Method, device and equipment for reconstructing minute-level carbon emission factor based on knowledge distillation

Abstract

The embodiment of the present disclosure relates to a kind of minute-level carbon emission factor reconstruction method, device and equipment based on knowledge distillation, method includes: obtaining multiple first training samples and multiple second training samples, one first training sample includes the load of thermal power unit, coal consumption, fixed carbon and corresponding carbon emission factor true value in preset duration, one second training sample is obtained by removing fixed carbon from one first training sample;Using multiple first training samples to train teacher model to obtain target teacher model, and obtaining the first carbon emission factor mapping value that teacher model outputs for each first training sample in training process;Based on multiple second training samples and multiple first carbon emission factor mapping value, student model is trained to obtain carbon emission factor mapping model, so that, the load of thermal power unit and coal consumption in preset duration are input into the carbon emission factor mapping model, i.e. the carbon emission factor data of preset duration can be acquired with period.

Description

Technical Field

[0001] This disclosure relates to the fields of power systems, transfer learning, and carbon emission technology, and in particular to a method, apparatus, and equipment for reconstructing minute-level carbon emission factors based on knowledge distillation. Background Technology

[0002] Thermal power units are the main contributors to carbon emissions in the power industry, and their transformation towards cleaner and lower-carbon technologies is urgently needed. Carbon emission factors are a core indicator for quantifying the carbon emission levels of thermal power units and assessing emission reduction effectiveness. These factors can be obtained through flue gas analysis, fuel analysis, and IPCC guidelines. Flue gas analysis directly measures the composition and flow rate of flue gas at the chimney outlet to calculate the carbon emission factor in real time. This method offers strong real-time performance but is costly and not widely used. Fuel analysis calculates the carbon emission factor based on coal quality, which is relatively low-cost and widely used, but its accuracy, real-time performance, and time resolution are highly dependent on coal quality data. IPCC guidelines provide fixed carbon emission factors, which are the lowest-cost, but their accuracy is low and their time resolution is only one year, failing to dynamically reflect the actual carbon emission factors of the units. Therefore, while the latter two methods are low-cost and widely used, they suffer from low time resolution of the carbon emission factor, making them difficult to match the actual operational needs of the power system.

[0003] Currently, power system dispatching schedules unit output in 15-minute increments. To achieve precise emission reduction, the temporal resolution of carbon emission factors should be increased to the 15-minute level. However, methods for obtaining minute-level carbon emission factor data are currently lacking. Some scholars have addressed the lack of minute-level data by establishing regression equations between load, coal quality, and carbon emission factors; however, such methods rely on high-temporal-resolution coal quality data as input and are unsuitable for scenarios where coal quality data is missing. Summary of the Invention

[0004] To solve the above-mentioned technical problems, or at least partially solve them, this disclosure provides a method, apparatus, and equipment for reconstructing minute-level carbon emission factors based on knowledge distillation.

[0005] In a first aspect, embodiments of this disclosure provide a minute-level carbon emission factor reconstruction method based on knowledge distillation, including: Multiple first training samples and multiple second training samples are obtained. Each first training sample includes the load, coal consumption, fixed carbon and corresponding carbon emission factor true values ​​of thermal power units within a preset time period. Each second training sample is obtained by removing the fixed carbon from a first training sample. The preset time period is in the minute range. The teacher model is trained using the multiple first training samples to obtain a trained target teacher model, and the first carbon emission factor mapping value output by the teacher model for each first training sample during the training process is obtained. The student model is trained based on the multiple second training samples and multiple first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model. The carbon emission factor mapping model is used to obtain the carbon emission factor of thermal power units at a preset time period.

[0006] Secondly, embodiments of this disclosure provide a method for obtaining carbon emission factors, the method comprising: Obtain the load and coal consumption of thermal power units within a preset time period; The load and coal consumption of the thermal power unit within the preset time period are input into a pre-trained carbon emission factor mapping model, and the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption of the thermal power unit within the preset time period is obtained. The carbon emission factor mapping model is trained by the minute-level carbon emission factor reconstruction method based on knowledge distillation as described in the first aspect. The carbon emission factor mapping value is determined as the carbon emission factor of the thermal power unit within the preset time period.

[0007] Thirdly, embodiments of this disclosure provide a minute-level carbon emission factor reconstruction device based on knowledge distillation, the device comprising: The sample acquisition module is used to acquire multiple first training samples and multiple second training samples. Each first training sample includes the load, coal consumption, fixed carbon and corresponding carbon emission factor true values ​​of thermal power units within a preset time period. Each second training sample is obtained by removing the fixed carbon from a first training sample. The preset time period is in the minute range. The first training module is used to train the teacher model using the multiple first training samples to obtain the trained target teacher model, and to obtain the first carbon emission factor mapping value output by the teacher model for each first training sample during the training process. The second training module is used to train the student model based on the multiple second training samples and multiple first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model. The carbon emission factor mapping model is used to obtain the carbon emission factor of the thermal power unit at a preset time period.

[0008] Fourthly, embodiments of this disclosure provide a carbon emission factor acquisition device, the device comprising: The first acquisition module is used to acquire the load and coal consumption of thermal power units within a preset time period; The second acquisition module is used to input the load and coal consumption of the thermal power unit within the preset time period into a pre-trained carbon emission factor mapping model, and to acquire the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption of the thermal power unit within the preset time period, wherein the carbon emission factor mapping model is trained by the minute-level carbon emission factor reconstruction method based on knowledge distillation as described in the first aspect. The determination module is used to determine the carbon emission factor mapping value as the carbon emission factor of the thermal power unit within the preset time period.

[0009] Fifthly, embodiments of this disclosure provide an electronic device, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the executable instructions to implement the knowledge distillation-based minute-level carbon emission factor reconstruction method as described in the first aspect, or to implement the carbon emission factor acquisition method as described in the second aspect.

