Method and device for predicting coal consumption of thermal power generating unit and program product

Abstract

The invention relates to a thermal power generating unit coal consumption prediction method and device and a program product. Comprising the steps of determining a target prediction time period, a target prediction moment and a target time scale; screening out a target operation data set matched with the target time scale and the target prediction time period from a preset operation data set corresponding to the target thermal power generating unit according to the target prediction moment; and inputting the target operation data set into a pre-trained target coal consumption prediction model to obtain the coal consumption corresponding to the target thermal power generating unit under the target time scale in the target prediction time period. Thus, the problem that a traditional single model is difficult to consider multi-time-dimension prediction precision can be effectively solved, meanwhile, errors caused by manual scale conversion are avoided, and full-cycle decision support from real-time level to strategic level is provided for fuel scheduling of a power plant.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of power generation system operation optimization, and particularly relates to a method and device for predicting coal consumption of a thermal power unit and a program product. BACKGROUND

[0002] In related technologies, deep learning and machine learning algorithms can be combined to predict the coal consumption of a thermal power unit. However, the above methods are mostly limited to single-time dimension analysis, either focusing on short-term minute-level data or focusing on long-term daily data, and ignoring the dynamic correlation of data at different time scales. SUMMARY

[0003] Therefore, the present disclosure provides a method and device for predicting coal consumption of a thermal power unit and a program product to solve the problems in related technologies.

[0004] In a first aspect, the present disclosure provides a method for predicting coal consumption of a thermal power unit, the method comprising: determining a target prediction period, a target prediction time and a target time scale; selecting, from a preset operation data set corresponding to a target thermal power unit, a target operation data set matching the target time scale and the target prediction period according to the target prediction time; and inputting the target operation data set into a pre-trained target coal consumption prediction model to obtain the coal consumption of the target thermal power unit at the target time scale within the target prediction period.

[0005] In a second aspect, the present disclosure provides a device for predicting coal consumption of a thermal power unit, applied to the method for predicting coal consumption of a thermal power unit as described in the first aspect, the device comprising: a determination module configured to determine a target prediction period, a target prediction time and a target time scale; an acquisition module configured to select, from a preset operation data set corresponding to a target thermal power unit, a target operation data set matching the target time scale and the target prediction period according to the target prediction time; and a prediction module configured to input the target operation data set into a pre-trained target coal consumption prediction model to obtain the coal consumption of the target thermal power unit at the target time scale within the target prediction period.

[0006] In a third aspect, the present disclosure provides a computer program product, wherein the computer program / instructions are executed by a processor to implement the steps of the method for predicting coal consumption of a thermal power unit.

[0007] The at least one technical solution adopted by the embodiments of the present disclosure can achieve the following beneficial effects: by determining the target prediction period, the target prediction time and the target time scale, the target operation data set matched with the target time scale and the target prediction period is filtered from the preset operation data set corresponding to the target thermal power generating unit according to the target prediction time, and the target operation data set is input into the target coal consumption prediction model pre-trained to obtain the coal consumption of the target thermal power generating unit corresponding to the target time scale in the target prediction period. In this way, the target prediction period, the target prediction time and the target time scale can be accurately set, the matched target operation data set is filtered with the target time scale as the core, and the target operation data set is input into the pre-trained model, thereby realizing deep coupling of the prediction process and the time scale, and further outputting the coal consumption prediction result meeting the requirement of the target time scale, so as to effectively solve the problem that the traditional single model cannot consider the prediction accuracy of multiple time dimensions, and eliminate the error of manual scale conversion, thereby providing real-time to strategic-level full-cycle decision support for power plant fuel scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0008] The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings. The drawings provided in the present disclosure serve to provide a further understanding to the embodiments of the present disclosure, and constitute a part of the specification, which explains the present disclosure together with the embodiments of the present disclosure, but do not constitute a limitation to the present disclosure. In the drawings, the same reference numerals generally indicate the same components or steps throughout the drawings.

[0009] Figure 1 A flowchart of a prediction method of coal consumption of a thermal power generating unit provided for an exemplary embodiment of the present disclosure; Figure 2 A flowchart of data selection and collection in a power plant coal consumption prediction model prediction module provided for an exemplary embodiment of the present disclosure; Figure 3 A flowchart of model prediction module data condition screening provided for an exemplary embodiment of the present disclosure; Figure 4 A flowchart of a prediction method of coal consumption of a thermal power generating unit provided for an exemplary embodiment of the present disclosure; Figure 5 A structural schematic diagram of a prediction device of coal consumption of a thermal power generating unit provided for an embodiment of the present disclosure; Figure 6 A structural schematic diagram of an electronic device provided for an embodiment of the present disclosure; Figure 7 A structural schematic diagram of a computer system provided for an embodiment of the present disclosure; Figure 8A schematic diagram of a computer program product according to an embodiment of the present disclosure is provided. DETAILED DESCRIPTION

[0010] Embodiments of the present disclosure will be described in more detail with reference to the drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein, but rather the embodiments are provided so that the present disclosure can be more thoroughly and completely understood. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0011] It should be understood that each of the steps recited in the method embodiments of the present disclosure can be performed in different orders and / or in parallel. In addition, the method embodiments can include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0012] The term "comprising" and variations thereof as used herein are used inclusively, i.e., "comprising, but not limited to". The term "based on" is "based, at least in part, 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". Related definitions of other terms will be given in the description below. It should be noted that the concepts "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0013] It should be noted that the modification of "one" or "multiple" mentioned in the present disclosure is illustrative rather than limiting, and those skilled in the art should understand that, unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0014] The names of the messages or information exchanged between the devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0015] Under the dual background of the continuous growth of global energy demand and the accelerated promotion of low-carbon transformation, coal-fired power plants are facing the urgent need to improve operating efficiency and reduce environmental impact. Therefore, precise prediction of coal consumption of thermal power units in coal-fired power plants can not only optimize the combustion process and improve the power generation efficiency of power plants, but also provide decision-making basis for intelligent fuel scheduling, thereby effectively reducing operating costs and avoiding market fluctuation risks. Especially in the face of complex working conditions such as load fluctuation and extreme weather, high-precision coal consumption prediction can become a key technical support for fine operation and management of coal-fired power plants.

[0016] In the related art, deep learning can be combined with machine learning algorithms to predict the coal consumption of a thermal power unit. For example, a deep learning method based on a convolutional neural network can extract and process features from multi-source historical operation data of the coal combustion process of the thermal power unit, thereby predicting the coal consumption of the thermal power unit. Further, an optimization method combining random forest regression and genetic algorithm can be used to adjust and optimize the operating parameters, thereby further improving the accuracy of the coal consumption prediction of the thermal power unit. In the related art, some methods can use time series similarity analysis to indirectly calculate the coal consumption by matching historical operation data for load prediction.

[0017] The above-mentioned prediction method for coal consumption of a thermal power unit in the related art has two major pain points. On the one hand, existing prediction methods are mostly limited to single-time dimension analysis, focusing on either short-term minute-level data or long-term daily data, while ignoring the dynamic correlation of data at different time scales. On the other hand, a large number of prediction models still rely on traditional linear regression algorithms, which usually assume a fixed proportional relationship between coal consumption and operating parameters. However, in actual coal combustion processes, there are complex nonlinear couplings between parameters such as steam temperature and thermal power unit load. For example, when the unit load increases from 30% to 70%, the coal consumption growth rate is not uniform. This makes it difficult for prediction models to accurately match the actual coal consumption fluctuations in actual operating scenarios such as frequent load changes and multi-condition switching of thermal power units, making it difficult to meet the coal consumption control precision requirements of coal-fired power plants.

