Comprehensive energy optimization method and system based on improved neural network

Abstract

The invention discloses a comprehensive energy optimization method and system based on an improved neural network, and the method comprises the following steps: building an improved neural network model according to a pre-collected multi-modal data set by taking the minimization of the total operation cost, the minimization of the emission of carbon and other pollutants and the maximization of the overall energy utilization efficiency as targets; predicting the comprehensive energy by using an improved neural network model, comparing a result output by the improved neural network model with a set target, and establishing a comprehensive energy optimization model in combination with constraint conditions; according to the comprehensive energy optimization model, various optimization schemes are generated according to requirements and supply of different energy forms, and the scheme with the minimum performance index fluctuation under all disturbance scenes is selected as the optimal scheme to optimize comprehensive energy in combination with risk assessment. The optimization precision and efficiency of the comprehensive energy system are improved, the energy demand and supply fluctuation are dynamically dealt with, and effective balance among economical efficiency, environmental protection and energy utilization efficiency is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy management and optimization, and more specifically, relates to a comprehensive energy optimization scheduling method and system based on an improved neural network. Background Art

[0002] Against the backdrop of growing global energy demand and increasingly serious environmental problems, efficient operation and sustainable development of energy systems have become key issues. As an important means to address this challenge, the Integrated Energy System (IES) integrates multiple energy forms such as electricity, heat, cooling, and gas, aiming to achieve optimal allocation and utilization of energy resources by coordinating and optimizing the production, transmission, and consumption of different energy forms. IES can not only improve energy utilization efficiency and reduce energy waste, but also promote the access and use of renewable energy, reduce dependence on fossil energy, and thus reduce the environmental burden.

[0003] However, as the complexity of energy systems increases, traditional planning and optimization methods face many challenges. In particular, existing methods often fail to provide efficient and comprehensive solutions in response to complex and changing energy demand and supply situations. This is because traditional methods are usually based on linear or simplified models and cannot fully consider the nonlinear relationships and dynamic characteristics between components in the energy system. Therefore, new methods and technologies are needed to improve the planning and optimization of energy systems.

[0004] Existing integrated energy system planning methods mainly include linear programming (LP), mixed integer linear programming (MILP), dynamic programming (DP), etc. These methods have solved the optimization problems in energy systems to a certain extent, but they also have some shortcomings: Linear programming (LP): It is suitable for situations where the relationship between components in the system can be described linearly, but it has limited effect when facing nonlinear or complex energy conversion processes. Mixed integer linear programming (MILP): It can handle problems with integer decision variables, such as the optimization of equipment start and stop states. However, this method has high computational complexity and low solution efficiency when dealing with large-scale problems. Dynamic programming (DP): It is suitable for time series decision problems, but when dealing with large-scale energy systems, the dimension of the state space will increase significantly, resulting in an increased computational burden.

[0005] Although these traditional methods are effective in specific situations, they are difficult to cope with the ever-changing demands and environmental conditions in energy systems due to their reliance on model assumptions and limitations in computational complexity. Therefore, exploring more efficient optimization methods is an important direction of current research.

[0006] In recent years, deep learning and neural network technologies have made significant progress and demonstrated strong performance in areas such as image recognition and natural language processing. The core advantages of these technologies lie in their powerful feature extraction and nonlinear modeling capabilities, which enable neural networks to learn underlying laws and patterns from complex data. This advantage makes the application of neural networks in energy system optimization a new research direction. Although neural networks have shown potential in energy system optimization, traditional neural network models still have some problems when dealing with high-dimensional and variable energy systems, such as overfitting, unstable training, and high computational complexity. Summary of the invention

[0007] In order to solve the deficiencies in the prior art, the present invention provides a comprehensive energy optimization method and system based on an improved neural network to deal with the multi-objective optimization problem in IES planning.

[0008] The present invention adopts the following technical solution.

[0009] A first aspect of the present invention provides a comprehensive energy optimization method based on an improved neural network, comprising the following steps:

[0010] With the goal of minimizing total operating costs, minimizing carbon emissions and other pollutant emissions, and maximizing overall energy efficiency, an improved neural network model is constructed based on pre-collected multimodal datasets;

[0011] Use the improved neural network model to predict comprehensive energy, compare the output of the improved neural network model with the set target, and establish a comprehensive energy optimization model in combination with constraint conditions;

[0012] According to the comprehensive energy optimization model, a variety of optimization schemes are generated for the demand and supply of different energy forms. Combined with risk assessment, the scheme with the smallest fluctuation of performance indicators under all disturbance scenarios is selected as the best scheme to optimize the comprehensive energy.

[0013] Preferably, the improved neural network model is as follows:

[0014]

[0015] Where:

[0016] W l and b l Represent the weight and bias of each layer respectively; σ represents the activation function; α i represents the weight of the attention mechanism; W o and b o Represents the weights and biases of the output layer.

