Low-carbon demand response resource allocation method based on dynamic carbon emission factors

Abstract

The invention discloses a low-carbon demand response resource allocation method based on a dynamic carbon emission factor. The method comprises the following steps: 1, solving to obtain the dynamic carbon emission factor; 2, performing adaptive clustering on the dynamic carbon emission factor data and the power consumption data based on a Dirichlet process hybrid model to obtain clustering labels of carbon emission and power consumption in different time periods and probability distribution of the clustering labels; 3, on the basis of historical data and the result of the step 2, predicting the electricity consumption and the dynamic carbon emission factor in the next 24 hours by using double LSTMs (Long Short Term Memory); 4, constructing an optimization model responding to the carbon emission reduction demand based on the fixed daily electricity consumption constraint and the calculation result of the step 3; and 5, on the basis of rigid constraint and flexible constraint conditions, solving by using the demand response optimization model to obtain electricity demand and carbon emission in the next 24 hours for configuring an electricity scheme. According to the method, the contribution of the demand response measure to carbon emission reduction can be accurately evaluated, collaborative optimization of the user side and the power grid side can be realized, and the carbon emission reduction benefit is maximized.

Description

Technical Field

[0001] The present invention relates to the technical field of low-carbon dispatching of power systems, and particularly to a method for optimizing the allocation of low-carbon demand response resources based on dynamic carbon emission factors. Background Art

[0002] With the increasingly serious global climate change problem, as one of the main sources of carbon emissions, the power system has higher requirements for its low-carbon transformation. The traditional demand response mechanism mainly focuses on the economy and reliability of the power system, and suppresses load fluctuations and reduces system operation costs by motivating users to adjust their electricity consumption behaviors, but fails to take carbon emissions as the core consideration factor. At the same time, existing carbon emission assessments mostly adopt static carbon emission factors. This method assumes that the carbon emission intensity of power generation units remains constant at different times, ignoring the impacts of factors such as the fluctuations of renewable energy output, dynamic load changes, and grid operation status on real-time carbon emissions, resulting in a significant deviation between the carbon emission assessment results and the actual situation.

[0003] Currently, the power system faces the following main problems: First, the traditional demand response mechanism overemphasizes economic benefits and only guides user behaviors through electricity price signals, failing to fully consider the carbon emission characteristics under different spatio-temporal conditions and being difficult to effectively support the low-carbon transformation of the power system; Second, the calculation method of static carbon emission factors cannot accurately reflect the real-time operation state of the power system. Especially in the context of large-scale integration of renewable energy, the volatility and intermittency characteristics on the power generation side and the dynamic changes on the load side make carbon emissions show obvious spatio-temporal differences; Third, there is a lack of a resource optimization allocation method that deeply integrates dynamic carbon emission factors with demand response. Existing dispatching models often take economy as the main optimization goal and fail to establish an effective coupling mechanism between carbon emissions and power dispatching, resulting in the low-carbon dispatching effect being difficult to meet the requirements of low carbon emissions.

[0004] Therefore, developing a low-carbon demand response resource optimization allocation technology based on dynamic carbon emission factors has important theoretical value and practical significance. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for allocating low-carbon demand response resources based on dynamic carbon emission factors. By constructing a dynamic carbon emission factor model considering the time dimension, the optimal time-series dispatching of electricity load is realized under the constraint of fixed electricity consumption to minimize the overall carbon emissions of the system.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a method for allocating low-carbon demand response resources based on dynamic carbon emission factors, including the following steps:

[0008] S1. Integrate the marginal carbon emission factor and the average carbon emission factor to obtain the dynamic carbon emission factor;

[0009] S2. Based on the Dirichlet process mixture model, perform adaptive clustering on the dynamic carbon emission factor data and the electricity consumption data to obtain the clustering labels and their probability distributions of carbon emissions and electricity consumption in different time periods, realizing the intelligent identification and classification of carbon emission characteristics and electricity consumption characteristics in different time periods;

[0010] S3. Based on the historical data and the results of S2, use the dual LSTM deep learning network to predict the electricity consumption and the dynamic carbon emission factor in the next 24 hours;

[0011] S4. Under the constraint of the fixed daily electricity consumption, construct a demand response optimization model with the goal of maximizing carbon emission reduction based on the calculation results of S3;

[0012] S5. Based on the rigid constraint and the flexible constraint conditions, use the demand response optimization model to solve and obtain the electricity demand and carbon emissions in the next 24 hours, which are used to configure the electricity consumption plan.

