A high-energy-consumption enterprise electric carbon factor dynamic prediction method and device, electronic equipment and storage medium

Abstract

The application discloses a kind of high energy-consuming enterprise electric carbon factor dynamic prediction method, device, electronic equipment and storage medium, belong to electric carbon factor dynamic prediction field, the method includes: obtaining the power plant generation plan data of target enterprise to be predicted period, external power supply electricity purchase plan data, production working condition plan parameter and surrounding weather forecast data, constructs multivariate time series feature set;Multivariate time series feature set is input into electric carbon factor prediction model, and the electric carbon factor of target enterprise to be predicted period is predicted;Based on the electric carbon factor obtained by prediction, the energy optimization of target enterprise is carried out;By implementing the application, the problem that the lack of effective use of the difference information of the superior and inferior individuals in the population in the prior art, leading to the hyperparameter optimization process of electric carbon factor prediction model is difficult to obtain global optimal solution, thereby causing the problem of insufficient prediction accuracy can be solved.

Description

Technical Field

[0001] This invention relates to the field of dynamic prediction technology of carbon dioxide factor, specifically to a method, device, electronic device, and storage medium for dynamic prediction of carbon dioxide factor in high-energy-consuming enterprises. Background Technology

[0002] High-energy-consuming enterprises, as key targets of carbon emission management, face increasingly stringent pressures regarding carbon quota compliance and carbon asset management. The electricity carbon factor, or carbon dioxide emissions per unit of electricity, is a crucial indicator for calculating an enterprise's indirect emissions. For enterprises with complex energy structures that own their own power plants while simultaneously purchasing electricity, the electricity carbon factor is influenced by a combination of factors, including the power generation strategy of their own power plants, the external power grid structure, the enterprise's production conditions, and meteorological conditions, exhibiting strong nonlinear and dynamic fluctuation characteristics.

[0003] However, existing technologies mostly employ deep learning-based time series prediction models. The predictive performance of deep learning time series prediction models largely depends on the configuration of hyperparameters such as learning rate, number of neurons, and window size. Currently, the selection of hyperparameters mainly relies on trial and error based on human experience or traditional grid search and conventional evolutionary algorithms. Human experience often lacks universality and is inefficient; while conventional evolutionary algorithms, during the search process, lack effective utilization of the differences between individuals of different quality in the population, such as superior and inferior individuals, causing the algorithm to easily get trapped in local optima in the later stages of optimization, making it difficult to find a better combination of hyperparameters. This results in limited accuracy of the final generated carbon factor prediction results, making it difficult to meet the needs of refined management. Summary of the Invention

[0004] This invention provides a method, device, electronic device, and storage medium for dynamic prediction of carbon dioxide factors in high-energy-consuming enterprises. It can solve the problem in the prior art that the lack of effective utilization of the differences between superior and inferior individuals in the population leads to the hyperparameter optimization process of the carbon dioxide factor prediction model being prone to getting trapped in local optima and making it difficult to obtain the global optimal solution, thus resulting in insufficient prediction accuracy.

[0005] An embodiment of the present invention provides a method for dynamic prediction of the carbon factor of high-energy-consuming enterprises, comprising: Obtain the target company's self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding weather forecast data for the forecast period; Based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production condition plan parameters and the surrounding meteorological forecast data, a multivariate time series feature set for the period to be predicted is constructed. Input the multivariate time series feature set into the preset electric carbon factor prediction model to generate the electric carbon factor of the target enterprise for the period to be predicted; Based on the carbon factor, at least one of the following is adjusted: the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, and the production operating condition plan parameters, in order to optimize the energy consumption of the target enterprise. The training of the electric carbon factor prediction model includes: Obtain the multivariate time series feature set of the target enterprise for historical periods and the corresponding historical carbon dioxide factor; initialize the carbon dioxide factor prediction model and generate a hyperparameter population; wherein, the hyperparameter population contains multiple sets of hyperparameters; The electric carbon factor prediction model is iteratively trained based on a multivariate temporal feature set from historical periods and a hyperparameter population until a preset iteration termination condition is met, thus determining the globally optimal hyperparameter combination. In each iteration, the multivariate temporal feature set from historical periods is input into the electric carbon factor prediction model configured with hyperparameters from each group in the hyperparameter population to obtain training prediction results. The fitness value is determined based on the error between the training prediction results and the corresponding historical electric carbon factors, and dominant and suboptimal individuals are selected based on the fitness value. A new generation of hyperparameter population is generated based on the numerical differences between dominant and suboptimal individuals to update the hyperparameter population. Configure the electric carbon factor prediction model based on the globally optimal hyperparameter combination, and generate the trained electric carbon factor prediction model.

[0006] Furthermore, the process of constructing a multivariate time-series feature set for the period to be predicted based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters, and the surrounding meteorological forecast data includes: Based on the preset timestamp, the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operation plan parameters and the surrounding meteorological forecast data are resampled to generate corresponding time series data with uniform time granularity. The time series data is subjected to dimensionless standardization processing to generate dimensionless time series data; Based on a preset time window, a multivariate time series feature set for the period to be predicted is constructed using dimensionless time series data.

[0007] Furthermore, the iterative evolution training of the electric carbon factor prediction model based on the multivariate time-series feature set and hyperparameter population of historical periods until a preset iteration termination condition is met, to determine the globally optimal hyperparameter combination, includes: Each hyperparameter individual in the hyperparameter population generated after initializing the electric carbon factor prediction model is configured to use the electric carbon factor prediction model. The multivariate time series feature set of historical periods is input into the configured electric carbon factor prediction model to obtain the initial training prediction result. The fitness value of each hyperparameter individual is determined based on the error between the initial training prediction result and the historical electric carbon factor. Repeatedly perform parameter evolution iteration operations until the preset iteration termination condition is met, and determine the individual with the best fitness value in the current hyperparameter population as the global optimal hyperparameter combination; The parameter evolution iteration operation includes: For each hyperparameter individual in the current hyperparameter population, a preset number of sample individuals are randomly selected from the current hyperparameter population. The individual with the best fitness value among the sample individuals is determined as the dominant individual, and the individual with the worst fitness value is determined as the suboptimal individual. The initial hyperparameter population is the hyperparameter population generated after initializing the electric carbon factor prediction model. The decay factor is determined based on the current number of iterations and a random step size vector is generated. At the same time, the base evolution vector is determined from the current hyperparameter population according to the preset probability strategy. Candidate hyperparameter individuals are generated using the base evolution vector, the decay factor, the random step size vector, the current hyperparameter individual, the dominant individual, and the disadvantaged individual. The candidate hyperparameter individuals are used to configure the electric carbon factor prediction model, and the multivariate time series feature set of historical periods is input into the configured electric carbon factor prediction model to obtain candidate training prediction results. The candidate fitness value of the candidate hyperparameter individual is determined based on the error between the candidate training prediction result and the corresponding historical carbon factor. If the difference between the candidate fitness value and the best fitness value in the current hyperparameter population satisfies the preset dynamic selection threshold condition, then the candidate hyperparameter individual is determined as the corresponding next-generation hyperparameter individual; otherwise, the current hyperparameter individual is determined as the corresponding next-generation hyperparameter individual. Once the next generation of hyperparameter individuals for all hyperparameter individuals in the current hyperparameter population has been determined, a next generation hyperparameter population is formed based on the determined next generation hyperparameter individuals, and the current hyperparameter population is updated using the next generation hyperparameter population.