[0010] In a sixth aspect, embodiments of this disclosure provide a computer-readable storage medium storing a computer program for implementing the minute-level carbon emission factor reconstruction method based on knowledge distillation as described in the first aspect, or for implementing the carbon emission factor acquisition method as described in the second aspect.

[0011] The technical solution provided in this disclosure has the following advantages compared with the prior art: The carbon emission factor acquisition scheme provided in this disclosure involves acquiring multiple first training samples and multiple second training samples. Each first training sample includes the load, coal consumption, fixed carbon, and corresponding true values ​​of the carbon emission factor for a thermal power unit within a preset time period. Each second training sample is obtained by removing fixed carbon from a first training sample. The preset time period is on the order of minutes. The teacher model is trained using these multiple first training samples to obtain a trained target teacher model. The first carbon emission factor mapping value output by the teacher model for each first training sample during the training process is also acquired. Based on the multiple second training samples and the multiple first carbon emission factor mapping values, the carbon emission factor is further processed... The student model is trained to obtain a trained carbon emission factor mapping model. Then, by inputting the load and coal consumption of the thermal power unit within a preset time period into this carbon emission factor mapping model, the carbon emission factor of the thermal power unit within the preset time period can be obtained. This realizes the acquisition of carbon emission factor data with a preset time period as the period, and achieves the acquisition of carbon emission factors with a preset time period as the time resolution at the minute level. Moreover, only the load and coal consumption of the thermal power unit are needed to obtain high time resolution carbon emission factors, without relying on high time resolution coal quality data as input. This solves the problem that minute-level carbon emission factor data cannot be reconstructed when coal quality data is missing. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0013] Figure 1 A flowchart illustrating a minute-level carbon emission factor reconstruction method based on knowledge distillation, provided as an exemplary embodiment of this disclosure; Figure 2 A flowchart illustrating a minute-level carbon emission factor reconstruction method based on knowledge distillation, provided as another exemplary embodiment of this disclosure; Figure 3 A schematic diagram of the training process of a student model provided in a specific embodiment of this disclosure; Figure 4 A schematic flowchart of a method for obtaining carbon emission factors provided as an exemplary embodiment of this disclosure; Figure 5 A schematic diagram of a minute-level carbon emission factor reconstruction device based on knowledge distillation provided in an embodiment of this disclosure; Figure 6 This is a schematic diagram of the structure of a carbon emission factor acquisition device provided in an embodiment of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0016] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0017] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0018] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0019] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0020] To address the issue that existing carbon emission factor reconstruction methods cannot be applied to scenarios where coal quality data is missing, this disclosure proposes a minute-level reconstruction method for carbon emission factors in thermal power units based on improved knowledge distillation. This solves the problem of being unable to reconstruct minute-level carbon emission factor data when coal quality data is missing. This solution is based on an improved knowledge distillation algorithm. By transferring knowledge of coal quality mapping relationships from the teacher model to the student model, the student model learns these mapping relationships and outputs minute-level carbon emission factors in preset time intervals. This achieves high temporal resolution carbon emission factors even in scenarios where coal quality data is missing.

[0021] The following detailed explanation, in conjunction with the accompanying drawings, details the method, apparatus, and equipment for reconstructing minute-level carbon emission factors based on knowledge distillation provided in this disclosure.

[0022] Figure 1 This is a flowchart illustrating a knowledge distillation-based minute-level carbon emission factor reconstruction method provided as an exemplary embodiment of the present disclosure. The method can be executed by a knowledge distillation-based minute-level carbon emission factor reconstruction device provided in the embodiments of the present disclosure. The device can be implemented by software and / or hardware and can be integrated into an electronic device.

[0023] like Figure 1 As shown, this knowledge distillation-based minute-level carbon emission factor reconstruction method may include the following steps: Step 101: Obtain multiple first training samples and multiple second training samples. A first training sample includes the load, coal consumption, fixed carbon, and corresponding true values ​​of carbon emission factors of thermal power units within a preset time period. A second training sample is obtained by removing fixed carbon from a first training sample. The preset time period is in the minute range.

[0024] The specific value of the preset duration can be set according to actual needs. For example, if the current power system dispatching arranges unit output in 15-minute units, the preset duration can be set to 15 minutes (min) to improve the time resolution of carbon emission factors to the 15-minute level and achieve precise emission reduction.

[0025] In this embodiment of the disclosure, multiple first training samples are obtained. Each first training sample includes the load, coal consumption, fixed carbon, and corresponding true values ​​of carbon emission factors of the thermal power unit within a preset time period. The load refers to the arithmetic mean of the measured active power of the thermal power unit within the preset time period (e.g., within a continuous 15-minute time window). The coal consumption refers to the net coal mass (in tons) entering the boiler combustion system within the same preset time period (e.g., within the same 15-minute window). The fixed carbon refers to the received basis fixed carbon content (Car) (in %) of the coal batch within the same preset time period (e.g., 15 minutes). The true value of the carbon emission factor refers to the actual value of the carbon emission factor generated by the thermal power unit within the same preset time period under that time window. For example, taking a preset time period of 15 minutes as an example, the load, coal consumption, fixed carbon, and true carbon emission factor data of the thermal power unit within a continuous 15-minute time window are obtained to obtain a first training sample.

[0026] In this embodiment of the disclosure, the second training sample is matched with the first training sample. A first training sample is obtained by removing the fixed carbon data. For example, if a first training sample includes the load, coal consumption, fixed carbon, and actual carbon emission factor data of a thermal power unit within a continuous 15-minute time window, then the corresponding second training sample includes the load, coal consumption, and actual carbon emission factor data from the first training sample, but does not include fixed carbon.

[0027] Step 102: Train the teacher model using multiple first training samples to obtain the trained target teacher model, and obtain the first carbon emission factor mapping value output by the teacher model for each first training sample during the training process.