[0013] In some embodiments, the marginal carbon emission factor is:

[0014]

[0015] Wherein,

[0016]

[0017] The average carbon emission factor AEF t Reflects the average carbon emission level of the entire power grid at a specific time point. The average carbon emission factor is:

[0018]

[0019] Wherein,

[0020]

[0021] The Is a piecewise function used to determine the capacity utilization rate of the power plant.

[0022] When It means that when the cumulative installed capacity is less than the reserved capacity, the power plant operates at full load.

[0023] When It means that when the previous cumulative installed capacity has reached or exceeded the reserved capacity, the power plant stops operating.

[0024] In other cases, it means partial load operation, operating according to the ratio of the remaining demand to the installed capacity.

[0025] where ε p is the unit carbon emission of the specific fuel of the p-th power plant p, p = 1, 2,..., P; η T is the transmission efficiency; is the residual electricity load at time t; is the installed capacity of the i-th generating unit; is the installed capacity of power plant p; is the capacity utilization rate; is the electricity generated by each fuel type f at time step t, and F is the set of all power generation fuel types.

[0026] In some embodiments, the dynamic carbon emission factor is:

[0027] CEF t = αAEF t +(1 - α)MET t ;

[0028] where α is the weight coefficient, α ∈ (0, 1).

[0029] In some embodiments, S2 includes the following steps:

[0030] S21. Construct a Dirichlet process mixture model;

[0031] S22. Train the Dirichlet process mixture model by Gibbs sampling method until the model converges;

[0032] S23. Based on the trained Dirichlet process mixture model, obtain the clustering labels and their probability distributions for each time period;

[0033] S24. Identify time periods with similar carbon emission characteristics according to the clustering results of S23.

[0034] In some embodiments, the Dirichlet process mixture model is:

[0035]

[0036] The process of generating carbon emission factor data is:

[0037]

[0038] The process of generating electricity consumption data is:

[0039]

[0040] where G is the random probability measure; DP(α, G0) is the Dirichlet process, α is the concentration parameter, and G0 is the base measure; θ i is the θ-thi Parameters of the observed data; x i is a random variable representing the observed data; F(θ i ) is the likelihood function of the observed data; π i is the mixing function satisfying ; N(μ i , ∑ i ) is the Gaussian distribution, and μ i , ∑ i are the mean and covariance matrix of the i-th component respectively; ω j is the mixing weight; v j , Λ j are the mean and covariance matrix of the j-th component respectively.

[0041] In some embodiments, in S3, after obtaining the feature categories of the data at each moment in the previous 7 days, when constructing the input of the LSTM visual neural network prediction model, this feature category information is fused with the original data. The input feature vector of the LSTM deep learning network for predicting the dynamic carbon emission factor is:

[0042] X cf (f) = [cf t , class cf (t)];

[0043] where class cf (t) = argmin i∈[1,k] d cf (t, i);

[0044] The input feature vector of the LSTM deep learning network for predicting the electricity consumption is:

[0045] X ec (f) = [ec t , class ec (t)];

[0046] where class ec (t) = argmin i∈[1,m] d ec (t, i);

[0047] In the formula, cf t is the dynamic carbon emission factor data at time t; class cf (t) is the feature category to which the dynamic carbon emission factor data at time t belongs; ec t is the electricity consumption data at time t; class ec (t) is the feature category to which the electricity consumption data at time t belongs; k represents the total number of feature categories of the dynamic carbon emission factor data; m represents the total number of feature categories of the electricity consumption data.

[0048] In some embodiments, in S4, the demand response optimization model with the goal of maximizing carbon emission reduction is as follows:

[0049]

[0050] In the formula, CEF t is the predicted 24-hour dynamic carbon emission factor; is the increase in electricity consumption; is the decrease in electricity consumption. The increase in electricity consumption is equal to the decrease in electricity consumption, that is, the electricity consumption is fixed.

[0051] In some embodiments, in S5, the rigid constraints include: daily electricity consumption conservation constraint, power limit constraint, and equipment safe operation constraint; the flexible constraints include load change rate constraint and user comfort constraint.