[0008] Furthermore, determining the fitness value of each hyperparameter individual based on the error between the initial training prediction result and the historical electrocarbon factor includes: For each hyperparameter individual, calculate the mean absolute error, root mean square error, and mean absolute percentage error between the initial training prediction result and the historical carbon factor corresponding to the current hyperparameter individual. Calculate the sum of the mean absolute error, the root mean square error, and the mean absolute percentage error, and determine the sum as the fitness value of the current hyperparameter individual.

[0009] Furthermore, the step of determining the decay factor and generating a random step size vector based on the current iteration number, and simultaneously determining the basis evolution vector from the current hyperparameter population according to a preset probability strategy, includes: The iteration progress value is calculated based on the ratio of the current number of iterations to the preset maximum number of iterations. Based on the preset decay function, the decay factor is calculated according to the iteration progress value; A random vector containing random perturbation information is generated based on a preset random distribution function, and the random vector is determined as a random step size vector; wherein, the dimension of the random vector is consistent with the dimension of the current hyperparameter individual; A strategy selection random number between 0 and 1 is generated according to a preset uniform distribution function, and it is determined whether the strategy selection random number falls into a preset advantage guidance probability interval. If the random number selected by the strategy falls within the advantage-guided probability interval, then the individual with the best fitness value in the current hyperparameter population is determined as the base evolution vector. If the random number selected by the strategy does not fall within the advantage guidance probability range, then a hyperparameter individual is randomly selected from the current hyperparameter population to be determined as the base evolution vector.

[0010] Furthermore, the step of generating candidate hyperparameter individuals using the base evolution vector, the decay factor, the random step size vector, the current hyperparameter individual, the dominant individual, and the suboptimal individual includes: Calculate the numerical difference between the dominant and the disadvantaged individuals to generate an evolutionary difference vector; The evolutionary difference vector is weighted according to the decay factor to obtain the weighted evolutionary vector; Calculate the vector sum of the base evolution vector, the weighted evolution vector, and the random step size vector, and determine the calculated vector as the candidate hyperparameter individual.

[0011] Furthermore, the historical carbon factor of the target company for a given period can be obtained through the following methods: Acquire the target enterprise's historical power generation data from its self-owned power plant, carbon emission factors of its self-owned power plant, historical power purchase data from external sources, and carbon emission factors of external sources within a historical period. Calculate the product of the historical power generation data of the self-owned power plant and the carbon emission factor of the self-owned power plant to generate the carbon emissions of the self-owned power plant; The carbon emissions of the external power source are generated by multiplying the historical electricity purchase data of the external power source with the carbon emission factor of the external power source. Calculate the sum of the carbon emissions from the self-owned power plant and the carbon emissions from the external power source to generate a first calculation result; Calculate the sum of the historical power generation data of the self-owned power plant and the historical power purchase data of the external power source to generate a second calculation result; Calculate the ratio of the first calculation result to the second calculation result to generate the historical carbon factor for the historical period.

[0012] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0013] One embodiment of the present invention provides a dynamic prediction device for the carbon factor of high-energy-consuming enterprises, comprising: a data acquisition module, a feature construction module, a prediction generation module, an energy consumption optimization module, and a model training module; The data acquisition module is used to acquire the target enterprise’s self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding weather forecast data for the forecast period. The feature construction module is used to construct a multivariate time series feature set for the period to be predicted based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters and the surrounding meteorological forecast data. The prediction generation module is used to input a multivariate time series feature set into a preset electric carbon factor prediction model to generate the electric carbon factor of the target enterprise for the period to be predicted. The energy optimization module is used to adjust at least one of the following based on the carbon factor: the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, and the production operating condition plan parameters, so as to optimize the energy consumption of the target enterprise. The model training module is used to acquire the multivariate time-series feature set of the target enterprise for historical periods and the corresponding historical carbon dioxide factor; initialize the carbon dioxide factor prediction model and generate a hyperparameter population; wherein the hyperparameter population contains multiple sets of hyperparameters; iteratively evolve and train the carbon dioxide factor prediction model according to the multivariate time-series feature set of historical periods and the hyperparameter population until a preset iteration termination condition is met, and determine the globally optimal hyperparameter combination; wherein, in each iteration, the multivariate time-series feature set of historical periods is input into the carbon dioxide factor prediction model configured by each set of hyperparameters in the hyperparameter population to obtain training prediction results; determine the fitness value according to the error between the training prediction result and the corresponding historical carbon dioxide factor, and select dominant and suboptimal individuals based on the fitness value; generate a new generation of hyperparameter population to update the hyperparameter population according to the numerical difference information between dominant and suboptimal individuals; configure the carbon dioxide factor prediction model according to the globally optimal hyperparameter combination, and generate the trained carbon dioxide factor prediction model.

[0014] Based on the above method embodiments, the present invention provides corresponding electronic device embodiments.

[0015] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the dynamic prediction method for the carbon factor of high-energy-consuming enterprises as described in any of the above-described method embodiments.

[0016] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0017] One embodiment of the present invention provides a storage medium storing a computer program thereon, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described method embodiments of the dynamic prediction method for the carbon factor of high-energy-consuming enterprises.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method, apparatus, electronic device, and storage medium for dynamic prediction of carbon factor for high-energy-consuming enterprises. The method acquires data on the target enterprise's self-owned power plant power generation plan, external power purchase plan, production operating condition plan parameters, and surrounding meteorological forecast data for the period to be predicted, constructing a multivariate time-series feature set. This multivariate time-series feature set is then input into an carbon factor prediction model to predict the carbon factor for the target enterprise during the period to be predicted. Based on the predicted carbon factor, the self-owned power plant power generation plan, external power purchase plan, and production operating condition plan are optimized and adjusted to optimize the target enterprise's energy consumption. Specifically, the carbon factor prediction model is optimized using a hyperparameter iterative evolution training method based on historical multivariate time-series feature sets and historical carbon factors to determine the globally optimal hyperparameter combination, resulting in the trained carbon factor prediction model.

