Network load interaction and collaborative optimization method, network load interaction and collaborative optimization device and equipment

Abstract

The invention discloses a network load interaction collaborative optimization method, a network load interaction collaborative optimization device and network load interaction collaborative optimization equipment, and relates to the field of power systems. The network load interaction effect can be optimized, and the benefit of energy scheduling can be improved. The method comprises the steps of collecting multi-source data of an energy supply side and an energy demand side of the high-energy-consumption industry; inputting the multi-source data into the trained load prediction model to obtain a grade level and a load prediction value; determining a plurality of variables and an objective function of network load interaction and constraint conditions of the variables, and optimizing the variables based on the constraint conditions, the objective function and the grade level and the load prediction value of the energy supply side to obtain an optimal variable combination; and inputting the optimal variable combination into a benefit evaluation model based on the game theory to obtain an evaluation index of the optimal variable combination, and determining a network load interaction energy distribution scheme through the evaluation index.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power systems, in particular to a method and device for coordinated optimization of grid-load interaction. BACKGROUND

[0002] As the core carrier of energy consumption and power load, the load characteristics of high energy-consuming industries show significant complexity and uniqueness. High energy-consuming industrial load has typical characteristics such as strong production correlation, large fluctuation amplitude, and differentiated demand for energy grade. In the production process, there is a rigid demand for high-grade energy, and the power load will fluctuate greatly and change nonlinearly due to production rhythm, process adjustment, and other factors. If there is a lack of effective control strategy, it will have a significant impact on the stability of the power grid. At the same time, frequent start-stop operations of equipment will exacerbate the fluctuation degree of power load, and the operation efficiency of some systems directly restricts the overall energy consumption level. According to the statistics of power grid operation data, during the large-scale equipment start-stop of high energy-consuming enterprises, the local power grid voltage fluctuation amplitude can reach ±5%, which poses a potential risk to the safe and stable operation of the power system.

[0003] Currently, there are two major technical bottlenecks in load prediction and grid-load interaction for high energy-consuming industrial users: first, the accuracy of load prediction needs to be optimized and improved; second, the evaluation system of grid-load interaction has limitations. The existing evaluation method of grid-load interaction has problems such as single evaluation dimension and insufficient quantification of multivariate constraints, and there is currently a lack of effective source-grid-load collaborative operation mechanism. When developing a grid-load interaction scheme, the power grid often proposes requirements unilaterally, and the enterprise responds passively, lacking a process of deep consultation and joint decision-making between the two parties. This lack of coordination makes it difficult to find the best balance point between power grid operation stability and enterprise economic benefits, and it is easy to deviate to the optimal condition of one side. According to the power grid research data, only 25% of the existing grid-load interaction schemes can meet the requirements of power grid operation stability and enterprise economic benefits, and the evaluation results show the characteristics of "unilateral optimization", which results in poor actual application effect of a large number of schemes. SUMMARY

[0004] The present application provides a method and device for coordinated optimization of grid-load interaction, which constructs a high-precision load prediction model that integrates multi-source data, establishes an evaluation system of grid-load interaction by combining multivariate constraint optimization and game theory, improves the adaptability of grid-load interaction to actual application, and optimizes the effect of energy dispatching.

[0005] In a first aspect, the present application provides a method for coordinated optimization of grid-load interaction, comprising: collecting multi-source data of the energy supply side and demand side of high energy-consuming industries; input the multi-source data of the energy supply side into the trained load prediction model to obtain the grade level and the load prediction value of the energy supply side, and input the multi-source data of the demand side into the load prediction model to obtain the grade level and the load prediction value of the demand side; determine a plurality of variables and a target function of the grid-load interaction, and a constraint condition of the plurality of variables, optimize the variables based on the constraint condition, the target function, and the grade level and the load prediction value of the energy supply side to obtain an optimal variable combination; input the optimal variable combination into a benefit evaluation model based on game theory to obtain an evaluation index of the optimal variable combination, and determine an energy distribution scheme of the grid-load interaction through the evaluation index.

[0006] In the grid-load interaction collaborative optimization method provided in the embodiment, the accuracy of the load prediction model is improved through multi-source data, and the energy distribution scheme is optimized through multi-variable constraints, so that the grid-load interaction simultaneously considers the energy supply side and the demand side, achieves comprehensive optimization, and introduces game theory to construct a bilateral equilibrium of the grid side and the enterprise side, comprehensively evaluates the grid-load interaction scheme, and finally outputs an optimal grid-load interaction scheme that takes into account the grid operation safety and enterprise benefits, improves the effect of the grid-load interaction scheme in actual application, and further reduces the cost and improves the benefit of energy dispatching.

[0007] In a second aspect, the present application provides a grid-load interaction collaborative optimization device, comprising: a data acquisition module configured to acquire multi-source data of an energy supply side and a demand side of a high-energy-consuming industry; a load prediction module configured to input the multi-source data of the energy supply side into a trained load prediction model to obtain the grade level and the load prediction value of the energy supply side, and input the multi-source data of the demand side into the load prediction model to obtain the grade level and the load prediction value of the demand side; a multi-variable optimization module configured to determine a plurality of variables and a target function of the grid-load interaction, and a constraint condition of the plurality of variables, optimize the variables based on the constraint condition, the target function, and the grade level and the load prediction value of the energy supply side to obtain an optimal variable combination; a scheme evaluation module configured to input the optimal variable combination into a benefit evaluation model based on game theory to obtain an evaluation index of the optimal variable combination, and determine an energy distribution scheme of the grid-load interaction through the evaluation index.

[0008] In a third aspect, the present application provides an electronic device, which comprises a memory and one or more processors. The memory stores one or more computer programs comprising instructions, which, when executed by the processor, cause the electronic device to perform the grid-load interaction collaborative optimization method of the first aspect.

[0009] In a fourth aspect, the present application provides a computer-readable storage medium, which stores instructions. When the instructions are executed on an electronic device, the electronic device performs the method for collaborative optimization of payload interaction as in the first aspect.

[0010] In a fifth aspect, the present application provides a computer program product, which, when executed on an electronic device, causes the electronic device to perform the method for collaborative optimization of payload interaction as in the first aspect.

[0011] It can be understood that the beneficial effects achieved by the above-provided device for collaborative optimization of payload interaction, electronic device, computer-readable storage medium, and computer program product can refer to the beneficial effects in the first aspect, which will not be described here again. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 A flowchart of the method for collaborative optimization of payload interaction provided by the embodiments of the present application is shown; Figure 2 A structural diagram of the device for collaborative optimization of payload interaction provided by the embodiments of the present application is shown; Figure 3 A structural diagram of the electronic device provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0013] In order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the same items or similar items with basically the same functions and effects are distinguished by using “first”, “second”, etc. For example, the first chip and the second chip are only used to distinguish different chips, and do not limit the sequence. Those skilled in the art can understand that “first”, “second”, etc. do not limit the quantity and execution sequence, and “first”, “second”, etc. also do not necessarily mean different. It should be noted that in the embodiments of the present application, “exemplary” or “for example” means to serve as an example, illustration or description. Any embodiment or design scheme described as “exemplary” or “for example” in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the use of “exemplary” or “for example” is intended to present the relevant concept in a specific manner. In the embodiments of the present application, “at least one” means one or more, and “multiple” means two or more.