[0017] Preferably, constructing an improved neural network model based on a pre-collected multimodal dataset includes:

[0018] The collected multimodal data is input into the neural network model. The multimodal data is received through the input layer. The output of the input layer is as follows:

[0019] X(t)=[x1(t),x2(t),...,x N (t)]

[0020] Where:

[0021] N represents the number of input features;

[0022] x i (t) represents the value of the i-th feature at time t, i = 1, 2, ..., N;

[0023] The input data features are extracted through multiple hidden layers, including convolutional layers and recurrent layers; the network is assumed to contain L layers, and the output of the lth layer is as follows:

[0024] H l (t) = f(W l H l-1 (t)+b l )

[0025] Where:

[0026] H l (t) represents the output of the lth layer at time t, l = 1, 2, ..., L, where H0(t) = X(t);

[0027] W l Represents the weight matrix of the lth layer, which is used to linearly transform the input features;

[0028] b l Represents the bias vector of the lth layer, which is used to adjust the output of the neural network model;

[0029] f(·) represents the activation function;

[0030] The convolution layer is used to extract local features of the input data, and the weight of the convolution kernel is set to K. l ,,The output of the convolutional layer is as follows:

[0031] C l (t) = f(K l *H l-1 (t)+b l )

[0032] Where:

[0033] * indicates convolution operation; K l Represents the weight matrix of the convolution kernel;

[0034] The recurrent layer is used to capture the temporal dependency of time series data. The output of the recurrent layer is as follows:

[0035] D l (t) = LSTM (D l-1 (t), D l (t-1), W l , b l )

[0036] Where:

[0037] D l (t) represents the output of the lth recurrent layer at time t;

[0038] Set the attention weight to α i (t), the output A(t) of the attention layer is as follows:

[0039]

[0040] Where:

[0041] v i is the learning parameter of the attention mechanism; h i (t) is the output of the hidden layer;

[0042] The fusion layer is used to integrate multimodal features. The output F(t) of the fusion layer is calculated as the weighted sum of multimodal features as follows:

[0043]

[0044] Where:

[0045] wm is the weight of the mth modal feature; M is the number of modal features;

[0046] The output layer makes decisions based on the output of the fusion layer. The output layer is assumed to have K neurons, and the output O(t) is as follows:

[0047] O(t)=g(W o F(t)+b0)

[0048] Where:

[0049] W o is the weight matrix of the output layer; b o is the bias vector of the output layer; g(·) is the activation function of the output layer.

[0050] Preferably, the goal is to minimize total operating costs, minimize carbon emissions and other pollutant emissions, and maximize overall energy efficiency, including:

[0051] Minimizing the total operating cost includes minimizing fuel costs, minimizing equipment operation and minimizing maintenance costs. The total cost C is calculated as follows: total :

[0052] C total =C fuel (t)+C OM (t)

[0053] C fuel (t) = p fuel ·F(t)

[0054] C OM (t) = c OM ·T run (t)

[0055] Where:

[0056] C fuel (t) represents the fuel cost; p fuel represents the price per unit of fuel;

[0057] C OM (t) represents the operation and maintenance cost of the equipment; T run (t) and represents the operation time of the equipment in time t; F(t) represents the fuel consumption;

[0058] Minimize carbon emissions and other pollutant emissions through fuel consumption and emission factor calculations:

[0059] The carbon emission factor e of the fuel is calculated as follows: fuel :

[0060] minE emission (t) = min[e fuel ·F(t)]

[0061] Where:

[0062] E emission (t) represents carbon emissions;

[0063] Maximizing overall energy efficiency is expressed by the ratio of total energy production to energy consumption. The total energy input E is calculated as follows: input (t):

[0064]

[0065] Where:

[0066] η system (t) represents the overall energy efficiency; E use (t) represents the amount of energy effectively utilized.

[0067] Preferably, the comprehensive energy optimization model is as follows:

[0068]

[0069] Where:

[0070] C total Represents the total operating cost of comprehensive energy; C fuel (t) represents the fuel cost; C OM (t) represents the operation and maintenance cost of the equipment; E carbon (t) represents the costs related to carbon emissions and other pollutant emissions.

[0071] Preferably, the results output by the improved neural network model are compared with the set target, as follows:

[0072] L(θ)=λ1C total (t)+λ2E carbon (t)-λ3η system (t)

[0073] Where:

[0074] θ is the parameter set of the neural network; λ1, λ2, λ3 are the weight coefficients of each target, which are used to balance the relationship between different targets; C total represents the total operating cost of comprehensive energy; η system (t) represents the overall energy efficiency; E carbon (t) express the emission-related costs, based on the amount of carbon emissions and other pollutants emitted;

[0075] The neural network model is trained through the back propagation algorithm and optimizer, and the parameter θ is continuously updated until the comparison result of the output of the improved neural network model converges with the set target. The formula is as follows:

[0076]

[0077] Where:

[0078] η is the learning rate; Represents the changing trend of the improved neural network model output relative to each parameter.