[0019] This invention addresses the problem that existing carbon factor prediction models lack effective utilization of population difference information during hyperparameter optimization and are prone to getting trapped in local optima. It introduces a population update mechanism based on the numerical differences between dominant and suboptimal individuals during iterative evolutionary training, using these differences to guide the search direction and escape local search traps. This population update mechanism helps the model converge to a better hyperparameter combination more quickly, thus overcoming the limitations of manual experience or conventional algorithm parameter configuration and effectively improving the accuracy and robustness of dynamic carbon factor prediction for high-energy-consuming enterprises. Simultaneously, based on the predicted carbon factor, adjustments are made to the power generation plan data of self-owned power plants, the power purchase plan data of external power sources, and the production operating condition plan parameters, realizing the physical feedback of the prediction results in the industrial system and providing reliable technical support for energy optimization in high-energy-consuming enterprises. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for dynamic prediction of carbon factors in high-energy-consuming enterprises, provided by an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of a dynamic prediction device for the carbon factor of high-energy-consuming enterprises provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, to address the problem in existing technologies that lack effective utilization of information on differences between superior and inferior individuals in a population, leading to the hyperparameter optimization process of the electrocarbon factor prediction model easily getting trapped in local optima and making it difficult to obtain the global optimal solution, thus resulting in insufficient prediction accuracy, an embodiment of the present invention provides a dynamic prediction method for electrocarbon factors in high-energy-consuming enterprises, comprising at least the following steps: Step S1: Obtain the target enterprise's self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding weather forecast data for the forecast period. Specifically, when implementing the dynamic prediction method for the carbon factor of high-energy-consuming enterprises provided by this invention, the first step is to acquire data, namely, to obtain the target enterprise's self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding meteorological forecast data for the period to be predicted. The target enterprise here specifically refers to a high-energy-consuming industrial enterprise that owns its own power plant and also needs to connect to an external power source for electricity purchase. Such enterprises are characterized by complex energy structures and large load fluctuations. The period to be predicted refers to the future time range within which the user wants to predict the carbon factor value.

[0024] Specifically, self-owned power plant generation plan data refers to the predetermined power generation sequence or power generation arrangement of internal generating units during the forecast period, formulated by the target enterprise based on its production plan and energy dispatch strategy. Self-owned power plants typically include coal-fired units, gas-fired units, or new energy units such as photovoltaic and wind power built within the enterprise. Different types of units have different carbon emission characteristics. External power purchase plan data refers to the electricity load data or electricity purchase data that the target enterprise plans to purchase from the external power grid during the forecast period. Self-owned power plant generation plan data and external power purchase plan data together constitute the target enterprise's future power supply structure. Since the carbon emission intensity of self-owned power plants and the carbon emission intensity of the external power grid usually differ significantly, these two types of data are the basic physical quantities for calculating the combined electricity carbon emission intensity.

[0025] Production operating condition planning parameters refer to the expected production and operational status data of a target enterprise during the forecast period. These parameters specifically include, but are not limited to, start-up and shutdown schedules for major energy-consuming equipment, projected production load rates, target output, or expected energy consumption per unit of product. These parameters directly determine the target enterprise's total electricity demand and its fluctuation patterns, thereby influencing the ratio of output from its own power plant to externally purchased electricity through the power balance relationship. For example, during peak production periods, the target enterprise may need to simultaneously operate its own power plant at full capacity and purchase a large amount of external electricity; while during off-peak periods, it may be able to meet its demand solely through its own power plant.

[0026] Surrounding meteorological forecast data refers to the weather forecast values ​​for the target company's geographical location during the forecast period. Specifically, this data includes temperature, humidity, solar radiation intensity, wind speed, and air pressure. Meteorological conditions affect the target company's total electricity consumption by influencing the load on auxiliary equipment such as air conditioning and heating. Furthermore, for target companies with self-owned photovoltaic or wind power units, meteorological conditions directly determine the upper limit of the output of these renewable energy units, thus altering the proportion of clean energy in the power structure. By collecting source data from these four dimensions, this invention comprehensively covers the key factors affecting changes in the carbon factor from the power supply side, load demand side, and environmental impact side. This provides a complete data foundation for subsequently constructing a high-fidelity multivariate time-series feature set, ensuring the comprehensiveness and accuracy of dynamic prediction of the carbon factor from the source.

[0027] Step S2: Based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operation plan parameters, and the surrounding meteorological forecast data, construct a multivariate time series feature set for the period to be predicted; In a preferred embodiment, the step of constructing a multivariate time-series feature set for the period to be predicted based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters, and the surrounding meteorological forecast data includes: Based on the preset timestamp, the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operation plan parameters and the surrounding meteorological forecast data are resampled to generate corresponding time series data with uniform time granularity. The time series data is subjected to dimensionless standardization processing to generate dimensionless time series data; Based on a preset time window, a multivariate time series feature set for the period to be predicted is constructed using dimensionless time series data.

[0028] Specifically, in implementing the dynamic prediction method for the carbon factor of high-energy-consuming enterprises provided by this invention, after obtaining the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters, and the surrounding meteorological forecast data for the target enterprise during the prediction period, the step of constructing a multivariate time-series feature set is performed. Since the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters, and the surrounding meteorological forecast data come from different data sources, such as the enterprise's internal distributed control system, power trading system, and meteorological service platform, the data from different sources often have different physical dimensions, data acquisition frequencies, and numerical orders of magnitude. Directly inputting the original heterogeneous data into the carbon factor prediction model will lead to difficulty in model training convergence or a decrease in prediction accuracy. Therefore, constructing a multivariate time-series feature set for the prediction period based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters, and the surrounding meteorological forecast data is a key data processing step in converting the original multi-source heterogeneous data into data that meets the input requirements of the carbon factor prediction model.

[0029] Specifically, the process of constructing a multivariate time-series feature set for the period to be predicted includes three sub-steps: resampling, dimensionless standardization, and feature set construction. First, resampling is performed. Data on self-owned power plant generation plans, external power purchase plans, production operating parameters, and surrounding weather forecasts may have different time intervals in their original state; for example, weather data may be updated hourly, while production operating parameters may be recorded every minute. Using a preset timestamp as a reference, resampling is performed on the self-owned power plant generation plans, external power purchase plans, production operating parameters, and surrounding weather forecasts. For original data with sampling frequencies inconsistent with the preset timestamp, a linear interpolation algorithm is used to calculate the value at the target time. The specific interpolation formula is as follows: in, This indicates the interpolated value that needs to be calculated at the target preset timestamp; Indicates the target's preset timestamp; This indicates the closest time point in the original data that is before the target preset timestamp; This indicates the time point in the original data that is closest to and after the target preset timestamp; Indicates at a point in time The corresponding original data value; Indicates at a point in time The corresponding original data values. Through resampling, the original data of different frequencies are mapped onto a unified time axis, generating corresponding time-series data with uniform time granularity, ensuring the synchronization of multivariate data in the time dimension.