[0052] In some embodiments, in order to ensure that the total electricity consumption remains unchanged before and after demand response regulation, the daily electricity consumption conservation constraint is:

[0053]

[0054] Considering the carrying capacity and safe operation requirements of the power system, the actual load in each time period must be controlled within the range allowed by the system. The power limit constraint is:

[0055]

[0056] To protect the safe operation of electrical equipment, the amplitude of a single load adjustment needs to be restricted. The equipment safe operation constraint is:

[0057]

[0058] To avoid the impact of violent load fluctuations on the power grid, the load change rate between adjacent time periods needs to be restricted. The load change rate constraint is:

[0059]

[0060] Considering the user experience and controlling the impact of load adjustment on users' daily life and production activities, the user comfort constraint is:

[0061]

[0062] In the formula, is the increase in electricity consumption; is the decrease in electricity consumption; is the reference load; P max 、P minare the maximum and minimum allowable loads respectively; is the maximum allowable power increase; is the maximum allowable power decrease; R r represents the maximum allowable difference in power change between adjacent time periods; β is the load adjustment proportionality coefficient acceptable to users.

[0063] On the other hand, the present invention provides a low-carbon demand response resource allocation system based on dynamic carbon emission factors, using the above method, including the following modules:

[0064] Dynamic carbon emission factor calculation module: calculates the dynamic carbon emission factor by using the marginal carbon emission factor and the average carbon emission factor;

[0065] Clustering module: adaptively clusters the dynamic carbon emission factor data and the electricity consumption data based on the Dirichlet process mixture model to obtain the clustering labels and their probability distributions of carbon emissions and electricity consumption in different time periods;

[0066] Prediction module: based on historical data and the results of S2, uses a dual LSTM deep learning network to predict the electricity consumption and dynamic carbon emission factors for the next 24 hours;

[0067] Demand response optimization module: based on the fixed daily electricity consumption constraint, constructs a carbon emission reduction demand response optimization model based on the calculation results of S3;

[0068] Demand response scheduling module: based on rigid constraints and flexible constraint conditions, uses the demand response optimization model to solve for the electricity demand and carbon emissions for the next 24 hours, which are used to configure the electricity consumption plan.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] Aiming at the problem that the demand response methods in the prior art mainly focus on electricity prices and electricity consumption and ignore the dynamic characteristics of carbon emissions. The present invention introduces dynamic carbon emission factors into demand response optimization, enabling users to flexibly adjust their electricity consumption behaviors according to the real-time carbon emission intensity of the power grid and achieving precise carbon emission reduction.

[0071] To solve the problem of inaccurate identification of the carbon emission characteristics and electricity consumption characteristics of the power system. The present invention can accurately capture the carbon emission and electricity consumption patterns in different time periods through the adaptive clustering method of the Dirichlet process mixture model, providing more reliable data support for demand response decision-making.

[0072] The present invention uses a dual LSTM deep learning network to realize the joint prediction of electricity consumption and dynamic carbon emission factors for 24 hours, significantly improving the prediction accuracy and providing more accurate inputs for demand response optimization.

[0073] To solve the problem that it is difficult to balance user experience and environmental protection benefits in demand response optimization. The present invention establishes a complete constraint system including multi-dimensional constraints such as user comfort and equipment safety, and ensures the actual usage needs of users while achieving the maximum carbon emission reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0075] Figure 2 It is a schematic diagram comparing the total carbon emissions before and after optimization in Embodiment 1 of the present invention;

[0076] Figure 3 It is a schematic diagram comparing the carbon emissions in 24 hours before and after optimization in Embodiment 1 of the present invention;

[0077] Figure 4 It is a schematic diagram of the increase and decrease of the load in 24 hours in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0079] Embodiment 1:

[0080] Please refer to Figure 1 , a low-carbon demand response resource allocation method based on dynamic carbon emission factors. By constructing a dynamic carbon emission factor evaluation model considering spatio-temporal characteristics, it accurately depicts the carbon emission characteristics of the power system under different operating states, and introduces the carbon emission index as a core element into the demand response optimization decision. It can not only accurately evaluate the contribution of demand response measures to carbon emission reduction, but also achieve the collaborative optimization of the user side and the grid side, maximizing the carbon emission reduction benefit while ensuring the economy and reliability of the system. At the same time, this embodiment provides a new technical path for the low-carbon transformation of the power system, which helps to improve the low-carbon operation level of the power system and promote the in-depth participation of demand-side resources in system regulation.