[0014] It should be noted that “at the time of” in the embodiments of the present application can be at the moment when a certain condition occurs, or can be within a period of time after a certain condition occurs, which is not limited in the embodiments of the present application.

[0015] The implementation of the embodiments will be described in detail below with reference to the drawings.

[0016] The embodiment provides a network-load interaction collaborative optimization method. The network-load interaction collaborative optimization method can be applied to various electronic devices such as a computer (PC), a tablet computer, a virtual reality / augmented reality device, a wearable device, an industrial computer, and a vehicle-mounted device. The network-load interaction collaborative optimization method can also be applied to a server, a cloud, and a server cluster, and the embodiment does not specially limit this.

[0017] Figure 1 A flowchart of the network-load interaction collaborative optimization method provided by the embodiment is shown.

[0018] As shown in Figure 1 , the network-load interaction collaborative optimization method can include the following steps. Step 101: Collecting multi-source data of an energy supply side and a demand side of a high-energy-consumption industry.

[0019] The energy supply side includes a power grid side and some power supply enterprises. The demand side includes high-energy-consumption industrial users, i.e., energy-consuming users.

[0020] The multi-source data includes four types of core data, i.e., production process data of the energy-consuming users, meteorological and environmental data, economic and scheduling data, and power grid operation data, and the data update frequency is ≤1 minute.

[0021] The production process data includes production equipment operation parameters (such as a steelmaking furnace temperature and a rolling mill rotating speed), production batch information (such as a steelmaking batch duration and product yield), and each grade energy consumption data (high grade: blast furnace gas 1200℃ consumption, medium grade: waste heat steam 300℃ consumption, and low grade: circulating water 50℃ consumption).

[0022] The meteorological and environmental data include outdoor temperature, humidity, and light intensity (which affect workshop cooling / heating load) and pollutant emission concentration (which affect environmental protection equipment load).

[0023] The economic and scheduling data include enterprise order quantity, production plan scheduling information (temporary production increase / decrease), power grid side electricity price policy (peak-valley electricity price), and carbon trading price.

[0024] The power grid operation data include power grid power supply voltage, frequency, and line transmission capacity (which affect network-load interaction feasibility).

[0025] Step 102: Inputting the multi-source data of the energy supply side into a trained load prediction model to obtain grade levels and load prediction values of the energy supply side, and inputting multi-source data of the demand side into the load prediction model to obtain grade levels and load prediction values of the demand side.

[0026] In order to unify the data, the collected multi-source data is preprocessed, specifically including: extracting the abnormal data in the multi-source data, correcting the abnormal data; filling the missing data of the corrected data, and fusing the filled data by using the Bayesian network to obtain a multi-variable set; inputting the multi-variable set into the trained load prediction model.

[0027] (1) Abnormal data correction: An isolated forest algorithm is used, which isolates data points by constructing multiple isolated trees. Assuming that the data set D contains n samples {X1, X2, X3,..., Xn}, the isolated tree T is recursively divided into two subsets by randomly selecting a sub-sample, randomly selecting an attribute q and a split point p of the attribute (p is between the minimum and maximum values of the attribute), until the subset contains only one sample or reaches the preset maximum tree depth. The path length h(x) of sample x in isolated tree T represents the difficulty of its isolation, and the shorter the path, the easier the sample is to be isolated, and the higher the abnormality. The final abnormality score calculation formula is:

[0028] Where E(h(x)) is the average path length of sample x in multiple isolated trees, and c(n) is the expected path length when the sample size is n. When s(x, n) approaches 1, the average path length of the sample is much shorter than the expected path length of the normal sample, and it is determined as abnormal data; when the score approaches 0, the sample is more likely to be normal data.

[0029] After identifying the abnormal data, it can be corrected by the adjacent value, the mean value.

[0030] (2) Missing data filling Multiple imputation method is used, and statistical model (such as regression model) is established based on observed data. Assuming that the variable with missing values is Y and the complete variable is X, the regression equation is established , a plurality of imputation data sets are generated by multiple simulations, and the imputation results are combined. Let M be the number of simulations, and the imputation value obtained by the mth simulation is , the final imputation value is , and the error rate is ≤3%.

[0031] Wherein, is the predicted value of the variable Y missing value by the regression model, is the intercept term of the regression equation, is the regression coefficient, which represents the influence degree of variable X on variable Y, is a random error term, which represents the part that the model cannot explain. is the m-th simulation result of the i-th sample and the j-th missing value.

[0032] (3) Data fusion The data fusion algorithm based on Bayesian network is a directed acyclic graph, and the nodes represent variables and the edges represent the probability dependence between variables. Assuming that there is a variable set {X1, X2,..., X n}, the joint probability distribution can be expressed as:

[0033] The probability distribution of each variable is updated by calculating the conditional probability, the data conflict is eliminated, and a unified "high energy-consuming industrial multi-source data feature library" is constructed.

[0034] Wherein, is the probability of {X1, X2,..., X n} occurring simultaneously; represents the probability of variable X i occurring under the condition that the parent node set occurs; ) is the parent node set of variable X i , that is, the set of other variables that have a direct probability impact on variable X i .

[0035] The load prediction model is constructed by a deep neural network, which is used to predict the load value and the grade level of energy. The embodiment further comprises: obtaining multi-source historical data of high energy-consuming industrial users, the multi-source historical data comprising production data, meteorological environment data, economic scheduling data and power grid operation data; inputting the multi-source historical data into a deep neural network, the deep neural network comprising a convolutional layer and a long short-term memory network; extracting feature representation of the multi-source historical data through the convolutional layer; inputting the feature representation into the long short-term memory network to obtain output data of the long short-term memory network, the output data comprising a grade level and a load prediction value; determining the difference between the output data and the true label of the high energy-consuming industrial user through mean square error, and according to the difference, using an adaptive learning rate optimization algorithm and a back propagation algorithm to iteratively update the parameters of the deep neural network to obtain the load prediction model.

[0036] Considering that the load data of high energy-consuming industrial users has spatiotemporal coupling characteristics, the input data matrix, i.e. multi-source historical data, can include time series features, represented as , which contains rich time-series characteristics and multivariate correlation information. Among them, T represents the time step, which is used to capture the dynamic trend of load data over time; N is the dimension of data characteristics, which covers key factors affecting load, such as industrial production process parameters, environmental temperature, and price fluctuations; C is the number of channels, which can distinguish different types of data sources, such as real-time monitoring data and historical statistical data.