[0079] Preferably, the constraints include the balance between electricity demand and supply, the balance between heat demand and supply, the balance between gas demand and supply, the start and stop status of equipment and the power output range, the transmission capacity limit of the power grid and the heat grid, and the emission limit;

[0080] The balance of electricity demand and supply, heat demand and supply, and gas demand and supply are as follows:

[0081] Taking the supply and demand matching of various energy forms as the core constraint of the comprehensive energy optimization model, L e (t), L h (t), L g (t) are the demand for electricity, heat and gas respectively, P gen (t), Q chp (t), G supply (t) supply of electricity, heat and gas respectively;

[0082] The electricity demand and supply balance is calculated as follows:

[0083] P gen (t)+P chp (t) = L e (t)

[0084] Where:

[0085] P gen (t) represents the output power of the power generation equipment; P chp (t) represents the power output of the cogeneration equipment; L e (t) represents the power demand;

[0086] The heat demand and supply balance is calculated as follows:

[0087] Q chp (t)+Q hp (t) = L h (t)

[0088] Where:

[0089] Q chp (t) represents the heat output of the cogeneration equipment; Q hp (t) represents the heat energy output of the heat pump; L h (t) represents the thermal energy demand;

[0090] The gas demand and supply balance is calculated as follows:

[0091] G supply (t) = L g (t)

[0092] Where:

[0093] G supply (t) represents the gas supply; L g (t) represents the gas demand.

[0094] Preferably, the start / stop state and power output range of the device include:

[0095] Set u i (t) is the start / stop state of device i, P i，min(t) and P i，max (t) represents the minimum and maximum output power of the equipment respectively:

[0096] The start-stop state constraint is calculated as follows:

[0097] u i (t)∈{0,1}

[0098] The power output range constraint is calculated as follows:

[0099] P i，min (t)·u i (t)≤P i (t)≤P i，max (t)·u i (t)

[0100] Where:

[0101] P i，min (t) represents the minimum output power of the equipment; P i，max (t) represents the maximum output power of the equipment;

[0102] u i (t) indicates the start and stop status of the equipment, u i (t) = 1 means the device is turned on, u i (t) = 1 means the device is turned off;

[0103] P i (t) represents the actual output power of the equipment;

[0104] Transmission capacity limitations of the electricity and heat networks include:

[0105] Taking into account the maximum transmission capacity of the power grid and the heat grid, set P ij (t) is the power transfer from node i to node j, Q ij (t) is the amount of heat energy transferred;

[0106] The power grid transmission capacity limit is calculated as follows:

[0107] 0≤P ij (t)≤P ij，max

[0108] Where:

[0109] P ij，max Indicates the maximum transmission capacity of the power transmission network;

[0110] The heat network transmission capacity limit is calculated as follows:

[0111] 0≤Q ij (t)≤Q ij，max

[0112] Where:

[0113] Q ij (t) represents the amount of heat energy transferred from node i to node j; Q ij，max Indicates the maximum transmission capacity of the heat transmission network;

[0114] Emissions limits include:

[0115] Setting E max Indicates the maximum emission:

[0116] E carbon (t)≤E max

[0117] Where:

[0118] E carbon (t) represents the carbon emissions at time t.

[0119] Preferably, the risk assessment process includes:

[0120] Generate multiple disturbance scenarios based on historical data and external factors, including extreme weather conditions and sudden load changes;

[0121] For each disturbance scenario, evaluate the performance indicators of different optimization schemes generated by the integrated energy optimization model, including cost, emission and supply-demand balance; analyze the performance fluctuations of each scheme under different scenarios and record the fluctuation range of each scheme;

[0122] Calculate risk metrics for each option under all scenarios, including:

[0123] Calculate the average underperformance of the solution under all scenarios;

[0124] The performance fluctuation range of the scheme in each scenario is evaluated through the variance numerical value.

[0125] The second aspect of the present invention provides a comprehensive energy optimization system based on an improved neural network, comprising: a neural network model building and training module, an optimization model building and solution generation module, and a risk assessment and optimal solution selection module;

[0126] The neural network model building and training module is used to build and train improved neural network models by processing multimodal datasets;

[0127] The optimization model building and solution generation module is used to build a comprehensive energy optimization model, with the goal of minimizing operating costs, carbon emissions and maximizing energy efficiency, to generate multiple optimization solutions. By solving the optimization model, it takes into account the demand and supply matching of different energy forms and provides a variety of optimization solutions;

[0128] The risk assessment and optimal solution selection module is used to perform risk assessment on the generated optimization solutions and analyze the stability and robustness of each solution under different disturbance scenarios; based on the evaluation results, the solution with the smallest performance fluctuation under various disturbance scenarios is selected.

