A method for decision-making on power coal purchase quantity based on deep learning

Abstract

The present invention discloses a method for decision-making on the purchase amount of electric coal based on deep learning, which belongs to the technical field of electric coal purchase. The historical regional electricity consumption characteristics of the power supply area are learned through a deep learning algorithm, so that the real-time regional electricity consumption characteristics can be identified in the subsequent process, and the predicted daily load can be determined. The final daily coal consumption can be determined according to the predicted daily load. Finally, whether electric coal purchase is needed and the amount of electric coal to be purchased can be determined according to the daily coal consumption. Analysis is performed from the load side to make the prediction more accurate and simple, which can effectively improve the decision-making efficiency of the purchase amount of electric coal and replace the traditional manual operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power coal procurement, and specifically relates to a power coal procurement quantity decision method based on deep learning. Background Art

[0002] With the rapid development of my country's economy, the demand for electricity is growing. As the main fuel for thermal power generation, the purchase volume of thermal coal has a significant impact on the production cost and economic benefits of power companies. Thermal coal refers to coal used for thermal power generation, which is the main fuel source for thermal power plants. The quality of thermal coal has a direct impact on the efficiency and environmental impact of thermal power generation. The traditional decision-making method for thermal coal procurement mainly relies on manual experience, which has problems such as inaccurate prediction and low efficiency. Therefore, it is of great significance to study a decision-making method for thermal coal procurement based on deep learning. Summary of the invention

[0003] The present invention provides a method for deciding the purchase quantity of electric coal based on deep learning, so as to solve the problems of inaccuracy and low efficiency caused by relying on manual experience for purchasing electric coal in the prior art.

[0004] A method for decision-making on the purchase quantity of thermal coal based on deep learning, comprising:

[0005] Collect the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant, and determine the unit load coal consumption according to the daily load and the daily coal consumption;

[0006] Collect historical regional electricity consumption characteristics of the power supply area and build a deep learning model for predicting daily load, and use the historical regional electricity consumption characteristics to train the deep learning model to obtain a trained deep learning model;

[0007] Collecting real-time regional power consumption characteristics of the power supply area, and using the real-time regional power consumption characteristics as input of the trained deep learning model to obtain predicted daily load;

[0008] Based on the predicted load and unit load coal consumption, the predicted coal consumption is determined, and the coal purchase quantity is determined based on the predicted coal consumption.

[0009] Furthermore, the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant are collected, and the unit load coal consumption is determined according to the daily load and the daily coal consumption, including:

[0010] Collect the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant;

[0011] The daily coal consumption of the power plant is divided by the daily load of the power supply area corresponding to the power plant to obtain the unit load coal consumption.

[0012] Furthermore, historical regional electricity consumption characteristics of the power supply area are collected, including:

[0013] Collect the daily average temperature, daily average humidity, rainfall, season type and date type of the power supply area to obtain the historical regional electricity consumption characteristics, and collect the daily electricity consumption corresponding to the historical regional electricity consumption characteristics;

[0014] The date types include working days, holidays, and non-working days that are not holidays.

[0015] Furthermore, a deep learning model for predicting daily load is constructed, including: constructing a BP deep learning model to obtain a deep learning model for predicting daily load.

[0016] Furthermore, the historical regional electricity consumption characteristics are used to train the deep learning model, and the trained deep learning model is obtained, including:

[0017] The historical regional electricity consumption characteristics are constructed as the model input vector corresponding to the deep learning model, and the daily electricity consumption corresponding to the historical regional electricity consumption characteristics is used as the expected value to obtain training data;

[0018] Initialize a coding population for the hyperparameters of the deep learning model; wherein the coding population includes multiple coding individuals, each coding individual includes part or all of the hyperparameters to be trained of the deep learning model;

[0019] The training data is used to obtain the fitness corresponding to each coding individual in the coding population, and according to the fitness of each coding individual, the coding individual with the largest fitness is determined as the global optimal individual, and the coding individual with the smallest fitness is determined as the global worst individual;