[0018] To address the above problems, the present disclosure provides a prediction method for coal consumption of a thermal power unit. The method can first preprocess historical operation data by removing outliers, filling missing values, and selecting steady states, etc. to ensure data quality. Then, through a multi-scale analysis method, the time series is divided into datasets of different time granularities such as hourly, daily, and weekly scales to meet the needs of different prediction scenarios. Next, with the help of the self-attention mechanism of the initial prediction model, long-term dependencies and nonlinear features in the time series are effectively captured to obtain a target coal consumption prediction model. At the same time, the hyperparameters of the target coal consumption prediction model are automatically tuned through Bayesian optimization and genetic algorithm, further improving the prediction accuracy and generalization ability of the target coal consumption prediction model. In addition, the system realizes real-time data processing and prediction by interfacing with SIS, and finally provides accurate and real-time coal consumption prediction results for power plants, providing strong support for fuel procurement and dispatching optimization decisions.

[0019] Figure 1 A flowchart of a prediction method for coal consumption of a thermal power unit is provided for an exemplary embodiment of the present disclosure. As shown in Figure 1 , it specifically includes: In S101, a plurality of original running data corresponding to different time instants of the target thermal power generating unit in a target prediction period is obtained, and the plurality of original running data is preprocessed to obtain a plurality of to-be-processed running data.

[0020] In some embodiments, a plurality of key influence factors affecting the coal consumption of the target thermal power generating unit can be determined first, such as key influence factors that are strongly correlated with the coal consumption prediction, such as unit load, coal feeder coal supply, and ambient temperature; then, original running data corresponding to each key influence factor and actual coal consumption corresponding to different time points can be collected at a fixed time interval within the target prediction period. Here, all original running data included in the target prediction period can be obtained from the Supervisory Information System (SIS) of the coal-fired power plant, and then the original running data corresponding to each key influence factor can be selected from all original running data. The key influence factors of the embodiments of the present disclosure are shown in Table 1.

[0021] Table 1 Key Influence Factor Table

[0022] wherein j is an identifier corresponding to the coal feeder, n is an identifier corresponding to the last coal feeder, j = 1, 2, …, n.

[0023] In some embodiments, since there may be interference information caused by abnormal fluctuations or unstable working conditions in the original running data, the original running data can be preprocessed using a Gaussian Mixture Model and a k-means algorithm, etc., to obtain a plurality of to-be-processed running data. For example, the distribution characteristics of the original running data can be identified using a Gaussian Mixture Model first, and then the data clusters can be divided by combining the k-means algorithm, so as to accurately exclude outliers and separate the to-be-processed running data of the target thermal power generating unit under stable running conditions. Here, the Gaussian Mixture Model (GMM) is a probability model that assumes that all data points are generated by mixing a plurality of components following a Gaussian distribution (normal distribution). The k-means algorithm is a commonly used clustering analysis method, and its core idea is to divide the data into k clusters, so that the sum of the squared distances of each data point to the cluster center to which it belongs is minimized.

[0024] It should be understood that in actual application, the specific key influence factors can not be limited, that is, the original running data can involve a plurality of influence factors, and these influence factors include not only key influence factors but also other influence factors. At this time, the original running data corresponding to all influence factors can also be preprocessed to obtain preprocessed to-be-processed running data. Here, one influence factor corresponds to one original running data or to-be-processed running data at one time point.

[0025] S102, constructing a time series dataset according to the plurality of to-be-processed operation data.

[0026] In some embodiments, the multi-dimensional to-be-processed operation data obtained after preprocessing can be arranged in chronological order into a structured time series dataset to provide a high-quality data basis for subsequent training and optimization of the coal consumption prediction model of the thermal power generating unit. For a time series, a group of data arranged in chronological order, each data corresponds to a unique time point, reflecting the change of a certain key influencing factor in the time dimension. Then, the time series dataset can be constructed according to the time series corresponding to each of the plurality of key influencing factors.

[0027] For example, the key influencing factors and the key influencing factors The time series corresponding to the target prediction period can be represented as:

[0028]

[0029] wherein, represents the to-be-processed operation data corresponding to the key influencing factor a at the 1st time point, the 1st time point can be the initial time point within the target prediction period, represents the to-be-processed operation data corresponding to the key influencing factor a at the 2nd time point, represents the to-be-processed operation data corresponding to the key influencing factor a at the mth time point, the mth time point can be the last time point within the target prediction period; represents the to-be-processed operation data corresponding to the key influencing factor b at the 1st time point, the 1st time point is the initial time point within the target prediction period, represents the to-be-processed operation data corresponding to the key influencing factor b at the 2nd time point, represents the to-be-processed operation data corresponding to the key influencing factor b at the mth time point, the mth time point can be the last time point within the target prediction period.

[0030] Based on this, the time series dataset can be constructed according to the key influencing factors and other key influencing factors The time series dataset can be represented as:

[0031] S103, training the initial prediction model using the time series dataset to obtain a target coal consumption prediction model. The initial prediction model of the embodiments of the present disclosure can be a Transformer model, a Long Short-Term Memory Network (LSTM), and a Gated Recurrent Unit (GRU). The Transformer model is a special recurrent neural network (RNN) that can effectively handle long-term dependencies in time series data; the GRU is a variant of the recurrent neural network (RNN) that controls the flow of information through a gating mechanism, which can effectively handle long-term dependencies in time series data and is commonly used in time series prediction, natural language processing and other fields.

[0032] In some embodiments, based on the constructed time series dataset, a Transformer model combined with a Bayesian optimization algorithm can be used to predict the coal consumption of the power plant; wherein in order to perform multi-time scale coal consumption prediction, the processed time series dataset is divided according to different time window lengths and time intervals. Based on this, the time window length L and the time interval Δt can be adjusted so that the final target coal consumption prediction model can be constructed for different time scales such as hourly, daily and weekly scales. It should be understood that the Transformer model is a deep learning model based on self-attention mechanism, which is used to capture long-range dependencies and nonlinear features in time series data to achieve high-precision prediction of target thermal power unit coal consumption; and the Bayesian optimization algorithm is a global search strategy for optimizing model hyperparameters, which efficiently finds the optimal hyperparameter combination by constructing a probability model (Bayesian model) of the objective function to improve the coal consumption prediction accuracy of the Transformer model.

[0033] Specifically, assuming that the window length of the sliding window is L, the time interval is Δt, and the window sliding step is s, when the time series dataset is divided, the continuous data segment is intercepted according to the set window length based on the sliding window mechanism, and a plurality of overlapping or non-overlapping subsequences are generated by sliding the window with a fixed step s, and finally a window sequence composed of each time step data is obtained, thereby realizing multi-scale processing of time series data and providing input data of different time granularities for subsequent coal consumption prediction model training.

[0034] For example, assuming that the window length of the sliding window is L, the time interval is , and the window sliding step is s, the time series dataset The data is divided into windows to obtain the time step data. After division for:

[0035] in, For the time series subset corresponding to the t-th time step, For the first The time series subset corresponding to each time step For the first The time series subset corresponding to each time step A subset of a time series containing L time steps starting from time step t; Here, q represents the identifier of the key impact factor, and The key influencing factor 'a' corresponds to the unprocessed runtime data at time point 't'. This represents the operational data to be processed corresponding to the key influencing factor b at time point t. This represents the unprocessed runtime data corresponding to the key influencing factor q at time point t.

[0036] Here, by setting the sliding step size s, the time window can slide on the original time series at a fixed step size, generating overlapping or non-overlapping subsets of the time series. For example, if the length of the time window is L and the sliding step size s=1, then each new time window moves forward by only 1 time step, forming a highly overlapping subset of the time series; if s=L, then the time windows do not overlap.