[0129] Compared with the prior art, the present invention has the following beneficial effects:

[0130] (1) Through the improved neural network model, the nonlinear relationship and complex dynamic characteristics between various energy forms in the integrated energy system are fully considered; compared with traditional linear programming and mixed integer linear programming methods, it has stronger nonlinear modeling capabilities, effectively improves the accuracy of energy system optimization, and achieves higher computational efficiency through multi-objective optimization;

[0131] (2) The present invention can dynamically adjust and predict based on the comprehensive energy optimization model by modeling and optimizing multimodal data sets; it solves the complex energy demand and supply fluctuations that traditional methods cannot cope with, and ensures the stability and robustness of the system under different disturbance scenarios;

[0132] (3) The present invention integrates the objectives of total operating cost, carbon emissions and energy utilization efficiency into a comprehensive optimization framework. By combining the improved neural network model with risk assessment, it achieves a balance between economic benefits and environmental protection goals among different optimization schemes, and selects the scheme with the smallest fluctuation in performance indicators as the final optimization result, ensuring the efficient operation and environmental friendliness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0133] Figure 1 is the typical daily load curve in winter;

[0134] Figure 2 is the convergence of the upper and lower bounds (UB and LB) during the iteration process;

[0135] Figure 3 It is a neural network structure diagram;

[0136] Figure 4 Flowchart of a comprehensive energy optimization method based on improved neural network. DETAILED DESCRIPTION

[0137] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0138] Figure 1This is a typical daily load curve in winter. Through analysis, it can be seen that the electric load shows a typical "morning and evening peak" feature, which is in line with the changing law of electricity demand in winter, especially during periods of high demand for heating and lighting. The gas load and heat load are relatively stable, but the gas load fluctuates during the day, which may be related to the daytime heating demand or other gas demand. In terms of load, the electric load is the largest, followed by the heat load, and the gas load is relatively small, which reflects the main structure of energy use in the region.

[0139] Figure 2 is the upper and lower bounds (U B and L B ) convergence. As can be seen from the figure, the optimization algorithm goes through multiple stages in the convergence process. The upper and lower bounds eventually converge, indicating that the algorithm has found a better solution. B ) and the lower bound (LB) are already quite close in the early stage (about 60 iterations), which means that the convergence speed is fast and there is less room for improvement in the later stage. Overall, this shows that the optimization algorithm has good convergence when dealing with this problem.

[0140] Figure 3 The picture shows a typical feedforward neural network structure, including input layer, hidden layer and output layer.

[0141] like Figure 4 As shown, Example 1 of the present invention provides a comprehensive energy optimization method based on an improved neural network, comprising the following steps:

[0142] With the goal of minimizing total operating costs, minimizing carbon emissions and other pollutant emissions, and maximizing overall energy efficiency, an improved neural network model is constructed based on pre-collected multimodal datasets;

[0143] Preferably, the improved neural network model is as follows:

[0144]

[0145] Where:

[0146] W l and b l Represent the weight and bias of each layer respectively; σ represents the activation function; α i represents the weight of the attention mechanism; W o and b o Represents the weights and biases of the output layer.

[0147] Preferably, constructing an improved neural network model based on a pre-collected multimodal dataset includes:

[0148] The collected multimodal data is input into the neural network model. The multimodal data is received through the input layer. The output of the input layer is as follows:

[0149] X(t)=[x1(t),x2(t),…,x N (t)]

[0150] Where:

[0151] N represents the number of input features;

[0152] x i (t) represents the value of the i-th feature at time t, i = 1, 2, ..., N;

[0153] The input data features are extracted through multiple hidden layers, including convolutional layers and recurrent layers; the network is assumed to contain L layers, and the output of the lth layer is as follows:

[0154] H l (t) = f(W l H l-1 (t)+b l )

[0155] Where:

[0156] H l (t) represents the output of the lth layer at time t, l = 1, 2, ..., L, where H0(t) = X(t);

[0157] W l Represents the weight matrix of the lth layer, which is used to linearly transform the input features;

[0158] b l Represents the bias vector of the lth layer, which is used to adjust the output of the neural network model;

[0159] f(·) represents the activation function;

[0160] The convolution layer is used to extract local features of the input data, and the weight of the convolution kernel is set to K. l ,,The output of the convolutional layer is as follows:

[0161] C l (t) = f(K l *H l-1 (t)+b l )

[0162] Where:

[0163] * indicates convolution operation; K l Represents the weight matrix of the convolution kernel;

[0164] The recurrent layer is used to capture the temporal dependency of time series data. The output of the recurrent layer is as follows:

[0165] D l (t) = LSTM (D l-1 (t), D l (t-1), W l , b l )

[0166] Where:

[0167] D l (t) represents the output of the lth recurrent layer at time t;

[0168] Set the attention weight to α i (t), the output A(t) of the attention layer is as follows:

