Thermal power plant intelligent power generation method and system based on power supply area load prediction

Abstract

The invention discloses a thermal power plant intelligent power generation method and system based on power supply area load prediction, and belongs to the technical field of power generation. A power supply area load prediction model is constructed through a deep learning model, and the power supply area load prediction model is trained by using load training data; the method comprises the following steps: acquiring a trained power supply area load prediction model, performing load prediction on a power supply area by adopting the trained power supply area load prediction model, determining a power supply area load prediction result, and finally automatically generating a power generation plan of a thermal power plant on the basis of the power supply area load prediction result. And the power generation plan of the thermal power plant is transmitted to the equipment specified by the staff to perform intelligent power generation of the thermal power plant, so that the power generation efficiency can be effectively improved, and energy waste in the power generation process is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of power generation, and in particular relates to a smart power generation method and system for a thermal power plant based on load forecasting in a power supply area. Background Art

[0002] With the rapid development of the social economy and the continuous improvement of people's living standards, electricity demand continues to grow, and the power system faces increasing pressure. As one of the main sources of electricity supply in my country, the power generation efficiency and stability of thermal power plants are of great significance to ensuring the safe and stable operation of the power system. However, traditional thermal power plant power generation methods are often based on fixed power generation plans, which are difficult to adapt to the real-time changes in power load demand, resulting in low power generation efficiency and energy waste. Therefore, how to achieve smart power generation in thermal power plants and improve power generation efficiency and energy utilization has become a pressing issue. Summary of the Invention

[0003] The present invention provides a smart power generation method and system for a thermal power plant based on power supply area load forecasting, which is used to solve the problem that the existing technology adopts a fixed power generation plan, resulting in low power generation efficiency and energy waste.

[0004] In one aspect, the present invention provides a smart power generation method for a thermal power plant based on load forecasting in a power supply area, comprising: Collecting load-related data corresponding to the power supply area, and preprocessing the load-related data to obtain load training data; Constructing a power supply area load forecasting model using a deep learning model, and training the power supply area load forecasting model using the load training data to obtain a trained power supply area load forecasting model; Using the trained load forecasting model for the power supply area to forecast the load in the power supply area, and determining the load forecasting result for the power supply area; Based on the load forecast results of the power supply area, a power generation plan of the thermal power plant is automatically generated, and the power generation plan of the thermal power plant is transmitted to the equipment designated by the staff to carry out smart power generation of the thermal power plant.

[0005] Furthermore, it also includes: During the power generation process, the deviation between the actual load of the power supply area and the load forecast result of the power supply area is calculated to determine the deviation value; When the deviation value is greater than a preset threshold, new load training data is generated, and the trained power supply area load forecasting model is updated according to the new load training data.

[0006] Furthermore, historical load data and meteorological data of the power supply area are collected to obtain load-related data corresponding to the power supply area, and the load-related data are preprocessed to obtain load training data, including: Collect historical load data, meteorological data, and date types of the power supply area to obtain load-related data corresponding to the power supply area; wherein the meteorological data includes seasonal data, weather data, and temperature data; Normalizing the load-related data corresponding to the power supply area to obtain the normalized load-related data; Based on the normalized load correlation data, M consecutive historical load data, M+1 meteorological data, and M+1 date types are used to construct a load influencing factor sample; and the M+1 historical load data is used to construct a load expectation label. The load influencing factor samples and their corresponding load expected labels are composed into load training data, and multiple load training data are repeatedly obtained.

[0007] Furthermore, a deep learning model is used to construct a power supply area load forecasting model, including: using a CNN-BP Net model to construct a power supply area load forecasting model.

[0008] Furthermore, the load training data is used to train the power supply area load forecasting model to obtain the trained power supply area load forecasting model, including: Initializing and encoding the parameters of the power supply area load forecasting model to determine a plurality of parameter codes; Based on the load training data, a loss function value corresponding to each parameter encoding is obtained, and the parameter encoding with the smallest loss function value is determined as the optimal parameter encoding; Based on the optimal parameter code, an initial search is performed on the parameter code using a search area selection mechanism to determine the parameter code after the initial search; Using a joint spiral search mechanism to perform accelerated search on the parameter code after the initial search, and determining the parameter code after the accelerated search; Performing a fine search on the parameter code after the accelerated search using an optimal fine search mechanism to determine the parameter code after the fine search; The Cauchy mutation search mechanism is used to perform mutation search on the parameter coding after the fine search, and the parameter coding after the mutation search is determined; Determine whether the current number of training times has reached the maximum number of training times. If so, obtain the final parameters of the power supply area load forecasting model based on the parameter coding after the mutation search, and obtain the power supply area load forecasting model after training. Otherwise, return to the step of obtaining the optimal parameter coding.