[0081] The specific steps are as follows:

[0082] S1. Integrate the marginal carbon emission factor and the average carbon emission factor to obtain the dynamic carbon emission factor.

[0083] Search for and integrate multi-source data of the power system, including generator set characteristics, fuel emission levels, electricity consumption, etc. in a certain region and year. Use the piecewise linear method to perform piecewise linear fitting on the generator efficiency curve and transmission efficiency curve to obtain the power generation efficiency and transmission efficiency, and calculate the dynamic marginal carbon emission factor and average carbon emission factor from the available data.

[0084] The unit carbon emission of a specific fuel in a certain power plant is given by the following formula:

[0085]

[0086] In the formula, ε f is the carbon emission intensity of a certain fuel type f; is the power generation efficiency of power plant p for this fuel.

[0087] The marginal carbon emission factor MET that changes with time t is:

[0088]

[0089] Among them,

[0090]

[0091] In the formula, is the carbon emission of the power plant in a given time step.

[0092] The marginal carbon emission factor reflects the incremental carbon emission when the load changes. MET t represents the additional carbon emission brought about by an increase in the load by one unit.

[0093] The average carbon emission factor AEFt reflects the average carbon emission level of the entire power grid at a specific time point:

[0094]

[0095] Among them,

[0096]

[0097] The is a piecewise function used to determine the capacity utilization rate of the power plant.

[0098] When it means that when the cumulative installed capacity is less than the reserved capacity, the power plant operates at full load.

[0099] When it means that when the previous cumulative installed capacity has reached or exceeded the reserved capacity, the power plant stops operating.

[0100] In other cases, it represents partial load operation and operates according to the ratio of the remaining demand to the installed capacity.

[0101] The calculation of the dynamic carbon emission factor is expressed as:

[0102] CEF t = αAEF t +(1 - α)MET t ;

[0103] In the formula, α is the weight coefficient, α ∈ (0, 1), and is determined according to the degree of load change.

[0104] Calculating the dynamic carbon emission factor through the weighted fusion of the marginal carbon emission factor and the average carbon emission factor has significant advantages. This embodiment can capture both the instantaneous dynamic characteristics and the long-term steady-state characteristics of the power system: MET t reflects the short-term carbon emission impact brought by load changes and is suitable for evaluating the emission reduction effect of short-term regulation measures such as demand response; AEF t reflects the overall carbon emission level of the system and is suitable for evaluating the effect of long-term emission reduction strategies such as clean energy substitution. By adjusting the weight coefficient α, the influence degrees of short-term and long-term characteristics can be flexibly balanced according to the actual application scenario. When the load fluctuates violently, increase the weight of MET t to highlight the marginal effect, and when the system is relatively stable, increase the weight of AEF t to reflect the overall characteristics. This weighted fusion mechanism provides a more comprehensive and accurate carbon emission assessment framework and can provide a more reliable theoretical basis for the low-carbon scheduling decision-making of the power system.

[0105] S2. Based on the Dirichlet Process Mixture Model (DPMM), perform adaptive clustering on the dynamic carbon emission factor data and electricity consumption data to obtain the clustering labels and their probability distributions of carbon emissions and electricity consumption in different time periods, realizing the intelligent identification and classification of carbon emission characteristics and electricity consumption characteristics in different time periods. The Dirichlet Process Mixture Model is a non-parametric Bayesian model.

[0106] Aiming at the complex non-linear characteristics of the dynamic carbon emission factor and electricity consumption data, a non-parametric Bayesian clustering method is used for modeling and analysis.

[0107] Let the dynamic carbon emission factor data set calculated by S1 be: CEF = {cef1, cef2,..., cef n};

[0108] The electricity consumption data is: EC = {ec1, ec2,..., ec n}

[0109] where n is the number of samples.