[0037] Subsequently, these time series subsets can be used as the training set for the model, thereby reflecting the changes in the operating status of thermal power units at different times. By analyzing the relationship between load fluctuations and coal consumption within the time window, multi-scale data support can be provided for the prediction model, thus achieving more accurate coal consumption prediction.

[0038] In some embodiments, for each time window, when the length of the input time series is seq_length and the prediction time steps are pred_steps, the corresponding time series subset... It can be represented as:

[0039] For example, suppose the real-time monitoring system of the thermal power unit in coal-fired power plant A monitors every 10 minutes... Raw operational data is collected every few minutes. This raw data includes unit load, coal mill feed rate, ambient temperature, etc. We can assume there are 10 key influencing factors. We need to process a one-month historical dataset (approximately 4320 time steps, T=4320) using a sliding window. , for training the coal consumption prediction model.

[0040] The time window parameters are set as follows: The length of the time window L = 24, that is, each time window contains 24 time step data, corresponding to 4 hours of operation data; The sliding step s = 6, that is, each time window slides 6 time steps, corresponding to 1 hour; The time interval t = 10 minutes; Take the 100th time step (i.e. 100x10 = 1000 minutes, about 16.67 hours) as the starting point, and according to the formula:

[0041] The time series corresponding to the first time window is obtained , which contains 24 consecutive time step operation data starting from the 100th time step. Moving the sliding window by a step size s = 6, the time series corresponding to the next time window is obtained , which has 18 time step data overlapping between , that is, 24-6 = 18. Based on this, repeat the process until all data in are traversed, that is, a series of time series subsets ( ) can be generated.

[0042] In some embodiments, in order to ensure that the time intervals of the input time series and the prediction target time series are uniform, the time intervals of the time series and the prediction target time series can also be checked by calculating the time difference between adjacent data points. The specific calculation formula is:

[0043] Where data_index represents the time index in the time series subset. If all time intervals meet the expectation, that is, Δt = interval, then the time series is considered as a valid input, where interval represents the time interval of data sampling. Further, a plurality of valid input time series subsets can be divided into a test set and a training set to train the initial prediction model to obtain the target coal consumption prediction model.

[0044] In some embodiments, the training set can be used to train the initial prediction model with mean square error (MSE) as the loss function, the optimal model parameters are updated by comparing the validation loss, and the optimal hyperparameter combination is found by Bayesian optimization. Here, mean square error (MSE) is a commonly used indicator to measure the deviation degree of the predicted coal consumption of the model from the actual coal consumption.

[0045] Specifically, the initial prediction model can be trained by the training set, and the trained target coal consumption prediction model can be tested by the test set. Finally, the mean square error (MSE) of the predicted coal consumption and the actual coal consumption is calculated as a loss function to measure the prediction performance of the target coal consumption prediction model.

[0046] Y is the time series corresponding to the actual coal consumption, i.e., the sequence corresponding to the actual coal consumption of the coal consumption of the power plant; Y is the time series corresponding to the actual coal consumption, i.e., the sequence corresponding to the actual coal consumption of the coal consumption of the power plant; the actual coal consumption at the i-th time point; the predicted coal consumption at the i-th time point; and R represents the total number of data samples.

[0047] In some embodiments, during the process of training the initial prediction model by the training set, the validation loss calculated by the loss function can also be compared with the historical optimal validation loss. If the current validation loss is less than the historical optimal loss, the optimal loss value is updated, and the current corresponding model parameters are saved. If the current validation loss is greater than or equal to the historical optimal loss, the parameters of the historical optimal model are maintained unchanged, and at the same time, the model parameter iteration based on the training set is continued, the loss is tried to be reduced by adjusting the internal weight of the model, until the validation loss is improved or the maximum iteration number is reached. In this way, it can be ensured that each time the validation loss is improved, the record is updated and the model is saved, so as to retain the model version with the best performance.

[0048] After the above operation is completed, the Bayesian optimization method can also be used to carry out hyperparameter optimization work to find the optimal combination of hyperparameters to improve the target coal consumption prediction model. Based on this, the target coal consumption prediction model can be obtained through the above series of processes based on sequence data of different time scales, and the final target coal consumption prediction model can be applied to coal consumption prediction under different time intervals, which can accurately predict the future coal consumption trend of the power plant.

[0049] For example, in the coal consumption prediction, the time series data can be divided into multiple window sequences by a sliding window, which is used as a training set to train the initial prediction model. Different hyperparameter combinations can be tried each time, such as adjusting the number of hidden layer nodes of the neural network, changing the number of LSTM units, etc. After the training is completed, the prediction error of the model under this set of hyperparameters on the validation set can be recorded, and these error data and the corresponding hyperparameters constitute the samples required for optimization. The optimal hyperparameter combination found by Bayesian optimization can make the target coal consumption prediction model more accurately mine the potential relationship between parameters when processing multi-scale data divided by sliding windows, improve the prediction accuracy of future coal consumption, and provide strong support for optimizing fuel scheduling and reducing costs in power plants.

[0050] In S104, a target prediction time is obtained, a plurality of prediction operation data of different time scales are obtained from a preset database according to the target prediction time, and the plurality of prediction operation data are integrated to obtain a prediction operation data set.

[0051] In some embodiments, the operation data of the target thermal power unit can be collected in real time by SIS, and the collected operation data can be stored in a pre-set storage space. For example, the operation data can be stored in a pre-configured MySQL database, which can provide real-time data support for subsequent prediction. In the subsequent process, the target coal consumption prediction model can call the real-time operation data stored in the MySQL database as the input parameter of the subsequent model, and then the coal consumption prediction can be performed according to the coal consumption prediction model of different time scales such as time scale, day scale and week scale, to adapt to the needs of different scenarios such as fuel procurement and scheduling of coal-fired power plants. It should be understood that the current time operation data corresponding to the key influence factors can also be obtained, and the current time operation data corresponding to other parameters can also be obtained. The specific key influence factors are not limited. It should be understood that the MySQL database is a popular relational database management system (RDBMS), which is operated based on SQL (Structured Query Language).

[0052] Figure 2 A flowchart of data selection and collection in a power plant coal consumption prediction model prediction module is provided for an exemplary embodiment of the present disclosure. As shown in FIG. 1, the process of data selection and collection in the power plant coal consumption prediction model prediction module includes the following steps. Figure 2As shown, the time index of the required historical operation data can be determined in combination with the prediction requirements of different time scales, such as the hour scale, the day scale and the week scale, based on the time of performing the prediction plan. Subsequently, the system obtains the historical operation data corresponding to different time scales from the MySQL database according to the time index, and exports the historical operation data. After the export, the historical operation data corresponding to different time scales is merged to generate a preset operation data set of different time scales, forming a unified prediction input, so as to ensure the time pertinence and integrity of the data and provide accurate input data for subsequent coal consumption prediction. Specifically, steps S201 to S205 are included: S201: Obtain a target prediction time. Here, the target prediction time is the time of predicting the coal consumption of the target thermal power unit.

[0053] S202: Determine a target time scale. Here, the target time scale can be any one of the hour scale, the day scale or the week scale.

[0054] S203: Determine a time index according to the target time scale.

[0055] Here, when determining the time index according to the target time scale, the target time scale required for prediction is first determined, and then the time range of the required historical operation data is automatically calculated based on the time of performing the prediction and in combination with the length requirement of the historical operation data corresponding to the target time scale, to form a time index list, so as to accurately locate and obtain the historical operation data of the corresponding time period from the database.