[0169]

[0170] Where:

[0171] v i is the learning parameter of the attention mechanism; h i (t) is the output of the hidden layer;

[0172] The fusion layer is used to integrate multimodal features. The output F(t) of the fusion layer is calculated as the weighted sum of multimodal features as follows:

[0173]

[0174] Where:

[0175] w m is the weight of the mth modal feature; M is the number of modal features;

[0176] The output layer makes decisions based on the output of the fusion layer. The output layer is assumed to have K neurons, and the output O(t) is as follows:

[0177] O(t)=g(W o F(t)+b o )

[0178] Where:

[0179] W o is the weight matrix of the output layer; b o is the bias vector of the output layer; g(·) is the activation function of the output layer.

[0180] Preferably, the goal is to minimize total operating costs, minimize carbon emissions and other pollutant emissions, and maximize overall energy efficiency, including:

[0181] Minimizing the total operating cost includes minimizing fuel costs, minimizing equipment operation and minimizing maintenance costs. The total cost C is calculated as follows:total :

[0182] C total =C fuel (t)+C OM (t)

[0183] C fuel (t) = p fuel ·F(t)

[0184] C OM (t) = c OM ·T run (t)

[0185] Where:

[0186] C fuel (t) represents the fuel cost; p fuel represents the price per unit of fuel;

[0187] C OM (t) represents the operation and maintenance cost of the equipment; T run (t) and represents the operation time of the equipment in time t; F(t) represents the fuel consumption;

[0188] Minimize carbon emissions and other pollutant emissions through fuel consumption and emission factor calculations:

[0189] The carbon emission factor e of the fuel is calculated as follows: fuel :

[0190] minE emission (t) = min[e fuel ·F(t)]

[0191] Where:

[0192] E emission (t) represents carbon emissions;

[0193] Maximizing overall energy efficiency is expressed by the ratio of total energy production to energy consumption. The total energy input E is calculated as follows: input (t):

[0194]

[0195] Where:

[0196] η system (t) represents the overall energy efficiency; E use (t) represents the amount of energy effectively utilized.

[0197] Preferably, the comprehensive energy optimization model is as follows:

[0198]

[0199] Where:

[0200] C total represents the total operating cost of comprehensive energy; c fuel (t) represents the fuel cost; C OM (t) represents the operation and maintenance cost of the equipment; E carbon (t) represents the costs related to carbon emissions and other pollutant emissions.

[0201] Use the improved neural network model to predict comprehensive energy, compare the output of the improved neural network model with the set target, and establish a comprehensive energy optimization model in combination with constraint conditions;

[0202] Preferably, the results output by the improved neural network model are compared with the set target, as follows:

[0203] L(θ)=λ1C total (t)+λ2E carbon (t)-λ3η system (t)

[0204] Where:

[0205] θ is the parameter set of the neural network; λ1, λ2, λ3 are the weight coefficients of each target, which are used to balance the relationship between different targets; C total represents the total operating cost of comprehensive energy; η system (t) represents the overall energy efficiency; E carbon (t) represents emission-related costs, based on the amount of carbon emissions and other pollutants emitted.

[0206] The neural network model is trained through the back propagation algorithm and optimizer, and the parameter θ is continuously updated until the comparison result of the output of the improved neural network model converges with the set target. The formula is as follows:

[0207]

[0208] Where:

[0209] η is the learning rate; Represents the changing trend of the improved neural network model output relative to each parameter.

[0210] Preferably, the constraints include the balance between electricity demand and supply, the balance between heat demand and supply, the balance between gas demand and supply, the start and stop status of equipment and the power output range, the transmission capacity limit of the power grid and the heat grid, and the emission limit;

[0211] The balance of electricity demand and supply, heat demand and supply, and gas demand and supply are as follows:

[0212] Taking the supply and demand matching of various energy forms as the core constraint of the comprehensive energy optimization model, L e (t), L h (t), L g (t) are the demand for electricity, heat and gas respectively, P gen (t), Q chp (t), G supply (t) supply of electricity, heat and gas respectively;

[0213] The electricity demand and supply balance is calculated as follows:

[0214] P gen (t)+P chp (t) = L e (t)

[0215] Where:

[0216] P gen (t) represents the output power of the power generation equipment; P chp (t) represents the power output of the cogeneration equipment; L e (t) represents the power demand;

[0217] The heat demand and supply balance is calculated as follows:

[0218] Q chp (t)+Q hp (t) = L h (t)

[0219] Where:

[0220] Q chp (t) represents the heat output of the cogeneration equipment; Q hp (t) represents the heat energy output of the heat pump; L h (t) represents the thermal energy demand;

[0221] The gas demand and supply balance is calculated as follows:

[0222] G supply (t) = L g (t)

[0223] Where:

[0224] G supply (t) represents the gas supply; L g (t) represents the gas demand.