[0030] Secondly, dimensionless standardization is performed on time-series data with uniform time granularity. Because time-series data contains variables with different physical meanings, such as power, temperature, and yield, the numerical ranges of these variables vary greatly. If used directly, variables with larger values ​​will dominate the model gradient update, masking the important characteristics of smaller variables. Dimensionless standardization of time-series data is typically performed using the Z-Score standardization method, and the specific standardization calculation formula is as follows: in, Indicates the first The characteristic variable at the th ... Dimensionless standardized values ​​at a given time point; Indicates the first The characteristic variable at the th ... The original values ​​at each time point; Indicates the first The arithmetic mean of the characteristic variables over the entire statistical period; Indicates the first The standard deviation of each feature variable over the entire statistical period. Through this processing, various variables in the dimensionless time series data are transformed to the same order of magnitude or follow a standard normal distribution, thereby accelerating the convergence speed of the carbon factor prediction model and improving training stability.

[0031] Finally, based on a preset time window, a multivariate time series feature set for the period to be predicted is constructed using dimensionless time series data. The change in the electrocarbon factor depends not only on the current input state but also on the evolution of the state at historical moments. Using a sliding window technique, the dimensionless time series data is truncated according to a preset time window length, generating a series of feature matrices containing continuous time steps. The specific formula for constructing the feature set is as follows: in, This indicates the completed construction for the time to be predicted. The multivariate temporal feature set input matrix; This represents the total number of feature variables; This indicates the preset time window length, which is the number of historical time steps the model backtracks. This indicates the target time at which prediction is currently needed; Indicates the first The characteristic variable at the th ... The dimensionless standardized values ​​at each time point are obtained. Through the above steps, the original discrete data can be transformed into structured, standardized, and high-quality feature data containing time-series dependencies. This effectively eliminates the inconsistencies in frequency and dimensions of multi-source data, laying a solid data foundation for the electric carbon factor prediction model to capture complex nonlinear dynamic laws.

[0032] Step S3: Input the multivariate time series feature set into the preset electric carbon factor prediction model to generate the electric carbon factor of the target enterprise for the period to be predicted. In a preferred embodiment, training the electric carbon factor prediction model includes: Obtain the multivariate time series feature set of the target enterprise for historical periods and the corresponding historical carbon dioxide factor; initialize the carbon dioxide factor prediction model and generate a hyperparameter population; wherein, the hyperparameter population contains multiple sets of hyperparameters; The electric carbon factor prediction model is iteratively trained based on a multivariate temporal feature set from historical periods and a hyperparameter population until a preset iteration termination condition is met, thus determining the globally optimal hyperparameter combination. In each iteration, the multivariate temporal feature set from historical periods is input into the electric carbon factor prediction model configured with hyperparameters from each group in the hyperparameter population to obtain training prediction results. The fitness value is determined based on the error between the training prediction results and the corresponding historical electric carbon factors, and dominant and suboptimal individuals are selected based on the fitness value. A new generation of hyperparameter population is generated based on the numerical differences between dominant and suboptimal individuals to update the hyperparameter population. Configure the electric carbon factor prediction model based on the globally optimal hyperparameter combination, and generate the trained electric carbon factor prediction model.

[0033] In a preferred embodiment, the historical carbon factor of the target enterprise for a historical period is obtained by the following method: Acquire the target enterprise's historical power generation data from its self-owned power plant, carbon emission factors of its self-owned power plant, historical power purchase data from external sources, and carbon emission factors of external sources within a historical period. Calculate the product of the historical power generation data of the self-owned power plant and the carbon emission factor of the self-owned power plant to generate the carbon emissions of the self-owned power plant; The carbon emissions of the external power source are generated by multiplying the historical electricity purchase data of the external power source with the carbon emission factor of the external power source. Calculate the sum of the carbon emissions from the self-owned power plant and the carbon emissions from the external power source to generate a first calculation result; Calculate the sum of the historical power generation data of the self-owned power plant and the historical power purchase data of the external power source to generate a second calculation result; Calculate the ratio of the first calculation result to the second calculation result to generate the historical carbon factor for the historical period.

[0034] In a preferred embodiment, the step of iteratively training the electric carbon factor prediction model based on a multivariate time-series feature set and hyperparameter population from historical periods until a preset iteration termination condition is met, and determining the globally optimal hyperparameter combination, includes: Each hyperparameter individual in the hyperparameter population generated after initializing the electric carbon factor prediction model is configured to use the electric carbon factor prediction model. The multivariate time series feature set of historical periods is input into the configured electric carbon factor prediction model to obtain the initial training prediction result. The fitness value of each hyperparameter individual is determined based on the error between the initial training prediction result and the historical electric carbon factor. Repeatedly perform parameter evolution iteration operations until the preset iteration termination condition is met, and determine the individual with the best fitness value in the current hyperparameter population as the global optimal hyperparameter combination; The parameter evolution iteration operation includes: For each hyperparameter individual in the current hyperparameter population, a preset number of sample individuals are randomly selected from the current hyperparameter population. The individual with the best fitness value among the sample individuals is determined as the dominant individual, and the individual with the worst fitness value is determined as the suboptimal individual. The initial hyperparameter population is the hyperparameter population generated after initializing the electric carbon factor prediction model. The decay factor is determined based on the current number of iterations and a random step size vector is generated. At the same time, the base evolution vector is determined from the current hyperparameter population according to the preset probability strategy. Candidate hyperparameter individuals are generated using the base evolution vector, the decay factor, the random step size vector, the current hyperparameter individual, the dominant individual, and the disadvantaged individual. The candidate hyperparameter individuals are used to configure the electric carbon factor prediction model, and the multivariate time series feature set of historical periods is input into the configured electric carbon factor prediction model to obtain candidate training prediction results. The candidate fitness value of the candidate hyperparameter individual is determined based on the error between the candidate training prediction result and the corresponding historical carbon factor. If the difference between the candidate fitness value and the best fitness value in the current hyperparameter population satisfies the preset dynamic selection threshold condition, then the candidate hyperparameter individual is determined as the corresponding next-generation hyperparameter individual; otherwise, the current hyperparameter individual is determined as the corresponding next-generation hyperparameter individual. Once the next generation of hyperparameter individuals for all hyperparameter individuals in the current hyperparameter population has been determined, a next generation hyperparameter population is formed based on the determined next generation hyperparameter individuals, and the current hyperparameter population is updated using the next generation hyperparameter population.

[0035] In a preferred embodiment, determining the fitness value of each hyperparameter individual based on the error between the initial training prediction result and the historical electric carbon factor includes: For each hyperparameter individual, calculate the mean absolute error, root mean square error, and mean absolute percentage error between the initial training prediction result and the historical carbon factor corresponding to the current hyperparameter individual. Calculate the sum of the mean absolute error, the root mean square error, and the mean absolute percentage error, and determine the sum as the fitness value of the current hyperparameter individual.