[0009] Furthermore, based on the optimal parameter code, an initial search is performed on the parameter code using a search area selection mechanism, and the parameter code after the initial search is determined to be:

[0010] in, Indicates the z The first training m parameter encoding, Indicates the m Parameter encoding after the initial search, represents the optimal parameter encoding, m =1,2,…,NP, NP represents the total number of parameter codes, represents a natural constant, represents a natural constant, represents pi, b represents the position selection trajectory control factor, Indicates the random position search range control factor between [-1,1], Indicates the z Mean encoding during training.

[0011] Furthermore, a joint spiral search mechanism is used to perform an accelerated search on the parameter code after the initial search, and the parameter code after the accelerated search is determined to be:

[0012]

[0013]

[0014]

[0015] in, Indicates the z The first training k Parameter encoding after the initial search, Indicates the k Parameter encoding after an accelerated search, k =1,2,…,NP, represents the position transformation factor, Indicates the z The first training k +1 parameter encoding after the initial search, represents the first search control factor, represents the second search control factor, sin represents the sine function, Indicates the preset maximum number of training times. Indicates the maximum value of the position transformation factor, Indicates the minimum value of the position transformation factor, Indicates parameter encoding The corresponding first spiral search angle, and , represents the spiral search angle control parameter, which is set to a constant between [5,10]. represents the first random number between (0,1), Indicates parameter encoding The corresponding first spiral search radius, and , represents the spiral search radius control parameter and is set to a constant between [0.5, 2]. represents the second random number between (0,1), Indicates the z The first training j The parameter encoding after the initial search corresponds to the first spiral search radius, Indicates the z The first training j The parameter encoding after the initial search corresponds to the first spiral search angle, and || indicates the absolute value sign. Furthermore, the optimal fine search mechanism is used to perform a fine search on the parameter code after the accelerated search, and the parameter code after the fine search is determined to be:

[0016]

[0017]

[0018]

[0019] in, Indicates the z The first training n Parameter encoding after an accelerated search, Indicates the n Parameter encoding after a fine search, n =1,2,…,NP, represents the inertia weight, represents the third search control factor, represents the fourth search control factor, represents the first weighting parameter, represents the second weighting parameter, and as well as are all set to constants between [1,2]. Indicates parameter encoding The corresponding second spiral search angle, and , represents the third random number between (0,1), represents the second spiral search radius, and = ; Indicates the z The first training j The second spiral search angle corresponding to the parameter encoding after the accelerated search, Indicates the z The first training j The second spiral search radius corresponding to the parameter encoding after the accelerated search, represents the maximum value of the inertia weight, represents the minimum value of the inertia weight, Represents the inverse cosine function.

[0020] Furthermore, the Cauchy mutation search mechanism is used to perform mutation search on the parameter encoding after the fine search, and the parameter encoding after the mutation search is determined to be:

[0021] in, Indicates the z The first training h Parameter encoding after a fine search, Indicates the h The parameter encoding after the mutation search, h =1,2,…,NP, Represents a random number generated by standard Cauchy mutation.

[0022] On the other hand, the present invention provides a smart power generation system for a thermal power plant based on load forecasting in a power supply area, comprising: a data acquisition module, a deep learning module, a load forecasting module, and a smart power generation module; The data acquisition module is used to collect load-related data corresponding to the power supply area, and pre-process the load-related data to obtain load training data; The deep learning module is used to construct a power supply area load forecasting model using a deep learning model, and train the power supply area load forecasting model using the load training data to obtain the power supply area load forecasting model after training; The load forecasting module is used to use the trained power supply area load forecasting model to perform load forecasting on the power supply area and determine the power supply area load forecasting result; The smart power generation module is used to automatically generate a power generation plan for a thermal power plant based on the load forecast results of the power supply area, and transmit the power generation plan to a device designated by the staff to perform smart power generation in the thermal power plant.

