Coal feeder control method, device and storage medium based on multi-modal thermal power energy consumption analysis

Abstract

The present invention proposes a coal feeder control method, device and storage medium based on multimodal thermal power energy consumption analysis. The method includes: collecting the coal burning quantity C, power generation Ge and external transmission Oe in a historical period to form a first historical vector data set H1=(C, Ge, Oe), obtaining the steam input temperature It and output temperature Ot of the steam turbine to form a second historical vector data set H2=(It, Ot), obtaining the historical weather data Hw and environmental images of the external transmission destination to form a third historical vector data set H3=(Hw, Im), obtaining the weather data Nw of the external transmission destination at the current moment and the future weather data Fw, and collecting M current environmental images of the destination; training a multimodal adjustment model based on the collected data; inputting Nw, Fw and M environmental images into the trained multimodal adjustment model to obtain the required coal burning quantity FC in the future period, and controlling the coal feeder based on FC. The amount of coal burned is saved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a coal feeder control method, device and storage medium based on multi-modal thermal power energy consumption analysis. Background Art

[0002] Abstract: In order to reduce energy consumption and environmental pollution, existing measures mainly include coal gasification transformation, waste heat utilization, combustion regulation, flue gas drying, pressure difference reduction, etc. These technologies are relatively traditional technologies. In the existing technology, some power plants have adopted artificial intelligence technology to predict future electricity consumption and then accurately control the coal input.

[0003] However, in the prior art, the method of predicting the amount of coal burned is to use a single neural network structure to make predictions based on historical data. Due to the large amount of data, the training time of the neural network is relatively long, the prediction accuracy will decrease, and the time required to retrain the neural network will be even longer, which is not conducive to real-time use in power plants. Summary of the invention

[0004] The present invention proposes the following technical solutions to address one or more technical deficiencies in the above-mentioned prior art.

[0005] A coal feeder control method based on multi-modal thermal power energy consumption analysis, the method comprising:

[0006] The data collection step collects the coal burning quantity C, power generation Ge and external transmission Oe of a historical period to form a first historical vector data set H1=(C, Ge, Oe), obtains the steam input temperature It and output temperature Ot of the steam turbine to form a second historical vector data set H2=(It, Ot), obtains the historical weather data Hw and environmental images of the external transmission destination to form a third historical vector data set H3=(Hw, Im), obtains the weather data Nw of the external transmission destination at the current moment and the future weather data Fw, and collects M current environmental images of the destination, wherein C={C1, C2, ..., C n}, Ge = {Ge1, Ge2, ..., Ge n}, Oe={Oe1, Oe2,..., Oe n}, It={It1, It2,..., It n}, Ot={Ot1, Ot2,..., Ot n}, Hw={Ot1, Ot2,...,Ot n}, Im={Im1, Im2,..., Im n}, C i ,Ge i , Oe i、It i , Ot i , Hw i 、Im i They represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5;

[0007] A multimodal adjustment model training step, wherein the multimodal adjustment model includes first, second and third branch networks, wherein the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), and the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the weight parameters of the trained first branch network and the second branch network are adjusted based on the training result of the third branch network, and the first, second and third branch networks after the weight parameters are adjusted constitute the trained multimodal adjustment model;

[0008] The coal feeder control step collects the weather data Nw of the destination of the external power transmission at the current moment, the future weather data Fw, and the M environmental images, and inputs them into the trained multimodal adjustment model to obtain the required coal burning quantity FC in the future period, and generates a control signal of the coal feeder based on the required coal burning quantity FC in the future period, and controls the speed and coal supply quantity of the coal feeder based on the control signal.

[0009] Furthermore, the first and second branch networks are LSTM networks, and the third branch network is a CNN network.

[0010] Furthermore, the amount of coal required to be burned in the future period of time FC is the amount of coal required to be burned in the next 48 hours, FC = {FC1, FC2, ..., FC 48}, FC1, FC2, ..., FC 48 They respectively represent the amount of coal needed to be burned in the next 1, 2, ..., 48 hours.

[0011] Furthermore, the operation of adjusting the weight parameters of the trained first branch network and the second branch network based on the training results of the third branch network is: adjusting the weight parameters of the trained second branch network based on the training results of the third branch network, using the first test data set to test whether the performance of the second branch network after adjusting the weight parameters is better than that of the second branch network before the weight adjustment; if so, adjusting the weight parameters of the trained first branch network based on the training results of the third branch network and the weight parameters of the adjusted second branch network after training, using the second test data set to test whether the performance of the first branch network after adjusting the weight parameters is better than that of the first branch network before the weight adjustment; if so, the weight parameter adjustment ends.