[0110] Since the carbon emission characteristics in the power system are affected by multiple factors such as the power source structure and the output of renewable energy, and the power load shows strong randomness and volatility, traditional clustering methods are difficult to effectively mine the potential laws in the data. Therefore, the Dirichlet process mixture model (DPMM) is used for clustering analysis. The Dirichlet process mixture model DPMM is expressed as:

[0111]

[0112] The generation process of carbon emission factor data is:

[0113]

[0114] The generation process of power consumption data is:

[0115]

[0116] The posterior distribution of the Dirichlet process mixture model DPMM is inferred by Gibbs sampling:

[0117]

[0118] In the formula, θ i is; θ -i is all parameters except θ i ; f(x i |θ i ) is the likelihood function; δ(θ j ) is the Dirac function; x 1:n is all observed data sequences; x i is the i-th observed data; α is the mixing weight parameter; G0 is the base distribution.

[0119] Through the Dirichlet process mixture model DPMM, the clustering results of carbon emission factor data CEF clusters ={CEF1, CEF2,..., CEF k} and power consumption data EC clusters ={EC1, EC2,..., EC m} can be obtained, where k and m are the optimal clustering numbers determined adaptively.

[0120] This embodiment fully considers the dynamics and uncertainty of carbon emissions in the power system, and realizes adaptive identification of carbon emission patterns through non-parametric Bayesian methods, providing a new analysis tool for building a demand response mechanism based on dynamic carbon emission factors. The Dirichlet process mixed model DPMM can effectively capture and reflect the impact of multi-dimensional time and social factors on carbon emissions when clustering dynamic carbon emission factors. In the time dimension, the difference in carbon emission patterns between weekdays and holidays, as well as the changes in electricity load caused by seasonal changes, can be identified; in the intraday time series, the emission characteristics of typical time periods such as morning and evening peaks and intensive industrial production periods can be distinguished. At the same time, it can also reflect the carbon emission fluctuations caused by special factors such as major social activities (such as important events, large-scale exhibitions), extreme weather events, and public emergencies. In addition, the adaptive clustering characteristics of the Dirichlet process mixed model DPMM enable it to capture the impact of medium- and long-term social development trends such as regional industrial structure adjustment, changes in energy consumption habits, and the popularization of new energy vehicles on carbon emission patterns, providing data support for the formulation of more accurate demand response strategies. The results can be used to guide users to flexibly adjust their electricity consumption behavior according to the carbon emission characteristics of different time periods and promote the low-carbon operation of the power system.

[0121] S3, based on historical data and the results of S2, uses a dual LSTM deep learning network to predict electricity consumption and dynamic carbon emission factors for the next 24 hours.

[0122] When constructing the input features of the LSTM deep learning network prediction model, it is first necessary to determine the feature categories of the historical data of the previous 7 days. These historical data need to be separately calculated with the feature class centers obtained by clustering the Dirichlet process mixture model DPMM to determine the feature category to which they belong.

[0123] For dynamic carbon emission factor data, assume that the Dirichlet process mixture model DPMM clustering obtains k feature class centers C = {C1, C2, ..., C k}, then the distance between the carbon emission factor data at each time t and the center of the i-th cluster can be expressed as:

[0124] d cf (t,i)=‖cf t -c i ‖;

[0125] For the carbon emission factor data at time t, the characteristic category to which it belongs can be determined by the following method:

[0126] class cf (t) = argmin i∈[1,k] d cf (t,i);

[0127] Similarly, for the electricity consumption data, assuming that the Dirichlet process mixture model DPMM clustering obtains m characteristic class centers, the distance between the electricity consumption data at each moment t and the j-th class center can be expressed as:

[0128] d ec (t,j) = ||ec t -e j ||;

[0129] The corresponding characteristic category is determined as:

[0130] class ec (t) = argmin i∈[1,m] d ec (t,i);

[0131] Through the above calculations, the characteristic categories of the data at each moment in the previous 7 days can be obtained. When constructing the input of the LSTM deep learning network prediction model, this characteristic category information is fused with the original data. For the LSTM model used for dynamic carbon emission factor data prediction, the input feature vector can be expressed as:

[0132] X cf (f) = [cf t ,class cf (t)];

[0133] The input feature vector of the LSTM model for electricity consumption prediction is:

[0134] X ec (f) = [ec t ,class ec (t)];

[0135] In the design of the LSTM deep learning network structure, LSTM deep learning networks for electricity consumption prediction and carbon emission factor prediction are respectively constructed. Each network adopts a three-layer LSTM architecture: the first layer is configured with 128 LSTM units, mainly responsible for extracting basic features from complex input sequences; the second layer sets 64 units to fuse and optimize the features extracted by the first layer; the third layer contains 32 units for deep feature extraction and abstraction. Finally, the predicted values of electricity consumption and carbon emission factors for the next 24 hours are output through the fully connected layer. During the training process, all input data is standardized to convert features with different dimensions to the same scale range to improve the training effect and prediction accuracy of the model.