[0037] The convolutional (CNN) layer as the core component of feature extraction processes the input data through convolution kernels . The first two dimensions (k x k) correspond to the "local window size" of the input data (which can be understood as local extraction of "time step + feature dimension"), and the last two dimensions represent the mapping of "input channel number → output channel number", which conforms to the operation logic of multi-channel convolution (each output channel is obtained by weighting all input channels through the corresponding convolution kernel). It can be flexibly adjusted according to the complexity requirements of the model to generate new feature representations. The convolution operation formula is:

[0038] where represents the value of the output feature map at position (i, j) in the two-dimensional space and the c'th channel, reflecting the feature expression at that specific position and channel after convolution operation; m, n: respectively, the offset index of the convolution kernel in the horizontal and vertical directions, 0 ≤ m < k, 0 ≤ n < k, k is the size of the convolution kernel; represents the value of the input data matrix at position (i + m, j + n) and the c'th channel, which is the source of the convolution operation; represents the weight coefficient of the convolution kernel at horizontal offset m, vertical offset n, input channel c, and output channel c', which determines the extraction strength of the input data at different positions and channel features; is the bias term corresponding to the c'th output channel, which is a learnable parameter used to adjust the baseline value of the convolution operation result and enhance the expression ability of the model.

[0039] LSTM (Long Short-Term Memory Network) as a variant of recurrent neural network (RNN), effectively solves the gradient vanishing and gradient explosion problems of traditional RNN when processing long sequence data, and is particularly suitable for capturing long-term dependencies in time series data. The core structure of the LSTM layer consists of a forget gate , an input gate , an output gate , and a cell state , which work together to achieve selective memory and update of information. The specific calculation formula is as follows: Forget gate Decide which information to discard from the cell state at the previous time step The output is a vector between 0 and 1, which is weighted and activated by the Sigmoid function σ on the input The concatenation of the hidden state at the previous time step And the input at the current time step The weighted sum and activation of the input gate

[0040] Input gate Used to control the flow of new information into the cell state at the current time step, also based on the Sigmoid function to activate the weighted input:

[0041] At the same time, the hyperbolic tangent function Generates a candidate cell state, Used to store potential updates to the cell state at the current time step:

[0042] Cell state As the "memory channel" of LSTM, it updates under the joint action of the forget gate and the input gate: first, the forget gate Selectively retains part of the cell state at the previous time step Then the input gate Fuses the filtered candidate state To achieve a dynamic balance between "forgetting" and "remembering" information:

[0043] Output gate Decide the content of the hidden state Output at the current time step, first use the Sigmoid function to determine the output weight, then use the hyperbolic tangent function to transform the updated cell state Finally, element-wise multiplication to get the final output:

[0044]

[0045] Where σ is the Sigmoid function, its value range is (0, 1), commonly used for output probability distribution; Where Is the output gate weight matrix, The output gate is biased, and the parameter definition logic of the forget gate and the input gate is consistent. tanh is the hyperbolic tangent function, the value range is (-1, 1), and the data can be mapped to the appropriate range; W and b are the weight matrix and bias vector of each calculation unit, respectively, and the parameters are optimized through model training; The hidden state is used to pass information between time steps; the symbol represents element-wise multiplication, which realizes fine filtering and fusion of information.

[0046] Train the model with multi-source historical data, set the number of training iterations according to the size of the data set, use the EarlyStopping method to prevent overfitting, and stop training when the validation set loss does not decrease for a certain number of times. At the same time, use the learning rate decay strategy to ensure the integrity of the training strategy. Use mean square error (MSE) as the loss function:

[0047] Among them, is the true load value, is the predicted load value.

[0048] Use the adaptive moment estimation (Adam) optimizer combined with the back propagation algorithm to update the weights and bias parameters of the deep neural network. Through dynamic adjustment of the learning rate mechanism, the model convergence efficiency is effectively improved. After the model training is completed, based on the real-time acquisition of the multi-variable set, the precise prediction of the high-energy industrial user's sub-grade load can be realized. According to the energy quality characteristics, it is divided into three grades of high, medium and low: high-grade energy (including electricity, natural gas, etc.) is mainly used to drive the operation of core production equipment; medium-grade energy (including steam, waste heat, etc.) is mainly used to serve auxiliary production links; low-grade energy (such as waste heat, biomass fuel, etc.) is mainly used for preheating and basic heating process. In addition, the model outputs the total load prediction value, which provides reliable data basis for net load interaction decision-making.

[0049] Taking the multi-variable set of the energy supply side as input, the grade level and load prediction value of the energy supply side can be obtained; taking the multi-variable set of the energy demand side as input, the grade level and load prediction value of the demand side can be obtained.

[0050] Step 103: Determine the multiple variables and objective function of net load interaction, and the constraint conditions of the multiple variables, optimize the variables based on the constraint conditions, the objective function, and the grade level and load prediction value of the energy supply side, to obtain the optimal variable combination. The grade levels of the energy supply side and the demand side include high, medium and low, and the multiple variables include energy supply amount of different grade levels, load demand amount of different grade levels, unit carbon emission amount and energy utilization efficiency; the constraint conditions include that the energy supply amount and the load demand amount meet the supply balance condition, the unit carbon emission amount is less than a first preset value, and the energy cascade utilization efficiency is greater than a second preset value.

[0051] The defined core variables include high, medium and low grade energy supply amounts P1, P2 and P3, such as grid power supply amount P1, enterprise self-provided energy such as photovoltaic direct supply amount P2, waste heat and energy supply amount P3, and sub-grade load demand (high grade load P4, medium grade load P5 and low grade load P6), unit product carbon emission amount (C) and energy cascade utilization efficiency (η).

[0052] The energy supply amount and the load demand amount meet the supply balance condition, that is, the total supply balance P1+P2+P3=P4+P5+P6 is met. The constraint conditions further include hard constraints, that is, conditions that must be met, including grid safety constraints, production constraints and carbon efficiency constraints, and soft constraints, including energy efficiency constraints and economic constraints.

[0053] The grid safety constraints are: load fluctuation amplitude ΔP≤±5% (in line with the grid GB / T15945 standard) and voltage deviation ΔU≤±2%.

[0054] The production constraints are: production batch load P4≥90% rated value (to avoid interruption of steelmaking furnace and other equipment).

[0055] The carbon efficiency constraints are: unit product carbon emission amount C≤industry benchmark value×90% (such as steel industry ton steel carbon emission≤1.6t).

[0056] The energy efficiency constraints are: energy cascade utilization efficiency η≥85%.

[0057] The economic constraints are: enterprise total energy cost (grid power purchase cost + self-provided energy operation and maintenance cost - carbon trading income)≤baseline period cost×90%. Profit_C is the carbon trading income, Cost1 is the grid power purchase cost, and Cost2 is the self-provided energy operation and maintenance cost.

[0058] An improved particle swarm optimization (IPSO) algorithm is adopted, and a target function is constructed to maximize the comprehensive benefits of grid-load interaction, and the formula is as follows:

[0059] Wherein, Profit_C is the carbon trading income, Cost1 is the grid electricity purchase cost, Cost2 is the self-provided energy cost; η_target is the energy efficiency target value, C represents the actual carbon emission of the energy-consuming user, C_benchmark is the carbon efficiency benchmark value, ΔP is the grid load fluctuation, and ΔP_max is the maximum grid load fluctuation allowed; is a weight coefficient.