[0028] In this embodiment, the teacher model can be trained using multiple first training samples to obtain a trained teacher model (referred to as the target teacher model for ease of description and distinction). During the training process, for the first training sample input to the teacher model, the teacher model will perform carbon emission factor mapping and output the corresponding carbon emission factor mapping result (referred to as the first carbon emission factor mapping value for ease of description and distinction). In this embodiment, the first carbon emission factor mapping value output by the teacher model for each first training sample is obtained during each training session.

[0029] In the teacher-student model, the teacher model typically refers to a complex model that is high-performance, high-capacity, has a large number of parameters, and is well-trained. Its main function is to provide "knowledge" for the lightweight student model. In this embodiment, the teacher model can adopt a commonly used large language model that has high confidence output capability, multi-level representation capability, high task adaptability, stability, and robustness. This disclosure does not impose specific restrictions on the model structure of the teacher model.

[0030] In one optional embodiment of this disclosure, when training the target teacher model, a first sample subset for the current training round can be obtained from multiple first training samples. The number of first training samples included in the first sample subset can be set according to actual needs, and this disclosure does not limit this. The load, coal consumption, and fixed carbon of each first training sample in the first sample subset are input into the teacher model to perform carbon emission factor mapping, and the first carbon emission factor mapping value output by the teacher model is obtained. Based on the true value of the carbon emission factor and the first carbon emission factor mapping value corresponding to each first training sample in the first sample subset, a loss calculation is performed to obtain the teacher loss value. For example, the loss function used to calculate the teacher loss value can be mean squared error (MSE). If the teacher loss value is greater than a first preset value, the model parameters of the teacher model are updated, and the sample subset for the next training round is obtained for iterative training until the calculated teacher loss value is less than or equal to the first preset value, or the number of training rounds reaches the first threshold, and the training is completed, resulting in a trained target teacher model. The specific values ​​of the first preset value and the first threshold can be set according to actual needs, and this disclosure does not limit this.

[0031] Taking a preset duration of 15 minutes as an example, a first training sample includes 15 minutes of load + 15 minutes of coal consumption + 15 minutes of carbon fixation (input to the teacher model), and the corresponding 15 minutes of true carbon emission factor (target of the teacher model, denoted as ). Y_ truth Before training the teacher model, the model training parameters need to be set, including the number of training epochs e, the number of samples B processed by the teacher model in one forward + backpropagation (i.e., batch_size, representing the number of the first training samples contained in the first sample subset), the learning rate learn_rate (denoted as lr), and the first count threshold E. The training process of the teacher model is as follows: 1. Randomly initialize teacher model parameters

[0032] 2. for e=1 to E do 3. for each batch ∈D do 4. Calculate the teacher model output: Y_teacher =

[0033] 5. Calculate the loss value:

[0034] 6. Backpropagation:

[0035] 7. Update teacher model parameters:

[0036] 8. end for 9. end for 10. Return the trained teacher model and its output.

[0037] In the above training process, X j and Y j To form a first training sample, X j Includes 15 minutes of load + 15 minutes of coal consumption + 15 minutes of fixed carbon. Y j for X j The corresponding 15-minute carbon emission factor true value, D includes multiple first training samples. When calculating the teacher loss value of the teacher model, it is based on the first carbon emission factor mapping value output by the teacher model and the corresponding... Y_truth The teacher loss value is calculated using the MSE function, and the model parameters of the teacher model are adjusted through backpropagation until the number of training iterations reaches the first threshold, thus obtaining the trained target teacher model.

[0038] Step 103: Train the student model based on multiple second training samples and multiple first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model. The carbon emission factor mapping model is used to obtain the carbon emission factor of thermal power units at a preset time period.

[0039] In this embodiment of the disclosure, after the target teacher model is trained, the student model is trained using the first carbon emission factor mapping value output by the teacher model for each first training sample and multiple second training samples during the training process, to obtain the trained target student model. The trained target student model serves as the carbon emission factor mapping model, which is used to obtain the carbon emission factor of the thermal power unit at a preset time period. In other words, the trained carbon emission factor mapping model can achieve high time resolution carbon emission factor acquisition at the minute level, and the time resolution of the carbon emission factor mapping model is the preset time period.

[0040] The student model is a lightweight model that can adopt the same model structure as the teacher model, only with different depth / width. For example, the teacher model consists of a 12-layer BERT model, while the student model consists of a 6-layer BERT model. By adopting the same model structure as the teacher model, it is easier to achieve layer alignment distillation and improve training stability.

[0041] When training the carbon emission factor mapping model, multiple second training samples are input into the student model in batches. The student model performs carbon emission factor mapping based on the input second training samples and outputs the corresponding carbon emission factor mapping result (referred to as the second carbon emission factor mapping value for ease of description and distinction). Loss is calculated based on the second carbon emission factor mapping value, the first carbon emission factor mapping value, and the true value of the carbon emission factor, resulting in the student model's loss value. Training is complete when the loss value is less than or equal to a second preset value, or when the number of training iterations reaches a second threshold, resulting in a trained carbon emission factor mapping model. The first carbon emission factor mapping value output by the teacher model is used to calculate the soft-label loss, thereby transferring the knowledge from the large teacher model to the small student model.

[0042] It should be noted that, in this embodiment, in order for the model to learn the temporal relationship between data, when obtaining the first training sample and the second training sample, the first training sample can be sorted according to the timestamp of the sample data in the first training sample, and input into the teacher model for model training in the order of the timestamp from earliest to latest. Similarly, the second training sample obtained based on the first training sample is also sorted according to the timestamp and input into the student model in order for model training.

