Comprehensive energy system low-carbon scheduling method considering energy-carbon coupling

Abstract

The invention discloses an integrated energy system low-carbon scheduling method considering energy-carbon coupling, and the method is based on a carbon emission flow theory, is combined with the strong fitting capability of a neural network, proposes a carbon flow constraint learning method, converts a complex mapping relation between power flow and carbon flow into mixed integer linear constraint, and achieves the low-carbon scheduling of an integrated energy system. And effective embedding of the carbon flow constraint in the optimization model is realized. Meanwhile, in order to reduce the structural complexity of the neural network, a sparse training strategy is introduced, the model parameter scale is effectively compressed, a ReLU activation function is linearized through an improved large-M method, and a cut plane constraint is introduced to gradually tighten a feasible region, so that the solving efficiency of an optimization model is remarkably improved. And finally, embedding the carbon flow constraint model into the optimal scheduling problem of the integrated energy system, exciting the carbon emission reduction consciousness of the load side, and promoting the load side to perform low-carbon energy consumption adjustment by guiding the demand response behavior of the load side based on the carbon signal of the load side, thereby realizing low-carbon scheduling under energy-carbon coordination and reducing the overall carbon emission level of the system.

Description

Technical Field

[0001] The present invention relates to the field of optimized scheduling of integrated energy systems, and in particular to a low-carbon scheduling method for integrated energy systems taking energy-carbon coupling into consideration. Technical Background

[0002] As global attention to climate change continues to increase, reducing greenhouse gas emissions has become an important trend in global energy development. The integrated energy system has the advantages of cascaded energy utilization and multi-energy complementary integration. It can efficiently utilize various types of energy, thereby reducing carbon emissions. It is an important way to accelerate the realization of the "carbon neutrality" goal. In the process of promoting low-carbon sustainable development and realizing the transition to a low-carbon economy, the analysis and statistics of carbon emissions are particularly important. The carbon emission flow theory is a carbon emission tracking method based on the distribution of energy flow in the network. It regards carbon emissions as a virtual network flow that accompanies the flow of energy from the source end to the load end. Compared with traditional macro-statistical methods and full life cycle methods, the carbon emission flow theory has a clearer definition of carbon emission responsibilities on the energy consumption side, and can accurately track and trace the specific flow of carbon emissions.

[0003] Currently, carbon emission flow models have been widely used to analyze load-side carbon emissions, in order to stimulate the carbon reduction potential of demand-side resources and promote low-carbon operation of the system. Most existing studies calculate carbon emission flows based on the results of optimal power flow solutions, without co-optimizing carbon emission flows and power flows. Carbon emission flows are closely related to power flows, but due to the complex nonlinear coupling relationship between the two, it is difficult to explicitly model carbon flows as tractable constraints in the optimization model. Therefore, existing carbon emission flow analysis mainly focuses on the calculation of node carbon potential. Scheduling optimization algorithms that consider energy-carbon flow constraints all iterate low-carbon demand response and energy flow scheduling separately, that is, based on system cost scheduling to obtain power flow results, then calculate the corresponding carbon emission flow, form a carbon signal to guide the demand-side response, and then re-schedule. This method fails to achieve the co-optimization of carbon emission flows and power flows, and the resulting scheduling scheme is difficult to guarantee global optimality under energy-carbon flow constraints.

[0004] In order to solve the above problems, introducing data-driven energy-carbon flow constraints into the optimization scheduling model is one of the research directions of energy scheduling algorithms with the goal of low-carbon economy. In recent years, the "Optimization with Constraint Learning" (OCL) framework and technology have emerged, aiming to solve the constraints that are difficult to explicitly model in real optimization problems. This method uses a deep neural network with a ReLU activation function to learn these constraints from data, and accurately converts the trained neural network model into a mixed integer linear program, so that the fitting relationship of the constraints is effectively embedded in the optimization model. However, as the scale of the fully connected neural network expands, the number of its parameters increases exponentially, which significantly increases the computational complexity of the subsequent energy-carbon coordinated optimization. For this reason, sparse neural networks have become an effective alternative. By introducing structural pruning or sparse constraints during the training process, the redundant connections in the network are effectively reduced. Compared with the fully connected structure, sparse neural networks reduce the parameter scale and computational cost, while achieving performance comparable to or even better than the original neural network. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the existing technology and propose a low-carbon scheduling method for an integrated energy system that takes energy-carbon coupling into consideration.

[0006] The object of the present invention is achieved through the following technical solution: a low-carbon scheduling method for an integrated energy system considering energy-carbon coupling, the method comprising:

[0007] The AC power flow model of the power grid, the equivalent power flow model of the unified heat network supply and return water, and the natural gas network model are coupled through the energy conversion equipment model to construct an energy flow model for the coupled power-heat-gas network. The operating power flow constraints of the energy flow model are used as constraints, and the minimization of operating costs is used as the objective function to construct an integrated energy system optimization model that does not consider the carbon cost on the load side.

[0008] Based on the carbon emission flow theory and the energy flow distribution of the integrated energy system, the system carbon emission flow model is constructed by applying the proportional sharing principle and the energy merging principle. The system carbon emission flow model is used to calculate the carbon flow rate of each load node according to the load and unit output size to obtain the training data set.