[0060] The population size is set to 100, the iteration number is set to 500, and the adaptive inertia weight (0.9 at the beginning of iteration and 0.4 at the later stage) is used to obtain the optimal variable combination (such as the energy supply scheme of P1=30%, P2=40%, and P3=30%) that satisfies all constraints. A hybrid constraint processing mechanism of "penalty function + feasibility rule" is introduced to impose an exponential penalty on the solution that violates the hard constraint (for example, if the carbon emission exceeds 1%, the penalty coefficient is increased by 2 times); for the solution that satisfies the hard constraint, the solution close to the soft constraint target value is preferentially retained.

[0061] Exemplarily, the weight of each variable is determined through the attention mechanism, and the variable is weighted; the value of the objective function is solved by using the weighted variable, and the variable is optimized through the improved particle swarm optimization algorithm until the value of the objective function meets the preset condition.

[0062] The carbon trading income Profit_C, the grid electricity purchase cost Cost1, the energy-consuming user self-provided energy operation and maintenance cost Cost2, the actual carbon emission C of the energy-consuming user, and the grid load fluctuation ΔP are calculated by using the weighted variable, and then the value of the objective function is obtained.

[0063] The attention mechanism is introduced in the embodiment to focus on the key features that have the greatest impact on the optimization target, so that the algorithm preferentially adjusts the core variables and improves the optimization efficiency. The specific calculation process is as follows: The input features are , T is the time step, d is the hidden layer dimension, and P4, P5, P6, Cost1, Profit_C, etc. are six core features.

[0064] The attention score is calculated as follows: .

[0065] Wherein, is the input feature; and are trainable weight matrices and biases for linear transformation of ; tanh maps the transformed features to the interval (-1, 1) to increase the nonlinear expression capability; is a trainable weight vector, which is finally compressed into a single scalar score by doing an inner product with the output of tanh. The normalized attention weight is calculated as follows:

[0066] wherein T is the total number of time steps; is a summation variable, iterating over all time steps from 1 to T; is the result of summing the attention scores over all time steps after exponentiation.

[0067] The output of the attention mechanism is:

[0068] i.e. a weighted sum of "weight x feature", the weighted key features are input into the IPSO algorithm, so that the optimization process prioritizes the peak segment electricity purchase cost, high-grade load and other core variables. The final determined variables high-grade energy supply P1, medium-grade energy supply P2, low-grade energy supply P3, high-grade load demand P4, medium-grade load demand P5, low-grade load demand P6, unit carbon emission C and energy cascade utilization efficiency η are the optimal variable combination.

[0069] Step 104: input the optimal variable combination into the benefit evaluation model based on game theory to obtain an evaluation index of the optimal variable combination, and determine an energy distribution scheme of the grid-interactive load through the evaluation index.

[0070] The core demand of the grid side benefit is to ensure power supply safety and stability and reduce dispatching and peak shaving cost. The quantitative calculation of these two demands needs to take the output results of the load forecasting model as the core input to form a logical closed loop of load forecasting guiding benefit quantification. The benefit evaluation model comprehensively coordinates between the demand for power grid stability and the optimization of energy efficiency, carbon emission and economic benefit maximization of energy-consuming user side, and evaluates the optimal variable combination. Specifically, based on the grid safety index, energy-consuming user economy index, energy utilization efficiency, load regulation response speed and initial weight combination, the first benefit function of the grid and the second benefit function of the energy-consuming user are determined; the weight parameter combination when the first benefit function and the second benefit function meet the preset condition is solved based on the game theory equilibrium solving mechanism; when the change amplitude of each weight in the weight parameter combination is less than the preset threshold, the benefit evaluation model is obtained.

[0071] The "total load prediction curve", "sub-grade load prediction value" and "load fluctuation trend prediction" output by the load forecasting model correspond to the quantitative calculation of the three indexes in the grid side benefit function, which are the core data support for the benefit function to have actual guiding significance. The specific application scenarios are as follows: The grid side benefit function, i.e. the first benefit function (G): highlights power supply stability and peak shaving cost, and the formula is as follows:

[0072] Wherein, ΔP is the grid load fluctuation, which is the fluctuation amount calculated based on the load prediction curve and the optimized load, ΔP = | optimized load - predicted load |; ΔP_max is the maximum load fluctuation allowed by the grid, ΔP_max = 5% x P, P is the arithmetic mean of the 24-hour total load prediction value output by the load prediction model. ΔU represents the change amount of the grid voltage, which is obtained by the load curve optimized through the power flow calculation. The predicted P4, P5, P6 time series data are input during the power flow calculation, so as to accurately calculate the voltage change amount in different time periods; ΔU_max is the maximum voltage change allowed, the sum of the absolute values of the positive and negative deviations of 35 kV and above is not more than 10% of the nominal voltage; the deviation of 20 kV and below is ± 7% of the nominal voltage, and the fluctuation range of 220V user is + 7%, -10%; Cost_grid is the grid peak shaving cost, Cost_grid0 is the benchmark peak shaving cost; λ1, λ2, λ3 are the internal weights of the grid side, and λ1 + λ2 + λ3 = 1, λ1 + λ2 ≥ 60%, the priority of the grid safety is higher than the income.

[0073] The energy-consuming user side benefit function, i.e. the second benefit function (E), highlights economic benefits, energy efficiency and carbon efficiency, and the formula is as follows:

[0074] Wherein, Cost0 is the energy-consuming user benchmark energy cost, which is used as a reference standard for measuring changes in energy costs; μ1, μ2, μ3 are internal weights of the energy-consuming user, wherein μ1 ≥ 40%, reflecting the principle of giving priority to economic benefits.

[0075] In the economic benefit item, Cost1 is calculated based on the sub-grade load prediction value, combined with the optimized grid power supply ratio and peak-valley electricity price; Cost2 is the self-provided energy operation cost: the operation cost corresponding to the optimized photovoltaic direct supply amount P2 and the waste heat supply amount P3; Profit_C is the carbon trading income: based on the optimized unit product carbon emission amount C, combined with the carbon trading price; Cost0 is the enterprise benchmark energy cost: the cost benchmark calculated based on the original load prediction value.

[0076] In the energy efficiency item, η is the energy cascade utilization efficiency, which is calculated based on the grade attribute of the sub-grade load prediction (P4 is high grade and P6 is low grade), combined with the optimized energy matching scheme (P4 matches grid power and P6 matches waste heat), and the actual energy efficiency is, for example, η = 88%; The energy efficiency target value, i.e. the soft constraint of 85%.

[0077] In the carbon efficiency item, C is the actual unit product carbon emission amount, and C_benchmark is the carbon efficiency benchmark value, i.e. the industry benchmark of the hard constraint.