[0036] In a preferred embodiment, the step of determining the decay factor and generating a random step size vector based on the current iteration number, and simultaneously determining the basis evolution vector from the current hyperparameter population according to a preset probability strategy, includes: The iteration progress value is calculated based on the ratio of the current number of iterations to the preset maximum number of iterations. Based on the preset decay function, the decay factor is calculated according to the iteration progress value; A random vector containing random perturbation information is generated based on a preset random distribution function, and the random vector is determined as a random step size vector; wherein, the dimension of the random vector is consistent with the dimension of the current hyperparameter individual; A strategy selection random number between 0 and 1 is generated according to a preset uniform distribution function, and it is determined whether the strategy selection random number falls into a preset advantage guidance probability interval. If the random number selected by the strategy falls within the advantage-guided probability interval, then the individual with the best fitness value in the current hyperparameter population is determined as the base evolution vector. If the random number selected by the strategy does not fall within the advantage guidance probability range, then a hyperparameter individual is randomly selected from the current hyperparameter population to be determined as the base evolution vector.

[0037] In a preferred embodiment, generating candidate hyperparameter individuals using the base evolution vector, the decay factor, the random step size vector, the current hyperparameter individual, the dominant individual, and the suboptimal individual includes: Calculate the numerical difference between the dominant and the disadvantaged individuals to generate an evolutionary difference vector; The evolutionary difference vector is weighted according to the decay factor to obtain the weighted evolutionary vector; Calculate the vector sum of the base evolution vector, the weighted evolution vector, and the random step size vector, and determine the calculated vector as the candidate hyperparameter individual.

[0038] Specifically, after constructing the multivariate time-series feature set for the period to be predicted, a prediction step is executed, which involves inputting the multivariate time-series feature set into a pre-defined carbon factor prediction model to generate the carbon factor for the target company during the predicted period. To extract deep nonlinear features and capture long-term dependencies from the multivariate time-series feature set, the pre-defined carbon factor prediction model preferably adopts a Bidirectional Long Short-Term Memory (BiLSTM-Attention) network architecture based on an attention mechanism. Specifically, the carbon factor prediction model includes an input layer, a BiLSTM network layer, an attention mechanism layer, and a fully connected output layer. The input layer receives a standardized set of multivariate temporal features. The bidirectional long short-term memory (LSTM) network layer contains forward and backward LSTM units, extracting temporal features from both the forward and reverse time series directions, generating a bidirectional temporal feature vector containing past and future contextual information to avoid information loss. Specifically, for each LSTM unit, the flow of information is controlled through forget gates, input gates, and output gates. The sigmoid activation function controls the opening and closing of the gates, and the tanh activation function updates the cell state, thereby achieving the memorization and forgetting of long-distance historical information. The attention mechanism layer performs weighted calculations on the bidirectional temporal feature vectors, automatically assigning corresponding weight coefficients based on the different degrees of influence of different time steps on the electrocarbon factor, strengthening the feature response of key time steps, suppressing irrelevant noise interference, and generating a weighted feature vector. The fully connected output layer uses activation functions to perform linear transformations and nonlinear mappings on the weighted feature vectors, ultimately outputting the predicted electrocarbon factor value of the target company in the predicted time period.

[0039] However, the predictive performance of the carbon emission factor prediction model is highly dependent on the configuration of hyperparameters, which specifically include the number of neurons in the bidirectional long short-term memory network layer, the number of network layers, the learning rate, and the training batch size. To address the difficulty of manual parameter tuning, this invention proposes a model training method based on an improved evolutionary strategy. Before training begins, training data needs to be prepared, namely, obtaining the multivariate time-series feature set of the target enterprise for historical periods and the corresponding historical carbon emission factor. The historical carbon emission factor is not directly measured but needs to be calculated based on the target enterprise's energy structure. Specifically, historical data on the target enterprise's self-owned power plant generation, carbon emission factor of the self-owned power plant, historical data on external power purchases, and carbon emission factor of the external power source are obtained. The carbon emissions of the self-owned power plant are calculated by multiplying the historical data on self-owned power plant generation and the carbon emission factor of the self-owned power plant, and the carbon emissions of the external power source are calculated by multiplying the historical data on external power purchases and the carbon emission factor of the external power source. Subsequently, the sum of carbon emissions from the self-owned power plant and carbon emissions from external power sources is calculated as the first result (i.e., total carbon emissions). The sum of historical power generation data from the self-owned power plant and historical electricity purchase data from external power sources is calculated as the second result (i.e., total electricity consumption). Finally, the ratio of the first result to the second result is calculated to generate the historical carbon factor for the historical period. This calculation process ensures the accuracy of the physical meaning of the labeled data and can truly reflect the "carbon dioxide emissions per unit of electricity".

[0040] After preparing the training data, the electric carbon factor prediction model is initialized, and a hyperparameter population is generated. The hyperparameter population contains multiple sets of hyperparameters, each set corresponding to a single hyperparameter individual, representing a potential model configuration. Subsequently, the electric carbon factor prediction model is iteratively trained based on the multivariate temporal feature set from historical periods and the hyperparameter population until a preset iteration termination condition is met (e.g., reaching the maximum number of iterations or error convergence), determining the globally optimal hyperparameter combination. In each iteration, the electric carbon factor prediction model is configured using each hyperparameter individual in the hyperparameter population, and the multivariate temporal feature set from historical periods is input into the configured electric carbon factor prediction model to obtain initial training prediction results. To comprehensively evaluate the quality of hyperparameter individuals, the fitness value of each hyperparameter individual is determined based on the error between the initial training prediction results and historical electric carbon factors. Specifically, for each hyperparameter individual, the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) between the initial training prediction results and historical electric carbon factors for the current hyperparameter individual are calculated. Calculate the sum of the mean absolute error, root mean square error, and mean absolute percentage error, and determine the sum as the fitness value of the current hyperparameter individual. The smaller the fitness value, the smaller the model prediction error under that hyperparameter individual configuration, and the higher the individual quality.

[0041] After determining the fitness value of each hyperparameter individual, a parametric evolution iteration operation is performed to update the population. First, for each hyperparameter individual in the current hyperparameter population (as the target individual), a predetermined number of sample individuals (e.g., 3 randomly selected) are randomly drawn from the current hyperparameter population. The individual with the best fitness value among these sample individuals is identified as the dominant individual, and the individual with the worst fitness value is identified as the suboptimal individual. Using the difference between dominant and suboptimal individuals to guide the evolutionary direction helps the population quickly approach the optimal solution region. Next, the numerical difference between dominant and suboptimal individuals is calculated, generating an evolutionary difference vector.

[0042] Simultaneously, to balance the algorithm's global exploration capability and local exploitation capability, control parameters need to be determined during the evolutionary process. The decay factor is determined based on the current iteration count. Specifically, the iteration progress value is calculated based on the ratio of the current iteration count to the preset maximum iteration count, and the decay factor is calculated based on a preset decay function. The formula for calculating the decay factor is as follows: in, Indicates the attenuation factor; Indicates the current iteration number; This represents the preset maximum number of iterations. The decay function causes the decay factor to decrease non-linearly with increasing iteration count, maintaining a large value in the early stages of iteration to expand the search range, and decreasing it in the later stages of iteration for finer searching.