[0136] The establishment of a demand response optimization model using the predicted data for the next 24 hours can help users plan and adjust their electricity consumption behavior in advance. Through predictive power load management, users can shift high-power consumption activities from peak electricity consumption periods to off-peak periods, not only avoiding high electricity price periods but also reducing carbon emissions.

[0137] S4. Based on the fixed daily electricity consumption constraint, construct a carbon emission reduction demand response optimization model based on the calculation results of S3.

[0138] Assume that the electricity consumption in the next 24 hours is fixed, and establish a demand response optimization model with the goal of maximizing carbon emission reduction based on the data predicted by S3. The objective function is:

[0139]

[0140] Among them, The increase in electricity consumption is equal to the decrease in electricity consumption, that is, the electricity consumption is fixed.

[0141] Achieve carbon emission reduction by intelligently adjusting the electricity consumption time period. This method not only ensures the user experience but also makes full use of the difference in carbon emission intensity of power generation at different time periods of the power grid, maximizing the emission reduction effect while keeping the total electricity consumption unchanged.

[0142] S5. Based on the rigid constraints and flexible constraint conditions, use the demand response optimization model to solve for the electricity demand and carbon emissions in the next 24 hours for configuring the electricity consumption plan.

[0143] In the process of optimizing the allocation of low-carbon demand response resources, the construction of the constraint system is the key to ensuring the feasibility of the plan. These constraints can be divided into two categories: rigid constraints and flexible constraints.

[0144] Rigid constraints include the following types of constraints:

[0145] (1) Daily electricity consumption conservation constraint:

[0146] Ensure that the total electricity consumption remains unchanged before and after the demand response adjustment.

[0147]

[0148] (2) Power limit constraint:

[0149] Considering the carrying capacity and safe operation requirements of the power system, the actual load in each time period must be controlled within the range allowed by the system:

[0150]

[0151] The power limit constraint prevents the system from being overloaded or operating unstably due to load regulation.

[0152] (3) Equipment safe operation constraints:

[0153] In order to protect the safe operation of electrical equipment, the range of single load adjustment needs to be limited.

[0154]

[0155] Existing methods often ignore the bearing capacity of equipment, which can easily cause equipment damage. This embodiment takes into account the equipment safety constraints to extend the service life of the equipment and reduce the risk of equipment failure.

[0156] Flexible constraints include the following constraints:

[0157] (1) Load change rate constraint:

[0158] In order to avoid the impact of severe load fluctuations on the power grid, it is necessary to limit the load change rate between adjacent time periods:

[0159]

[0160] The load change rate constraint helps maintain the smooth operation of the system and reduce grid fluctuations caused by rapid load changes. r The setting needs to consider the system's regulation capability and stability requirements.

[0161] (2) User comfort constraints:

[0162] This type of constraint mainly considers user experience and controls the impact of load adjustment on users' daily life and production activities:

[0163]

[0164] Where β is the load adjustment ratio coefficient acceptable to the user.

[0165] The user comfort constraint ensures that the load adjustment is within the acceptable range for the user, avoiding significant interference to the user's normal power consumption. Different β values ​​can be set according to the characteristics of different types of users (such as residential, commercial, and industrial users).

[0166] By reasonably setting and dynamically adjusting these constraints, the optimal low-carbon demand response scheduling plan can be achieved while ensuring system safety, user comfort and equipment reliability.

[0167] S4 and S5 together constitute an optimization solution algorithm, and ultimately the optimal low-carbon demand response resource scheduling based on dynamic carbon emission factors is obtained.