[0020] Based on the upper and lower limits corresponding to the coding individual, a nonlinear local search is performed on the coding individual in its neighborhood to obtain the coding individual after the local search;

[0021] For the coded individuals after the local search, all coded individuals are subjected to vortex search around the global optimal individual to obtain the coded individuals after the vortex search;

[0022] For the coding individual after the vortex search, the coding individual is made to learn the position information of the global optimal individual and the position information of the coding population to perform an extended search, thereby obtaining the coding individual after the extended search;

[0023] For the coding individuals after the extended search, use the global worst individual to perform random flight search on the coding individuals to obtain the coding individuals after the random flight search;

[0024] Repeat local search, vortex search, expansion search and random flight search until the training end conditions are met, re-obtain the global optimal individual, and decode the global optimal individual as the final hyperparameter of the deep learning model to obtain the trained deep learning model.

[0025] Furthermore, based on the upper and lower limits corresponding to the coding individual, a nonlinear local search is performed on the coding individual in its neighborhood to obtain the coding individual after the local search, including:

[0026] Determine the adaptive search step size based on the current number of training times;

[0027] Based on the upper and lower limits corresponding to the coded individuals, the search individuals formed by the search range are determined;

[0028] Generate a first random number between (0,1), and use the first random number, the adaptive search step size, and the search individual to determine the neighborhood search amount;

[0029] According to the neighborhood search amount, a local search is performed on the coding individual to obtain the coding individual after the local search.

[0030] Furthermore, for the coding individuals after the local search, all coding individuals are subjected to vortex search around the global optimal individual to obtain coding individuals after the vortex search, including:

[0031] For the encoded individuals after local search, the step size control factor is determined according to the current number of training times;

[0032] Generate a second random number between (0, 1), and determine the search direction using a cosine function based on the second random number;

[0033] For the coding individuals after local search, the difference between the coding individuals and the global optimal individual is obtained, and after the difference is controlled by the step size control factor, all coding individuals search around the global optimal individual in their respective search directions to form a vortex search, and the coding individuals after the vortex search are obtained.

[0034] Furthermore, for the coding individual after the vortex search, the coding individual is made to learn the position information of the global optimal individual and the position information of the coding population to perform an extended search, and the coding individual after the extended search is obtained, including:

[0035] For the coding individual after the vortex search, a third random number between (0, 1) is generated, and based on the third random number, a sine function is used to determine the coding individual position influence amount;

[0036] After fusing the position information of the global optimal individual and the position information of the coded population, the cosine function and random coefficients are used for interference to obtain the comprehensive position influence.

[0037] According to the coding individual position influence amount and the comprehensive position influence amount, the coding individual after the extended search is determined.

[0038] Furthermore, for the coding individuals after the extended search, a global worst individual is used to perform a random flight search on the coding individuals, and the coding individuals after the random flight search are obtained, including:

[0039] Based on the global worst individual, a random flying individual is determined;

[0040] For the coded individuals after the extended search, the random flight individuals are used to perform curve flight and the positions are adaptively adjusted to obtain the coded individuals after the random flight search;

[0041] Determine whether the fitness of the encoded individual after the random flight search increases. If so, accept the random flight search, otherwise reject the random flight search.

[0042] Further, based on the predicted load and the unit load coal consumption, the predicted coal consumption is determined, and the coal purchase quantity is determined based on the predicted coal consumption, including:

[0043] Multiplying the predicted load amount by the unit load coal consumption to obtain the predicted coal consumption;

[0044] Obtain the remaining amount of electric coal corresponding to the power plant, and divide the remaining amount of electric coal by the predicted electric coal consumption to obtain the number of days that the electric coal can be consumed;

[0045] Determine whether the number of days that electricity coal can be consumed is less than the preset required consumption days. If so, subtract the number of days that electricity coal can be consumed from the required consumption days to get the difference days, and multiply the predicted electricity coal consumption by the difference days to get the minimum electricity coal purchase quantity. Otherwise, it is determined that there is sufficient electricity coal and no purchase is required, and the electricity coal purchase quantity is determined to be zero.