[0043] Furthermore, to prevent the algorithm from getting trapped in local optima, random perturbations need to be introduced. A random vector containing random perturbation information is generated based on a preset random distribution function, and this random vector is then used as a random step size vector. The dimension of the random vector is consistent with the dimension of the current hyperparameter individual. Preferably, a Levy Flight strategy is used to generate the random step size vector, leveraging the long-tail distribution characteristics to increase the probability of escaping local optima.

[0044] Specifically, Levy's flight stride It can be generated by the following formula: in, and Let be a random variable that follows a normal distribution, i.e. , . The value is typically set to 1.5. The generated step size is then used. Construct the random step size vector by combining the step size factor .

[0045] Simultaneously, a base evolution vector is determined from the current hyperparameter population according to a preset probabilistic strategy. Specifically, a strategy selection random number between 0 and 1 is generated based on a preset uniform distribution function, and it is determined whether the strategy selection random number falls within a preset advantage guidance probability interval (e.g., [0, 0.5)). If the strategy selection random number falls within the advantage guidance probability interval, the individual with the best fitness value in the current hyperparameter population (the globally optimal individual) is determined as the base evolution vector to guide evolution using elite information; if the strategy selection random number does not fall within the advantage guidance probability interval, a hyperparameter individual is randomly selected from the current hyperparameter population to be determined as the base evolution vector to maintain population diversity.

[0046] Based on the vectors determined above, candidate hyperparameter individuals are generated. The calculations are performed using the base evolution vector, decay factor, random step size vector, dominant individuals, and suboptimal individuals. The specific formula for generating candidate hyperparameter individuals is as follows: in, This represents the generated candidate hyperparameter individuals; Represents a definite basis evolution vector; Indicates the attenuation factor; Indicates a clearly dominant individual; Indicates a clearly disadvantaged individual; This represents a vector with a fixed random step size.

[0047] After generating candidate hyperparameter individuals, the candidate hyperparameter individuals are used to configure the electric carbon factor prediction model. The multivariate time-series feature set from historical periods is input into the configured electric carbon factor prediction model to obtain candidate training prediction results, and the corresponding candidate fitness values ​​are calculated. If the difference between the candidate fitness value and the best fitness value in the current hyperparameter population satisfies a preset dynamic selection threshold condition (i.e., the candidate individual's performance is better than the current individual or it meets a specific acceptance criterion), then the candidate hyperparameter individual is determined as the corresponding next-generation hyperparameter individual; otherwise, the current hyperparameter individual is retained as the corresponding next-generation hyperparameter individual. In one embodiment, the dynamic selection threshold condition adopts a greedy selection strategy, that is: if the candidate fitness value is less than the fitness value of the current individual (smaller error), then the condition is satisfied. After traversing all hyperparameter individuals in the current hyperparameter population, the determined next-generation hyperparameter individuals are used to form the next-generation hyperparameter population, and the next-generation hyperparameter population replaces the current hyperparameter population, entering the next iteration. Repeat the above process until the iteration termination condition is met. The individual with the best fitness value in the final hyperparameter population is determined as the global optimal hyperparameter combination. This combination is then used to configure the electric carbon factor prediction model, resulting in the trained electric carbon factor prediction model.

[0048] Through the above methods, this invention can accurately predict the future carbon emission factors of high-energy-consuming enterprises by combining actual planned data from self-owned power plants and external power sources with production and meteorological factors, and utilizing a bidirectional long short-term memory network model deeply optimized by an evolutionary strategy. In particular, during model training, by introducing the difference information between dominant and disadvantaged individuals to guide population evolution, and combining an adaptive decay factor and a stochastic step size strategy, it effectively solves the problems of traditional hyperparameter optimization methods easily getting trapped in local optima and having low search efficiency. This significantly improves the generalization ability and prediction accuracy of the prediction model, providing reliable data support for enterprises to manage carbon quotas and optimize energy use.

[0049] Step S4: Based on the carbon factor, adjust at least one of the following: the self-owned power plant power generation plan data, the external power purchase plan data, and the production operating condition plan parameters, in order to optimize the energy consumption of the target enterprise. Specifically, after obtaining the carbon factor of the target enterprise for the forecast period, the predicted one-dimensional time series values ​​need to be converted into physical execution commands to control the operating status of the industrial site's electrical equipment and generator sets. The specific parameter adjustment process is divided into three dimensions: power supply ratio adjustment, production load spatiotemporal shift, and multi-energy coupling and collaborative scheduling, thus covering different dimensions of actual energy consumption scenarios.

[0050] Within the dimension of power supply ratio adjustment, a cost optimization model is established with the goal of minimizing total carbon emissions. Based on the electricity carbon factor and the target company's preset carbon emission quota values, the expected carbon quota gap for the forecast period is calculated. A greater than zero expected carbon quota gap indicates that the target company faces the risk of penalties for exceeding emission limits. A carbon emission cost accounting formula is constructed for both self-owned power plants and external power sources: In the formula, This represents the target company's expected total carbon emissions during the forecast period. This indicates the active power supplied by the self-owned generating units corresponding to the adjusted power generation plan data of the self-owned power plant; This represents the inherent power generation carbon emission intensity coefficient of the target company's internal self-owned generating units; This indicates the value of the active power supplied by the external power source corresponding to the adjusted external power purchase plan data; This represents the carbon factor corresponding to the period to be predicted.

[0051] With the goal of minimizing the expected total carbon emissions, control commands are dynamically issued while meeting total load demand. When the carbon factor is greater than the inherent carbon emission intensity coefficient, the control command increases the active power supplied by the self-owned generating units and proportionally reduces the active power supplied by external sources, i.e., increasing the load factor of internal generating units to replace external high-carbon electricity. When the carbon factor is less than the inherent carbon emission intensity coefficient, the control command reduces the fuel feed of the self-owned generating units to reduce the active power supplied by them, and simultaneously increases the active power supplied by external sources from the grid via tie lines.

[0052] Within the spatiotemporal shift dimension of production load, flexible physical adjustments are made to the planned parameters of production conditions. Peak and trough points of the electrocarbon factor are extracted at different time intervals. After extraction, a time-stamped start-up and shutdown control sequence for production equipment is generated. For flexible high-energy-consuming equipment with energy storage characteristics or production tasks that allow for delays, the start-up time of this equipment is forcibly postponed, shifting high-energy-consuming production conditions to the trough time interval when the electrocarbon factor is low. For rigid production equipment requiring continuous operation, physical commands for condition reduction are issued, lowering the motor speed and target output value during the peak time interval of the electrocarbon factor. By performing overload production operations during the trough time interval, the output loss during the peak time interval is compensated for, ensuring that the overall production target for the entire cycle is met.