[0009] Based on the training data set, a sparse neural network model is used to fit the nonlinear coupling relationship between tidal current and carbon flow;

[0010] Construct a comprehensive energy system optimization model taking into account demand response, including system operation flow constraints, carbon flow constraints, and demand response load constraints as constraints. The flow constraint is an energy flow model coupled with electricity, heat, and gas networks. The carbon flow constraint is a mixed integer linear constraint obtained by equivalent conversion of the sparse network model using the Big-M method. A step-by-step carbon price mechanism is used to describe the carbon cost of each node load.

[0011] The integrated energy system optimization model taking demand response into account is solved: each time a node relaxation solution is obtained, a cutting plane constraint is added according to the improved linearization method of the ReLU neuron large-M method; it is judged whether the error of the current optimization solution meets the standard. If not, the solution is continued; if so, the final optimization result is obtained.

[0012] Beneficial effects of the present invention:

[0013] The present invention proposes a low-carbon scheduling method for an integrated energy system that takes into account energy-carbon coupling. Since the calculation of carbon emission flow is highly nonlinear and non-convex, the complex mapping relationship between tidal current and carbon flow cannot be directly converted into an easily handled constraint in the optimal scheduling of an integrated energy system. The present invention utilizes the strong fitting ability of neural networks to propose a method for learning carbon flow constraints, converting complex mapping relationships into mixed integer linear constraints, thereby effectively embedding carbon flow constraints into the optimization model. At the same time, the fully connected neural network is sparsely trained to reduce the parameter scale and model complexity of the neural network and reduce the computational burden of the subsequent optimization model; and the large-M method is improved to linearize the ReLU neurons, effectively and reasonably add cutting plane constraints in the process of solving the scheduling optimization, gradually tighten the feasible domain of the problem, thereby reducing the optimization search space, and compared with the traditional large-M method, the solution speed of the optimization model will be significantly improved. Finally, the carbon flow constraint model is embedded in the optimal scheduling problem of the integrated energy system. The step-by-step carbon price mechanism is used to describe the carbon cost of each node load, stimulate the carbon emission reduction awareness on the load side, guide the load side to respond to low-carbon demand under its own carbon signal perception, and realize energy-carbon coordinated low-carbon scheduling, thereby reducing the overall carbon emissions of the integrated energy system. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a calculation flow chart of a low-carbon scheduling method for an integrated energy system considering energy-carbon coupling provided in an embodiment of the present invention.

[0015] Figure 2 Schematic diagram of a low-carbon scheduling device for an integrated energy system considering energy-carbon coupling provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings.

[0017] like Figure 1 As shown, the present invention provides a low-carbon scheduling method for an integrated energy system considering energy-carbon coupling, and the specific process includes:

[0018] S1. Under different load parameter settings, solve the integrated energy system optimization model without considering the load-side carbon cost, and calculate the carbon flow rate of each load node based on the carbon emission flow theory to generate a training data set;

[0019] S2. Train a sparse neural network to accurately fit the mapping relationship between tidal currents and carbon flows. After training, determine whether the test error meets the standard. If not, adjust the training parameters and retrain. If so, convert the model into a mixed integer linear constraint using the Big-M method and embed it into the integrated energy system optimization model.

[0020] S3. Solve the integrated energy system optimization model that takes demand response into account. During the solution process, each time a node relaxation solution is obtained, a cutting plane constraint is added according to the improved linearization method of the ReLU neuron large-M method; determine whether the error of the current optimization solution meets the standard. If not, continue to solve; if so, obtain the final optimization result.

[0021] Said S1 comprises:

[0022] S1.1. Under different load parameter settings, solve the integrated energy system optimization model without considering the load-side carbon cost:

[0023] The integrated energy system optimization scheduling model is established without considering the load-side carbon cost, with the goal of minimizing the total system cost and the system operation flow constraint as the constraint condition. The model objective function is shown as follows:

[0024] minC=C Buy,e +C Buy,g +C Gen

[0025] Where C Buy,e The cost of electricity purchase is the power purchased in the power subsystem multiplied by the time-of-use electricity price; C Buy,g is the gas purchase cost, which is the gas purchase amount multiplied by the gas price in the natural gas subsystem; C Gen It is the unit operating cost, including CHP unit, EB, etc. in the system.

[0026] The system operation power flow constraints are as follows:

[0027] The AC power flow model of the power grid includes the node power balance constraints:

[0028]

[0029] Where, P PG,i With Q PG,i represents the active and reactive output of the unit at node i, P L,i With Q L,i Represents the active power and reactive power of the load at node i, P ij With Q ij are the active and reactive power flowing from node i to node j, l ki is the square of the current on line ki, rki with x ki They represent the resistance and reactance of line ki respectively, j:i→j means that node j is the outflow node of node i, k:k→i means that node k is the inflow node of node i, N e is the number of grid nodes;

[0030] The voltage relationship between the two ends of the line is:

[0031] Where, v i is the square of the voltage at node i;

[0032] The power formula at the head end of the branch is: In radial networks, the lossless relaxation of this equation using a second-order cone is:

[0033] The equivalent power flow model for unified heat supply and return water network includes:

[0034] Based on a heat network flow model that includes both thermal and hydraulic models, an equivalent flow model for both supply and return water in the heat network is established. First, a flow calculation based on the thermal and hydraulic models is performed on the heat network, ignoring the delay of the heat network pipelines. A mass flow control method is used, where the water flow rate is constant and the supply water temperature is adjusted to meet heating demand.