[0043] The minute-level carbon emission factor reconstruction method based on knowledge distillation provided in this disclosure involves acquiring multiple first training samples and multiple second training samples. Each first training sample includes the load, coal consumption, fixed carbon, and corresponding true values ​​of carbon emission factors for a thermal power unit within a preset time period. Each second training sample is obtained by removing fixed carbon from a first training sample. The preset time period is in the minute range. A teacher model is trained using these multiple first training samples to obtain a trained target teacher model. The first carbon emission factor mapping value output by the teacher model for each first training sample during training is also obtained. Based on the multiple second training samples and multiple first carbon emission factors... The sub-mapping values ​​are used to train the student model to obtain a trained carbon emission factor mapping model. Then, by inputting the load and coal consumption of the thermal power unit within a preset time period into this carbon emission factor mapping model, the carbon emission factor of the thermal power unit within the preset time period can be obtained. This achieves the acquisition of carbon emission factor data with a preset time period as the cycle, and realizes the acquisition of carbon emission factors with a preset time period as the time resolution at the minute level. Moreover, only the load and coal consumption of the thermal power unit are needed to obtain the high time resolution carbon emission factor, without relying on high time resolution coal quality data as input. This solves the problem that minute-level carbon emission factor data cannot be reconstructed when coal quality data is missing.

[0044] Knowledge distillation (KD) uses a total loss function composed of the soft-label loss from the weighted teacher model output and the hard-label loss from the original data. It transfers knowledge from a large teacher model to a small student model and can be used for classification and regression tasks. (1) In the classification task, a temperature parameter T is introduced. The outputs of the student model and the teacher model are softened by the temperature-inclusive softmax function, and the soft loss is calculated by KL divergence. The hard loss of the student model output and the original data is calculated by the cross-entropy function, and the weighted sum is used to form the total loss.

[0045] (2) In the regression task, the output of the teacher model is directly used as the soft label or the hidden state of the teacher model is distilled to calculate the soft loss and hard loss respectively, and the weighted sum is used to form the total loss.

[0046] This disclosure also provides an improved knowledge distillation algorithm that considers the time-varying trend characteristics of carbon emission factors and can adaptively adjust the correction value of fixed carbon to the carbon emission factors. Thus, in an optional embodiment of this disclosure, such as... Figure 2 As shown, based on the foregoing embodiments, step 103 may include the following sub-steps: Step 201: Obtain the second sample subset for the current training round from multiple second training samples.

[0047] The number of second training samples included in the second sample subset can be set according to actual needs, and this disclosure does not impose any restrictions on its specific value. In addition, the size of the second sample subset can be the same as or different from the size of the first sample subset when training the teacher model, and this disclosure does not impose any restrictions on this either.

[0048] Step 202: Input the load and coal consumption of each second training sample in the second sample subset into the student model to perform carbon emission factor mapping, and obtain the second carbon emission factor mapping value output by the student model.

[0049] In this embodiment, a second training sample includes the load, coal consumption, and corresponding true values ​​of carbon emission factors of thermal power units within a preset time period. The load and coal consumption are input into the student model, while the true values ​​of the carbon emission factors serve as the training objective for calculating the loss value. In this embodiment, the load and coal consumption of each second training sample in the second sample subset are input into the student model. The student model performs carbon emission factor mapping based on this input and outputs the corresponding carbon emission factor mapping result (referred to as the second carbon emission factor mapping value).

[0050] Step 203: Determine the first soft output of the student model for each second training sample in the second sample subset based on the second carbon emission factor mapping value, and determine the second soft output of the teacher model for the target first training sample based on the first carbon emission factor mapping value output by the teacher model for each target first training sample corresponding to each second training sample in the second sample subset.

[0051] It is understood that in the embodiments of this disclosure, the second training sample is obtained by removing the fixed carbon parameter from the first training sample, that is, each second training sample has a corresponding first training sample. Thus, for each second training sample in the subset of second samples, there is a corresponding first training sample (referred to as the target first training sample for ease of description and distinction).

[0052] In this embodiment, for each second training sample in the second sample subset, the soft output of each second training sample is determined based on the second carbon emission factor mapping value output by the student model (referred to as the first soft output for ease of description and distinction), and for each target first training sample, the soft output of each target first training sample is determined based on the first carbon emission factor mapping value output by the teacher model for each target first training sample (referred to as the second soft output for ease of description and distinction).

[0053] Soft output refers to the smooth probability distribution output by the model (teacher model, student model) after temperature scaling of the mapping value of a single sample.

[0054] In one optional embodiment of this disclosure, when determining the first soft output of each second training sample, preset temperature parameters and time-varying step size parameters can be obtained first. For the i-th second training sample in the second sample subset, neighboring samples associated with the i-th second training sample are determined based on the time-varying step size parameters. Based on the second carbon emission factor mapping values ​​corresponding to the neighboring samples, the neighboring mapping mean of the i-th second training sample is determined. The neighboring mapping mean refers to the average value of the second carbon emission factor mapping values ​​corresponding to each neighboring sample of the i-th second training sample. Finally, the first soft output of the i-th second training sample is determined based on the temperature parameters, the time-varying step size parameters, the second carbon emission factor mapping value of the i-th second training sample, and the neighboring mapping mean.

[0055] The neighboring samples associated with the i-th second training sample refer to the number of time-varying step-size parameters preceding the i-th second training sample and the number of time-varying step-size parameters following the i-th second training sample. For example, assuming the time-varying step-size parameter is K, the neighboring samples associated with the i-th second training sample include the K second training samples preceding the i-th second training sample and the K second training samples following the i-th second training sample. If i=1, the neighboring samples associated with it only include the K adjacent second training samples following it; if the i-th second training sample is the last second training sample in the subset of second samples, the neighboring samples associated with it only include the K adjacent second training samples preceding it.

[0056] It is understood that in this embodiment of the disclosure, the calculation method of the first soft output is similar to that of the second soft output.