[0056] For example, taking the execution of the week scale coal consumption prediction at 14:00 on June 15, 2025 as an example, it can be known that the week scale prediction needs to cover 7 days of historical operation data, i.e. 168 hours; then, taking the prediction time 2025-06-15 14:00 as a reference, 168 hours are traced back to determine that the time range of the historical operation data is from 2025-06-08 14:00 to 2025-06-15 13:59. Based on this, the time index of each minute in this period (such as 2025-06-08 14:00:00, 2025-06-08 14:01:00, ……, 2025-06-15 13:59:00) can be automatically generated.

[0057] S204: Obtain historical data matching the target time scale according to the target time scale. Specifically, S204 includes S2041 to S2043.

[0058] S2041: When the target time scale is the hour scale, query and obtain the historical operation data corresponding to the hour scale in the SQL database according to the small time interval time column index.

[0059] S2042: When the target time scale is day scale, the historical running data corresponding to the day scale is queried and obtained in the SQL database according to the day interval time column index.

[0060] S2043: When the target time scale is week scale, the historical running data corresponding to the week scale is queried and obtained in the SQL database according to the week interval time column index.

[0061] In some embodiments, for steps S2041-S2043, the corresponding historical running data can be directly queried and obtained in the SQL database according to the time index, and steps S2041-S2043 can be processed in parallel or in a preset order.

[0062] S205: The historical running data corresponding to the hour scale, the historical running data corresponding to the day scale, and the historical running data corresponding to the week scale are merged to obtain a preset running data set. Here, the merging can be to merge all the small time interval data, the day interval data, and the week interval data into one file.

[0063] S105, inputting the target running data set in the preset running data set matching the target prediction scale into the pre-trained target coal consumption prediction model to obtain the coal consumption prediction result corresponding to the target prediction scale.

[0064] Figure 3 A flowchart of a model prediction module data condition screening method is provided for an exemplary embodiment of the present disclosure. As shown in Figure 3 After obtaining the target running data set, the historical running data in the target running data set can be screened and processed to eliminate data that may contain errors, be incomplete or irrelevant, and ensure that the data format is consistent with the model input format to ensure the effectiveness of the model training.

[0065] S301, whether the data amount of the target running data set meets a preset data amount.

[0066] In some embodiments, the time point of the required historical running data can be calculated based on the target prediction time, an index list is generated, and the corresponding historical running data is queried from the MySQL database, and different preset data amounts are set for different time scales, for example: 168 time step data (corresponding to 7 days x 24 hours) are required for hour scale prediction; 20 time step data (about 3 weeks) are required for day scale prediction; 4 time step data (about 1 month) are required for week scale prediction.

[0067] Here, if the data result in the preset data amount is empty or the data amount of the preset running data set does not satisfy the preset value, such as the hourly scale data amount < 168, step S304 is executed.

[0068] S302, when the data amount of the preset running data set satisfies the preset value, it is determined whether the preset running data set is in a steady state running interval.

[0069] In some embodiments, a clustering technique (such as a Gaussian mixture model, k-means) can be used to denoise the historical running data in the test data set, and filter out the time period in which the thermal power unit is stably running. In this way, the interference of non-steady state conditions, such as start-stop furnace, severe load fluctuation, etc., on the prediction accuracy can be excluded, and the representativeness of the data features is ensured. When the preset running data set is not in the steady state running interval, step S304 is executed.

[0070] S303, when the data amount of the preset running data set satisfies the preset value, and the preset running data set is the data set corresponding to the steady state running interval, it is determined whether the data in the preset running data set is continuous.

[0071] In some embodiments, the time index in the test data set can be traversed, and the time difference between adjacent data points can be calculated If there is a discontinuous time interval, such as a 2-minute interval between two time points, or a missing data point, the missing time is prompted and step S304 is executed.

[0072] Specifically, when the data amount of the preset running data set satisfies the preset value, the preset running data set is the data set corresponding to the steady state running interval, and the data in the preset running data set is continuous, step S305 is executed; otherwise, step S304 is executed.

[0073] S304, the preset running data set is empty.

[0074] In some embodiments, only when the data amount of the preset running data set meets the standard, is in the steady state interval, and the time sequence is continuous, the test data set will be determined to be valid and passed to the subsequent model prediction link; otherwise, an empty preset running data set is passed to S305, and the empty preset running data set is output in step S305, so that the prediction deviation caused by invalid data can be avoided.

[0075] S305, output the preset running data set.

[0076] S306, determine whether the preset running data set is empty.

[0077] In some embodiments, for the input preset running data set, it is first determined whether the preset running data set is empty. An empty set indicates that the required data for the current prediction time does not meet the requirements of the target coal consumption prediction model, and prediction cannot be performed. At this time, null values are directly filled in the prediction result, that is, S308 is executed.

[0078] Specifically, in the case where the preset running data set is empty, a null value prediction result is output, that is, S308 is executed, and the prediction result output by step S308 is null.

[0079] S307, in the case where the preset running data set is not empty, the preset running data set is input into the target coal consumption prediction model for processing to obtain a prediction result.

[0080] In some embodiments, if the preset running data set is not empty, it is input into the model for processing: first, the configuration file corresponding to the current time scale is read to obtain the input requirements, parameter settings and other information of the model under the time scale; then the data set is loaded, and the MinMax standardization method is used to pre-process the data in the preset running data set to linearly transform the data to the [0, 1] interval to eliminate the dimension influence of different features; then the pre-trained Transformer model file corresponding to the time scale is loaded, the data set after standardization processing is input into the model, the self-attention mechanism of the model is used to capture the long-time dependence and nonlinear characteristics in the time series, and the coal consumption prediction is completed; finally, the prediction result is output in the form of DataFrame to clearly and structurally pass to the subsequent prediction result writing and other processing links.

[0081] S308, output the prediction result.

[0082] In some embodiments, the finally obtained prediction result can also be written into a MySQL database according to a set format for front-end display and power plant fuel procurement and operation scheduling analysis. Specifically, the database format can be set as: execution batch, unit number, prediction time scale, prediction time and prediction value. The final prediction result is written into the local server MySQL database table.

[0083] In summary, the embodiments of the present disclosure can first perform preprocessing such as outlier rejection, missing value filling and steady state screening on historical operation data to ensure data quality; then through a multi-scale analysis method, the time series is divided into different time granularity datasets such as time scale, daily scale and weekly scale to meet the needs of different prediction scenarios; then with the help of the self-attention mechanism of the initial prediction model, the long-time dependence and nonlinear characteristics in the time series are effectively captured to obtain the target coal consumption prediction model; at the same time, the hyperparameters of the target coal consumption prediction model are automatically optimized through Bayesian optimization and genetic algorithm, further improving the prediction accuracy and generalization ability of the target coal consumption prediction model. In addition, the system realizes real-time data processing and prediction through the interface with SIS, and finally provides accurate and real-time coal consumption prediction results for power plants, and provides strong support for fuel procurement and dispatching optimization decision-making.

[0084] The method for predicting coal consumption of a thermal power unit provided by the embodiments of the present disclosure can be executed by a terminal or a chip applied to the terminal.

[0085] For example, the terminal can include one or more of a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, a Personal Digital Assistant (PDA), and a wearable device based on augmented reality (AR) and / or virtual reality (VR) technology, and the like. The terminal can also include, but is not limited to, a remote control device, a wearable device, a street lamp, a smart terminal of a household appliance, and the like, and the embodiments of the present disclosure do not make specific limitations thereto.

[0086] Figure 4 A flowchart of a method for predicting coal consumption of a thermal power unit is provided for an exemplary embodiment of the present disclosure. As shown in Figure 4 The method for predicting coal consumption of a thermal power unit includes: S401, determining a target prediction period, a target prediction time and a target time scale.