[0046] The present invention provides a method for making decisions on the purchase amount of electric coal based on deep learning. The deep learning algorithm is used to learn the historical regional electricity consumption characteristics of the power supply area, so that the real-time regional electricity consumption characteristics can be identified in the future, and the predicted daily load can be determined. The final daily coal consumption can be determined according to the predicted daily coal consumption. Finally, whether electric coal purchase is needed and the amount of electric coal to be purchased can be determined according to the daily coal consumption. Analysis from the load side makes the prediction more accurate and simple, which can effectively improve the decision-making efficiency of the electric coal purchase amount and replace the traditional manual operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0048] Figure 1 A flowchart of a method for deciding the amount of power coal purchase based on deep learning provided in an embodiment of the present invention.

[0049] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0050] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0051] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0052] like Figure 1 As shown, an embodiment of the present invention provides a method for deciding the purchase amount of electric coal based on deep learning, comprising:

[0053] S1. Collect the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant, and determine the unit load coal consumption according to the daily load and the daily coal consumption;

[0054] S2. Collect historical regional electricity consumption characteristics of the power supply area and build a deep learning model for predicting daily load, and use the historical regional electricity consumption characteristics to train the deep learning model to obtain a trained deep learning model;

[0055] The historical regional electricity consumption characteristics of the power supply area can be some factors that affect the power load, such as the power consumption of the previous N days and the date, temperature, season and other influencing factors of the day. By learning these influencing factors, prediction can be achieved.

[0056] S3. Collecting the real-time regional power consumption characteristics of the power supply area, and using the real-time regional power consumption characteristics as the input of the trained deep learning model to obtain the predicted daily load;

[0057] After the deep learning model is trained with historical regional electricity consumption characteristics, the deep learning model has the ability to recognize features, so it can collect real-time regional electricity consumption characteristics for electricity consumption prediction.

[0058] S4. Determine the predicted power coal consumption based on the predicted load and the power coal consumption per unit load, and determine the power coal purchase quantity based on the predicted power coal consumption.

[0059] The number of days the remaining coal can be consumed can be determined based on the predicted coal consumption, and the coal purchase quantity can be determined based on this number of days, so that the demand side can be converted to make the result more accurate.

[0060] The present invention provides a method for making decisions on the purchase amount of electric coal based on deep learning. The deep learning algorithm is used to learn the historical regional electricity consumption characteristics of the power supply area, so that the real-time regional electricity consumption characteristics can be identified in the future, and the predicted daily load can be determined. The final daily coal consumption can be determined according to the predicted daily coal consumption. Finally, whether electric coal purchase is needed and the amount of electric coal to be purchased can be determined according to the daily coal consumption. Analysis from the load side makes the prediction more accurate and simple, which can effectively improve the decision-making efficiency of the electric coal purchase amount and replace the traditional manual operation.

[0061] In the embodiment of the present invention, the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant are collected, and the unit load coal consumption is determined according to the daily load and the daily coal consumption, including:

[0062] Collect the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant;

[0063] The daily coal consumption of the power plant is divided by the daily load of the power supply area corresponding to the power plant to obtain the unit load coal consumption.

[0064] In the embodiment of the present invention, the historical regional electricity consumption characteristics of the power supply area are collected, including:

[0065] Collect the daily average temperature, daily average humidity, rainfall, season type and date type of the power supply area to obtain the historical regional electricity consumption characteristics, and collect the daily electricity consumption corresponding to the historical regional electricity consumption characteristics;

[0066] The date types include working days, holidays, and non-working days that are not holidays.

[0067] Optionally, for any day, the historical regional electricity consumption characteristics of that day and the daily electricity consumption of the previous N days can be used as sample data, and the daily electricity consumption of that day can be used as the expected output, so that the sample data can be used as input and the expected output can be used as the true value label to train the deep learning model, so that the deep learning model has data analysis capabilities.

[0068] After collecting the real-time regional electricity consumption characteristics, it is also necessary to build the input of the deep learning model together with the daily electricity consumption of the previous N days to achieve prediction.