[0057] As an example, suppose the obtained temperature parameter is... T The time-varying step size parameter is K Then the first soft formula can be calculated using the following formula (1), and the second soft output can be calculated using the following formula (2): (1) (2) In the above formulas (1)-(2), This represents the first soft output of the i-th second training sample. This represents the second carbon emission factor mapping value of the i-th second training sample. Let represent the mean of the neighboring mappings of the i-th second training sample; This represents the second soft output of the target first training sample corresponding to the i-th second training sample. This represents the first carbon emission factor mapping value of the target first training sample corresponding to the i-th second training sample. Let represent the mean of the neighboring mappings of the target first training sample corresponding to the i-th second training sample.

[0058] In this embodiment of the disclosure, by introducing a temperature parameter T and a time-varying step size parameter when determining the soft outputs corresponding to the teacher model and the student model respectively, the time-varying trend characteristics of the carbon emission factor of the thermal power unit are considered. This achieves the purpose of improving the soft label of knowledge distillation and softening the outputs of the teacher model and the student model. As a result, the correction value of fixed carbon to the carbon emission factor can be adaptively adjusted, thereby improving the rationality and accuracy of the output results of the final carbon emission factor mapping model.

[0059] Step 204: Based on the first soft output, second soft output, second carbon emission factor mapping value and carbon emission factor true value corresponding to each second training sample in the second sample subset, determine the student loss value of the student model.

[0060] In this embodiment, after determining the first soft output of the student model for each second training sample in the second sample subset, and the second soft output of the teacher model for the target first training sample corresponding to each second training sample in the second sample subset, the student loss value of the student model is further determined based on the obtained first soft output, second soft output, second carbon emission factor mapping value and carbon emission factor true value.

[0061] It is understandable that, since the second training sample corresponds one-to-one with the first training sample, the teacher model's second soft output for the target first training sample can also be understood as the second soft output for the second training sample.

[0062] In one optional embodiment of this disclosure, when determining the student loss value for the current training round of the student model, the first loss value of the student model can be determined based on the first soft output and the second soft output corresponding to each second training sample in the second sample subset; and the second loss value of the student model can be determined based on the second carbon emission factor mapping value and the true value of the carbon emission factor corresponding to each second training sample in the second sample subset; then, the first loss value and the second loss value are weighted and summed to obtain the student loss value.

[0063] The first loss value is determined based on the soft output and can also be called the soft loss; the second loss value is determined based on the carbon emission factor mapping value and the true value of the carbon emission factor and can also be called the hard loss. The first loss value can be calculated using the following formula (3), the second loss value can be calculated using the following formula (4), and the student loss value can be calculated using the following formula (5): (3) (4) (5) In the above formulas (2)-(5), This represents the first loss value. Indicates the second soft output. Indicates the first soft output. This represents the second loss value. This represents the true value of the carbon emission factor. This represents the mapping value of the second carbon emission factor. This represents the student's loss value. Indicates the weighting coefficient. The specific value can be set according to actual needs, and this disclosure does not impose any restrictions on it.

[0064] Step 205: If the student loss value is greater than the second preset value, update the model parameters of the student model and obtain the sample subset of the next training round for iterative training until the calculated student loss value is less than or equal to the second preset value, or the training round reaches the second threshold, and the training is completed, and the trained carbon emission factor mapping model is obtained.

[0065] The specific values ​​of the second preset value and the second threshold number can be set according to actual needs, and this disclosure does not impose any restrictions on them.

[0066] In this embodiment, after calculating the student loss value for this round of training, it is compared with the second preset value. If the student loss value is greater than the second preset value, the model parameters of the student model are updated, and a sample subset for the next training round is obtained for iterative training. The student loss value is recalculated and compared with the second preset value until the calculated student loss value is less than or equal to the second preset value, or the training round reaches the second threshold. The training is then completed, and the trained carbon emission factor mapping model is obtained.

[0067] Combination Figure 3The diagram illustrates the training process of the student model. It shows that the first training sample X_teacher is input into the teacher model for training. The first carbon emission factor mapping value Y_teacher generated by the teacher model is used to calculate the second soft output Y_teacher_soft during the student model training process. The second training sample X_student is input into the student model for training. The second carbon emission factor mapping value Y_student generated by the student model is used to calculate the first soft output Y_student_soft during the student model training process. Y_teacher_soft and Y_student_soft are used to calculate the first loss value soft_loss. Y_student and the true carbon emission factor value Y_truth are used to calculate the second loss value hard_loss. Then, based on soft_loss and hard_loss, the student loss value total_loss is calculated. Based on total_loss, backpropagation is used to adjust the parameters of the student model for iterative training, ultimately obtaining the trained student model as the carbon emission factor mapping model.

[0068] Taking a preset duration of 15 minutes as an example, multiple second training samples constitute a training sample set D. Each second training sample includes 15 minutes of load + 15 minutes of coal consumption (the input to the student model) and the corresponding 15 minutes of carbon emission factor truth value (the student model's target, denoted as Y_truth). Before training the student model, it is necessary to set the model training parameters, including the training round e, the number of samples B processed by the student model in one forward + backpropagation (i.e., batch_size, representing the number of second training samples contained in the second sample subset), the learning rate learn_rate (denoted as lr), the second threshold E, and the first carbon emission factor mapping value corresponding to each second training sample in the first training round. The training process of the student model is as follows: 1. Randomly initialize student model parameters

[0069] 2. for e=1 to E do 3. for each batch ∈D do 4. Calculate the student model output: Y_student=

[0070] 5. Calculate the soft output of the student model:

[0071] 6. Calculate the soft output of the teacher model:

[0072] 7. Calculate soft loss:

[0073] 8. Calculate the hard loss:

[0074] 9. Calculate the total loss function:

[0075] 10. Backpropagation:

[0076] 11. Update student model parameters:

[0077] 12. end for 13. end for 14. Return the trained student model, which is the carbon emission factor mapping model.