[0087] In some embodiments, the determined target prediction period, target prediction time and target time scale can lay a clear time dimension framework for subsequent coal consumption prediction work; at the same time, the clear target prediction period defines the time range of the prediction, which gives the prediction a clear boundary; the target prediction time accurately anchors the starting reference point of the prediction, providing a core reference for data screening and analysis; the target time scale determines the granularity of the prediction, whether it is a sub-scale or clock scale for short-term accurate regulation, or a day scale or week scale for long-term planning, can make the subsequent data screening and model prediction work revolve around this time characteristic, effectively improve the pertinence and adaptability of the prediction, and ensure that the prediction result meets the application requirements of different scenarios.

[0088] It should be understood that the target prediction time herein can be a time in the historical process, and can also be the current time, and the target time scale and the target prediction period can be set according to actual conditions. For example, the target time scale can include hour scale, day scale and week scale, and the target prediction period can be one week, one month or one year, etc.

[0089] S402, according to the target prediction time, the target running data set matched with the target time scale and the target prediction period is screened out from the preset running data set corresponding to the target thermal power unit.

[0090] In some embodiments, the target prediction time can be used as the starting point for screening, and the target time scale and the period can be combined to accurately screen the target running data set that meets the prediction requirements from the preset running data set, avoiding irrelevant data interference, ensuring that the data input into the model is highly consistent with the time characteristics of the prediction target, and thus the data quality and the accuracy and effectiveness of the model prediction can be effectively improved.

[0091] For example, assuming that in the power plant coal consumption prediction scenario, the target prediction period is set as 2025 / 7 / 1 0:00-2025 / 7 / 1 24:00, the target prediction time is 2025 / 7 / 1 0:00, and the target time scale is 1 hour. Since the running data of the target thermal power unit can be collected in real time through SIS in general, and the collected running data can be stored in a pre-set storage space, and the running data collected at different times can be sorted in chronological order in the storage space, and then the preset running data set can be constructed according to the running data corresponding to different times.

[0092] Based on this, starting from 0:00 on July 1, the data within 1 hour before and after each hour on July 1 can be screened out according to the 1-hour time scale, such as the data in each period of 0:00-1:00, 1:00-2:00, etc., and finally the target running data set in units of hours covering the entire prediction period is formed, which is used for subsequent coal consumption prediction model analysis.

[0093] S403, input the target operation data set into the pre-trained target coal consumption prediction model to obtain the corresponding coal consumption of the target thermal power generating unit at the target time scale within the target prediction period.

[0094] In some embodiments, inputting the target operation data set into the pre-trained target coal consumption prediction model is a key link for converting data value into prediction results; relying on the deep mining ability of the model for data characteristics, combined with the data set matched with the target time scale and period, the model can output high-precision prediction results that meet actual needs. This process not only guarantees the accuracy of the time dimension of the prediction results, but also improves the prediction reliability through model training optimization, which can provide scientific basis for energy management, cost control and other decisions.

[0095] For example, assuming that in a certain thermal power plant, the target prediction period is set to August 15, 2025, 8:00-12:00, the target time scale is 30 minutes, and the target operation data set containing unit power, steam pressure, environmental temperature and other parameters is obtained through screening. Then the target operation data set can be input into the pre-trained target coal consumption prediction model, the target coal consumption prediction model can automatically extract the correlation characteristics of each operation data and coal consumption in the target operation data set, and finally output the corresponding coal consumption prediction results every 30 minutes in this period, i.e. predicted coal consumption. For example, the predicted coal consumption from 8:00 to 8:30 is 5.2 tons, and the predicted coal consumption from 8:30 to 9:00 is 5.5 tons, etc. to help the power plant plan fuel reserves and unit scheduling in advance.

[0096] As can be seen, the embodiments of the present disclosure can determine the target prediction period, the target prediction time and the target time scale; according to the target prediction time, the target operation data set matched with the target time scale and the target prediction period is screened from the pre-set operation data set corresponding to the target thermal power generating unit; the target operation data set is input into the pre-trained target coal consumption prediction model to obtain the corresponding coal consumption of the target thermal power generating unit at the target time scale within the target prediction period. In this way, by accurately setting the target prediction period, the target prediction time and the target time scale, the matched target operation data set is screened with the target time scale as the core, and the target operation data set is input into the pre-trained model, the deep coupling of the prediction process and the time scale is realized, and then the coal consumption prediction results meeting the requirements of the target time scale can be output, thereby effectively solving the problem that the traditional single model cannot consider the prediction accuracy of multiple time dimensions, and eliminating the error of manual scale conversion, providing real-time to strategic-level full-cycle decision support for power plant fuel scheduling.

[0097] In some embodiments, the target time scale includes any one of a hour scale, a week scale and a day scale, and the target running data set matched with the target time scale is filtered from the preset running data set according to the target prediction time point and the target time scale, including: obtaining a plurality of historical running data corresponding to different time points within the target prediction time period from the preset database, and filtering the plurality of historical running data according to different time scales; constructing a running data set corresponding to different time scales according to the filtering result, and constructing a preset running data set according to the running data set corresponding to different time scales; and filtering the target running data set matched with the target time scale from the preset running data set.

[0098] It can be seen that the data processing flow of the embodiments of the present disclosure realizes fine management and efficient use of historical running data through hierarchical filtering and integration. First, the historical running data within the target prediction time period is extracted from the preset database to provide raw materials for subsequent analysis; then, the data is filtered based on different time scales to construct a multi-dimensional running data set and form a preset running data set covering multiple time granularities, thereby ensuring the comprehensiveness and flexibility of the data; finally, the target running data set is accurately filtered from the preset data set according to the target time scale, so as to ensure that the data input into the prediction model is highly consistent with the prediction demand in the time dimension, effectively improve the pertinence of the data and the accuracy of the prediction result, and lay a solid data foundation for subsequent coal consumption prediction and other work. Here, the specific data processing process can be referred to the corresponding content in the foregoing Figure 2

[0099] In some embodiments, filtering the target running data set matched with the target time scale from the preset running data set includes: filtering a candidate running data set matched with the target time scale from the preset running data set; and when the data amount of the candidate running data set meets a preset data amount, and / or the candidate running data set is a data set corresponding to a steady-state running interval, and / or a plurality of data in the candidate running data set is continuous, determining that the candidate running data set is the target running data set.

[0100] ​Specifically, after obtaining the preset operation data set, the historical operation data in the preset operation data set can also be screened and processed to eliminate data that may contain errors, incompleteness or irrelevance and ensure that the data format is consistent with the model input format to ensure the effectiveness of subsequent model training. Then, the candidate operation data set can be selected from the preset operation data set, and the data range conforming to the target time scale is preliminarily framed to narrow the screening range. The multiple judgment conditions such as "preset data amount", "steady state operation interval" and "data continuity" perform deep checking from the dimensions of data size, working condition stability and data integrity. Only when the data amount is sufficient, in a steady state operation state, and the data is not missing and interrupted, it is determined as the target operation data set, effectively avoiding the prediction deviation caused by insufficient data amount, working condition fluctuation or data fault, providing stable and reliable input data for the subsequent coal consumption prediction model, and ensuring the accuracy and reliability of the prediction result. Here, the specific data screening process can refer to the above Figure 3 Corresponding content.

[0101] In some embodiments, the target operation data set is input into the pre-trained target coal consumption prediction model, and the method further comprises: determining a plurality of key influence factors, and obtaining corresponding original operation data of the plurality of key influence factors at different time points from the preset database; constructing a training set according to the original operation data of the key influence factors at different time points; and training the initial prediction model using the training set to obtain the target coal consumption prediction model.