[0168] like Figure 2As shown in Fig. -4, first, the electricity load data and grid carbon emission factor data for the next 24 hours are obtained through a prediction algorithm. Multiply the predicted hourly electricity consumption by the carbon emission factor for the corresponding period to obtain the baseline carbon emission situation before optimization. Subsequently, the 24-hour electricity load is redistributed through an optimized scheduling algorithm to obtain an optimized electricity consumption plan while ensuring electricity demand.

[0169] Figure 2 The bar chart in [reference] compares and shows the changes in the total carbon emissions in 24 hours before and after the optimized scheduling. The total carbon emissions of the system before optimization were approximately 1.33 units. After the load optimized scheduling, the total carbon emissions decreased to approximately 1.28 units, achieving a reduction effect of approximately 3.8%. Figure 3 It details the comparison of the carbon emissions per hour before and after optimization, and it can be intuitively seen the emission reduction effects in different periods. Figure 4 Then, it shows the specific increase and decrease changes in the carbon emissions in each period within 24 hours through a change analysis chart to help analyze the effect distribution of the optimized scheduling. Through this optimized scheduling scheme based on prediction data, it is possible to make full use of the periods with lower grid carbon emission factors to arrange the electricity load, thereby achieving a reduction in overall carbon emissions.

[0170] Embodiment 2

[0171] A low-carbon demand response resource allocation system based on dynamic carbon emission factors, using the above method, includes the following modules:

[0172] Dynamic carbon emission factor calculation module: Calculate the dynamic carbon emission factor using the marginal carbon emission factor and the average carbon emission factor;

[0173] Clustering module: Based on the Dirichlet process mixture model, perform adaptive clustering on the dynamic carbon emission factor data and electricity consumption data to obtain the clustering labels and their probability distributions of carbon emissions and electricity consumption in different periods;

[0174] Prediction module: Based on historical data and the results of S2, use a dual LSTM deep learning network to predict the electricity consumption and dynamic carbon emission factors for the next 24 hours;

[0175] Demand response optimization module: Based on the fixed daily electricity consumption constraint, construct a carbon emission reduction demand response optimization model based on the calculation results of S3;

[0176] Demand response scheduling module: Based on rigid constraints and flexible constraint conditions, use the demand response optimization model to solve for the electricity demand and carbon emissions for the next 24 hours, which are used to configure the electricity consumption plan.

[0177] A low-carbon demand response resource allocation system based on dynamic carbon emission factors of the present invention can be installed in a computer device. The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a low-carbon demand response resource allocation program based on dynamic carbon emission factors. Among them, the memory includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. The processor is the control core of the electronic device, connecting various components of the entire computer device through various interfaces and lines, and by running or executing programs or modules stored in the memory, and calling data stored in the memory, to execute various functions of the computer device and process data.

[0178] The module described in the present invention refers to a series of computer program segments that can be executed by the processor of a computer device and can complete fixed functions, and are stored in the memory of the computer device.

[0179] After considering the specification and practicing the content disclosed herein, those skilled in the art will readily think of other embodiments of the present application. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include common general knowledge or conventional technical means in the technical field not disclosed in the present application.

Claims

1. A low-carbon demand response resource allocation method based on dynamic carbon emission factors, characterized in that: The following steps are involved: S1. The dynamic carbon emission factor is obtained by integrating the marginal carbon emission factor and the average carbon emission factor; S2. Adaptively cluster the dynamic carbon emission factor data and electricity consumption data based on the Dirichlet process mixture model to obtain the clustering labels and probability distribution of carbon emissions and electricity consumption in different time periods; S3, based on historical data and the results of S2, uses a dual LSTM deep learning network to predict the electricity consumption and dynamic carbon emission factors for the next 24 hours; S4: Based on the fixed daily electricity consumption constraint and the calculation results of S3, an optimization model for demand response is constructed with the goal of maximizing carbon emission reduction; S5. Based on rigid constraints and flexible constraints, the demand response optimization model is used to solve the electricity demand and carbon emissions for the next 24 hours, which are used to configure the electricity plan.

2. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 1, characterized in that: The marginal carbon emission factor is: in, The average carbon emission factor is: in, In the formula, ε p is the unit carbon emission of the specific fuel of the p-th power plant, p = 1, 2, ..., P; η T For transmission efficiency; is the residual power load at time t; is the installed capacity of the i-th generator set; is the installed capacity of power plant p; is the capacity utilization; is the amount of electricity generated by each fuel type f at time step t, and F is the set of all power generation fuel types.

3. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 2, characterized in that: The dynamic carbon emission factor is: CEF t =αAEF t +(1-α)MET t ; In the formula, α is the weight coefficient, α∈(0,1).

4. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 1, characterized in that: S2 includes the following steps: S21, construct a Dirichlet process mixture model; S22, training the Dirichlet process mixture model by Gibbs sampling method until the model converges; S23, based on the trained Dirichlet process mixture model, obtaining cluster labels and their probability distributions for each time period; S24. Identify time periods with similar carbon emission characteristics based on the clustering results of S23.

5. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 4, characterized in that: The Dirichlet process mixture model is: The process of generating carbon emission factor data is as follows: The process of generating electricity consumption data is as follows: Where G is the random probability measure; DP(α,G0) is the Dirichlet process, α is the concentration parameter, G0 is the benchmark measure; θ i is the θth i Parameters of observation data; x i is a random variable representing the observed data; F(θ i ) is the likelihood function of the observed data; π i To satisfy The mixing function of i ,∑ i ) is Gaussian distribution, μ i ,∑ i are the mean and covariance matrix of the i-th component respectively; ω j is the mixing weight; v j ,Λ j are the mean and covariance matrices of the j-th component, respectively.

6. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 1, characterized in that: In S3, the input feature vector of the LSTM deep learning network for predicting dynamic carbon emission factors is: X cf (f)=[cf t ,class cf (t)]; Among them, class cf (t) = argmin i∈[1,k] d cf (t,i); The input feature vector of the LSTM deep learning network for predicting electricity consumption is: X ec (f)=[ec t ,class ec (t)]; Among them, class ec (t) = argmin i∈[1,m] d ec (t,i); In the formula, cf t is the dynamic carbon emission factor data at time t; class cf (t) is the characteristic category to which the dynamic carbon emission factor data at time t belongs; ec t is the electricity consumption data at time t; class ec (t) is the characteristic category to which the electricity consumption data at time t belongs; k represents the total number of characteristic categories of dynamic carbon emission factor data; and m represents the total number of characteristic categories of electricity consumption data.

7. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 1, characterized in that: In S4, the demand response optimization model with the goal of maximizing carbon emission reduction is: Where, CEF t is the dynamic carbon emission factor; To increase the amount of electricity used; To reduce electricity consumption, The increase in electricity consumption is equal to the decrease in electricity consumption, that is, the electricity consumption is fixed.

8. The low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 1 is characterized in that: In S5, the rigid constraints include: daily electricity conservation constraints, power limit constraints, and equipment safe operation constraints; the flexible constraints include load change rate constraints and user comfort constraints.

9. A low-carbon demand response resource allocation method based on dynamic carbon emission factors according to claim 8, characterized in that: The daily electricity consumption conservation constraint is: The power limit constraint is: The equipment safety operation constraints are: The load change rate constraint is: The user comfort constraint is: In the formula, To increase the amount of electricity used; To reduce electricity consumption; is the reference load; R max , P min are the maximum and minimum loads allowed, respectively; is the maximum permissible power increase; is the maximum permissible power reduction; R r It represents the maximum allowable difference in power change between adjacent time periods; β is the load adjustment proportional coefficient acceptable to the user.

10. A low-carbon demand response resource allocation system based on dynamic carbon emission factors, using the method according to any one of claims 1 to 9, characterized in that: Includes the following modules: Dynamic carbon emission factor calculation module: The dynamic carbon emission factor is calculated using the marginal carbon emission factor and the average carbon emission factor; Clustering module: Based on the Dirichlet process mixture model, dynamic carbon emission factor data and electricity consumption data are adaptively clustered to obtain cluster labels and probability distributions of carbon emissions and electricity consumption in different time periods; Prediction module: Based on historical data and the results of S2, a dual LSTM deep learning network is used to predict the electricity consumption and dynamic carbon emission factors for the next 24 hours; Demand response optimization module: Based on the fixed daily power consumption constraint, a carbon emission reduction demand response optimization model is constructed based on the calculation results of S3; Demand response scheduling module: Based on rigid constraints and flexible constraints, the demand response optimization model is used to solve the electricity demand and carbon emissions for the next 24 hours, which are used to configure the electricity consumption plan.

Images