[0012] Furthermore, the operation of training the third branch network based on the third historical vector data set H3 = (Hw, Im) is: using a convolutional layer to extract Im = {Im1, Im2, ..., Im n}, and compare the feature map of each image with the corresponding historical weather Hw = {Ot1, Ot2, ..., Ot n} Perform feature fusion to form a fusion vector Hm = {Hm1, Hm2, ..., Hm n}, use the fusion vector Hm to train the third branch network.

[0013] Furthermore, the operation of adjusting the weight parameters of the trained second branch network based on the training result of the third branch network is: using the third test data to input the trained third branch network to obtain the predicted output F3 (Hm j ), each F3(Hm j ) corresponds to the label value Y3 j , where 1≤j≤m, m≥2;

[0014] The weight parameters of the second branch network after adjustment training are:

[0015]

[0016] Wherein, W2′ represents the matrix of weight parameters of the second branch network after adjustment, and W2 represents the matrix of weight parameters of the second branch network before adjustment;

[0017] Furthermore, the operation of adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and adjusting the weight parameters of the trained second branch network is:

[0018] When the performance of the second branch network after adjusting the weight parameters in the first test data set is better than that of the second branch network before adjusting the weights, the prediction output F2((C j ,Ge j , Oe j )), each F2((C j ,Ge j , Oe j ))The corresponding label value is Y2 j ;

[0019]

[0020] Among them, W1′ represents the matrix of weight parameters of the first branch network after adjustment, and W1 represents the matrix of weight parameters of the first branch network before adjustment.

[0021] The present invention also proposes a coal feeder control device based on multi-modal thermal power energy consumption analysis, the device comprising:

[0022] The data collection unit collects the amount of coal burned C, the amount of power generated Ge and the amount of external transmission Oe in a historical period to form a first historical vector data set H1=(C, Ge, Oe), obtains the steam input temperature It and the output temperature Ot of the steam turbine to form a second historical vector data set H2=(It, Ot), obtains the historical weather data Hw and the environmental image of the destination of the external transmission to form a third historical vector data set H3=(Hw, Im), obtains the weather data Nw of the destination of the external transmission at the current moment and the future weather data Fw, and collects the current M environmental images of the destination, wherein C={C1, C2, ..., C n}, Ge = {Ge1, Ge2, ..., Ge n}, Oe={Oe1, Oe2,..., Oe n}, It={It1, It2,..., It n}, Ot={Ot1, Ot2,..., Ot n}, Hw={Ot1, Ot2,...,Ot n}, Im={Im1, Im2,..., Im n}, C i ,Ge i , Oe i 、It i , Ot i , Hw i 、Im i They represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5;

[0023] A multimodal adjustment model training unit, wherein the multimodal adjustment model includes first, second and third branch networks, wherein the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), and the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the weight parameters of the trained first branch network and the second branch network are adjusted based on the training result of the third branch network, and the first, second and third branch networks after the weight parameters are adjusted constitute the trained multimodal adjustment model;

[0024] The coal feeder control unit collects the weather data Nw of the destination of the external power transmission at the current moment, the future weather data Fw and the M environmental images and inputs them into the trained multimodal adjustment model to obtain the required coal burning quantity FC in the future period, generates a control signal of the coal feeder based on the required coal burning quantity FC in the future period, and controls the speed and coal supply quantity of the coal feeder based on the control signal.

[0025] Furthermore, the first and second branch networks are LSTM networks, and the third branch network is a CNN network.

[0026] Furthermore, the amount of coal required to be burned in the future period of time FC is the amount of coal required to be burned in the next 48 hours, FC = {FC1, FC2, ..., FC 48}, FC1, FC2, ..., FC 48 They respectively represent the amount of coal needed to be burned in the next 1, 2, ..., 48 hours.

[0027] Furthermore, the operation of adjusting the weight parameters of the trained first branch network and the second branch network based on the training results of the third branch network is: adjusting the weight parameters of the trained second branch network based on the training results of the third branch network, using the first test data set to test whether the performance of the second branch network after adjusting the weight parameters is better than that of the second branch network before the weight adjustment; if so, adjusting the weight parameters of the trained first branch network based on the training results of the third branch network and the weight parameters of the adjusted second branch network after training, using the second test data set to test whether the performance of the first branch network after adjusting the weight parameters is better than that of the first branch network before the weight adjustment; if so, the weight parameter adjustment ends.

[0028] Furthermore, the operation of training the third branch network based on the third historical vector data set H3 = (Hw, Im) is: using a convolutional layer to extract Im = {Im1, Im2, ..., Im n}, and compare the feature map of each image with the corresponding historical weather Hw = {Ot1, Ot2, ..., Ot n} Perform feature fusion to form a fusion vector Hm = {Hm1, Hm2, ..., Hm n}, use the fusion vector Hm to train the third branch network.