[0078] Through the above training process, B second training samples are input into the student model each time to train the model, and the total loss value is calculated. Based on the total loss value, the model parameters of the student model are updated through backpropagation. When the training rounds reach the second threshold E, the training ends and the trained carbon emission factor mapping model is obtained.

[0079] This disclosure discloses a minute-level carbon emission factor reconstruction method based on knowledge distillation. It obtains a second sample subset for the current training round from multiple second training samples, inputs the load and coal consumption of each second training sample in the second sample subset into a student model for carbon emission factor mapping, and obtains the second carbon emission factor mapping value output by the student model. Based on the second carbon emission factor mapping value, it determines the first soft output of the student model for each second training sample in the second sample subset. It also determines the second soft output of the teacher model for the target first training sample based on the first carbon emission factor mapping value output by the teacher model for each second training sample corresponding to the target first training sample in the second sample subset. The first soft output, second soft output, second carbon emission factor mapping value, and true value of the carbon emission factor corresponding to the second training sample are used to determine the student loss value of the student model. If the student loss value is greater than the second preset value, the model parameters of the student model are updated, and a subset of samples for the next training round is obtained for iterative training until the calculated student loss value is less than or equal to the second preset value, or the number of training rounds reaches the second threshold. The training is then completed, and a trained carbon emission factor mapping model is obtained. In this way, the knowledge of coal quality mapping relationship in the teacher model is transferred to the student model, enabling the student model to learn the mapping relationship including coal quality, thereby outputting minute-level carbon emission factors and achieving high time resolution carbon emission factor acquisition.

[0080] Corresponding to the above-described knowledge distillation-based minute-level carbon emission factor reconstruction method, this disclosure also provides a carbon emission factor acquisition method, which utilizes the carbon emission factor mapping model trained through the above embodiments to achieve minute-level carbon emission factor acquisition.

[0081] Figure 4 A schematic flowchart of a carbon emission factor acquisition method provided in an exemplary embodiment of this disclosure is shown below. Figure 4 As shown, the method for obtaining carbon emission factors may include the following steps: Step 301: Obtain the load and coal consumption of the thermal power unit within a preset time period.

[0082] The specific value of the preset duration is consistent with the time window length of the training samples used in the carbon emission factor mapping model. For example, if the time window length of the training samples is 15 minutes, then in the actual model application scenario, the preset duration is 15 minutes, that is, the load and coal consumption of the thermal power unit within 15 minutes are obtained, so as to obtain the carbon emission factor in 15-minute units.

[0083] Step 302: Input the load and coal consumption within a preset time period into the pre-trained carbon emission factor mapping model, and obtain the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption within the preset time period. The carbon emission factor mapping model is trained by the minute-level carbon emission factor reconstruction method based on knowledge distillation described in the above embodiment.

[0084] In this embodiment, the load and coal consumption within a preset time period are input into a pre-trained carbon emission factor mapping model. The carbon emission factor mapping model performs carbon emission factor mapping based on the input data and outputs the corresponding carbon emission factor mapping value.

[0085] Step 303: Determine the carbon emission factor mapping value as the carbon emission factor of the thermal power unit within a preset time period.

[0086] In this embodiment, after obtaining the carbon emission factor mapping value output by the carbon emission factor mapping model, it can be determined as the carbon emission factor of the thermal power unit within the preset time period.

[0087] In practical applications, the time resolution can be preset. Each time, the load and coal consumption of thermal power units within the preset time window are obtained. The corresponding carbon emission factor is obtained by using the carbon emission factor mapping model, so that the time resolution of the carbon emission factor is provided to the preset time (minute level), which helps to achieve precise emission reduction.

[0088] The carbon emission factor acquisition method of this disclosure acquires the load and coal consumption of a thermal power unit within a preset time period, inputs the load and coal consumption within the preset time period into a pre-trained carbon emission factor mapping model, and acquires the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption within the preset time period. The carbon emission factor mapping value is then determined as the carbon emission factor of the thermal power unit within the preset time period. Thus, only the load and coal consumption within the preset time period are needed to acquire the carbon emission factor of the thermal power unit, avoiding dependence on high-time-resolution coal quality data. Furthermore, it achieves carbon emission factor acquisition with a preset time period of minutes as the time resolution, enabling the time resolution of the carbon emission factor to match the actual operating requirements of the power system, which helps to achieve precise emission reduction.

[0089] To achieve the above embodiments, this disclosure also provides a minute-level carbon emission factor reconstruction device based on knowledge distillation.

[0090] Figure 5 This is a schematic diagram of a minute-level carbon emission factor reconstruction device based on knowledge distillation provided in an embodiment of the present disclosure. The device is implemented in software and / or hardware and can be integrated into an electronic device.

[0091] like Figure 5 As shown, the knowledge distillation-based minute-level carbon emission factor reconstruction device 40 may include: a sample acquisition module 410, a first training module 420, and a second training module 430.

[0092] The sample acquisition module 410 is used to acquire multiple first training samples and multiple second training samples. A first training sample includes the load, coal consumption, fixed carbon and corresponding carbon emission factor true values ​​of the thermal power unit within a preset time period. A second training sample is obtained by removing fixed carbon from a first training sample. The preset time period is in the minute range. The first training module 420 is used to train the teacher model using multiple first training samples to obtain the trained target teacher model, and to obtain the first carbon emission factor mapping value output by the teacher model for each first training sample during the training process. The second training module 430 is used to train the student model based on multiple second training samples and multiple first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model. The carbon emission factor mapping model is used to obtain the carbon emission factor of thermal power units at a preset time period.