[0053] In the multi-energy coupled coordinated scheduling dimension, to prevent the parameter adjustment process from breaking the safe and stable operation boundary of the physical power grid, equipment safety constraints and power balance constraints are introduced. The extreme values ​​of the physical ramp rate of generator units in the self-owned power plant are extracted, and the unit output constraint formula is constructed: In the formula, This represents the difference in adjustment of the active power supplied by the self-contained generating unit within two adjacent forecast time steps; This indicates the maximum permissible ramp rate indicated on the nameplate of the physical generator set; This represents the time interval between two adjacent prediction time steps.

[0054] Simultaneously, establish the physical constraint formula for power balance: In the formula, This represents the total active power actually consumed by the target enterprise's production terminals under the adjusted production operating condition planning parameters; This indicates the power loss in the power distribution network within the factory area.

[0055] During parameter adjustment, all calculated control commands are sent to the target company's programmable logic controller and on-site measurement and control terminal to complete the valve opening adjustment and switching of the physical execution components.

[0056] By combining multi-dimensional physical closed-loop adjustment operations on the power supply side and the production load side, direct linkage and interaction between electricity carbon emission prediction data and physical industrial equipment is realized, effectively reducing the actual carbon compliance costs of target enterprises in the production and manufacturing process, and improving the low-carbon emission reduction benefits of high-energy-consuming enterprises throughout the entire operation cycle.

[0057] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0058] like Figure 2 As shown, an embodiment of the present invention provides a dynamic prediction device for the carbon factor of high-energy-consuming enterprises, including: a data acquisition module, a feature construction module, a prediction generation module, an energy consumption optimization module, and a model training module; The data acquisition module is used to acquire the target enterprise’s self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding weather forecast data for the forecast period. The feature construction module is used to construct a multivariate time series feature set for the period to be predicted based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters and the surrounding meteorological forecast data. The prediction generation module is used to input a multivariate time series feature set into a preset electric carbon factor prediction model to generate the electric carbon factor of the target enterprise for the period to be predicted. The energy optimization module is used to adjust at least one of the following based on the carbon factor: the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, and the production operating condition plan parameters, so as to optimize the energy consumption of the target enterprise. The model training module is used to acquire the multivariate time-series feature set of the target enterprise for historical periods and the corresponding historical carbon dioxide factor; initialize the carbon dioxide factor prediction model and generate a hyperparameter population; wherein the hyperparameter population contains multiple sets of hyperparameters; iteratively evolve and train the carbon dioxide factor prediction model according to the multivariate time-series feature set of historical periods and the hyperparameter population until a preset iteration termination condition is met, and determine the globally optimal hyperparameter combination; wherein, in each iteration, the multivariate time-series feature set of historical periods is input into the carbon dioxide factor prediction model configured by each set of hyperparameters in the hyperparameter population to obtain training prediction results; determine the fitness value according to the error between the training prediction result and the corresponding historical carbon dioxide factor, and select dominant and suboptimal individuals based on the fitness value; generate a new generation of hyperparameter population to update the hyperparameter population according to the numerical difference information between dominant and suboptimal individuals; configure the carbon dioxide factor prediction model according to the globally optimal hyperparameter combination, and generate the trained carbon dioxide factor prediction model.

[0059] It should be noted that the embodiments of the device described above correspond to the embodiments of the present invention described above, and can realize the dynamic prediction method for the carbon factor of high-energy-consuming enterprises as described in any one of the above embodiments of the present invention. Furthermore, the embodiments of the device described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort.

[0060] Based on the above-described method embodiments of the present invention, a corresponding embodiment of an electronic device is provided.

[0061] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the dynamic prediction method for the carbon factor of high-energy-consuming enterprises according to any one of the present invention, or, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0062] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0063] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0064] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0065] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0066] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments; Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described methods for dynamic prediction of carbon factors in high-energy-consuming enterprises.

[0067] The aforementioned storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0068] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0069] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for dynamic prediction of carbon factor in high-energy-consuming enterprises, characterized in that, include: Obtain the target company's self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding weather forecast data for the forecast period; Based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production condition plan parameters and the surrounding meteorological forecast data, a multivariate time series feature set for the period to be predicted is constructed. Input the multivariate time series feature set into the preset electric carbon factor prediction model to generate the electric carbon factor of the target enterprise for the period to be predicted; Based on the carbon factor, at least one of the following is adjusted: the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, and the production operating condition plan parameters, in order to optimize the energy consumption of the target enterprise. The training of the electric carbon factor prediction model includes: Obtain the multivariate time series feature set of the target enterprise for historical periods and the corresponding historical carbon dioxide factor; initialize the carbon dioxide factor prediction model and generate a hyperparameter population; wherein, the hyperparameter population contains multiple sets of hyperparameters; The electric carbon factor prediction model is iteratively trained based on a multivariate temporal feature set from historical periods and a hyperparameter population until a preset iteration termination condition is met, thus determining the globally optimal hyperparameter combination. In each iteration, the multivariate temporal feature set from historical periods is input into the electric carbon factor prediction model configured with hyperparameters from each group in the hyperparameter population to obtain training prediction results. The fitness value is determined based on the error between the training prediction results and the corresponding historical electric carbon factors, and dominant and suboptimal individuals are selected based on the fitness value. A new generation of hyperparameter population is generated based on the numerical differences between dominant and suboptimal individuals to update the hyperparameter population. Configure the electric carbon factor prediction model based on the globally optimal hyperparameter combination, and generate the trained electric carbon factor prediction model.

2. The method for dynamic prediction of carbon factors in high-energy-consuming enterprises as described in claim 1, characterized in that, The process involves constructing a multivariate time-series feature set for the period to be predicted based on self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding meteorological forecast data. This set includes: Based on the preset timestamp, the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operation plan parameters and the surrounding meteorological forecast data are resampled to generate corresponding time series data with uniform time granularity. The time series data is subjected to dimensionless standardization processing to generate dimensionless time series data; Based on a preset time window, a multivariate time series feature set for the period to be predicted is constructed using dimensionless time series data.