[0035] In the hydraulic model, the node flow balance is expressed as:

[0036] A.m P =m N

[0037] Where m P is the pipeline mass flow rate, m N is the node mass flow, and “0” in the node-pipeline association matrix A indicates that the node is not connected to the pipeline, and +1 (-1) indicates that the node is the outflow (inflow) node of the pipeline.

[0038] The thermal model describes the relationship between supply and return water temperature and heat transfer:

[0039] Φ=C p ·m N ⊙(T s -T r )

[0040] Where Φ is the thermal power vector of the heat source or load node, C p is the specific heat capacity of water, T s and T r are the node supply water temperature and return water temperature respectively.

[0041] The relationship between the pipe inlet and outlet temperatures is expressed using the temperature drop equation:

[0042]

[0043] Where, T f and T t are the temperatures at the pipe inlet and outlet, T a is the ambient temperature, L is the pipe length, and λ is the total heat transfer coefficient per unit length of each pipe.

[0044] In the equivalent power flow model of the unified supply and return water of the heating network, the equivalent thermal power of the heating network pipes is expressed as the difference between the thermal powers of the supply and return water pipes:

[0045]

[0046] Where, P h,P,f and P h,P,t are the equivalent thermal power at the inlet and outlet of the heat network pipeline, T s,f and T s,t are the inlet and outlet temperatures of the water supply pipe, T r,f and T r,t are the return pipe inlet and outlet temperatures, respectively.

[0047] The natural gas grid model includes:

[0048] Similar to the power grid and heat network, the flow rate of each natural gas node satisfies the input and output balance. The flow balance equation of the natural gas network node is:

[0049] G·f P =f N

[0050] Where G is the node-pipeline association matrix, f P is the pipeline gas flow rate, f N is the node gas flow rate.

[0051] Without considering the slow inertia of natural gas transmission, the steady-state model of the natural gas network based on the Weymouth equation is adopted, that is, the gas flow rate f P is a nonlinear function of the node pressure at both ends of the pipeline:

[0052]

[0053] Where K ij is the constant of the natural gas pipeline ij, s P = +1 or -1 indicates the direction of the pipe flow, π i is the pressure value of node i.

[0054] The equivalent power vector P of c pipelines in the natural gas network g for:

[0055] P g=Bf P

[0056] Where B is the calorific value of natural gas, which is 10.45kWh / m 3 .

[0057] Energy conversion equipment models include:

[0058] The CHP unit is an important component for coupling electrical energy and thermal energy. It serves as the load end in the natural gas subsystem and as the source end in the power and thermal subsystem. Its operating characteristics are as follows:

[0059]

[0060] Where η CHP,e and η CHP,h is the electrical and thermal efficiency of the CHP unit, P CHP and H CHP is the electrical and thermal power output of the CHP unit, G CHP The energy contained in the natural gas input to the CHP unit, and It is the output and ramp rate constraint of CHP unit.

[0061] The electric boiler acts as the load end in the power subsystem and as the source end in the thermal subsystem. Its operating characteristics are as follows:

[0062]

[0063] Where H EB and P EB Output thermal power and input electrical power for electric boiler, and is the electric boiler input power constraint, η EB is the thermal efficiency of the electric boiler.

[0064] After constructing a comprehensive energy system optimization model that does not consider the carbon cost on the load side, different load parameters are set for solution to obtain different scheduling results, which are subsequently used as the input of the training data set.

[0065] S1.2. Based on the carbon emission flow theory, calculate the carbon flow rate of each load node and generate a training data set:

[0066] Based on the energy flow distribution of the integrated energy system, the carbon emission flow is calculated by applying the proportional sharing principle and the energy merging principle. The carbon emission intensity calculation formula of the grid node is:

[0067]

[0068] Where, is the carbon emission intensity of grid node i, is the carbon emission intensity of branch l, ρ e,S is the carbon emission intensity of the generator set, P e,l is the active power on branch l, is the set of branches flowing into node i, P e,S,i is the output of the unit at node i.

[0069] According to the principle of proportional sharing, the carbon emission intensity of the branch in the power grid is Equal to the carbon emission intensity of the branch flowing into node i:

[0070]

[0071] Similarly, the calculation formula for the node carbon intensity of the equivalent heat network is:

[0072]

[0073] Where, is the carbon emission intensity of heating network node i, is the carbon emission intensity of pipeline l, ρ h,S is the carbon emission intensity of the heating unit, P h,S,i is the heating power of the heat source at node i, P h,l is the equivalent thermal power on pipe l, is the set of pipes flowing into node i.

[0074] The calculation formula for the node carbon intensity of the natural gas grid is:

[0075]

[0076] Where, is the carbon emission intensity of gas grid node i, is the carbon emission intensity of pipeline l, ρ g,S is the carbon emission intensity of gas source, usually 0.2kgCO2 / kWh, P g,S,i is the equivalent power of the gas source at node i, P g,l is the equivalent power on pipe l, is the set of pipes flowing into node i.

[0077] For energy conversion devices that couple multiple energies in integrated energy systems, they are divided into single-input single-output devices and single-input multiple-output devices, and their CEF characteristics are analyzed separately.