[0023] The present invention provides a smart power generation method and system for a thermal power plant based on power supply area load forecasting. The method constructs a power supply area load forecasting model through a deep learning model, and uses load training data to train the power supply area load forecasting model to obtain the trained power supply area load forecasting model. The trained power supply area load forecasting model is then used to perform load forecasting on the power supply area, and the power supply area load forecasting result is determined. Finally, based on the power supply area load forecasting result, a power generation plan of the thermal power plant is automatically generated, and the power generation plan of the thermal power plant is transmitted to the equipment designated by the staff to perform smart power generation of the thermal power plant, which can effectively improve the power generation efficiency and thus avoid energy waste in the power generation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] 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.

[0025] Figure 1 A flowchart of a method for intelligent power generation in a thermal power plant based on load forecasting in the power supply area provided in an embodiment of the present invention.

[0026] Figure 2 A schematic structural diagram of a smart power generation system for a thermal power plant based on power supply area load forecasting provided in an embodiment of the present invention.

[0027] Among them, 21-data acquisition module, 22-deep learning module, 23-load forecasting module, and 24-smart power generation module.

[0028] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0029] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0030] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0031] like Figure 1As shown, an embodiment of the present invention provides a smart power generation method for a thermal power plant based on power supply area load forecasting, comprising: S11. Collect load-related data corresponding to the power supply area, and pre-process the load-related data to obtain load training data; In an embodiment of the present invention, historical load data and meteorological data of a power supply area are collected to obtain load-related data corresponding to the power supply area, and the load-related data are preprocessed to obtain load training data, including: Collect historical load data, meteorological data and date types of the power supply area to obtain load-related data corresponding to the power supply area; wherein, the meteorological data includes seasonal data, weather data and temperature data; seasonal data may include spring, summer, autumn or winter, weather data may include rainy, sunny, cloudy or snowy days, date types may include working days or non-working days, and date types may also include holidays. It is worth noting that when a certain day is a holiday, the holiday is directly located, otherwise it is determined as a working day or a non-working day according to the date; meteorological data can be obtained through the forecast of the Meteorological Bureau for prediction.

[0032] Normalizing the load-related data corresponding to the power supply area to obtain the normalized load-related data; Based on the normalized load correlation data, M consecutive historical load data, M+1 meteorological data, and M+1 date types are used to construct a load influencing factor sample; and the M+1 historical load data is used to construct a load expectation label. The load influencing factor samples and their corresponding load expected labels are composed into load training data, and multiple load training data are repeatedly obtained.

[0033] S12. Using a deep learning model to build a power supply area load forecasting model, and using the load training data to train the power supply area load forecasting model to obtain a trained power supply area load forecasting model; Deep learning models such as convolutional neural networks and long short-term memory networks can be used to build a power supply area load forecasting model, and then intelligent optimization algorithms such as gradient descent algorithm and particle swarm algorithm can be used to train the power supply area load forecasting model, so as to obtain the trained power supply area load forecasting model.

[0034] S13, using the trained power supply area load forecasting model to perform load forecasting on the power supply area, and determining a load forecast result for the power supply area; Using the trained power supply area load forecasting model to perform load forecasting on the power supply area can include: collecting real-time load influencing factor data of the power supply area, and using the trained power supply area load forecasting model to identify the real-time load influencing factor data to obtain the power supply area load forecasting result.

[0035] It is worth noting that since the load influencing factor samples have been normalized, the real-time load influencing factor data also needs to be normalized, and the data structure of the real-time load influencing factor data and the load influencing factor samples should be the same to ensure accurate data identification.

[0036] S14. Based on the load forecast result of the power supply area, a power generation plan of the thermal power plant is automatically generated, and the power generation plan of the thermal power plant is transmitted to the equipment designated by the staff to perform smart power generation of the thermal power plant.

[0037] For example, the load forecast for a power supply area generally measures power usage. Therefore, this can be used as the required power generation capacity to determine the power generation plan for thermal power plants, thereby improving power generation efficiency and energy utilization. Considering the reality of energy loss during power transmission and load fluctuations in the power supply area, a fault-tolerant load can be set. By adding the power supply area load and the fault-tolerant load, the power consumption in the power supply area can be determined. With the power consumption information, the power generation plan can be determined to generate power that matches the power consumption.