3. The method for dynamic prediction of carbon factors in high-energy-consuming enterprises as described in claim 2, characterized in that, The step of iteratively training the electric carbon factor prediction model based on a multivariate time-series feature set and hyperparameter population from historical periods until a preset iteration termination condition is met, and determining the globally optimal hyperparameter combination, includes: Each hyperparameter individual in the hyperparameter population generated after initializing the electric carbon factor prediction model is configured to use the electric carbon factor prediction model. The multivariate time series feature set of historical periods is input into the configured electric carbon factor prediction model to obtain the initial training prediction result. The fitness value of each hyperparameter individual is determined based on the error between the initial training prediction result and the historical electric carbon factor. Repeatedly perform parameter evolution iteration operations until the preset iteration termination condition is met, and determine the individual with the best fitness value in the current hyperparameter population as the global optimal hyperparameter combination; The parameter evolution iteration operation includes: For each hyperparameter individual in the current hyperparameter population, a preset number of sample individuals are randomly selected from the current hyperparameter population. The individual with the best fitness value among the sample individuals is determined as the dominant individual, and the individual with the worst fitness value is determined as the suboptimal individual. The initial hyperparameter population is the hyperparameter population generated after initializing the electric carbon factor prediction model. The decay factor is determined based on the current number of iterations and a random step size vector is generated. At the same time, the base evolution vector is determined from the current hyperparameter population according to the preset probability strategy. Candidate hyperparameter individuals are generated using the base evolution vector, the decay factor, the random step size vector, the current hyperparameter individual, the dominant individual, and the disadvantaged individual; The candidate hyperparameter individuals are used to configure the electric carbon factor prediction model, and the multivariate time series feature set of historical periods is input into the configured electric carbon factor prediction model to obtain candidate training prediction results. The candidate fitness value of the candidate hyperparameter individual is determined based on the error between the candidate training prediction result and the corresponding historical carbon factor. If the difference between the candidate fitness value and the best fitness value in the current hyperparameter population satisfies the preset dynamic selection threshold condition, then the candidate hyperparameter individual is determined as the corresponding next-generation hyperparameter individual; otherwise, the current hyperparameter individual is determined as the corresponding next-generation hyperparameter individual. Once the next generation of hyperparameter individuals for all hyperparameter individuals in the current hyperparameter population has been determined, a next generation hyperparameter population is formed based on the determined next generation hyperparameter individuals, and the current hyperparameter population is updated using the next generation hyperparameter population.

4. The method for dynamic prediction of carbon factors in high-energy-consuming enterprises as described in claim 3, characterized in that, The step of determining the fitness value of each hyperparameter individual based on the error between the initial training prediction result and the historical electric carbon factor includes: For each hyperparameter individual, calculate the mean absolute error, root mean square error, and mean absolute percentage error between the initial training prediction result and the historical carbon factor corresponding to the current hyperparameter individual. Calculate the sum of the mean absolute error, the root mean square error, and the mean absolute percentage error, and determine the sum as the fitness value of the current hyperparameter individual.

5. The method for dynamic prediction of carbon factors in high-energy-consuming enterprises as described in claim 4, characterized in that, The step of determining the decay factor and generating a random step size vector based on the current iteration number, and simultaneously determining the basis evolution vector from the current hyperparameter population according to a preset probability strategy, includes: The iteration progress value is calculated based on the ratio of the current number of iterations to the preset maximum number of iterations. Based on the preset decay function, the decay factor is calculated according to the iteration progress value; A random vector containing random perturbation information is generated based on a preset random distribution function, and the random vector is determined as a random step size vector; wherein, the dimension of the random vector is consistent with the dimension of the current hyperparameter individual; A strategy selection random number between 0 and 1 is generated according to a preset uniform distribution function, and it is determined whether the strategy selection random number falls into a preset advantage guidance probability interval. If the random number selected by the strategy falls within the advantage-guided probability interval, then the individual with the best fitness value in the current hyperparameter population is determined as the base evolution vector. If the random number selected by the strategy does not fall within the advantage guidance probability range, then a hyperparameter individual is randomly selected from the current hyperparameter population to be determined as the base evolution vector.

6. The method for dynamic prediction of carbon factors in high-energy-consuming enterprises as described in claim 5, characterized in that, The step of generating candidate hyperparameter individuals using the base evolution vector, the decay factor, the random step size vector, the current hyperparameter individual, the dominant individual, and the suboptimal individual includes: Calculate the numerical difference between the dominant and the disadvantaged individuals to generate an evolutionary difference vector; The evolutionary difference vector is weighted according to the decay factor to obtain the weighted evolutionary vector; Calculate the vector sum of the base evolution vector, the weighted evolution vector, and the random step size vector, and determine the calculated vector as the candidate hyperparameter individual.

7. The method for dynamic prediction of carbon factors in high-energy-consuming enterprises as described in claim 1, characterized in that, Historical carbon factor of the target company for a historical period can be obtained through the following methods: Acquire the target enterprise's historical power generation data from its self-owned power plant, carbon emission factors of its self-owned power plant, historical power purchase data from external sources, and carbon emission factors of external sources within a historical period. Calculate the product of the historical power generation data of the self-owned power plant and the carbon emission factor of the self-owned power plant to generate the carbon emissions of the self-owned power plant; The carbon emissions of the external power source are generated by multiplying the historical electricity purchase data of the external power source with the carbon emission factor of the external power source. Calculate the sum of the carbon emissions from the self-owned power plant and the carbon emissions from the external power source to generate a first calculation result; Calculate the sum of the historical power generation data of the self-owned power plant and the historical power purchase data of the external power source to generate a second calculation result; Calculate the ratio of the first calculation result to the second calculation result to generate the historical carbon factor for the historical period.

8. A dynamic prediction device for the carbon factor of high-energy-consuming enterprises, characterized in that, include: The module includes a data acquisition module, a feature construction module, a prediction generation module, an energy optimization module, and a model training module. The data acquisition module is used to acquire the target enterprise’s self-owned power plant power generation plan data, external power purchase plan data, production operating condition plan parameters, and surrounding weather forecast data for the forecast period. The feature construction module is used to construct a multivariate time series feature set for the period to be predicted based on the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, the production operating condition plan parameters and the surrounding meteorological forecast data. The prediction generation module is used to input a multivariate time series feature set into a preset electric carbon factor prediction model to generate the electric carbon factor of the target enterprise for the period to be predicted. The energy optimization module is used to adjust at least one of the following based on the carbon factor: the power generation plan data of the self-owned power plant, the power purchase plan data of the external power source, and the production operating condition plan parameters, so as to optimize the energy consumption of the target enterprise. The model training module is used to acquire the multivariate time-series feature set of the target enterprise for historical periods and the corresponding historical carbon dioxide factor; initialize the carbon dioxide factor prediction model and generate a hyperparameter population; wherein the hyperparameter population contains multiple sets of hyperparameters; iteratively evolve and train the carbon dioxide factor prediction model according to the multivariate time-series feature set of historical periods and the hyperparameter population until a preset iteration termination condition is met, and determine the globally optimal hyperparameter combination; wherein, in each iteration, the multivariate time-series feature set of historical periods is input into the carbon dioxide factor prediction model configured by each set of hyperparameters in the hyperparameter population to obtain training prediction results; determine the fitness value according to the error between the training prediction result and the corresponding historical carbon dioxide factor, and select dominant and suboptimal individuals based on the fitness value; generate a new generation of hyperparameter population to update the hyperparameter population according to the numerical difference information between dominant and suboptimal individuals; configure the carbon dioxide factor prediction model according to the globally optimal hyperparameter combination, and generate the trained carbon dioxide factor prediction model.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the dynamic prediction method for the carbon factor of high-energy-consuming enterprises as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform the dynamic prediction method for the carbon factor of high-energy-consuming enterprises as described in any one of claims 1 to 7.

Images