[0078] 1) Single-input single-output equipment: Taking EB as an example, the equipment input carbon emissions are allocated to the output energy flow, that is:

[0079]

[0080] Where, and are the carbon emission intensity of EB input and output ports respectively, and are the EB input electrical power and output thermal power, η EB is the EB energy conversion efficiency.

[0081] 2) Single-input, multi-output equipment: Taking the CHP unit as an example, the equipment input and output carbon emission balance is:

[0082]

[0083] Where, and are the carbon intensity and equivalent gas power of the CHP unit input port, and are the carbon intensity of the electrical and thermal output ports of the CHP unit, and They are the electrical and thermal output power of the CHP unit respectively.

[0084] For the allocation of carbon emissions among multiple outputs, the efficiency method is used, which assumes that the carbon emission intensity of the electrical output port and the thermal output port is inversely proportional to the corresponding energy efficiency:

[0085]

[0086] Where η e,CHP and η h,CHP They are the electrical and thermal efficiencies of the CHP unit respectively.

[0087] According to the above formula, the carbon intensity of the electrical and thermal output ports of the CHP unit are:

[0088]

[0089] The unified matrix calculation formula for carbon flow in the electricity-heat-gas network is as follows:

[0090] According to the principle of proportional sharing, the carbon intensity of all outflow lines from a node is equal to the carbon intensity of that node. If the carbon intensity of each node in the integrated energy system can be calculated, the carbon flow rate of all lines and load nodes can be calculated based on the carbon intensity and power flow results of each node. Therefore, solving the carbon intensity of each node is the primary goal. For the carbon emission flow of the electricity-heat-gas integrated energy system, a unified matrix form is used for calculation.

[0091] From the system carbon flow model, we can see that the calculation formulas for the carbon intensity of electricity, heat, and gas grid nodes are the same. The calculation formula for the carbon intensity of each node is:

[0092]

[0093] Where Π∈{E,H,G} represents the integrated energy system including electricity, heat and gas networks.

[0094] Considering the relationship between line carbon intensity and node carbon intensity, the numerator in the above formula can be rewritten in matrix form as follows:

[0095]

[0096] Where, is an N-dimensional row vector whose i-th element is 1 and the rest of the elements are 0, Represents the branch power flow distribution matrix, for If active power p flows from node i to node j, then represents the N-dimensional node carbon intensity column vector, Represents the K×N-dimensional source output power distribution matrix, where K is the number of source terminals. Represents the K-dimensional source-end carbon intensity column vector.

[0097] Define the node active flux matrix Its diagonal elements represent the absolute value of the active power flowing into the node in the direction of the power flow:

[0098]

[0099] Where, ξ N+K is an N+K-dimensional row vector whose elements are all 1.

[0100] Using the Matrix The calculation formula for node carbon intensity can be rewritten as:

[0101]

[0102] Expanding the above formula to an N-node system yields:

[0103]

[0104] Where, and are the load required power vector and its corresponding carbon flow rate vector respectively, and ⊙ represents the multiplication of the corresponding elements of the matrix.

[0105] According to the above calculation results, the load and unit output size are used as input, and the load carbon flow rate is used as output to construct a training data set.

[0106] The S2 includes:

[0107] S2.1. Train a sparse neural network to accurately fit the mapping relationship between tidal currents and carbon flows. After training, determine whether the test error meets the standard:

[0108] From the system's unified matrix calculation formula, it can be seen that the carbon flow rate of each load By matrix The matrix M is determined by the source output and load demand of each node. The network power distribution is determined. and The mapping relationship between them is as follows:

[0109]

[0110] Where, f c Represents the matrix M to load carbon flow rate The function of the mapping relationship, f p Representation matrix A function that maps the matrix M to the matrix M.

[0111] Since carbon flow calculation is highly nonlinear and nonconvex, and The mapping relationship between tidal currents and carbon flows cannot be converted into easily tractable constraints in the optimization model. By utilizing the strong fitting ability of neural networks, a carbon flow constraint learning method is proposed to convert the complex mapping relationship between tidal currents and carbon flows into tractable mixed integer linear constraints.

[0112] Using sparse neural network models to accurately fit the complex nonlinear coupling relationship between tidal currents and carbon flows includes:

[0113] The neural network model mainly includes the input layer Hidden layer (ReLU-based neurons), output layer The specific formula of the model is as follows:

[0114]

[0115] Where: x = v 0 , ReLU(x):=max{0,x}

[0116] Where, is the output of the i-th neuron in the l-th layer network, w and b are the neural network weight parameters and bias parameters respectively, [L] represents the set of L hidden layers of the neural network, N l is the number of neurons in the lth layer, [N l ] represents the set of neurons in the lth layer, x is the input of the neural network, and y is the output of the neural network.

[0117] Based on the neural network model structure, the mean square error (MSE) loss function L is minimized. MSE As the goal, the stochastic gradient descent algorithm is used to optimize the parameters in the neural network model so that the predicted value Gradually approaching the actual value y, as shown below:

[0118]

[0119] Where N MSE is the number of training data samples, f NN A function that maps input x to output y.