[0078] Then, the equilibrium weight is solved by Nash equilibrium game. In the interaction scenario of power system and high energy-consuming users, there is a natural difference between the interests of the grid side and the user side. In order to scientifically quantify the weights of both sides, a dynamic weight allocation model is constructed by using Nash equilibrium game theory. The specific implementation steps are as follows: 1) Initial weight setting Based on the operation characteristics of power system and the power consumption law of enterprise production, the four core indicators affecting the grid-load interaction are the grid safety and stability α, the enterprise power consumption economy β, the energy utilization efficiency γ, and the load regulation response speed δ.

[0079] α is the grid safety and stability and the load regulation response speed δ, which jointly support the grid side interest function. The calculation formula of the grid safety and stability α is: α=

[0080] ΔU is the actual voltage deviation. Based on the high-grade load P4 concentration period predicted by the sub-grade load, combined with the optimized load curve, the ΔU_max is obtained by the power flow calculation. The closer this value is to 1, the better the grid safety and stability.

[0081] The calculation formula of the load regulation response speed δ is: δ=

[0082] ΔP is the actual load fluctuation amplitude, and ΔP_max is the maximum load fluctuation allowed by the grid. The faster the load regulation response speed is, the smaller ΔP is, and the closer this value is to 1.

[0083] β is the core calculation basis of enterprise power consumption economy, which is the economic benefit item in the enterprise side interest function. The calculation formula is:

[0084] is the carbon trading income, Cost1 is the grid power purchase cost, Cost2 is the new energy operation and maintenance cost, and μ1 is the internal weight of energy-consuming users. This item directly reflects the enterprise power consumption economy, and the higher the β weight is, the greater the proportion of this item in the comprehensive benefit is.

[0085] γ is the core calculation basis of energy utilization efficiency, which is the energy efficiency item in the enterprise side interest function. The direct correlation formula is as follows:

[0086] is the energy cascade utilization efficiency, η_target is the energy efficiency target value, and μ2 is the internal weight of energy-consuming users. When η ≥ η_target, this item ≥ 1, and the higher the γ weight is, the better the comprehensive benefit of the energy efficiency standard scheme is.

[0087] The power grid side focuses on ensuring the stability of power supply, and needs to set the corresponding weights of four indicators: a-power grid safety and stability, β-enterprise electricity economy, γ-energy utilization efficiency, and δ-load regulation response speed . The initial weights can be randomly allocated or allocated according to experience, for example, (a, β, γ, δ) = (0.2, 0.2, 0.2, 0.4) , highlighting the key role of load regulation response speed in the stable operation of the power grid, so δ weight is the highest, matching the high-grade load fluctuation risk predicted. Energy-consuming users pay more attention to electricity cost and production benefit, and the initial weight is set to (a, β, γ, δ) = (0.4, 0.2, 0.2, 0.2) , emphasizing the basic guarantee of power grid safety and stability for continuous production of enterprises, so β weight is the highest, matching the optimized cost saving goal.

[0088] 2) Balanced solution mechanism An iterative optimization model is established with the two-party benefit function as the constraint condition, where G represents the comprehensive benefit function of the power grid side, including indicators such as power supply reliability improvement and peak regulation cost reduction; E represents the comprehensive benefit function of the energy-consuming user side, covering elements such as electricity cost saving and production plan flexibility adjustment.

[0089] Through a nonlinear programming algorithm, within the constraint space that satisfies “G ≥ G min and E ≥ E min ” (G min is the minimum interest threshold for the power grid to maintain safe operation, and E min is the basic interest bottom line for enterprises to guarantee production benefit), the gradient descent method is used for multiple rounds of iterative calculation. When the change amplitude of each indicator weight in the last three iterations is less than , it is considered that the system has reached a stable state, and the final balanced weight combination is output (such as (a, β, γ, δ) = (0.2, 0.2, 0.2, 0.4) ). This result not only guarantees the basic demand of the power grid side for load regulation, but also meets the reasonable electricity cost control target of enterprises, achieving the Pareto optimal of the interests of both the power grid and the load.

[0090] Substitute the determined optimal variable combination into the comprehensive benefit evaluation model based on the game theory weighting construction to obtain the values of the four indicators: a-power grid safety and stability, β-enterprise electricity economy, γ-energy utilization efficiency, and δ-load regulation response speed. After standardizing the values of the four indicators to the 0-1 dimension, the comprehensive benefit value F (i.e. the final evaluation indicator) is calculated. By comparing the comprehensive benefit value F with the minimum threshold Fmin=0.8 set by historical data, the differentiated decision-making process is executed.

[0091]

[0092] α, β, γ, δ are the values of the four indicators: power grid safety and stability, enterprise electricity economy, energy utilization efficiency, and load regulation response speed; corresponding weights of the four indexes.

[0093] When the evaluation index is greater than the evaluation threshold, the optimal variable combination is determined as the energy distribution scheme; when the evaluation index is not greater than the evaluation threshold, the variables are re-optimized until the evaluation index of the optimized optimal variable combination is greater than the evaluation threshold.

[0094] When F≥Fmin, the network-load interaction scheme output mechanism is triggered. The network-load interaction scheme of the trinity of "energy supply + load management + benefit target" is output, and the scheme is optimized in combination with the time dimension and the grade dimension. The energy supply distribution scheme needs to adjust the supply proportion based on the period characteristics, and reduce the cost by using low-price electricity in the valley segment; the load fine management scheme needs to be based on the principle of energy grade matching, P4 (high-grade load) is preferentially matched with blast furnace gas and grid electricity; P5 (medium-grade) is matched with photovoltaic direct supply; and P6 (low-grade) is matched with residual heat and energy, to ensure that η≥85%. The energy efficiency-carbon efficiency-economic benefit index is realized. In the energy efficiency dimension, the energy cascade utilization efficiency target is improved, the energy flow is optimized, and the energy conversion loss is reduced; in the carbon efficiency dimension, the unit product carbon emission intensity reduction index is set to promote green and low-carbon production; in the economic benefit dimension, the energy cost reduction is quantified, and the enterprise energy cost is reduced through reasonable energy allocation. When F<Fmin, it indicates that the scheme needs to be improved, and the iterative optimization mechanism is started. The core variable that reduces the benefit needs to be located based on sensitivity analysis, and the variable that has the greatest impact on F is identified. If the β score is low (the economic benefit is insufficient), the core problem is that Cost1 is too high; if Cost1 is too high, the P2 supply proportion in the peak segment can be increased (from 40% to 50%), and the P1 proportion can be reduced, and the IPSO algorithm is optimized, and the "variable adjustment-optimization calculation-benefit evaluation" optimization process is re-executed until F≥0.8, to ensure that the scheme finally meets the interests of both parties.