[0038] In an embodiment of the present invention, the following further comprises: During the power generation process, the deviation between the actual load of the power supply area and the load forecast result of the power supply area is calculated to determine the deviation value; When the deviation value is greater than a preset threshold, new load training data is generated, and the trained power supply area load forecasting model is updated according to the new load training data.

[0039] The embodiment of the present invention updates the trained load forecasting model for the power supply area through new load training data, which can make the prediction accuracy of the trained load forecasting model for the power supply area higher and higher, and further improve the power generation efficiency.

[0040] In an embodiment of the present invention, a deep learning model is used to construct a power supply area load forecasting model, including: using a CNN-BP NET (convolutional neural network-back propagation neural network) model to construct a power supply area load forecasting model.

[0041] It is worth noting that the above-mentioned deep learning model is only a preferred example of the embodiment of the present invention, and other deep learning models can also be used to construct the power supply area load forecasting model. For example, only CNN can be used to construct the power supply area load forecasting model.

[0042] In an embodiment of the present invention, the load training data is used to train the power supply area load forecasting model to obtain the trained power supply area load forecasting model, including: Initializing and encoding the parameters of the power supply area load forecasting model to determine a plurality of parameter codes; For example, the parameters of the power supply area load forecasting model are generally the connection weights between network layers. These connection weights have corresponding upper and lower limits. Therefore, the parameters can be initialized between the upper and lower limits, and the initialized parameters can be encoded into a vector to obtain parameter encoding. After repeated initialization, multiple parameter encodings can be obtained.

[0043] Based on the load training data, a loss function value corresponding to each parameter encoding is obtained, and the parameter encoding with the smallest loss function value is determined as the optimal parameter encoding; For example, after the parameters of the parameter encoding are applied to the power supply area load forecasting model, the load influencing factor sample is used as input, and the load expected label corresponding to the load influencing factor sample is used as the expected output to obtain the cross entropy loss function or root mean square loss function corresponding to the parameter encoding.

[0044] Based on the optimal parameter code, an initial search is performed on the parameter code using a search area selection mechanism to determine the parameter code after the initial search; Using a joint spiral search mechanism to perform accelerated search on the parameter code after the initial search, and determining the parameter code after the accelerated search; Performing a fine search on the parameter code after the accelerated search using an optimal fine search mechanism to determine the parameter code after the fine search; The Cauchy mutation search mechanism is used to perform mutation search on the parameter coding after the fine search, and the parameter coding after the mutation search is determined; Determine whether the current number of training times has reached the maximum number of training times. If so, obtain the final parameters of the power supply area load forecasting model based on the parameter coding after the mutation search, and obtain the power supply area load forecasting model after training. Otherwise, return to the step of obtaining the optimal parameter coding.

[0045] Obtaining the final parameters of the power supply area load forecasting model based on the parameter coding after the variation search may include: re-determining the optimal parameter coding based on the parameter coding after the variation search, and using the parameters in the re-determined optimal parameter coding as the final parameters of the power supply area load forecasting model.

[0046] Optionally, after a parameter code changes, the parameter code can be processed for out-of-bounds errors. For example, if a parameter in a dimension of the parameter code exceeds an upper limit, the exceeding parameter can be set to its corresponding upper limit. If a parameter in a dimension of the parameter code falls below a lower limit, the exceeding parameter can be set to its corresponding lower limit.

[0047] Since the existing gradient descent algorithm is prone to poor training results and falling into local optimality during the optimization process, the embodiment of the present invention provides a new training algorithm to improve the global search capability and overall training effect, and ultimately ensure the accuracy of load forecasting.

[0048] In the embodiment of the present invention, based on the optimal parameter code, an initial search is performed on the parameter code using a search area selection mechanism, and the parameter code after the initial search is determined to be:

[0049] in, Indicates the z The first training m parameter encoding, Indicates the m Parameter encoding after the initial search, represents the optimal parameter encoding, m =1,2,…,NP, NP represents the total number of parameter codes, represents a natural constant, represents a natural constant, represents pi, b represents the position selection trajectory control factor, Indicates the random position search range control factor between [-1,1], Indicates the z The mean encoding during training is used, that is, the parameter of each dimension in the mean encoding is the mean of the parameters of all parameters encoded in the same dimension.