[0120] Through neural network model training, learning and The complex mapping relationship between them can realize the accurate reasoning of the carbon flow rate of each node load, as shown in the following formula:

[0121]

[0122] Where, This is the inference result of the neural network model for the load carbon flow rate.

[0123] During the training of a fully connected neural network model, internal connections within the neural network are iteratively discarded and activated to reduce the number of parameters after the model is equivalent, alleviating the computational burden of the optimization scheduling model. The sparse training algorithm discards some connections based on weights and activates new connections using instantaneous gradient information. After updating the connections, training continues using the new neural network until the next update. The main components of sparse training include: 1) sparse distribution, 2) update strategy, 3) connection discarding, and 4) connection activation.

[0124] 1) Sparse distribution: sparsity s l ∈(0,1) is defined as the connection drop rate of layer l, and the sparsity of each layer is s l Keep consistent with the total sparsity S of the model. Initialize the first layer as a dense layer, that is, s 1 =0, the remaining layers are initialized as sparse layers, and the sparse training of each layer adopts a unified method.

[0125] 2) Update plan: After the model is initialized, the update interval is ΔT, some connections are discarded based on the neural network weight amplitude, and new connections are activated based on the neural network parameter gradient amplitude. The cosine annealing algorithm is used to define the score function f for updating the connection. decay :

[0126]

[0127] Where α is the initial fraction of updated connections, T end is the number of iterations of sparse training.

[0128] 3) Connection discarding: During the training process, since connections with smaller weight amplitudes have a smaller impact on the training loss, unnecessary connections in this part of the neural network can be discarded according to the update plan. Specifically, every ΔT steps, through ArgTopK(-|w l |,f decay (t)(1-s l )N l ) selects the connections to discard, where ArgTopK(ν,k) returns the set of indices of the top k largest elements in vector ν.

[0129] 4) Connection activation: After the connection drop step, inactive connections with high magnitude gradients are selected for reactivation. These weight parameters reflect the strong response connections between neurons in the back propagation. Specifically, by ArgTopK(|grad(w l )|,f decay (t)(1-s l )N l ) to select the connection to activate. Newly activated connections are initialized to 0 and therefore do not affect the output of the network.

[0130] After the training is completed, the test error of the neural network model is evaluated to see if it meets the standard. If not, the training parameters are adjusted and retrained. If it meets the standard, the subsequent process is continued.

[0131] S2.2. If the model test error meets the standard, the model is converted into a mixed integer linear constraint using the Big-M method and embedded into the integrated energy system optimization model;

[0132] Due to the strong nonlinear characteristics of neural networks, it is difficult for the model to directly participate in the operation and scheduling of integrated energy systems. The nonlinear part of the neural network model is the ReLU activation function of the hidden layer neurons. The optimization relaxation strategy Big-M method can be used to convert the ReLU activation function into a mixed integer linear constraint, which can be applied to the optimization scheduling model. Specifically, for neurons based on the ReLU function, the independent variable of the ReLU function is an affine function f(x) = b + w T x, where w is a column vector of weight parameters, linearized using the Big-M method:

[0133]

[0134] Where M L and M U are the upper and lower limits of all possible values ​​of x, respectively, and z is an auxiliary 0-1 variable. When f(x) ≥ 0, z is 1 and the output v is f(x); when f(x) < 0, z is 0 and the output v is 0.

[0135] For the trained sparse neural network model, the big-M method equivalent transformation based on the ReLU activation function converts ReLU(x) into a set of mixed integer linear constraints, and then and The complex mapping relationship between is converted into a tractable mixed integer linear constraint to achieve carbon flow constraint learning. The specific formula is as follows, where the input x represents Output represent

[0136]

[0137] S3. Solve the integrated energy system optimization model that takes demand response into account. During the solution process, each time a node relaxation solution is obtained, a cutting plane constraint is added according to the improved linearization method of the ReLU neuron large-M method; determine whether the error of the current optimization solution meets the standard. If not, continue to solve; if so, obtain the final optimization result.

[0138] The S3 includes:

[0139] S3.1. Build and solve a comprehensive energy system optimization model that takes demand response into account:

[0140] A comprehensive energy system optimization model that takes demand response into account is established, with the goal of minimizing the total system cost. It uses system operation flow constraints, carbon flow constraints, and demand response load constraints as constraints. A tiered carbon price mechanism is added to tap the carbon emission reduction potential on the load side and guide the demand response on the load side under carbon potential awareness. The model objective function is shown below:

[0141] minC=C Buy,e +C Buy,g +C Gen +C DR +C CB

[0142] Where C Buy,e is the electricity purchase cost, C Buy,g is the gas purchase cost, C Gen is the unit operating cost, C DR Compensation cost for demand response, C CB is the carbon emission cost on the load side.

[0143] 1) Demand response compensation cost C DR :

[0144]

[0145] Where, and are the compensation cost coefficients for reducing and transferring unit power load, and are the load reduction and transfer power of node n at time t respectively.

[0146] According to actual needs, the load adjustment amount allowed by demand response is constrained, that is, the adjustable load ratio range is given, as shown in the following formula:

[0147]

[0148] Where, is the load size of node n in period t before demand response is implemented; cut and λ mov They are the proportional coefficients of the load that can be reduced and transferred, respectively, and are determined by the specific load characteristics.