[0095] In the embodiment, the data statistics and analysis mechanism is started every 24 hours for the load prediction model, the multi-variable optimization model, and the benefit evaluation model. The system automatically collects the core operation data of the network-load interaction scheme, including the actual load prediction error, the energy cascade utilization efficiency, and the double-sided benefit value of the user side and the grid side. Based on the real-time data, the reverse optimization is performed from two dimensions of model optimization and weight adjustment. (1) Correcting the load prediction model A dynamic error assessment system is constructed. When the deviation between the actual load and the predicted load exceeds a 3% threshold, the system triggers a deep optimization process. Specifically, by analyzing the contribution of each production stage to the prediction error, the attention weights of the CNN-LSTM model are adjusted accordingly. For example, if a large prediction deviation occurs in the chemical reactor stage, the weights of features such as temperature, pressure, and raw material input are increased to enhance the model's ability to capture features of that stage. Simultaneously, historical data from similar operating conditions are introduced for transfer learning to further optimize model parameters and improve prediction accuracy. (2) Update the game equilibrium weights A dynamic monitoring mechanism for external conditions is established. When key external variables (such as a 50% increase in carbon prices, electricity price fluctuations exceeding 15%, or significant changes in the penetration rate of new energy sources) experience significant fluctuations, the system automatically initiates a game equilibrium recalculation process. Based on the new external environmental parameters, the weights of α (grid security and stability), β (enterprise electricity consumption economics), γ (energy utilization efficiency), and δ (load regulation response speed) are reallocated by solving the improved Nash equilibrium equation. By dynamically adjusting the weight allocation, the evaluation method for network-load interaction can be ensured to continuously adapt to the complex and ever-changing external environment and achieve a dynamic balance of interests among all parties.

[0096] In this embodiment, multi-source heterogeneous data, including enterprise production data, meteorological information, and electricity price fluctuations, are deeply integrated to construct a hybrid attention model based on CNN-LSTM. This model automatically extracts spatial features from the data using a convolutional neural network (CNN), captures temporal dependencies using a long short-term memory network (LSTM), and combines an attention mechanism to focus on key influencing factors, achieving accurate prediction of the load of high-energy-consuming industrial users. The effectiveness of traditional methods compared to this method is shown in Table 1. Compared to the 10%-15% error rate of traditional methods, the prediction error rate can be stably controlled within 5%. Simultaneously, it innovatively outputs load prediction values ​​by grade (such as base load, peak load, and valley load), providing high-precision data support for energy cascade utilization, energy storage system scheduling, and grid-load interaction strategy formulation.

[0097] Table 1 Comparison of Load Forecasting Accuracy

[0098] The embodiment also establishes a double-layer constraint mechanism to ensure the feasibility and efficiency of the grid-load interaction scheme. The hard constraint layer includes grid safety thresholds (such as voltage fluctuation range, frequency deviation), production continuity requirements (minimum downtime, equipment start-stop interval), and carbon efficiency policy indicators (upper limit of carbon emissions per unit of output), which are guaranteed to meet the mandatory standards through a constraint optimization algorithm. The soft constraint layer introduces energy efficiency optimization goals (such as comprehensive energy utilization rate improvement) and economic cost functions (including electricity costs and carbon trading costs), which are solved using a multi-objective particle swarm optimization algorithm to avoid conflicts between "compliance but low efficiency" and "high efficiency but violation of regulations", achieving coordinated optimization of safety, production, environmental protection, and economy.

[0099] Table 2 Double-layer constraint mechanism

[0100] Next, taking a large-scale steel enterprise with an output of 100 million tons as the application object, the engineering practice of this embodiment is carried out, and the specific implementation process is as follows: I. Multi-source data collection A distributed multi-dimensional data collection system is constructed, relying on industrial Internet of Things sensor networks, intelligent metering terminals, and external energy market data interfaces to realize real-time collection of key data. The system uses an edge computing architecture to complete data preprocessing and local storage at the data collection node, effectively reducing data transmission delay and ensuring data security. Step 1: Production process parameter collection Steelmaking furnace temperature monitoring: A dual-redundancy infrared temperature measurement system is used to monitor the temperature of the core area of the steelmaking furnace (1200-1300°C) with a sampling period of 10 minutes. The system integrates a blackbody radiation source self-calibration mechanism to ensure the continuity and accuracy of data collection. By establishing a linkage analysis model between temperature data and oxygen supply rate and charging rhythm, energy efficiency optimization in the smelting process is achieved. Rolling mill operation state monitoring: High-precision servo motor encoders are used to collect rolling mill main drive system (1500 r / min) speed data with millisecond-level precision. A multi-dimensional state monitoring matrix is constructed using vibration sensors and current transformers, and precise evaluation of equipment operating state is achieved through Fourier spectrum analysis, establishing a fault warning mechanism. Blast furnace gas metering: A multi-channel ultrasonic flowmeter with temperature and pressure compensation function is used to continuously meter blast furnace gas consumption (5000 m³ / h). By correlating gas consumption with blast furnace air volume and air pressure, a gas consumption prediction model is established to support energy dynamic scheduling. Step 2: Environmental and economic parameter collection Meteorological parameter monitoring: Multiple-element meteorological sensors are deployed at the meteorological station in the factory area to collect real-time data such as outdoor temperature, humidity, wind speed, and solar intensity (temperature fluctuation range 25-35°C). By establishing a meteorological- energy consumption coupling model, the impact of meteorological conditions on the energy consumption of cooling, ventilation, and other systems is analyzed to improve the accuracy of load forecasting. Power market data access: Interface with the provincial power trading platform to obtain peak and valley electricity prices (valley segment price 0.3 yuan / kWh, time period 23:00-7:00; peak segment price 0.8 yuan / kWh, time period 10:00-15:00 and 18:00-21:00), real-time electricity prices, and day-ahead electricity price data. Through electricity price trend analysis, combined with production planning, a time-of-use electricity strategy is developed to optimize electricity costs. Carbon emission data management: Access to the national carbon emission rights trading market, with synchronized carbon price data. Using a dual accounting mechanism of material balance method and emission factor method, carbon emission data from each production link of the enterprise is automatically collected to generate carbon emission reports, guiding the implementation of energy-saving and emission-reducing strategies by the enterprise. II. Load forecasting Step 1: Model training. A CNN-LSTM hybrid neural network model is used for load forecasting research. The CNN layer is used to extract the spatial features of the production equipment operating parameters, and the LSTM layer is used to process the long-term dependence relationship of time series data. Based on the enterprise's historical operating data at the minute level in the past three years (sample size over 100,000), the model is trained using the Adam optimization algorithm, with a learning rate of 0.001 and a training period of 50 rounds. To suppress the overfitting phenomenon, the Dropout regularization technique is introduced, randomly discarding 20% of the neuron connections. Step 2: Forecasting results. Achieve 15-minute resolution load forecasting for the next 24 hours. Among them, the high-grade load demand during the steelmaking operation period (8:00-18:00) reaches 200MW (electric arc furnace, continuous casting machine, and other equipment electricity), and the medium-grade load is 150MW (auxiliary equipment electricity); during the rolling mill production period (18:00-8:00 the next day), the medium-grade load is 180MW, and the low-grade load is 80MW (heating furnace heat preservation and equipment standby electricity). The prediction results are output in the form of a refined load curve, providing decision-making basis for energy scheduling. Step 3: Accuracy verification. The average absolute error (MAE), root mean square error (RMSE), and average absolute percentage error (MAPE) are used as evaluation indicators to verify the accuracy of the prediction results. The model prediction error rate is controlled at 4.2%, which is significantly better than the traditional ARIMA model (error rate 8.7%). By drawing scatter plots of predicted values and actual values and error distribution histograms, the model prediction performance is visually displayed. III. Multivariable optimization Step 1: Model construction Based on the improved particle swarm optimization (IPSO) algorithm, an energy scheduling optimization model is constructed, which contains 12 decision variables and 8 types of constraints (equipment operation constraints, power grid power interaction restrictions, carbon emission quota constraints, etc.). The decision variables include equipment start-stop time, running power, energy procurement quantity and distribution ratio, etc.; the constraints include minimum continuous operation time of equipment, power grid power fluctuation limit, total carbon emission control, etc. The objective function is to minimize the total cost, considering electricity charges, equipment operation and maintenance costs, and carbon emission costs. Step 2: Parameter setting The particle swarm size is set to 50, the maximum number of iterations is 200, and the inertia weight is linearly decreased from 0.9 to 0.4. A dynamic learning factor is introduced to improve the search performance, accelerating the global search speed in the early stage and enhancing the local optimization ability in the later stage. The elite preservation strategy is adopted to ensure that the best solution of each generation is passed to the next generation. Step 3: Optimization scheme The optimal energy scheduling scheme is obtained by model solving: in the valley period, the grid purchases 120 MW, the photovoltaic directly supplies 80 MW, and the waste heat and surplus energy is recycled for 50 MW; in the peak period, the photovoltaic directly supplies 100 MW, the waste heat and surplus energy is 80 MW, and the grid purchases 50 MW. Through sensitivity analysis, the influence of factors such as electricity price and carbon price on cost is studied, providing reference for enterprise decision-making. Four, game theory evaluation Step 1: Model construction A game evaluation model is constructed, which includes grid operators and enterprise users. The grid operators focus on ensuring power supply stability and reducing peak shaving costs, and their strategies include adjusting peak and valley electricity prices and issuing load instructions. Enterprise users focus on economic benefits and energy efficiency improvement, and their decisions include self-provided energy scheduling and production timing optimization. Combined with the hard constraint of "carbon emissions per unit product ≤ 90% of industry benchmark", the goal is to promote carbon emission reduction through carbon quota allocation and carbon policy guidance to influence enterprise cost and energy structure.