[0050] The search region selection mechanism provided by the embodiments of the present invention selects a search region for each parameter code, centered around the optimal parameter code. This selection is combined with the average position to maintain parameter code diversity and prevent rapid clustering, thereby improving the algorithm's global search capabilities. This also provides more directions for selecting the search region, further ensuring global search capabilities.

[0051] In an embodiment of the present invention, a joint spiral search mechanism is used to perform an accelerated search on the parameter code after the initial search, and the parameter code after the accelerated search is determined to be:

[0052]

[0053]

[0054]

[0055] in, Indicates the z The first training k Parameter encoding after the initial search, Indicates the k Parameter encoding after an accelerated search, k =1,2,…,NP, represents the position transformation factor, Indicates the z The first training k +1 parameter encoding after the initial search, represents the first search control factor, represents the second search control factor, sin represents the sine function, Indicates the preset maximum number of training times. Indicates the maximum value of the position transformation factor, Indicates the minimum value of the position transformation factor, Indicates parameter encoding The corresponding first spiral search angle, and , represents the spiral search angle control parameter, which is set to a constant between [5,10]. represents the first random number between (0,1), Indicates parameter encoding The corresponding first spiral search radius, and , represents the spiral search radius control parameter and is set to a constant between [0.5, 2]. represents the second random number between (0,1), Indicates the z The first training j The parameter encoding after the initial search corresponds to the first spiral search radius, Indicates the z The first training j The parameter encoding after the initial search corresponds to the first spiral search angle, and || indicates the absolute value sign. The joint spiral search mechanism provided in the embodiment of the present invention uses the idea of combining joint search and spiral search to search parameter encoding, and provides a relatively powerful global search capability in the early stage of the algorithm and a relatively powerful local search capability in the later stage of the algorithm. It can effectively balance the global search and local search of the algorithm and improve the algorithm training effect.

[0056] In the embodiment of the present invention, an optimal fine search mechanism is used to perform a fine search on the parameter code after the accelerated search, and the parameter code after the fine search is determined to be:

[0057]

[0058]

[0059]

[0060] in, Indicates the z The first training n Parameter encoding after an accelerated search, Indicates the n Parameter encoding after a fine search, n =1,2,…,NP, represents the inertia weight, represents the third search control factor, represents the fourth search control factor, represents the first weighting parameter, represents the second weighting parameter, and as well as are all set to constants between [1,2]. Indicates parameter encoding The corresponding second spiral search angle, and , represents the third random number between (0,1), represents the second spiral search radius, and = ; Indicates the z The first training j The second spiral search angle corresponding to the parameter encoding after the accelerated search, Indicates the z The first training j The second spiral search radius corresponding to the parameter encoding after the accelerated search, represents the maximum value of the inertia weight, represents the minimum value of the inertia weight, Represents the inverse cosine function.

[0061] The optimal fine search mechanism provided by the embodiment of the present invention has a larger weight coefficient in the early stage of the algorithm, that is, when the number of iterations is small, so that the algorithm can search near the optimal position more quickly, and the population has better global search ability at this time; in the later stage of the algorithm, that is, when the number of iterations is large, the weight coefficient is smaller, so that the algorithm can search carefully and slowly in the optimal position area, and the population gradually moves closer to the global optimal position, thereby improving the convergence ability of the algorithm and the local search ability.

[0062] In the embodiment of the present invention, a Cauchy mutation search mechanism is used to perform a mutation search on the parameter encoding after the fine search, and the parameter encoding after the mutation search is determined to be:

[0063] in, Indicates the z The first training h Parameter encoding after a fine search, Indicates the h The parameter encoding after the mutation search, h =1,2,…,NP, Represents a random number generated by standard Cauchy mutation.

[0064] Optionally, an annealing simulation algorithm or a greedy strategy may be used to control the Cauchy mutation search mechanism to ensure the training speed of the algorithm.

[0065] The Cauchy mutation search mechanism provided by the embodiment of the present invention can provide a powerful global search capability and ensure that the algorithm will not fall into a local optimum.

[0066] In summary, through the mutual cooperation of several mechanisms, the training effect of the algorithm can be effectively improved, thereby improving the accuracy of load prediction and ultimately improving power generation efficiency.