[0225] Preferably, the start / stop state and power output range of the device include:

[0226] Set u i (t) is the start / stop state of device i, P i，min (t) and P i，max (t) represents the minimum and maximum output power of the equipment respectively:

[0227] The start-stop state constraint is calculated as follows:

[0228] u i (t)∈{0,1}

[0229] The power output range constraint is calculated as follows:

[0230] P i，min (t)·u i (t)≤P i (t)≤P i，max (t)·u i (t)

[0231] Where:

[0232] P i，min (t) represents the minimum output power of the equipment; P i，max (t) represents the maximum output power of the equipment;

[0233] u i (t) indicates the start and stop status of the equipment, u i (t) = 1 means the device is turned on, u i (t) = 1 means the device is turned off;

[0234] P i (t) represents the actual output power of the equipment;

[0235] Transmission capacity limitations of the electricity and heat networks include:

[0236] Taking into account the maximum transmission capacity of the power grid and the heat grid, set P ij (t) is the power transfer from node i to node j, Q ij (t) is the amount of heat energy transferred;

[0237] The power grid transmission capacity limit is calculated as follows:

[0238] 0≤P ij (t)≤P ij，max

[0239] Where:

[0240] P ij，max Indicates the maximum transmission capacity of the power transmission network;

[0241] The heat network transmission capacity limit is calculated as follows:

[0242] 0≤Q ij (t)≤Q ij，max

[0243] Where:

[0244] Q ij (t) represents the amount of heat energy transferred from node i to node j; Q ij，max Indicates the maximum transmission capacity of the heat transmission network;

[0245] Emissions limits include:

[0246] Setting E max Indicates the maximum emission:

[0247] E carbon (t)≤E max

[0248] Where:

[0249] E carbon (t) represents the carbon emissions at time t.

[0250] According to the comprehensive energy optimization model, a variety of optimization schemes are generated for the demand and supply of different energy forms. Combined with risk assessment, the scheme with the smallest fluctuation of performance indicators under all disturbance scenarios is selected as the best scheme to optimize the comprehensive energy.

[0251] Preferably, the risk assessment process includes:

[0252] Generate multiple disturbance scenarios based on historical data and external factors, including extreme weather conditions and sudden load changes;

[0253] For each disturbance scenario, evaluate the performance indicators of different optimization schemes generated by the integrated energy optimization model, including cost, emission and supply-demand balance; analyze the performance fluctuations of each scheme under different scenarios and record the fluctuation range of each scheme;

[0254] Calculate risk metrics for each option under all scenarios, including:

[0255] Calculate the average underperformance of the solution under all scenarios;

[0256] The performance fluctuation range of the scheme in each scenario is evaluated through the variance numerical value.

[0257] Example 2 of the present invention provides a comprehensive energy optimization system based on an improved neural network, comprising: a neural network model building and training module, an optimization model building and solution generation module, and a risk assessment and optimal solution selection module;

[0258] The neural network model building and training module is used to build and train improved neural network models by processing multimodal datasets;

[0259] The optimization model building and solution generation module is used to build a comprehensive energy optimization model, with the goal of minimizing operating costs, carbon emissions and maximizing energy efficiency, to generate multiple optimization solutions. By solving the optimization model, it takes into account the demand and supply matching of different energy forms and provides a variety of optimization solutions;

[0260] The risk assessment and optimal solution selection module is used to perform risk assessment on the generated optimization solutions and analyze the stability and robustness of each solution under different disturbance scenarios; based on the evaluation results, the solution with the smallest performance fluctuation under various disturbance scenarios is selected.

[0261] The beneficial effects of the present invention are as follows:

[0262] (1) Through the improved neural network model, the nonlinear relationship and complex dynamic characteristics between various energy forms in the integrated energy system are fully considered; compared with traditional linear programming and mixed integer linear programming methods, it has stronger nonlinear modeling capabilities, effectively improves the accuracy of energy system optimization, and achieves higher computational efficiency through multi-objective optimization;

[0263] (2) The present invention can dynamically adjust and predict based on the comprehensive energy optimization model by modeling and optimizing multimodal data sets; it solves the complex energy demand and supply fluctuations that traditional methods cannot cope with, and ensures the stability and robustness of the system under different disturbance scenarios;

[0264] (3) The present invention integrates the objectives of total operating cost, carbon emissions and energy utilization efficiency into a comprehensive optimization framework. By combining the improved neural network model with risk assessment, it achieves a balance between economic benefits and environmental protection goals among different optimization schemes, and selects the scheme with the smallest fluctuation in performance indicators as the final optimization result, ensuring the efficient operation and environmental friendliness of the system.