[0101] Both sides have a benefit game in energy scheduling: tightening the load constraint of the grid may increase the cost of enterprises, and the concentrated production of enterprises in the valley period may exacerbate the pressure on the grid. Based on this, a game model is established, which quantifies the grid peak shaving cost, enterprise electricity purchase and carbon trading revenue, and carbon emission compliance rate as a revenue function, analyzes the revenue changes of strategy combinations such as "electricity price adjustment + energy optimization + carbon policy", and clearly defines the benefit balance point, providing a clear benefit boundary for subsequent Nash equilibrium solving and weight calculation.

[0102] Step 2: Weight calculation The Nash equilibrium solving method is used to determine the evaluation index weight vector. The payment matrix is constructed to describe the income of each party strategy combination, and the Nash equilibrium point is solved by using the iteration algorithm. Finally, the weight vector is determined: power grid safety and stability (a=0.35), enterprise electricity economy (b=0.2), energy utilization efficiency (g=0.2), and load regulation response speed (d=0.25), and the weight rationality is verified by consistency test. Step 3: Benefit evaluation Firstly, based on the optimal energy scheduling scheme output by the multivariate optimization module, a comprehensive benefit evaluation system including 4 dimensions and 15 indicators is constructed. According to the preset attribution of 15 indicators in four dimensions, the arithmetic mean of the standardized indicator value under each dimension is calculated. Finally, the comprehensive benefit value F=0.86 is calculated by weighted summation and determination of the equilibrium weight (set threshold F=0.8). Through sensitivity analysis, the key influencing factors are identified, which provides basis for scheme optimization. Five, operation effect verification A one-month field operation verification is carried out, and the optimization scheme implementation effect is compared and analyzed by real-time collection of enterprise energy consumption, production data and power grid interaction data. In terms of economic benefit, the enterprise energy procurement cost is reduced by 24% compared with the same period last year, and the monthly electricity fee expenditure is saved by more than 8 million yuan. Through the analysis of energy procurement structure and electricity consumption time distribution, the cost optimization path is determined and the investment recovery period is evaluated. In terms of environmental benefit, the unit product carbon emission is reduced from 1.8tCO2 to 1.1tCO2, with a decrease of 38%. Combined with carbon emission data and carbon market transaction situation, the emission reduction effect of each production link is analyzed and the change of carbon asset value is evaluated. In terms of power grid cooperation, the peak valley regulation capacity of power grid is improved, the power supply pressure in high load peak period is significantly reduced, and the power consumption capacity in low valley period is enhanced. The power grid peak shaving cost is reduced by 20%, the load fluctuation is controlled within ±4.1%, and the regional power grid operation stability is effectively improved. Through the establishment of feedback mechanism with power grid operator, the net load interaction strategy is continuously optimized. Further, the embodiment also provides a net load interaction coordination optimization device, which can be used to execute the above-mentioned net load interaction coordination optimization method. As Figure 2As shown, the grid-load interaction collaborative optimization device 200 specifically includes: a data acquisition module 201, used to collect multi-source data from the energy supply and demand sides of high-energy-consuming industries; a load forecasting module 202, used to input the multi-source data from the energy supply side into the trained load forecasting model to obtain the grade level and load forecast value of the energy supply side, and input the multi-source data from the demand side into the load forecasting model to obtain the grade level and load forecast value of the demand side; a multivariate optimization module 203, used to determine multiple variables and objective functions of grid-load interaction, as well as the constraints of the multiple variables, and optimize the variables based on the constraints, the objective function, and the grade level and load forecast value of the energy supply side to obtain the optimal variable combination; and a scheme evaluation module 204, used to input the optimal variable combination into a game theory-based benefit evaluation model to obtain the evaluation index of the optimal variable combination, and determine the energy allocation scheme of grid-load interaction through the evaluation index.

[0103] The specific details of each module or unit in the above-mentioned grid-load interaction collaborative optimization device have been described in detail in the corresponding grid-load interaction collaborative optimization method, so they will not be repeated here.

[0104] This application also provides an electronic device. Figure 3 A schematic diagram of the structure of an electronic device suitable for implementing embodiments of the present disclosure is shown. Figure 3 The electronic device 600 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0105] like Figure 3 As shown, the electronic device 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) 602 or a program loaded from a storage section 608 into a random access memory (RAM) 603. The RAM 603 also stores various programs and data required for system operation. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0106] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as necessary. A removable media 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 610 as necessary, so that a computer program read out therefrom is installed in the storage section 608 as necessary.