[0067] The present invention provides a smart power generation method for a thermal power plant based on power supply area load forecasting. The method constructs a power supply area load forecasting model through a deep learning model, and uses load training data to train the power supply area load forecasting model to obtain the trained power supply area load forecasting model. The trained power supply area load forecasting model is then used to perform load forecasting on the power supply area, and the power supply area load forecasting result is determined. Finally, based on the power supply area load forecasting result, a power generation plan of the thermal power plant is automatically generated, and the power generation plan of the thermal power plant is transmitted to the equipment designated by the staff to perform smart power generation of the thermal power plant, which can effectively improve the power generation efficiency and thus avoid energy waste in the power generation process.

[0068] like Figure 2As shown, the present invention provides a smart power generation system for a thermal power plant based on load forecasting in a power supply area, comprising: a data acquisition module 21, a deep learning module 22, a load forecasting module 23, and a smart power generation module 24; The data acquisition module 21 is used to collect load-related data corresponding to the power supply area, and pre-process the load-related data to obtain load training data; The deep learning module 22 is used to construct a power supply area load forecasting model using a deep learning model, and train the power supply area load forecasting model using the load training data to obtain a trained power supply area load forecasting model; The load forecasting module 23 is used to use the trained power supply area load forecasting model to perform load forecasting on the power supply area and determine the power supply area load forecasting result; The smart power generation module 24 is used to automatically generate a power generation plan for a thermal power plant based on the load forecast result of the power supply area, and transmit the power generation plan to a device designated by the staff to perform smart power generation of the thermal power plant.

[0069] Figure 2 The smart power generation system of a thermal power plant based on power supply area load forecasting shown can implement the above method and technical solution. Its principles and beneficial effects are similar and will not be repeated here.

[0070] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and variations can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A smart power generation method for a thermal power plant based on load forecasting in a power supply area, characterized in that: include: Collecting load-related data corresponding to the power supply area, and preprocessing the load-related data to obtain load training data; Constructing a power supply area load forecasting model using a deep learning model, and training the power supply area load forecasting model using the load training data to obtain a trained power supply area load forecasting model; Using the trained load forecasting model for the power supply area to forecast the load in the power supply area, and determining the load forecasting result for the power supply area; Based on the load forecast results of the power supply area, a power generation plan of the thermal power plant is automatically generated, and the power generation plan of the thermal power plant is transmitted to the equipment designated by the staff to carry out smart power generation of the thermal power plant.

2. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 1 is characterized in that: Also includes: During the power generation process, the deviation between the actual load of the power supply area and the load forecast result of the power supply area is calculated to determine the deviation value; When the deviation value is greater than a preset threshold, new load training data is generated, and the trained power supply area load forecasting model is updated according to the new load training data.

3. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 1 is characterized in that: Collect historical load data and meteorological data of the power supply area to obtain load-related data corresponding to the power supply area, and pre-process the load-related data to obtain load training data, including: Collect historical load data, meteorological data, and date types of the power supply area to obtain load-related data corresponding to the power supply area; wherein the meteorological data includes seasonal data, weather data, and temperature data; Normalizing the load-related data corresponding to the power supply area to obtain the normalized load-related data; Based on the normalized load correlation data, M consecutive historical load data, M+1 meteorological data, and M+1 date types are used to construct a load influencing factor sample; and the M+1 historical load data is used to construct a load expectation label. The load influencing factor samples and their corresponding load expected labels are composed into load training data, and multiple load training data are repeatedly obtained.

4. The smart power generation method for a thermal power plant based on power supply area load forecasting according to claim 1 is characterized in that: A deep learning model is used to build a power supply area load forecasting model, including: using a CNN-BP Net model to build a power supply area load forecasting model.

5. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 1 is characterized in that: The load training data is used to train the power supply area load forecasting model to obtain the trained power supply area load forecasting model, including: Initializing and encoding the parameters of the power supply area load forecasting model to determine a plurality of parameter codes; Based on the load training data, a loss function value corresponding to each parameter encoding is obtained, and the parameter encoding with the smallest loss function value is determined as the optimal parameter encoding; Based on the optimal parameter code, an initial search is performed on the parameter code using a search area selection mechanism to determine the parameter code after the initial search; Using a joint spiral search mechanism to perform accelerated search on the parameter code after the initial search, and determining the parameter code after the accelerated search; Performing a fine search on the parameter code after the accelerated search using an optimal fine search mechanism to determine the parameter code after the fine search; The Cauchy mutation search mechanism is used to perform mutation search on the parameter coding after the fine search, and the parameter coding after the mutation search is determined; Determine whether the current number of training times has reached the maximum number of training times. If so, obtain the final parameters of the power supply area load forecasting model based on the parameter coding after the mutation search, and obtain the power supply area load forecasting model after training. Otherwise, return to the step of obtaining the optimal parameter coding.

6. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 5 is characterized in that: Based on the optimal parameter code, an initial search is performed on the parameter code using a search area selection mechanism, and the parameter code after the initial search is determined to be: in, Indicates the z The first training m parameter encoding, Indicates the m Parameter encoding after the initial search, represents the optimal parameter encoding, m =1,2,…,NP, NP represents the total number of parameter codes, represents a natural constant, represents a natural constant, represents pi, b represents the position selection trajectory control factor, Indicates the random position search range control factor between [-1,1], Indicates the z Mean encoding during training.

7. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 6 is characterized in that: The joint spiral search mechanism is used to accelerate the search of the parameter code after the initial search, and the parameter code after the accelerated search is determined to be: in, Indicates the z The first training k Parameter encoding after the initial search, Indicates the k Parameter encoding after an accelerated search, k =1,2,…,NP, represents the position transformation factor, Indicates the z The first training k +1 parameter encoding after the initial search, represents the first search control factor, represents the second search control factor, sin represents the sine function, Indicates the preset maximum number of training times. Indicates the maximum value of the position transformation factor, Indicates the minimum value of the position transformation factor, Indicates parameter encoding The corresponding first spiral search angle, and , represents the spiral search angle control parameter, which is set to a constant between [5,10]. represents the first random number between (0,1), Indicates parameter encoding The corresponding first spiral search radius, and , represents the spiral search radius control parameter and is set to a constant between [0.5, 2]. represents the second random number between (0,1), Indicates the z The first training j The parameter encoding after the initial search corresponds to the first spiral search radius, Indicates the z The first training j The parameter encoding after the initial search corresponds to the first spiral search angle, and || indicates the absolute value sign.

8. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 7 is characterized in that: The optimal fine search mechanism is used to perform a fine search on the parameter code after the accelerated search, and the parameter code after the fine search is determined to be: in, Indicates the z The first training n Parameter encoding after an accelerated search, Indicates the n Parameter encoding after a fine search, n =1,2,…,NP, represents the inertia weight, represents the third search control factor, represents the fourth search control factor, represents the first weighting parameter, represents the second weighting parameter, and as well as are all set to constants between [1,2]. Indicates parameter encoding The corresponding second spiral search angle, and , represents the third random number between (0,1), represents the second spiral search radius, and = ; Indicates the z The first training j The second spiral search angle corresponding to the parameter encoding after the accelerated search, Indicates the z The first training j The second spiral search radius corresponding to the parameter encoding after the accelerated search, represents the maximum value of the inertia weight, represents the minimum value of the inertia weight, Represents the inverse cosine function.

9. The smart power generation method for thermal power plants based on power supply area load forecasting according to claim 8, characterized in that: The Cauchy mutation search mechanism is used to perform mutation search on the parameter encoding after the fine search, and the parameter encoding after the mutation search is determined to be: in, Indicates the z The first training h Parameter encoding after a fine search, Indicates the h The parameter encoding after the mutation search, h =1,2,…,NP, Represents a random number generated by standard Cauchy mutation.

10. A smart power generation system for thermal power plants based on load forecasting in power supply areas, characterized in that: include: Data acquisition module, deep learning module, load forecasting module and smart power generation module; The data acquisition module is used to collect load-related data corresponding to the power supply area, and pre-process the load-related data to obtain load training data; The deep learning module is used to construct a power supply area load forecasting model using a deep learning model, and train the power supply area load forecasting model using the load training data to obtain the power supply area load forecasting model after training; The load forecasting module is used to use the trained power supply area load forecasting model to perform load forecasting on the power supply area and determine the power supply area load forecasting result; The smart power generation module is used to automatically generate a power generation plan for a thermal power plant based on the load forecast results of the power supply area, and transmit the power generation plan to a device designated by the staff to perform smart power generation in the thermal power plant.

Images