[0102] Specifically, a plurality of key influence factors affecting the coal consumption of the target thermal power unit can be determined first, such as unit load, coal feeder coal supply amount, environmental temperature and other key influence factors strongly related to coal consumption prediction; then, the original operation data corresponding to each key influence factor and the actual coal consumption at different time points can be collected at a fixed time interval within the target prediction period. Here, all original operation data included in the target prediction period can be obtained from the Supervisory Information System (SIS) of the coal-fired power plant, and then the original operation data corresponding to each key influence factor can be selected from all original operation data. The key influence factors of the embodiments of the present disclosure are shown in Table 1. Then, a training set can be constructed according to the original operation data of the key influence factors at different time points; and the initial prediction model is trained using the training set to obtain the target coal consumption prediction model.

[0103] In some embodiments, the training set is constructed according to the original operation data of the key influence factors at different time points, and the method further comprises: First, the original operation data is preprocessed to obtain the to-be-processed operation data.

[0104] Specifically, in view of the interference information that may exist in the original running data due to abnormal fluctuations or unstable working conditions, the original running data can be preprocessed by using a Gaussian mixture model and a mean clustering algorithm (k-means) to obtain a plurality of to-be-processed running data. For example, the distribution characteristics of the original running data can be identified by using the Gaussian mixture model, and then the data clusters are divided by combining the k-means algorithm, so as to accurately remove outliers and separate the to-be-processed running data of the target thermal power generating unit under stable working conditions.

[0105] It should be understood that in actual application, the specific key influence factor can also not be limited, that is, the original running data can involve a plurality of influence factors, which include not only the key influence factor but also other influence factors. At this time, the original running data corresponding to all influence factors can also be directly preprocessed to obtain a large amount of target running data. Here, one influence factor corresponds to one original running data or target running data at one time point.

[0106] Then, the to-be-processed running data is integrated in time sequence to obtain a time series data set.

[0107] Specifically, in some embodiments, the multi-dimensional to-be-processed running data obtained after preprocessing can be arranged in time sequence to form a structured time series data set, so as to provide a high-quality data basis for subsequent training and optimization of the thermal power generating unit coal consumption prediction model. For a time series, a group of data arranged in time sequence, each data corresponds to a unique time point, reflects the change of a certain key influence factor in the time dimension, and then the time series data set can be constructed according to the time series corresponding to a plurality of key influence factors.

[0108] For example, the key influence factors and the key influence factors The time series corresponding to the target prediction period can be represented as:

[0109]

[0110] wherein, represents the to-be-processed running data corresponding to the key influence factor a at the 1st time point, the 1st time point can be the initial time point in the target prediction period, represents the to-be-processed running data corresponding to the key influence factor a at the 2nd time point, represents the to-be-processed running data corresponding to the key influence factor a at the mth time point, the mth time point can be the last time point in the target prediction period; denotes the to-be-processed running data corresponding to the key influence factor b at a first time point, the first time point being an initial time point in the target prediction period, denotes the to-be-processed running data corresponding to the key influence factor b at a second time point, denotes the to-be-processed running data corresponding to the key influence factor b at an mth time point, the mth time point being a last time point in the target prediction period.

[0111] Based on this, the time series data set may be constructed according to the key influence factor , the key influence factor , and other key influence factors. The time series data set may be expressed as:

[0112] Then, the time series data set can be divided into a plurality of time window corresponding time series subsets according to a plurality of preset time windows.

[0113] Specifically, in order to perform multi-time scale coal consumption prediction, the processed time series data set may be divided according to different time window lengths and time intervals. Based on this, the target coal consumption prediction model finally constructed can be constructed to be applicable to different time scales such as hour scale, day scale, and week scale by adjusting the time window length L and the time interval Δt.

[0114] For example, assuming that the window length of the sliding window is L, the time interval is Δt, and the window sliding step is s, when the time series data set is divided, the sliding window mechanism is used as the basis, a continuous data segment is intercepted according to the set window length, and a plurality of overlapping or non-overlapping subsequences are generated by sliding the window at a fixed step s, and finally a window sequence composed of time step data is obtained , so as to realize multi-scale processing of the time series data and provide input data of different time granularities for subsequent coal consumption prediction model training.

[0115] For example, assuming that the window length of the sliding window is L, the time interval is Δt, and the window sliding step is s, the time series data set can be divided by the following formula combined with the sliding window to obtain the window sequence of time step data , and the divided is:

[0116] wherein, ​​a time series subset corresponding to the t-th time step, a time series subset corresponding to the t-th time step, a time series subset corresponding to the t-th time step, a time series subset corresponding to the t-th time step, a time series subset corresponding to the t-th time step, a time series subset containing L time steps of data starting from the t-th time step; Here, q represents the identification of the key influence factor, and the to-be-processed running data corresponding to the t-th time point of the key influence factor a, the to-be-processed running data corresponding to the t-th time point of the key influence factor b, the to-be-processed running data corresponding to the t-th time point of the key influence factor q.

[0117] Here, by setting the sliding step s, the time window can be slid on the original time series at a fixed step, generating overlapping or non-overlapping time series subsets. For example, if the length of the time window is L and the sliding step s = 1, each new time window only moves forward by 1 time step, forming a highly overlapping time series subset; if s = L, the time windows are non-overlapping.

[0118] Finally, a training set can be constructed according to the multiple time series subsets. That is, a certain number of time series subsets can be selected from different time series subsets, and these selected time series subsets can be used as a training set to train the initial prediction model.

[0119] In some embodiments, after dividing the time series data set according to the preset multiple time windows to obtain multiple time series subsets corresponding to the multiple time windows, and before constructing a training set according to the multiple time series subsets, the method further includes: when the multiple to-be-processed running data in the multiple time series subsets all satisfy the time interval uniformity condition, constructing a training set according to the multiple time series subsets. Here, the time interval uniformity condition includes that the time intervals corresponding to two adjacent to-be-processed running data in the time series subset are the same.

[0120] Specifically, in order to ensure that the time intervals of the input time series and the prediction target time series are uniform, the time difference between adjacent data points can also be calculated to verify whether the time intervals of the time series and the prediction target time series are consistent, and the specific calculation formula is:

[0121] Wherein, data_index represents the time index in the time series subset. If all time intervals meet the expectation, i.e. Δt=interval, then the time series is regarded as a valid input, wherein, interval represents the time interval of data sampling. Further, the time series subset of multiple valid inputs can be divided into a test set and a training set to train the initial prediction model to obtain the target coal consumption prediction model.

[0122] The initial prediction model is trained using the training set to obtain the target coal consumption prediction model. The method further comprises: constructing a validation set according to the original operation data corresponding to the key influence factors at different time points; determining the target model parameters of the target coal consumption prediction model using the Bayesian optimization method according to the validation set; and determining the target model parameters of the target coal consumption prediction model based on the prediction error of the validation set by a preset model parameter determination method.

[0123] In some embodiments, the initial prediction model can be trained using the training set with mean square error (MSE) as the loss function, the optimal model parameters are updated by comparing the validation loss, and the optimal hyperparameter combination is found by Bayesian optimization. Here, mean square error (MSE) is a commonly used index for measuring the deviation degree of the predicted coal consumption of the model from the actual coal consumption.

[0124] Specifically, the initial prediction model can be trained using the training set, and the trained target coal consumption prediction model can be tested using the test set. Finally, the mean square error (MSE) of the predicted coal consumption and the actual coal consumption is calculated as a loss function to measure the prediction performance of the target coal consumption prediction model:

[0125] Wherein, Y is the time series corresponding to the actual coal consumption, i.e. the sequence corresponding to the actual coal consumption of the coal consumption of the power plant; is the time series corresponding to the predicted coal consumption, i.e. the sequence corresponding to the predicted coal consumption obtained by the target coal consumption prediction model; the actual coal consumption at the i-th time point; the predicted coal consumption at the i-th time point; and R represents the total number of data samples.