[0149] 2) Load-side carbon emission cost C CB :

[0150] A tiered carbon pricing mechanism is used to divide the load-side carbon emission responsibility into multiple continuous intervals. The greater the carbon emissions, the higher the carbon price in the corresponding interval. This "high emission, high penalty" pricing mechanism for load-side carbon responsibility can better stimulate its carbon emission reduction potential and reduce carbon emissions through demand response, as shown below:

[0151]

[0152] δ CB,m =δ CB,base (1+(m-1)·σ)

[0153] Where C CB is the carbon emission cost on the load side, is the number of system load nodes, is the carbon emission cost of load node n, N CB is the number of steps of carbon price on the load side, T is a scheduling period of 24 hours, Δt is 1 hour, δ CB,m is the carbon price at step m, is the carbon flow rate of load node n in a scheduling cycle, δ CB,base is the carbon benchmark price, and σ is the step change rate.

[0154] The system operation flow constraints in this model are the same as those in S1.1.

[0155] S3.2. In the process of solving the integrated energy system optimization model taking demand response into account, each time a node relaxation solution is obtained, a cutting plane constraint is added according to the improved linearization method of the ReLU neuron large-M method; it is judged whether the error of the current optimization solution meets the standard. If not, the solution is continued. If it meets the standard, the final optimization result is obtained.

[0156] Improved linearization method of ReLU neuron large-M method:

[0157] When using the Big-M method to linearize neurons based on the ReLU function, the activation state of the ReLU is constructed only based on the upper and lower bounds of the interval of the affine function f(x), without paying attention to the value range of the input vector x. This results in the Big-M method linearizing the ReLU neurons to produce a convex relaxation space that is not tight enough and not precise enough, which greatly increases the time to search for the optimal solution during the solution process. Based on this, the upper and lower bounds of the input vector x are used to construct the improved Big-M method linearization constraints for the ReLU neurons:

[0158] v≥b+w T x

[0159]

[0160] Where supp(w) is the weight w i The set is not equal to 0, are the sets of minimum and maximum values ​​of input x respectively.

[0161] To ensure definition and gather:

[0162]

[0163] Where, They are and The elements in the collection, Input x i The minimum and maximum values ​​of .

[0164] However, since the number of constraints in the improved Big-M method is too large, direct use will lead to difficulties in solving the optimization model. It is observed that when I = supp(w) and When , they are equivalent to the second and third equations in the original large-M method, so we can start from the original large-M method. During the solution process, each time a node relaxation solution is obtained, a cutting plane constraint is added to gradually tighten the feasible domain of the problem and improve the solution efficiency. Define the set of violated constraints corresponding to the current relaxation solution

[0165]

[0166] Where, and Input x for the current relaxation solution i and the value of the 0-1 variable z.

[0167] Determine whether the current relaxation solution violates The constraint of the improved Big-M method corresponding to the set is violated. If it is violated, then this constraint is the most violated among the constraints corresponding to the I set family. Add this constraint. The specific mathematical expression is as follows:

[0168] if:

[0169]

[0170] Then add constraints:

[0171]

[0172] Finally, determine whether the error of the current optimization solution meets the standard. If not, continue to solve it. If it meets the standard, the final low-carbon scheduling result of the integrated energy system considering energy-carbon coupling is obtained.

[0173] Corresponding to the aforementioned embodiment of a low-carbon scheduling method for an integrated energy system taking energy-carbon coupling into consideration, the present invention also provides an embodiment of a low-carbon scheduling device for an integrated energy system taking energy-carbon coupling into consideration.

[0174] See also Figure 2 An embodiment of the present invention provides a low-carbon scheduling device for an integrated energy system that takes energy-carbon coupling into consideration, comprising a memory and one or more processors, wherein the memory stores executable code. When the processor executes the executable code, it is used to implement a low-carbon scheduling method for an integrated energy system that takes energy-carbon coupling into consideration in the above-mentioned embodiment.

[0175] The embodiment of the low-carbon scheduling device for an integrated energy system taking into account energy-carbon coupling provided by the present invention can be applied to any device with data processing capabilities, and the device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located reading the corresponding computer program instructions in the non-volatile memory into the memory for execution. From the hardware level, if Figure 2 As shown, it is a hardware structure diagram of any device with data processing capability in a low-carbon dispatching device of an integrated energy system considering energy-carbon coupling provided by the present invention, except Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0176] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0177] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present invention. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0178] An embodiment of the present invention also provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a low-carbon scheduling method for an integrated energy system considering energy-carbon coupling in the above embodiment is implemented.

[0179] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0180] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the low-carbon scheduling method for an integrated energy system that considers energy-carbon coupling.

[0181] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the contents disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of the present application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered merely as exemplary, and the true scope and spirit of the present application are indicated by the claims.

[0182] It should be understood that the above general description and the detailed description that follows are exemplary and explanatory only and do not limit the present application. The present application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from the scope of the present application. The scope of the present application is limited only by the appended claims.

