C i,j =|t device,i -t user,j |+λ·SED(v i ,v j ) Among them, C i,j is the cost matrix element of dynamic time warping, which represents the alignment cost between the time series of community device data and the time series of user behavior data; SED is the standardized Euclidean distance, λ is the weight coefficient; i is the time index of device data; j is the time index of user behavior data; t device,i is the device data timestamp, indicating the time of the i-th device data sampling point; t user,j is the timestamp of user behavior data, indicating the time of the jth user behavior data sampling point; Then, the outlier detection is performed based on the sliding window algorithm of Grubbs test, and outliers are eliminated; finally, the missing values ​​are filled by the spatiotemporal correlation construction method; the preprocessed basic data set D is obtained. raw ={D device ,D iot ,D env ,D user }.

3. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 2 is characterized in that: The high-dimensional feature vector set v final Including statistical features v statistical , frequency domain features v freq , device cluster characteristics v cluster And the deeply fused multimodal features h fused , v final =[v statistical ;v freq ;v cluster ;h fused ], as follows: Statistical characteristics include mean u, standard deviation σ and trend term β: Where T is the total length of the time series; x t is the power value of the time series at the tth moment; is the time center point of the time series; The frequency domain feature is the main frequency component extracted by Fourier transform (FFT) of the power time series data: Spatial aggregation of devices in the same area (such as buildings and charging stations) is performed to obtain device cluster characteristics: The deeply fused multimodal features are obtained based on the fusion of device, environment and user features.

4. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 3 is characterized in that: The deep fusion multimodal features are obtained by fusing device, environment, and user features, as follows: Get device characteristics in the community device 、Environmental characteristics env , user characteristics h use , and construct cross-attention weights for device, environment, and user features: Where Q = W Q h device For query, K = W K h user is a key, V=E V h env is the value, the superscript T indicates transposition, d k is the dimension of the key; W Q ,W K ,W V are the weight coefficients of query, key and value respectively; Attention(Q,K,V) is the attention output; h device is the device characteristic vector, including the power and load rate of the device; h user is the user feature vector, including user behavior preference and elasticity index; h env is the environmental feature vector, including light intensity and ambient temperature; The final multimodal features are obtained by dynamically weighting the multimodal features through the gating mechanism: g=σ(W g [h device ;h user ;h env ]) h fused =g☉h device +(1-g)☉h user ; Where g is the gating weight vector; Wg is the learnable gating parameter matrix; σ is the Sigmoid activation function; and ⊙ is element-wise multiplication.

5. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 1 is characterized in that: The S3 is specifically S31: From the high-dimensional feature vector set v final Extract multimodal key features; S32: Construct a new energy output prediction model based on the Transformer-base model, wherein the Transformer-base model includes a Transformer and a regression head; the Transformer embeds the high-dimensional feature vector v final After input, it is embedded into a fixed-dimensional space d model Then, the temporal and spatial dependencies between the input features are captured by the multi-head self-attention output, and the regression head regresses the Transformer output to calculate the output of the next h time steps. S33: Use uncertainty quantification methods to evaluate model output intervals.

6. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 5 is characterized in that: The uncertainty quantification method is used to evaluate the model output interval, as follows: The prediction confidence interval CI(t+h) reflects the model's confidence in the prediction of renewable energy output. and the prediction uncertainty σ(t+h) is expressed as: Where k′ is the confidence factor; Monte Carlo sampling is used for uncertainty estimation. In the inference phase, M forward propagations are performed to calculate the predicted mean and variance:

7. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 1 is characterized in that: The load demand elasticity prediction model uses a dual-branch deep network to simultaneously predict rigid loads and flexible boundaries as follows: The dual-branch deep network includes a shared encoder, a rigid load prediction branch, and a flexible boundary prediction branch; The shared encoder, the high-dimensional feature vector set v final Extract common features: h shared =MLP(v final ); Among them, MLP is a multi-layer perceptron; h shared is the common feature extracted by MLP; The rigid load prediction branch and the flexible boundary prediction branch are based on the common characteristics h shared Make a prediction: in, is the predicted value of rigid load; h shared The common features generated by the shared encoder serve as the input for rigid load prediction; W r is the weight matrix of the fully connected layer, which maps the implicit features after nonlinear processing to the predicted value of the rigid load; Ur is the transformation matrix; b r is a bias term that compensates for the linear deviation in the feature mapping process; ReLU(·) is the activation function; is the lower boundary of the predicted flexible load; is the upper boundary of the predicted flexible load; W min , W max is the weight matrix; output boundary After being constrained by Sigmoid mapping, it is normalized to the interval [0,1]; The error function L is obtained by jointly optimizing the rigidity error L rigid , flexible boundary tightness L flex and physical constraints L constraint : L=L rigid +λ1L flex +λ2L constraint ; Where N′ is the number of samples; λ1, λ2 are weight hyperparameters; is the predicted value of the rigid load of the i′th sample; is the actual rigid load value of the i′th sample; is the actual flexible load value of the i′th sample; The lower bound of the flexible load predicted for the i′th sample; is the upper bound of the flexible load predicted for the i-th sample; τ is the confidence level; () + is the positive part function, which is used to measure the uncovered part of the boundary; Monte Carlo Dropout is used to evaluate the uncertainty of boundary predictions. M forward propagations are performed during the inference phase with Dropout enabled to statistically analyze the distribution of flexible boundaries: Final output, rigid load prediction value Flexible load adjustment boundary 8. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 1 is characterized in that: The multi-objective optimization control is adopted based on the predicted confidence interval of renewable energy output and the rigid load and flexible load adjustment boundaries to obtain the optimal equipment control instruction set, as follows: Randomly initialize the initial control strategy of each device in the system (such as photovoltaic power, wind power, thermal power, and flexible load regulation power) to form a control instruction set; For each solution in the population, multi-objective values ​​are calculated based on the objective function and constraints: the objective function includes minimizing the total energy cost f1, minimizing the carbon footprint f2, and minimizing the net power fluctuation f3 of the grid, as follows: in, is the power value purchased from the grid at time t; is the power value sold to the grid at time t; is the charging power of device d at time t; is the discharge power of device d at time t; It is the time-of-use electricity purchase price; is the time-of-use electricity price; γ d is the efficiency loss coefficient of device d, which is related to the charge and discharge characteristics of the device itself; δ grid,t is the carbon emission factor of the power grid; δ d is the fossil fuel emission rate of device d; Constraints include power balance constraints, uncertainty in renewable energy output, unadjustable rigid load constraints, and flexible load adjustment boundary constraints: Among them, P new，t The actual output of new energy at time t; The power sold to the grid; t is the standard deviation of the prediction error; For solutions that exceed the upper and lower boundaries of flexible load regulation or the confidence interval of renewable energy output, a correction mechanism is used to limit them to the valid range; The fitness of each solution is calculated using a target trade-off strategy based on neighborhood decomposition; Apply crossover and mutation operations to improve the diversity of solutions and explore better solutions; After a preset round of iterations, the Pareto optimal solution set that meets the constraints is obtained.

9. The method for monitoring community demand response carbon emission reduction data based on Internet of Things technology according to claim 1 is characterized in that: The control is based on the optimal device control instruction set, and the certified emission reduction and user response logs are obtained, and carbon credits are generated, as follows: Define the baseline carbon emissions when users do not participate in regulation: in, is the historical average power purchase amount from the power grid in period t when the user does not participate in the optimization regulation; grid，t is the carbon emission factor of the regional power grid in the corresponding period; After optimizing the control based on the optimal device control instruction set, the actual power purchase amount of the optimized user in the t period is collected in real time through the smart meter. Calculate actual carbon emissions: Then the emission reduction ΔE reduce =E baseline -E actual ; Obtain records of user flexible load adjustment response logs, including time, adjustment amount, duration, and response speed, and generate carbon credits: in, is the emission reduction of user u; is the response speed of the g-th action of user u; is the number of user participations; α, β, and γ are the basic integral rate of emission reduction, the response speed gain coefficient, and the fixed reward for single participation in regulation, respectively; is the average response speed.

10. A community demand response carbon emission reduction data monitoring system based on Internet of Things technology, characterized by: It includes a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the community demand response carbon emission reduction data monitoring method based on Internet of Things technology as described in any one of claims 1 to 9.

Images