[0126] In some embodiments, during the process of training the initial prediction model with the training set, the validation loss calculated by the loss function can also be compared with the historical optimal validation loss; if the current validation loss is less than the historical optimal loss, the optimal loss value is updated, and the current corresponding model parameters are saved; if the current validation loss is greater than or equal to the historical optimal loss, the parameters of the historical optimal model are maintained unchanged, and at the same time, the model parameter iteration based on the training set will be continued, and the loss will be tried to be reduced by adjusting the internal weights of the model until the validation loss is improved or the maximum iteration number is reached. In this way, it can be ensured that each time the validation loss is improved, the record is updated and the model is saved, so as to retain the model version with the best performance.

[0127] After the above operations are completed, the Bayesian optimization method can also be used to carry out hyperparameter optimization work to find the optimal hyperparameter combination to improve the target coal consumption prediction model. Based on this, the target coal consumption prediction model can be obtained based on the sequence data of different time scales through the above series of processes, and the final target coal consumption prediction model can be applied to the prediction of coal consumption at different time intervals, which can accurately predict the future coal consumption trend of the power plant.

[0128] For example, in coal consumption prediction, time series data can be divided into multiple window sequences by a sliding window, which is used as a training set to train the initial prediction model. Different hyperparameter combinations can be tried each time, such as adjusting the number of hidden layer nodes of the neural network, changing the number of LSTM units, etc. After training is completed, the prediction error of the model under this set of hyperparameters on the validation set can be recorded, and these error data and the corresponding hyperparameters constitute the samples required for optimization. The optimal hyperparameter combination found by Bayesian optimization can make the target coal consumption prediction model more accurately mine the potential relationship between parameters when processing multi-scale data divided by sliding windows, improve the prediction accuracy of future coal consumption, and provide strong support for optimizing fuel scheduling and reducing costs of power plants.

[0129] It can be seen that, by determining the target prediction period, the target prediction time, and the target time scale, according to the target prediction time, the target running data set matched with the target time scale and the target prediction period is filtered from the preset running data set corresponding to the target thermal power generating unit, and the target running data set is input into the target coal consumption prediction model pre-trained to obtain the coal consumption of the target thermal power generating unit corresponding to the target time scale in the target prediction period. In this way, by accurately setting the target prediction period, the target prediction time, and the target time scale, the matched target running data set is filtered with the target time scale as the core, and the target running data set is input into the pre-trained model, the deep coupling of the prediction process and the time scale is realized, and then the coal consumption prediction result meeting the requirement of the target time scale can be output, thereby the problem that the traditional single model is difficult to consider the prediction accuracy of multiple time dimensions can be effectively solved, and the error of manual scale conversion is avoided, and real-time to strategic-level full-cycle decision support for power plant fuel scheduling is provided.

[0130] The above describes the scheme provided by the embodiments of the present disclosure from the perspective of the server. It can be understood that, in order to realize the above functions, the server comprises a hardware structure and / or a software module for executing each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present disclosure can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is realized by hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present disclosure.

[0131] The embodiments of the present disclosure can divide the functions of the server according to the above method examples, for example, each function module can be divided according to each function, or two or more functions can be integrated in one management module. The above integrated module can be realized in the form of hardware or in the form of a software function module. It should be noted that the division of the modules in the embodiments of the present disclosure is illustrative, and is only a logical function division. When actually implemented, there can be another division method.

[0132] In the case of dividing each function module according to each function, the exemplary embodiments of the present disclosure provide a thermal power generating unit coal consumption prediction device, which can be a server or a chip applied to a server. Figure 5 A structural schematic diagram of a thermal power generating unit coal consumption prediction device provided by an embodiment of the present disclosure is shown in FIG. 5. As shown in FIG. 5, the thermal power generating unit coal consumption prediction device 500 comprises: Figure 5 ​The determining module 501 is configured to determine a target prediction period, a target prediction time point, and a target time scale; The obtaining module 502 is configured to filter, from a preset running data set corresponding to the target thermal power unit according to the target prediction time point, a target running data set matching the target time scale and the target prediction period. The prediction module 503 is configured to input the target running data set into a pre-trained target coal consumption prediction model to obtain a coal consumption corresponding to the target thermal power unit at the target time scale within the target prediction period.

[0133] In an optional manner, the target time scale includes any one of a time scale of an hour, a week, and a day, the obtaining module 502 is further configured to obtain a plurality of historical running data corresponding to different time points within the target prediction period from a preset database, and filter the plurality of historical running data according to different time scales; construct a running data set corresponding to different time scales according to a filtering result, and construct the preset running data set according to the running data set corresponding to different time scales; and filter the target running data set matching the target time scale from the preset running data set.

[0134] In an optional manner, the obtaining module 502 is further configured to filter a candidate running data set matching the target time scale from the preset running data set; and when a data amount of the candidate running data set meets a preset data amount, and / or the candidate running data set is a data set corresponding to a steady-state running interval, and / or a plurality of data in the candidate running data set is continuous, determine that the candidate running data set is the target running data set.

[0135] In an optional manner, the prediction module 503 is further configured to input the target running data set into the pre-trained target coal consumption prediction model, and the method further includes: determining a plurality of key influence factors, and obtaining original running data of the plurality of key influence factors corresponding to different time points from a preset database; constructing a training set according to the original running data of the key influence factors corresponding to different time points; and training an initial prediction model by using the training set to obtain the target coal consumption prediction model.

[0136] In an optional mode, the coal consumption prediction device 500 of the thermal power generating unit comprises an obtaining module 504, which is configured to pre-process the original operation data to obtain to-be-processed operation data, sequentially integrate the to-be-processed operation data to obtain a time series data set, divide the time series data set according to a plurality of preset time windows to obtain a plurality of time series subsets corresponding to the plurality of time windows respectively, and construct the training set according to the plurality of time series subsets.

[0137] In an optional mode, the obtaining module 504 is further configured to construct the training set according to the plurality of time series subsets when the plurality of to-be-processed operation data in the plurality of time series subsets all satisfy a time interval uniformity condition.

[0138] In an optional mode, the time interval uniformity condition comprises that time intervals corresponding to two adjacent to-be-processed operation data in the time series subset are the same.

[0139] In an optional mode, the obtaining module 504 is further configured to construct a verification set according to the original operation data corresponding to the key influence factors at different time points, determine target model parameters of the target coal consumption prediction model by using a Bayesian optimization method according to the verification set, and determine the target model parameters of the target coal consumption prediction model by a preset model parameter determination method based on a prediction error of the verification set.

[0140] The embodiments of the present disclosure further provide an electronic device, which comprises at least one processor, a memory for storing at least one processor-executable instruction, and wherein the at least one processor is configured to execute the instruction to implement the steps of the above-mentioned method disclosed by the embodiments of the present disclosure.

[0141] Figure 6 The structural schematic diagram of the electronic device provided by an embodiment of the present disclosure is shown in FIG. 6. As shown in FIG. 6, the electronic device 600 comprises at least one processor 601 and a memory 602 coupled to the processor 601, and the processor 601 can execute the corresponding steps in the above-mentioned method disclosed by the embodiments of the present disclosure. Figure 6

[0142] ​The processor 601 can also be referred to as a central processing unit (CPU), which can be an integrated circuit chip that has the processing capability of signals. Each step in the method disclosed in the embodiments of the present disclosure can be completed by the integrated logic circuit of hardware or the instructions in the form of software in the processor 601. The processor 601 can be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly embodied as hardware code processing for execution, or executed by a combination of hardware and software modules in the code processing. The software module can be located in the memory 602, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, and other mature storage media in the art. The processor 601 reads the information in the memory 602, and completes the steps of the above method in combination with the hardware thereof.