[0107] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication section 609, and / or installed from the removable media 611. When the computer program is executed by the central processing unit (CPU) 601, the above-described functions defined in the embodiments of the present application are executed.

[0108] For example, when the computer program is executed by the central processing unit (CPU) 601, the following can be performed: collecting multi-source data of an energy supply side and a demand side of a high-energy-consuming industry; inputting the multi-source data of the energy supply side into a trained load prediction model to obtain a grade level and a load prediction value of the energy supply side, and inputting the multi-source data of the demand side into the load prediction model to obtain a grade level and a load prediction value of the demand side; determining a plurality of variables and a target function of grid-load interaction, and a constraint condition of the plurality of variables, optimizing the variables based on the constraint condition, the target function, and the grade level and the load prediction value of the energy supply side to obtain an optimal variable combination; inputting the optimal variable combination into a benefit evaluation model based on game theory to obtain an evaluation index of the optimal variable combination, and determining an energy distribution scheme of grid-load interaction through the evaluation index.

[0109] It should be noted that the computer-readable medium disclosed herein may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0110] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0111] The units described in the embodiments of the present disclosure can be implemented by software, or by hardware, or by a combination of software and hardware. The units described can also be located in a single processor. In some cases, the names of the units do not constitute a limitation on the units themselves.

[0112] As another aspect, the present application also provides a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist independently without being assembled into the electronic device. The computer readable medium carries one or more programs, which include instructions that, when executed by the electronic device, cause the electronic device to implement the method described in the above embodiments.

[0113] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, such division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units.

[0114] The above merely describes the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any changes or replacements within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for network-load interaction and collaborative optimization, characterized in that, include: Collect multi-source data on the energy supply and demand sides of energy-intensive industries; Multi-source data from the energy supply side is input into the trained load forecasting model to obtain the grade level and load forecast value of the energy supply side. Multi-source data from the demand side is input into the load forecasting model to obtain the grade level and load forecast value of the demand side. Determine multiple variables and objective function for grid-load interaction, as well as constraints on the multiple variables. Based on the constraints, the objective function, and the grade level and load forecast value of the energy supply side, optimize the variables to obtain the optimal variable combination. The optimal variable combination is input into a game theory-based benefit evaluation model to obtain the evaluation index of the optimal variable combination, and the energy allocation scheme of grid-load interaction is determined through the evaluation index.

2. The network-load interaction and collaborative optimization method according to claim 1, characterized in that, The process of inputting multi-source data from the energy supply side into the trained load forecasting model includes: Extract abnormal data from the multi-source data and correct the abnormal data; The corrected data is imputed for missing data, and a Bayesian network is used to fuse the imputed data to obtain a multivariate set. The multivariate set is input into the trained load prediction model.

3. The network-load interaction and collaborative optimization method according to claim 1, characterized in that, The quality levels on both the energy supply side and the demand side include high, medium, and low. The multiple variables include the energy supply at different quality levels, the load demand at different quality levels, the unit carbon emission, and the energy cascade utilization efficiency. The constraints include that the energy supply and load demand meet the supply balance condition, the unit carbon emission is less than a first preset value, and the energy cascade utilization efficiency is greater than a second preset value.

4. The network-load interaction and collaborative optimization method according to claim 1, characterized in that, The objective function is: Wherein, Profit_C is the carbon trading revenue, Cost1 is the grid electricity purchase cost, Cost2 is the self-supplied energy cost; η_target is the energy efficiency target value, C represents the actual carbon emissions of energy-consuming users, C_benchmark is the carbon efficiency benchmark value, ΔP is the grid load fluctuation, and ΔP_max is the maximum allowable load fluctuation of the grid. These are the weighting coefficients.

5. The network-load interaction and collaborative optimization method according to claim 1, characterized in that, Also includes: Acquire multi-source historical data of energy-consuming users, including production data, meteorological and environmental data, economic dispatch data, and power grid operation data; The multi-source historical data is input into a deep neural network, which includes convolutional layers and a long short-term memory network. Feature representations of the multi-source historical data are extracted using convolutional layers; The feature representation is input into a long short-term memory network to obtain the output data of the long short-term memory network, the output data including grade level and load prediction value; The difference between the output data and the actual labels of energy-consuming users is determined by the mean square error. Based on the difference, the parameters of the deep neural network are iteratively updated using an adaptive learning rate optimization algorithm and a backpropagation algorithm to obtain the load prediction model.

6. The network-load interaction and collaborative optimization method according to claim 1, characterized in that, The process of optimizing the variables based on the constraints, the objective function, and the energy supply-side quality level and load forecast values ​​to obtain the optimal variable combination includes: The weights of each variable are determined through an attention mechanism, and the variables are then weighted. The objective function is solved by weighting variables, and the variables are optimized by an improved particle swarm optimization algorithm until the value of the objective function meets the preset conditions.

7. The method for network-load interaction and collaborative optimization according to claim 1, characterized in that, Also includes: Based on power grid security indicators, energy-consuming user economic indicators, energy utilization efficiency, load regulation response speed, and initial weight combination, the first interest function of the power grid and the second interest function of energy-consuming users are determined. The combination of weight parameters when the first interest function and the second interest function satisfy the preset conditions is solved based on the game theory equilibrium solution mechanism. When the change in each weight in the weight parameter combination is less than a preset threshold, a benefit evaluation model is obtained.

8. The method for network-load interaction and collaborative optimization according to claim 1, characterized in that, The process of determining the energy allocation scheme for grid-load interaction through the evaluation indicators includes: When the evaluation index is greater than the evaluation threshold, the optimal combination of variables is determined as the energy allocation scheme; When the evaluation index is not greater than the evaluation threshold, the variables are re-optimized until the evaluation index of the optimal variable combination is greater than the evaluation threshold.

9. A network-load interactive collaborative optimization device, characterized in that, include: The data acquisition module is used to collect multi-source data from the energy supply and demand sides of energy-intensive industries; The load forecasting module is used to input multi-source data from the energy supply side into the trained load forecasting model to obtain the grade level and load forecast value of the energy supply side, and to input multi-source data from the demand side into the load forecasting model to obtain the grade level and load forecast value of the demand side. The multivariate optimization module is used to determine multiple variables and objective functions of grid-load interaction, as well as the constraints of the multiple variables. Based on the constraints, the objective function, and the grade level and load forecast value of the energy supply side, the module optimizes the variables to obtain the optimal combination of variables. The scheme evaluation module is used to input the optimal variable combination into a game theory-based benefit evaluation model to obtain the evaluation index of the optimal variable combination, and to determine the energy allocation scheme of grid-load interaction through the evaluation index.

10. An electronic device, characterized in that, The device includes a processor and a memory, the memory storing one or more computer programs, the one or more computer programs including instructions that, when executed by the electronic device, cause the electronic device to perform the network-load interaction collaborative optimization method according to any one of claims 1-8.

Images