[0069] It is worth noting that after constructing sample data and expected output, or after using real-time regional electricity consumption characteristics to construct the input of a deep learning model, the data can be normalized to reduce the complexity of data processing.

[0070] In an embodiment of the present invention, a deep learning model for predicting daily load is constructed, including: constructing a BP deep learning model to obtain a deep learning model for predicting daily load.

[0071] In an embodiment of the present invention, the deep learning model is trained using the historical regional electricity consumption characteristics to obtain the trained deep learning model, including:

[0072] A1. Construct the historical regional electricity consumption characteristics as the model input vector corresponding to the deep learning model, and take the daily electricity consumption corresponding to the historical regional electricity consumption characteristics as the expected value to obtain training data;

[0073] A2. Initialize a coding population for the hyperparameters of the deep learning model; wherein the coding population includes multiple coding individuals, each of which includes part or all of the hyperparameters to be trained of the deep learning model;

[0074] For example, the hyperparameters that need to be trained in the deep learning model are randomly initialized within their upper and lower limits, and these hyperparameters are encoded into vectors to obtain encoded individuals.

[0075] A3. Using training data to obtain the fitness corresponding to each coding individual in the coding population, and according to the fitness of each coding individual, determining the coding individual with the largest fitness as the global optimal individual, and determining the coding individual with the smallest fitness as the global worst individual;

[0076] For example, the training data may be used to obtain the root mean square error function value, and the error function value may be negative to obtain the fitness corresponding to the encoded individual.

[0077] A4. Based on the upper and lower limits corresponding to the coding individual, a nonlinear local search is performed on the coding individual in its neighborhood to obtain the coding individual after the local search;

[0078] A5. For the coded individuals after the local search, all coded individuals are subjected to vortex search around the global optimal individual to obtain the coded individuals after the vortex search;

[0079] A6. For the coding individual after the vortex search, the coding individual is made to learn the position information of the global optimal individual and the position information of the coding population to perform an extended search to obtain the coding individual after the extended search;

[0080] A7. For the coding individual after the extended search, use the global worst individual to perform a random flight search on the coding individual to obtain the coding individual after the random flight search;

[0081] A8. Repeat local search, vortex search, expansion search, and random flight search until the training end conditions are met, re-obtain the global optimal individual, and decode the global optimal individual as the final hyperparameter of the deep learning model to obtain the trained deep learning model.

[0082] In the embodiment of the present invention, based on the upper and lower limits corresponding to the coding individual, a nonlinear local search is performed on the coding individual in its neighborhood to obtain the coding individual after the local search, including:

[0083] Determine the adaptive search step size based on the current number of training times Among them, e represents the natural constant, T represents the maximum number of training times;

[0084] Based on the upper and lower limits corresponding to the coded individuals, determine the search individual X consisting of the search range SS =(X max -X min ), where X SS represents the search individual, X max represents the upper bound individual corresponding to the encoded individual, which includes the upper limit of each dimension hyperparameter; X min Represents the lower bound individuals corresponding to the encoded individuals, and the lower bound individuals include the lower limits of the hyperparameters in each dimension;

[0085] Generate a first random number r1 between (0,1), and use the first random number r1, the adaptive search step size s, and the search individual X SS , determine the neighborhood search volume: X SS *r1*s;

[0086] According to the neighborhood search amount, a local search is performed on the coding individual, and the coding individual obtained after the local search is: in, represents the i-th encoding individual in the t-th training process, Represents the encoded individual after local search i=1,2,…,I, where I represents the total number of coded individuals.

[0087] The local search provided by the embodiment of the present invention can enable each coding individual to explore within its neighborhood, thereby determining a better position near each coding individual.