[0265] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0266] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A comprehensive energy optimization method based on an improved neural network, characterized in that: The following steps are involved: With the goal of minimizing total operating costs, minimizing carbon emissions and other pollutant emissions, and maximizing overall energy efficiency, an improved neural network model is constructed based on pre-collected multimodal datasets; Use the improved neural network model to predict comprehensive energy, compare the output of the improved neural network model with the set target, and establish a comprehensive energy optimization model in combination with constraint conditions; According to the comprehensive energy optimization model, a variety of optimization schemes are generated for the demand and supply of different energy forms. Combined with risk assessment, the scheme with the smallest fluctuation of performance indicators under all disturbance scenarios is selected as the best scheme to optimize the comprehensive energy.

2. The comprehensive energy optimization method based on an improved neural network according to claim 1, characterized in that: The improved neural network model is as follows: Where: W l and b l Represent the weight and bias of each layer respectively; σ represents the activation function; α i Represents the weight of the attention mechanism; W o and b o Represents the weights and biases of the output layer.

3. The comprehensive energy optimization method based on an improved neural network according to claim 1 or 2, characterized in that: Build improved neural network models based on pre-collected multimodal datasets, including: The collected multimodal data is input into the neural network model. The multimodal data is received through the input layer. The output of the input layer is as follows: X(t)=[x1(t),x2(t),…,x N (t)] Where: N represents the number of input features; x i (t) represents the value of the i-th feature at time t, i = 1, 2, ..., N; The input data features are extracted through multiple hidden layers; assuming that the network contains L layers, the output of the lth layer is as follows: H l (t)=f(W l H l-1 (t)+b l ) Where: H l (t) represents the output of the lth layer at time t, l = 1, 2, ..., L, where H0(t) = X(t); W l Represents the weight matrix of the lth layer, which is used to linearly transform the input features; bl represents the bias vector of the lth layer, which is used to adjust the output of the neural network model; f(·) represents the activation function; The convolution layer is used to extract local features of the input data, and the weight of the convolution kernel is set to K. l， , the output of the convolutional layer is as follows: C l (t)=f(K l *H l-1 (t)+b l ) Where: * indicates convolution operation; K l Represents the weight matrix of the convolution kernel; The recurrent layer is used to capture the temporal dependency of time series data. The output of the recurrent layer is as follows: D l (t)=LSTM(D l-1 (t),D l (t-1),W l ,b l ) Where: D l(t) represents the output of the lth recurrent layer at time t; Set the attention weight to α i (t), the output A(t) of the attention layer is as follows: Where: v i is the learning parameter of the attention mechanism; h i (t) is the output of the hidden layer; The fusion layer is used to integrate multimodal features. The output F(t) of the fusion layer is calculated as the weighted sum of multimodal features as follows: Where: wm is the weight of the mth modal feature; M is the number of modal features; The output layer makes decisions based on the output of the fusion layer. The output layer is assumed to have K neurons, and the output O(t) is as follows: O(t)=g(W o F(t)+b o ) Where: W o is the weight matrix of the output layer; b o is the bias vector of the output layer; g(·) is the activation function of the output layer.

4. The comprehensive energy optimization method based on an improved neural network according to claim 1 or 2, characterized in that: The goal is to minimize total operating costs, minimize carbon emissions and other pollutants, and maximize overall energy efficiency, including: Minimizing the total operating cost includes minimizing fuel costs, minimizing equipment operation and minimizing maintenance costs. The total cost C is calculated as follows: total : C total =C fuel (t)+C OM (t) C fuel (t)=p fuel ·F(t) C OM (t)=c OM ·T run (t) Where: C fuel (t) represents the fuel cost; p fuel represents the price per unit of fuel; C OM (t) represents the operation and maintenance cost of the equipment; T run (t) and represents the operation time of the equipment in time t; F(t) represents the fuel consumption; Minimize carbon emissions and other pollutant emissions through fuel consumption and emission factor calculations: The carbon emission factor e of the fuel is calculated as follows: fuel : minE emission (t)=min[e fuel ·F(t)] Where: E emission (t) represents carbon emissions; Maximizing overall energy efficiency is expressed by the ratio of total energy production to energy consumption. The total energy input E is calculated as follows: input (t): Where: η system (t) represents the overall energy efficiency; E use (t) represents the amount of energy effectively utilized.

5. The comprehensive energy optimization method based on an improved neural network according to claim 1, characterized in that: The comprehensive energy optimization model is as follows: Where: C total represents the total operating cost of comprehensive energy; C fuel (t) represents the fuel cost; C OM (t) represents the operation and maintenance cost of the equipment; E carbon (t) represents the costs related to carbon emissions and other pollutant emissions.