Claims

1. A low-carbon scheduling method for an integrated energy system considering energy-carbon coupling, characterized in that: The method includes: The AC power flow model of the power grid, the equivalent power flow model of the unified heat network supply and return water, and the natural gas network model are coupled through the energy conversion equipment model to construct an energy flow model for the coupled power-heat-gas network. The operating power flow constraints of the energy flow model are used as constraints, and the minimization of operating costs is used as the objective function to construct an integrated energy system optimization model that does not consider the carbon cost on the load side. Based on the carbon emission flow theory and the energy flow distribution of the integrated energy system, the system carbon emission flow model is constructed by applying the proportional sharing principle and the energy merging principle. The system carbon emission flow model is used to calculate the carbon flow rate of each load node according to the load and unit output size to obtain the training data set. Based on the training data set, a sparse neural network model is used to fit the nonlinear coupling relationship between tidal current and carbon flow; Construct a comprehensive energy system optimization model taking into account demand response, including system operation flow constraints, carbon flow constraints, and demand response load constraints as constraints. The flow constraint is an energy flow model coupled with electricity, heat, and gas networks. The carbon flow constraint is a mixed integer linear constraint obtained by equivalent conversion of the sparse network model using the Big-M method. A step-by-step carbon price mechanism is used to describe the carbon cost of each node load. The integrated energy system optimization model taking demand response into account is solved: each time a node relaxation solution is obtained, a cutting plane constraint is added according to the improved linearization method of the ReLU neuron large-M method; it is judged whether the error of the current optimization solution meets the standard. If not, the solution is continued; if so, the final optimization result is obtained.

2. A low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 1, characterized in that: The energy conversion equipment model specifically includes: CHP unit and electric boiler operation constraints, the CHP unit acts as the load end in the natural gas subsystem and as the source end in the power and thermal subsystems, and acts as the load end in the power subsystem and as the source end in the thermal subsystem; The operating characteristics of the CHP unit include: Where η CHP,e and η CHP,h is the electrical and thermal efficiency of the CHP unit, P CHP and H CHP is the electrical and thermal power output of the CHP unit, G CHP The energy contained in the natural gas input to the CHP unit, and Constraints on CHP unit output and ramp rate; The operating characteristics of the electric boiler are: Where H EB and P EB Output thermal power and input electrical power for electric boiler, and is the electric boiler input power constraint, η EB is the thermal efficiency of the electric boiler.

3. A low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 2, characterized in that: The AC power flow model of the power grid is specifically: Where, P PG,i With Q PG,i represents the active and reactive output of the unit at node i, P L,i With Q L,i Represents the active power and reactive power of the load at node i, P ij With Q ij are the active and reactive power flowing from node i to node j, l ki is the square of the current on line ki, r ki with x ki They represent the resistance and reactance of line ki respectively, j:i→j means node j is the outflow node of node i, k:k→i means node k is the inflow node of node i, N e is the number of grid nodes; The voltage relationship between the two ends of the line is: Where, v i is the square of the voltage at node i; The power formula at the head end of the branch is: In radial networks, the lossless relaxation of this equation using a second-order cone is:

4. A low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 2, characterized in that: The equivalent power flow model for unified heat supply and return water network includes: The heat network is subjected to flow calculation based on the thermal and hydraulic models, without considering the delay of the heat network pipes. That is, the water flow rate is constant, and the heating demand is met by changing the water supply temperature: In the hydraulic model, the node flow balance is expressed as: A·m P =m N Where m P is the pipeline mass flow rate, m N is the node mass flow rate, "0" in the node-pipeline association matrix A indicates that the node is not connected to the pipeline, and +1 (-1) indicates that the node is the outflow (inflow) node of the pipeline; The thermal model describes the relationship between supply and return water temperature and heat transfer: Φ=C p ·m N ⊙(T s -T r ) Where Φ is the thermal power vector of the heat source or load node, C p is the specific heat capacity of water, T s and T r are the node supply water temperature and return water temperature respectively; The relationship between the pipe inlet and outlet temperatures is expressed using the temperature drop equation: Where, T f and T t are the temperatures at the pipe inlet and outlet, T a is the ambient temperature, L is the pipe length, and λ is the total heat transfer coefficient per unit length of each pipe; In the equivalent power flow model of the unified supply and return water of the heating network, the equivalent thermal power of the heating network pipes is expressed as the difference between the thermal power of the supply and return water pipes: Where, P h,P,f and P h,P,t are the equivalent thermal power at the inlet and outlet of the heat network pipeline, T s,f and T s,t are the inlet and outlet temperatures of the water supply pipe, T r,f and T r,t are the return pipe inlet and outlet temperatures, respectively.

5. The low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 2 is characterized in that: The natural gas grid model includes: The airflow balance equation at the natural gas grid node is: G·f P =f N Where G is the node-pipeline association matrix, f P is the pipeline gas flow rate, f N is the node gas flow rate; Without considering the slow inertia of natural gas transmission, the steady-state model of the natural gas network based on the Weymouth equation is adopted, that is, the gas flow rate f P is a nonlinear function of the node pressure at both ends of the pipeline: Where K ij is the constant of the natural gas pipeline ij, s P = +1 or -1 indicates the direction of the pipe flow, π i is the pressure value of node i; The equivalent power vector P of c pipelines in the natural gas network g for: P g =Bf P Where B is the calorific value of natural gas.