[0088] In the embodiment of the present invention, for the coding individuals after the local search, all coding individuals are subjected to vortex search around the global optimal individual to obtain the coding individuals after the vortex search, including:

[0089] For the encoded individuals after local search, the step size control factor is determined according to the current number of training times: Among them, η represents the step size control factor, η0 represents the maximum value of the step size control factor, e represents the natural constant, and T represents the maximum number of training times;

[0090] Generate a second random number r2 between (0,1), and based on the second random number r2, use the cosine function to determine the search direction: cos(2πr2);

[0091] For the coded individual after local search, obtain the difference between the coded individual and the global optimal individual After the step size control factor is used to control the difference, all the coding individuals search around the global optimal individual in their respective search directions to form a vortex search. The coding individuals after the vortex search are:

[0092]

[0093] in, represents the coded individual after the nth local search, Represents the coded individual after vortex search n=1,2,…,I。

[0094] The vortex search provided by the embodiment of the present invention can enable all encoding individuals to perform collaborative search along the currently known optimal position, which is more conducive to finding the optimal position in the solution space.

[0095] In an embodiment of the present invention, for a coding individual after vortex search, the coding individual is made to learn the position information of the global optimal individual and the position information of the coding population to perform an extended search, and the coding individual after the extended search is obtained, including:

[0096] For the coding individual after the vortex search, a third random number r3 between (0,1) is generated, and based on the third random number r3, the sine function is used to determine the coding individual position influence amount: Among them, || represents the absolute value, Represents the coded individual after the mth vortex search, m=1,2,…,I.

[0097] After fusing the position information of the global optimal individual and the position information of the coded population, the cosine function and random coefficients are used for interference, and the comprehensive position influence is obtained as follows: Among them, r4 represents a random number between (0,1), represents the global optimal individual, Represents the average individual corresponding to the coding population. The parameters of each dimension in the average individual are obtained by the mean of all coding individuals in this dimension. ξ1 represents the first fusion coefficient, ξ1 = -π + (1-τ) * 2π, π represents pi, τ represents the constant term, ξ2 represents the second fusion coefficient, ξ2=-π+τ*2π.

[0098] According to the coding individual position influence amount and the comprehensive position influence amount, the coding individual after the extended search is determined to be: in, Represents the coded individual after the extended search

[0099] The extended search provided by the embodiment of the present invention can enable each coding individual to explore more areas based on the currently known optimal position and the coding population position, which is not only conducive to improving the diversity of the algorithm, but also can improve the traversability of the position.

[0100] In the embodiment of the present invention, for the coding individuals after the extended search, a global worst individual is used to perform a random flight search on the coding individuals to obtain the coding individuals after the random flight search, including:

[0101] Based on the global worst individual, the random flight individual is determined as: in, represents a random flying individual, represents the coded individual after a random extended search, represents the global worst individual, μ represents the first random flight parameter, and η represents a random number between (0,2], Γ represents a gamma function, v represents a second random flight coefficient, and v ≈ N (0,1);

[0102] For the coded individuals after the extended search, the random flight individuals are used to perform curve flight and the positions are adaptively adjusted. The coded individuals after the random flight search are obtained as follows: in, represents the coded individual after the kth extended search, k = 1, 2, ..., I, represents the coded individual after random flight search λ represents the upper limit of random coefficient λ max and the lower bound λ of the random coefficient min , r5 represents a random number between (-1,1), ε represents that all elements are random individuals between (-1,1), and the total number of elements of ε is the same as the total number of elements of the coded individuals, γ t represents the adjustment parameter, and γ min represents the minimum value of the adjustment parameter, γ max Indicates the maximum value of the adjustment parameter.

[0103] Determine whether the fitness of the encoded individual after the random flight search increases. If so, accept the random flight search, otherwise reject the random flight search.

[0104] The random flight search of the embodiment of the present invention searches at a certain rotation angle, which avoids the generation of repeated individuals to the greatest extent, can effectively improve the local development ability of the algorithm, and adopts Levy flight to expand the group search range, helping the algorithm to jump out of the local optimum when necessary. This strategy can not only enhance the local development ability of the algorithm, but also effectively help the algorithm jump out of the local optimum and realize global search, thereby improving the algorithm search performance.

[0105] Compared with the prior art, the algorithm provided by the embodiment of the present invention not only has the effect of fast convergence speed, but also has the technical effect of being able to search for the global optimum, which can effectively improve the prediction effect of daily load.