[0143] In addition, various operations / processes according to the present disclosure, when implemented by software and / or firmware, can be downloaded from a storage medium or a network to a computer system with a special hardware structure, for example, Figure 7 The computer system 700 shown is intended to represent various forms of digital electronic computer equipment, such as a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic equipment can also represent various forms of mobile devices, such as a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are merely examples, and are not intended to limit the implementations of the present disclosure described and / or claimed herein. Figure 7 A structural schematic diagram of a computer system provided for an embodiment of the present disclosure.

[0144] The computer system 700 is intended to represent various forms of digital electronic computer equipment, such as a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic equipment can also represent various forms of mobile devices, such as a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are merely examples, and are not intended to limit the implementations of the present disclosure described and / or claimed herein.

[0145] As Figure 7As shown, the computer system 700 includes a computing unit 701 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 702 or a computer program loaded into a random access memory (RAM) 703 from a storage unit 708. Various programs and data required for the operation of the computer system 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0146] A plurality of components in the computer system 700 are connected to the I / O interface 705, including an input unit 706, an output unit 707, a storage unit 708, and a communication unit 709. The input unit 706 can be any type of device that can input information to the computer system 700, and can receive inputted digital or character information, and generate key signal inputs related to user settings and / or function controls of the electronic device. The output unit 707 can be any type of device that can present information, and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 708 can include, but is not limited to, a magnetic disk, an optical disk. The communication unit 709 allows the computer system 700 to exchange information / data with other devices through a network such as the Internet, and can include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, e.g., a Bluetooth™ device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0147] The computing unit 701 can be various general and / or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 701 performs various methods and processes described above. For example, in some embodiments, the above-described methods disclosed by embodiments of the present disclosure can be implemented as a computer software program tangibly embodied in a machine-readable medium, e.g., the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device via the ROM 702 and / or the communication unit 709. In some embodiments, the computing unit 701 can be configured to perform the above-described methods disclosed by embodiments of the present disclosure by any other appropriate means, e.g., by means of firmware.

[0148] The embodiment of the present disclosure further provides a computer readable storage medium, wherein when instructions in the computer readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the above method disclosed by the embodiment of the present disclosure.

[0149] The computer readable storage medium in the embodiment of the present disclosure can be a tangible medium, which can contain or store programs for use by or in connection with an instruction execution system, apparatus or device. The above computer readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the above. More specifically, the above computer readable storage medium can include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0150] The above computer readable medium can be included in the above electronic device; or can exist separately without being assembled into the electronic device.

[0151] Figure 8 A schematic diagram of a computer program product provided by an embodiment of the present disclosure is shown. As shown in the figure, the computer program product 800 includes a computer program 801, wherein the computer program 801 is executed by a processor to implement the above method disclosed by the embodiment of the present disclosure. Figure 8

[0152] In the embodiments of the present disclosure, computer program code for carrying out operations of the present disclosure can be written in one or more programming languages or combinations of the same, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on a user computer, partially on a user computer, as an independent software package, partially on a user computer and partially on a remote computer, or entirely on a remote computer or server. In the case involving a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer.

[0153] ​The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present disclosure. In this regard, each block in the flow diagrams or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flow diagrams, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0154] The modules, components or units described in the embodiments of the present disclosure can be implemented by software or by hardware. In some cases, the name of the module, component or unit does not constitute a limitation on the module, component or unit itself.

[0155] The functions described above can be performed by one or more hardware logic components. For example, non-limiting examples of hardware logic components that can be used 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), etc.

[0156] The above description is merely some embodiments of the present disclosure and a description of principles of technology used. It should be understood by those skilled in the art that the disclosed scope of the present disclosure is not limited to the technical solutions formed by the specific combinations of the above technical features, and also covers other technical solutions formed by any combinations of the above technical features or equivalent features without departing from the above disclosed concept. For example, the technical solutions formed by replacing the above features with technical features disclosed in the present disclosure (but not limited to) having similar functions.

[0157] Although some specific embodiments of the present disclosure have been described in detail by way of examples, it should be understood that the above examples are merely for illustration, and are not intended to limit the scope of the present disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A method for predicting coal consumption of thermal power units, characterized in that, include: Determine the target prediction period, target prediction time, and target time scale; Based on the target prediction time, a target operation dataset that matches the target time scale and the target prediction period is selected from the preset operation dataset corresponding to the target thermal power unit. The target operating dataset is input into a pre-trained target coal consumption prediction model to obtain the coal consumption of the target thermal power unit at the target time scale during the target prediction period.

2. The method according to claim 1, characterized in that, The target time scale includes any one of hourly, weekly, and daily scales. The step of selecting a target operating dataset that matches the target time scale and the target predicted time period from a preset operating dataset corresponding to the target thermal power unit, based on the target predicted time, includes: Multiple historical operation data corresponding to different times within the target prediction period are obtained from a preset database, and the multiple historical operation data are filtered according to different time scales; Based on the screening results, construct the running datasets corresponding to different time scales, and construct the preset running datasets based on the running datasets corresponding to different time scales; The target running dataset that matches the target time scale is selected from the preset running dataset.

3. The method according to claim 2, characterized in that, The step of selecting the target running dataset that matches the target time scale from the preset running dataset includes: Select candidate running datasets that match the target time scale from the preset running dataset; When the amount of data in the candidate running dataset meets the preset data amount, and / or the candidate running dataset is the dataset corresponding to the steady-state running interval, and / or multiple data in the candidate running dataset are consecutive, the candidate running dataset is determined as the target running dataset.

4. The method according to claim 1, characterized in that, The method of inputting the target running dataset into a pre-trained target coal consumption prediction model further includes: Multiple key influencing factors are identified, and the original operational data corresponding to the multiple key influencing factors at different times are obtained from a preset database. A training set is constructed based on the original operational data corresponding to the key influencing factors at different times. The initial prediction model is trained using the training set to obtain the target coal consumption prediction model.

5. The method according to claim 4, characterized in that, The method of constructing a training set based on the original operational data corresponding to the key influencing factors at different times further includes: The raw running data is preprocessed to obtain the running data to be processed; The data to be processed is integrated in chronological order to obtain a time-series dataset; The time series dataset is divided according to multiple preset time windows to obtain multiple time series subsets corresponding to the multiple time windows respectively; The training set is constructed based on multiple subsets of the time series.

6. The method according to claim 5, characterized in that, After dividing the time series dataset according to multiple preset time windows to obtain multiple time series subsets corresponding to the multiple time windows, before constructing the training set based on the multiple time series subsets, the method further includes: When multiple unprocessed data in multiple time series subsets all satisfy the time interval uniformity condition, the training set is constructed based on the multiple time series subsets.

7. The method according to claim 6, characterized in that, The time interval uniformity condition includes the fact that the time intervals corresponding to two adjacent data points to be processed in the time series subset are the same.

8. The method according to claim 4, characterized in that, The method of training the initial prediction model using the training set to obtain the target coal consumption prediction model further includes: A validation set is constructed based on the original operational data corresponding to the key influencing factors at different times. Based on the validation set, the target model parameters of the target coal consumption prediction model are determined using a Bayesian optimization method. Based on the prediction error of the validation set, the target model parameters of the target coal consumption prediction model are determined by a preset model parameter determination method.

9. A device for predicting coal consumption of thermal power units, characterized in that, include: The determination module is used to determine the target prediction period, the target prediction time, and the target time scale; The acquisition module is used to filter out the target operation dataset that matches the target time scale and the target prediction period from the preset operation dataset corresponding to the target thermal power unit according to the target prediction time. The prediction module is used to input the target operating dataset into a pre-trained target coal consumption prediction model to obtain the coal consumption of the target thermal power unit at the target time scale during the target prediction period.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 8.

Images