[0093] Optionally, the first training module 420 is also used for: Obtain the first sample subset of the current training round from multiple first training samples; The load, coal consumption, and fixed carbon of each first training sample in the first sample subset are input into the teacher model to perform carbon emission factor mapping, and the first carbon emission factor mapping value output by the teacher model is obtained. The teacher loss value is obtained by calculating the loss based on the true value of the carbon emission factor and the mapping value of the first carbon emission factor corresponding to each first training sample in the first sample subset. If the teacher loss value is greater than the first preset value, update the model parameters of the teacher model and obtain a sample subset for the next training round for iterative training until the calculated teacher loss value is less than or equal to the first preset value, or the number of training rounds reaches the first threshold, the training is completed, and the trained target teacher model is obtained.

[0094] Optionally, the second training module 430 includes: The first acquisition unit is used to acquire a subset of the second samples for the current training round from multiple second training samples; The second acquisition unit is used to input the load and coal consumption of each second training sample in the second sample subset into the student model for carbon emission factor mapping, and to acquire the second carbon emission factor mapping value output by the student model. The first determining unit is used to determine the first soft output of the student model for each second training sample in the second sample subset based on the second carbon emission factor mapping value, and to determine the second soft output of the teacher model for the target first training sample based on the first carbon emission factor mapping value output by the teacher model for each target first training sample corresponding to each second training sample in the second sample subset. The second determining unit is used to determine the student loss value of the student model based on the first soft output, the second soft output, the second carbon emission factor mapping value and the true value of the carbon emission factor corresponding to each second training sample in the second sample subset. The iteration module is used to update the model parameters of the student model when the student loss value is greater than the second preset value, and to obtain the sample subset of the next training round for iterative training until the calculated student loss value is less than or equal to the second preset value, or the training round reaches the second threshold, and the training is completed, resulting in a trained carbon emission factor mapping model.

[0095] Further optionally, the first determining unit is also used for: Obtain the preset temperature parameters and time-varying step size parameters; For the i-th second training sample in the second sample subset, the neighboring samples associated with the i-th second training sample are determined based on the time-varying step size parameter; Based on the second carbon emission factor mapping values ​​corresponding to adjacent samples, determine the mean of the adjacent mappings of the i-th second training sample; Based on temperature parameters, time-varying step size parameters, the second carbon emission factor mapping value of the i-th second training sample, and the mean of adjacent mappings, the first soft output of the i-th second training sample is determined.

[0096] Optionally, the second determining unit is also used for: Based on the first soft output and the second soft output corresponding to each second training sample in the second sample subset, the first loss value of the student model is determined. Based on the second carbon emission factor mapping value and the true value of the carbon emission factor corresponding to each second training sample in the second sample subset, the second loss value of the student model is determined. The student's loss value is obtained by weighted summation of the first and second loss values.

[0097] The knowledge distillation-based minute-level carbon emission factor reconstruction device for electronic devices provided in this disclosure can execute the knowledge distillation-based minute-level carbon emission factor reconstruction method provided in this disclosure, and has the corresponding functional modules and beneficial effects for executing the method. Content not described in detail in the device embodiments of this disclosure can be referred to the description in any method embodiment of this disclosure.

[0098] To achieve the above embodiments, this disclosure also provides a carbon emission factor acquisition device.

[0099] Figure 6 This is a schematic diagram of a carbon emission factor acquisition device provided in an embodiment of the present disclosure. The device is implemented in software and / or hardware and can be integrated into an electronic device.

[0100] like Figure 6 As shown, the carbon emission factor acquisition device 50 may include: a first acquisition module 510, a second acquisition module 520, and a determination module 530.

[0101] The first acquisition module 510 is used to acquire the load and coal consumption of the thermal power unit within a preset time period. The second acquisition module 520 is used to input the load and coal consumption of the thermal power unit within a preset time period into a pre-trained carbon emission factor mapping model, and to acquire the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption of the thermal power unit within the preset time period. The carbon emission factor mapping model is trained by the minute-level carbon emission factor reconstruction method based on knowledge distillation described in the above embodiment. The determination module 530 is used to determine the carbon emission factor mapping value as the carbon emission factor of the thermal power unit within a preset time period.

[0102] The carbon emission factor acquisition device for electronic devices provided in this disclosure can execute the carbon emission factor acquisition method provided in this disclosure, and has the corresponding functional modules and beneficial effects of executing the method. Content not described in detail in the device embodiments of this disclosure can be referred to the description in any method embodiment of this disclosure.

[0103] This disclosure also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the knowledge distillation-based minute-level carbon emission factor reconstruction method provided in any embodiment of this disclosure, or implements the carbon emission factor acquisition method provided in any embodiment of this disclosure.

[0104] According to one or more embodiments of this disclosure, this disclosure provides an electronic device, including: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the minute-level carbon emission factor reconstruction method based on knowledge distillation as provided in any embodiment of the present disclosure, or to implement the carbon emission factor acquisition method provided in any embodiment of the present disclosure.

[0105] According to one or more embodiments of the present disclosure, the present disclosure provides a computer-readable storage medium storing a computer program for implementing the minute-level carbon emission factor reconstruction method based on knowledge distillation as provided in any embodiment of the present disclosure, or implementing the carbon emission factor acquisition method provided in any embodiment of the present disclosure.

[0106] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can 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 can 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 can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, 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: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0107] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0108] 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 the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can 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.

[0109] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0110] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0111] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, 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 devices, magnetic storage devices, or any suitable combination of the foregoing.

[0112] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0113] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0114] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for reconstructing minute-level carbon emission factors based on knowledge distillation, characterized in that, The method includes: Multiple first training samples and multiple second training samples are obtained. Each first training sample includes the load, coal consumption, fixed carbon and corresponding carbon emission factor true values ​​of thermal power units within a preset time period. Each second training sample is obtained by removing the fixed carbon from a first training sample. The preset time period is in the minute range. The teacher model is trained using the multiple first training samples to obtain a trained target teacher model, and the first carbon emission factor mapping value output by the teacher model for each first training sample during the training process is obtained. The student model is trained based on the multiple second training samples and multiple first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model. The carbon emission factor mapping model is used to obtain the carbon emission factor of thermal power units at a preset time period.