[0106] In an embodiment of the present invention, based on the predicted load amount and the unit load coal consumption, the predicted coal consumption is determined, and the coal purchase amount is determined based on the predicted coal consumption, including:

[0107] Multiplying the predicted load amount by the unit load coal consumption to obtain the predicted coal consumption;

[0108] Obtain the remaining amount of electric coal corresponding to the power plant, and divide the remaining amount of electric coal by the predicted electric coal consumption to obtain the number of days that the electric coal can be consumed;

[0109] Determine whether the number of days that electricity coal can be consumed is less than the preset required consumption days. If so, subtract the number of days that electricity coal can be consumed from the required consumption days to get the difference days, and multiply the predicted electricity coal consumption by the difference days to get the minimum electricity coal purchase quantity. Otherwise, it is determined that there is sufficient electricity coal and no purchase is required, and the electricity coal purchase quantity is determined to be zero.

[0110] The required consumption days indicate how many days the power plant's specified power coal must meet the power generation demand. By taking the difference between this data, the power coal purchase quantity can be determined.

[0111] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0115] A person of ordinary skill in the art can understand that all or part of the steps in realizing the above-mentioned facts and methods can be completed by instructing the relevant hardware through a program, and the program involved or the program described can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.

[0116] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for decision-making on the purchase amount of electric coal based on deep learning, characterized in that: include: Collect the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant, and determine the unit load coal consumption according to the daily load and the daily coal consumption; Collect historical regional electricity consumption characteristics of the power supply area and build a deep learning model for predicting daily load, and use the historical regional electricity consumption characteristics to train the deep learning model to obtain a trained deep learning model; Collecting real-time regional power consumption characteristics of the power supply area, and using the real-time regional power consumption characteristics as input of the trained deep learning model to obtain predicted daily load; Based on the predicted load and unit load coal consumption, the predicted coal consumption is determined, and the coal purchase quantity is determined based on the predicted coal consumption.

2. The method for decision-making on power coal purchase quantity based on deep learning according to claim 1 is characterized in that: Collecting the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant, and determining the unit load coal consumption according to the daily load and the daily coal consumption, including: Collect the daily load of the power supply area corresponding to the power plant and the daily coal consumption corresponding to the power plant; The daily coal consumption of the power plant is divided by the daily load of the power supply area corresponding to the power plant to obtain the unit load coal consumption.

3. The method for decision-making on power coal purchase quantity based on deep learning according to claim 1, characterized in that: Collect historical regional electricity consumption characteristics of the power supply area, including: Collect the daily average temperature, daily average humidity, rainfall, season type and date type of the power supply area to obtain the historical regional electricity consumption characteristics, and at the same time collect the daily electricity consumption corresponding to the historical regional electricity consumption characteristics; The date types include working days, holidays, and non-working days that are not holidays.

4. The method for decision-making on power coal purchase quantity based on deep learning according to claim 1, characterized in that: Constructing a deep learning model for predicting daily load, including: constructing a BP deep learning model to obtain a deep learning model for predicting daily load.

5. The method for decision-making on power coal purchase quantity based on deep learning according to claim 1, characterized in that: The deep learning model is trained using historical regional electricity consumption characteristics to obtain the trained deep learning model, including: The historical regional electricity consumption characteristics are constructed as the model input vector corresponding to the deep learning model, and the daily electricity consumption corresponding to the historical regional electricity consumption characteristics is used as the expected value to obtain training data; Initialize a coding population for the hyperparameters of the deep learning model; wherein the coding population includes multiple coding individuals, each coding individual includes part or all of the hyperparameters to be trained of the deep learning model; The training data is used to obtain the fitness corresponding to each coding individual in the coding population, and according to the fitness of each coding individual, the coding individual with the largest fitness is determined as the global optimal individual, and the coding individual with the smallest fitness is determined as the global worst individual; Based on the upper and lower limits corresponding to the coding individual, a nonlinear local search is performed on the coding individual in its neighborhood to obtain the coding individual after the local search; For the coded individuals after the local search, all coded individuals are subjected to vortex search around the global optimal individual to obtain the coded individuals after the vortex search; For the coding individual after the vortex search, the coding individual is made to learn the position information of the global optimal individual and the position information of the coding population to perform an extended search, thereby obtaining the coding individual after the extended search; For the coding individuals after the extended search, use the global worst individual to perform random flight search on the coding individuals to obtain the coding individuals after the random flight search; Repeat local search, vortex search, expansion search and random flight search until the training end conditions are met, re-obtain the global optimal individual, and decode the global optimal individual as the final hyperparameter of the deep learning model to obtain the trained deep learning model.