6. The low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 1 is characterized in that: The system carbon emission flow model includes: Node carbon intensity calculation formula: Among them, N ，i is the node carbon emission intensity, is an N-dimensional row vector whose i-th element is 1 and the rest of the elements are 0, represents the branch power flow distribution matrix, represents the N-dimensional node carbon intensity column vector, represents the K-dimensional source end carbon intensity column vector, is the node active flux matrix, whose diagonal elements represent the absolute value of the active power flowing into the node in the direction of the power flow: Where, ξ N+K is an N+K-dimensional row vector whose elements are all 1; Expanding the above formula to an N-node system yields: Where, and are the load required power vector and its corresponding carbon flow rate vector respectively, and ⊙ represents the multiplication of the corresponding elements of the matrix.

7. The low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 1 is characterized in that: The use of a sparse neural network model to fit the nonlinear coupling relationship between tidal current and carbon flow specifically includes: The neural network model includes an input layer Hidden layer, output layer The specific formula of the model is as follows: Where: x = v 0 , ReLU(x):=max{0,x} Where, is the output of the i-th neuron in the l-th layer network, w and b are the neural network weight parameters and bias parameters respectively, [L] represents the set of L hidden layers of the neural network, N l is the number of neurons in the lth layer, [N l ] represents the set of neurons in the first layer, x is the input of the neural network, and y is the output of the neural network; Based on the neural network model structure, to minimize the mean square error loss function L MSE As the goal, the stochastic gradient descent algorithm is used to optimize the parameters in the neural network model so that the predicted value Gradually approaching the actual value y, as shown below: Where N MSE is the number of training data samples, f NN A function that represents the mapping from input x to output y; Through neural network model training, learning and The complex mapping relationship between them can realize the accurate reasoning of the carbon flow rate of each node load, as shown in the following formula: Where, This is the inference result of the neural network model for the load carbon flow rate.

8. A low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 7, characterized in that: During the training of the sparse neural network model, the internal connections of the neural network are iteratively discarded and activated, specifically including: Sparse distribution: sparsity s l ∈(0,1) is defined as the connection drop rate of layer l, and the sparsity of each layer is s l Keep consistent with the total sparsity S of the model; initialize the first layer as a dense layer, that is, s 1 =0, the remaining layers are initialized as sparse layers, and the sparse training of each layer adopts a unified method; Update plan: After the model is initialized, the update interval is ΔT, some connections are discarded based on the neural network weight amplitude, and new connections are activated based on the neural network parameter gradient amplitude. The cosine annealing algorithm is used to define the score function f for updating the connection decay : Where α is the initial fraction of updated connections, T end is the number of iterations of sparse training; Connection discarding: During the training process, since the connection with smaller weight amplitude has a smaller impact on the training loss, unnecessary connections in this part of the neural network are discarded according to the update plan; specifically, every ΔT step, through ArgTopK(-|w l |,f decay (t)(1-s l )N l ) selects the connections to be discarded, where ArgTopK(ν,k) returns the set of indices of the top k largest elements in vector ν; Connection activation: After the connection drop step, inactive connections with high magnitude gradients are selected for reactivation. These weight parameters reflect the strong response connections between neurons in the back propagation. Specifically, by ArgTopK(|grad(w l )|,f decay (t)(1-s l )N l ) selects the connections to activate; newly activated connections are initialized to 0 and therefore do not affect the output of the network.

9. The low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 1 is characterized in that: The comprehensive energy system optimization model taking demand response into account includes: Objective function: minC=C Buy,e +C Buy,g +C Gen +C DR +C CB Where C Buy,e is the electricity purchase cost, C Buy,g is the gas purchase cost, C Gen is the unit operating cost, C DR Compensation cost for demand response, C CB is the carbon emission cost on the load side; Demand response compensation cost C DR : Where, and are the compensation cost coefficients for reducing and transferring unit power load, and are the load reduction and transfer power of node n at time t respectively; According to actual needs, the load adjustment amount allowed by demand response is constrained, that is, the adjustable load ratio range is given, as shown in the following formula: Where, is the load size of node n in period t before demand response is implemented; cut and λ mov are the proportional coefficients of the load that can be reduced and transferred, respectively, which are determined by the specific load characteristics; Load-side carbon emission cost C CB : A tiered carbon pricing mechanism is adopted to divide the load-side carbon emission responsibility into multiple continuous intervals, and carbon emissions are reduced through demand response, as shown below: d CB,m =d CB,base (1+(m-1)·s) Where C CB is the carbon emission cost on the load side, is the number of system load nodes, is the carbon emission cost of load node n, N CB is the number of steps of carbon price on the load side, T is a scheduling period of 24 hours, Δt is 1 hour, δ CB,m is the carbon price at step m, is the carbon flow rate of load node n in a scheduling cycle, δ CB,base is the carbon benchmark price, and σ is the step change rate.

10. The low-carbon scheduling method for an integrated energy system considering energy-carbon coupling according to claim 1, characterized in that: Each time a node relaxation solution is obtained, according to the improved linearization method of the ReLU neuron large-M method, a cutting plane constraint is added, including: The set of violated constraints corresponding to the current relaxed solution Where, and Input x for the current relaxation solution i and the value of the 0-1 variable z; Input x i The minimum and maximum values ​​of w i is the network weight; Determine whether the current relaxation solution violates The constraint of the improved Big-M method corresponding to the set is violated. If it is violated, then this constraint is the most violated among the constraints corresponding to the I set family. Add this constraint. The specific mathematical expression is as follows: if: Then add constraints:

Images