6. The comprehensive energy optimization method based on an improved neural network according to claim 1 or 5, characterized in that: The results output by the improved neural network model are compared with the set goals, as follows: L(θ)=λ1C total (t)+λ2E carbon (t)-λ3η system (t) Where: θ is the parameter set of the neural network; λ1, λ2, λ3 are the weight coefficients of each target, which are used to balance the relationship between different targets; C total represents the total operating cost of the integrated energy source; η system (t) represents the overall energy efficiency; E carbon (t) express the emission-related costs, based on the amount of carbon emissions and other pollutants emitted; The neural network model is trained through the back propagation algorithm and optimizer, and the parameter θ is continuously updated until the comparison result of the output of the improved neural network model converges with the set target. The formula is as follows: Where: η is the learning rate; Represents the changing trend of the improved neural network model output relative to each parameter.

7. The comprehensive energy optimization method based on an improved neural network according to claim 1 or 5, characterized in that: Constraints include electricity demand and supply balance, heat demand and supply balance, gas demand and supply balance, equipment start and stop status and power output range, transmission capacity limitations of power grids and heat networks, and emission limits; The balance of electricity demand and supply, heat demand and supply, and gas demand and supply are as follows: Taking the supply and demand matching of various energy forms as the core constraint of the comprehensive energy optimization model, L e (t), L h (t), L g (t) are the demand for electricity, heat and gas respectively, P gen (t), Q chp (t), G supply (t) supply of electricity, heat and gas respectively; The electricity demand and supply balance is calculated as follows: P gen (t)+P chp (t)=L e (t) Where: P gen (t) represents the output power of the power generation equipment; P chp (t) represents the power output of the cogeneration equipment; L e (t) represents the power demand; The heat demand and supply balance is calculated as follows: Q chp (t)+Q hp (t)=L h (t) Where: Q chp (t) represents the heat output of the cogeneration equipment; Q hp (t) represents the heat energy output of the heat pump; L h (t) represents the thermal energy demand; The gas demand and supply balance is calculated as follows: G supply (t)=L g (t) Where: G supply (t) represents the gas supply; L g (t) represents the gas demand.

8. The comprehensive energy optimization method based on improved neural network according to claim 7, characterized in that: The equipment start / stop status and power output range include: Set u i (t) is the start / stop state of device i, P i，min (t) and P i，max (t) represents the minimum and maximum output power of the equipment respectively: The start-stop state constraint is calculated as follows: u i (t)∈{0,1} The power output range constraint is calculated as follows: P i，min (t)·u i (t)≤P i (t)≤P i，max (t)·u i (t) Where: P i，min (t) represents the minimum output power of the equipment; P i，max (t) represents the maximum output power of the equipment; u i (t) indicates the start and stop status of the equipment, u i (t) = 1 means the device is turned on, u i (t) = 1 means the device is turned off; P i (t) represents the actual output power of the equipment; Transmission capacity limitations of the electricity and heat networks include: Taking into account the maximum transmission capacity of the power grid and the heat grid, set P ij (t) is the amount of power transferred from node i to node j, Qij(t) is the amount of heat energy transferred; The power grid transmission capacity limit is calculated as follows: 0≤P ij (t)≤P ij，max Where: P ij，max Indicates the maximum transmission capacity of the power transmission network; The heat network transmission capacity limit is calculated as follows: 0≤Qij(t)≤Qij,max Where: Qij(t) represents the amount of heat energy transferred from node i to node j; Qij,max represents the maximum transmission capacity of the heat energy transmission network; Emissions limits include: Setting E max Indicates the maximum emission: E carbon (t)≤E max Where: E carbon (t) represents the carbon emissions at time t.

9. The comprehensive energy optimization method based on improved neural network according to claim 1, characterized in that: The risk assessment process includes: Generate multiple disturbance scenarios based on historical data and external factors, including extreme weather conditions and sudden load changes; For each disturbance scenario, evaluate the performance indicators of different optimization schemes generated by the integrated energy optimization model, including cost, emission and supply-demand balance; analyze the performance fluctuations of each scheme under different scenarios and record the fluctuation range of each scheme; Calculate risk metrics for each option under all scenarios, including: Calculate the average underperformance of the solution under all scenarios; The performance fluctuation range of the scheme in each scenario is evaluated through the variance numerical value.

10. A comprehensive energy optimization system based on an improved neural network, comprising: Neural network model building and training module, optimization model building and solution generation module, and risk assessment and optimal solution selection module; Running a comprehensive energy optimization method based on an improved neural network as described in any one of claims 1 to 9, characterized in that: The neural network model building and training module is used to build and train improved neural network models by processing multimodal datasets; The optimization model building and solution generation module is used to build a comprehensive energy optimization model, with the goal of minimizing operating costs, carbon emissions and maximizing energy efficiency, to generate multiple optimization solutions. By solving the optimization model, it takes into account the demand and supply matching of different energy forms and provides a variety of optimization solutions; The risk assessment and optimal solution selection module is used to perform risk assessment on the generated optimization solutions and analyze the stability and robustness of each solution under different disturbance scenarios; based on the evaluation results, the solution with the smallest performance fluctuation under various disturbance scenarios is selected.

Images