6. The method for decision-making on the amount of power coal purchased based on deep learning according to claim 5 is characterized in that: Based on the upper and lower limits corresponding to the coded individual, a nonlinear local search is performed on the coded individual in its neighborhood to obtain the coded individual after the local search, including: Determine the adaptive search step size based on the current number of training times; Based on the upper and lower limits corresponding to the coded individuals, the search individuals formed by the search range are determined; Generate a first random number between (0,1), and use the first random number, the adaptive search step size, and the search individual to determine the neighborhood search amount; According to the neighborhood search amount, a local search is performed on the coding individual to obtain the coding individual after the local search.

7. The method for decision-making on power coal purchase quantity based on deep learning according to claim 6 is characterized in that: For the coded individuals after the local search, all coded individuals are subjected to vortex search around the global optimal individual to obtain the coded individuals after the vortex search, including: For the encoded individuals after local search, the step size control factor is determined according to the current number of training times; Generate a second random number between (0, 1), and determine the search direction using a cosine function based on the second random number; For the coding individuals after local search, the difference between the coding individuals and the global optimal individual is obtained, and after the difference is controlled by the step size control factor, all coding individuals search around the global optimal individual in their respective search directions to form a vortex search, and the coding individuals after the vortex search are obtained.

8. The method for decision-making on power coal purchase quantity based on deep learning according to claim 7 is characterized in that: For the coding individual after the vortex search, the coding individual is made to learn the position information of the global optimal individual and the position information of the coding population to perform an extended search, and the coding individual after the extended search is obtained, including: For the coding individual after the vortex search, a third random number between (0, 1) is generated, and based on the third random number, a sine function is used to determine the coding individual position influence amount; After fusing the position information of the global optimal individual and the position information of the coded population, the cosine function and random coefficients are used for interference to obtain the comprehensive position influence. According to the coding individual position influence amount and the comprehensive position influence amount, the coding individual after the extended search is determined.

9. The method for decision-making on the amount of power coal purchased based on deep learning according to claim 8 is characterized in that: For the coding individuals after the extended search, use the global worst individual to perform random flight search on the coding individuals to obtain the coding individuals after the random flight search, including: Based on the global worst individual, a random flying individual is determined; For the coded individuals after the extended search, the random flight individuals are used to perform curve flight and the positions are adaptively adjusted to obtain the coded individuals after the random flight search; Determine whether the fitness of the encoded individual after the random flight search increases. If so, accept the random flight search, otherwise reject the random flight search.

10. The method for decision-making on power coal purchase quantity based on deep learning according to claim 9, characterized in that: Based on the predicted load and the unit load coal consumption, the predicted coal consumption is determined, and the coal purchase quantity is determined based on the predicted coal consumption, including: Multiplying the predicted load amount by the unit load coal consumption to obtain the predicted coal consumption; Obtain the remaining amount of electric coal corresponding to the power plant, and divide the remaining amount of electric coal by the predicted electric coal consumption to obtain the number of days that the electric coal can be consumed; Determine whether the number of days that electricity coal can be consumed is less than the preset required consumption days. If so, subtract the number of days that electricity coal can be consumed from the required consumption days to get the difference days, and multiply the predicted electricity coal consumption by the difference days to get the minimum electricity coal purchase quantity. Otherwise, it is determined that there is sufficient electricity coal and no purchase is required, and the electricity coal purchase quantity is determined to be zero.

Images