2. The method according to claim 1, characterized in that, The step of training the teacher model using the multiple first training samples to obtain the trained target teacher model includes: Obtain the first sample subset of the current training round from the plurality of first training samples; The load, coal consumption, and fixed carbon of each first training sample in the first sample subset are input into the teacher model to perform carbon emission factor mapping, and the first carbon emission factor mapping value output by the teacher model is obtained. Based on the true value of the carbon emission factor and the mapping value of the first carbon emission factor corresponding to each first training sample in the first sample subset, the loss is calculated to obtain the teacher loss value; If the teacher loss value is greater than the first preset value, the model parameters of the teacher model are updated, and a sample subset of the next training round is obtained for iterative training until the calculated teacher loss value is less than or equal to the first preset value, or the number of training rounds reaches the first threshold, the training is completed, and the trained target teacher model is obtained.

3. The method according to claim 1, characterized in that, The step of training the student model based on the plurality of second training samples and the plurality of first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model includes: Obtain the second sample subset for the current training round from the plurality of second training samples; The load and coal consumption of each second training sample in the second sample subset are input into the student model for carbon emission factor mapping, and the second carbon emission factor mapping value output by the student model is obtained. The student model determines the first soft output of each second training sample in the second sample subset based on the second carbon emission factor mapping value, and the teacher model determines the second soft output of the target first training sample based on the first carbon emission factor mapping value output by the teacher model for each second training sample in the second sample subset. Based on the first soft output, the second soft output, the second carbon emission factor mapping value and the true value of the carbon emission factor corresponding to each second training sample in the second sample subset, the student loss value of the student model is determined. If the student loss value is greater than the second preset value, the model parameters of the student model are updated, and a sample subset of the next training round is obtained for iterative training until the calculated student loss value is less than or equal to the second preset value, or the training round reaches the second threshold, the training is completed, and the trained carbon emission factor mapping model is obtained.

4. The method according to claim 3, characterized in that, The step of determining the first soft output of the student model for each of the second training samples in the second sample subset based on the second carbon emission factor mapping value includes: Obtain the preset temperature parameters and time-varying step size parameters; For the i-th second training sample in the second sample subset, neighboring samples associated with the i-th second training sample are determined based on the time-varying step size parameter; Based on the second carbon emission factor mapping values ​​corresponding to the adjacent samples, the mean value of the adjacent mappings of the i-th second training sample is determined; Based on the temperature parameter, the time-varying step size parameter, the second carbon emission factor mapping value of the i-th second training sample, and the mean of adjacent mappings, the first soft output of the i-th second training sample is determined.

5. The method according to claim 3, characterized in that, The step of determining the student loss value of the student model based on the first soft output, the second soft output, the second carbon emission factor mapping value, and the true value of the carbon emission factor corresponding to each second training sample in the second sample subset includes: Based on the first soft output and the second soft output corresponding to each second training sample in the second sample subset, the first loss value of the student model is determined; Based on the second carbon emission factor mapping value and the true value of the carbon emission factor corresponding to each second training sample in the second sample subset, the second loss value of the student model is determined. The student's loss value is obtained by weighted summation of the first loss value and the second loss value.

6. A method for obtaining carbon emission factors, characterized in that, The method includes: Obtain the load and coal consumption of thermal power units within a preset time period; The load and coal consumption within the preset time period are input into a pre-trained carbon emission factor mapping model, and the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption within the preset time period is obtained. The carbon emission factor mapping model is trained by the minute-level carbon emission factor reconstruction method based on knowledge distillation as described in any one of claims 1-5. The carbon emission factor mapping value is determined as the carbon emission factor of the thermal power unit within the preset time period.

7. A minute-level carbon emission factor reconstruction device based on knowledge distillation, characterized in that, The device includes: The sample acquisition module is used to acquire multiple first training samples and multiple second training samples. Each first training sample includes the load, coal consumption, fixed carbon and corresponding carbon emission factor true values ​​of thermal power units within a preset time period. Each second training sample is obtained by removing the fixed carbon from a first training sample. The preset time period is in the minute range. The first training module is used to train the teacher model using the multiple first training samples to obtain the trained target teacher model, and to obtain the first carbon emission factor mapping value output by the teacher model for each first training sample during the training process. The second training module is used to train the student model based on the multiple second training samples and multiple first carbon emission factor mapping values ​​to obtain a trained carbon emission factor mapping model. The carbon emission factor mapping model is used to map the carbon emission factors of thermal power units with the preset duration as the period.

8. A carbon emission factor acquisition device, characterized in that, The device includes: The first acquisition module is used to acquire the load and coal consumption of thermal power units within a preset time period; The second acquisition module is used to input the load and coal consumption of the thermal power unit within the preset time period into a pre-trained carbon emission factor mapping model, and to acquire the carbon emission factor mapping value output by the carbon emission factor mapping model based on the load and coal consumption of the thermal power unit within the preset time period, wherein the carbon emission factor mapping model is trained by the minute-level carbon emission factor reconstruction method based on knowledge distillation as described in any one of claims 1-5. The determination module is used to determine the carbon emission factor mapping value as the carbon emission factor of the thermal power unit within the preset time period.

9. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the knowledge distillation-based minute-level carbon emission factor reconstruction method according to any one of claims 1-5, or to implement the carbon emission factor acquisition method according to claim 6.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for implementing the minute-level carbon emission factor reconstruction method based on knowledge distillation as described in any one of claims 1-5, or for implementing the carbon emission factor acquisition method as described in claim 6.

Images