[0029] Furthermore, the operation of adjusting the weight parameters of the trained second branch network based on the training result of the third branch network is: using the third test data to input the trained third branch network to obtain the predicted output F3 (Hm j ), each F3(Hm j ) corresponds to the label value Y3 j , where 1≤j≤m, m≥2;

[0030] The weight parameters of the second branch network after adjustment training are:

[0031]

[0032] Wherein, W2′ represents the matrix of weight parameters of the second branch network after adjustment, and W2 represents the matrix of weight parameters of the second branch network before adjustment;

[0033] Furthermore, the operation of adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and adjusting the weight parameters of the trained second branch network is:

[0034] When the performance of the second branch network after adjusting the weight parameters in the first test data set is better than that of the second branch network before adjusting the weights, the prediction output F2((C j ,Ge j , Oe j )), each F2((C j ,Ge j , Oe j ))The corresponding label value is Y2 j ;

[0035]

[0036] Among them, W1′ represents the matrix of weight parameters of the first branch network after adjustment, and W1 represents the matrix of weight parameters of the first branch network before adjustment.

[0037] The present invention further provides a computer-readable storage medium, wherein the storage medium stores computer program code, and when the computer program code is executed by a computer, any of the above methods is executed.

[0038] The technical effect of the present invention is: a coal feeder control method, device and storage medium based on multi-modal thermal power energy consumption analysis of the present invention, the method includes: a data collection step S101, collecting the amount of coal burned C, the amount of power generation Ge and the amount of external transmission Oe in a historical period to form a first historical vector data set H1 = (C, Ge, Oe), obtaining the steam input temperature It and the output temperature Ot of the steam turbine to form a second historical vector data set H2 = (It, Ot), obtaining the historical weather data Hw and environmental images of the destination of the external transmission to form a third historical vector data set H3 = (Hw, Im), obtaining the weather data Nw of the destination of the external transmission at the current moment and the future weather data Fw, and collecting M current environmental images of the destination, wherein C = {C1, C2, ..., C n}, Ge = {Ge1, Ge2, ..., Ge n}, Oe={Oe1, Oe2,..., Oe n}, It={It1, It2,..., It n}, Ot={Ot1, Ot2,..., Ot n}, Hw={Ot1, Ot2,...,Ot n}, Im={Im1, Im2,..., Im n}, C i ,Ge i , Oe i 、It i , Ot i , Hw i 、Im iRespectively represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5; multimodal adjustment model training step S102, the multimodal adjustment model includes first, second, and third branch networks, the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the third branch network is adjusted based on the training result of the third branch network. The weight parameters of the first branch network and the second branch network after training are adjusted, and the trained multimodal adjustment model is formed based on the first, second branch networks and the third branch network after the weight parameters are adjusted; the coal feeder control step S103, the weather data Nw of the external transmission power destination at the current moment, the future weather data Fw and the M environmental images are input into the trained multimodal adjustment model to obtain the required coal burning quantity FC in the future period of time, and the control signal of the coal feeder is generated based on the required coal burning quantity FC in the future period of time, and the speed and coal feeding quantity of the coal feeder are controlled based on the control signal. The important inventive concept of the present invention is to form the first historical vector data set with the collected coal burning quantity C, power generation Ge and external transmission power Oe, obtain the steam input temperature It and output temperature Ot of the steam turbine to form the second historical vector data set, obtain the historical weather data Hw and environmental image of the external transmission power destination to form the third historical vector data set, and train different branches of the multimodal adjustment model respectively, so that there is no need to form the data into a large data vector for training a single neural network, and adjust the weight parameters of the trained first branch network and the second branch network based on the training result of the third branch network, that is, after the training of the three branch networks is completed, the joint parameters will be adjusted, which improves the training speed, and adds the image data of the destination during the training process. The image data more intuitively reflects the local climate conditions, and is integrated with multiple local images and weather data to make the trained third branch neural network more accurate, so that the subsequent control of the coal feeder is more precise, thereby saving the amount of coal burned, which not only reduces costs but also is beneficial to environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Other features, objects and advantages of the present application will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings.

[0040] Figure 1 The present invention is a flowchart of a coal feeder control method based on multi-modal thermal power energy consumption analysis according to an embodiment of the present invention.

[0041] Figure 2It is a structural diagram of a coal feeder control device based on multi-modal thermal power energy consumption analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0043] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0044] Figure 1 A coal feeder control method based on multi-modal thermal power energy consumption analysis of the present invention is shown, and the method comprises:

[0045] Data collection step S101, collects the coal burning quantity C, power generation Ge and external transmission Oe of a historical period to form a first historical vector data set H1 = (C, Ge, Oe), obtains the steam input temperature It and output temperature Ot of the steam turbine to form a second historical vector data set H2 = (It, Ot), obtains the historical weather data Hw and environmental images of the external transmission destination to form a third historical vector data set H3 = (Hw, Im), obtains the weather data Nw of the external transmission destination at the current moment and the future weather data Fw, and collects M current environmental images of the destination, where C = {C1, C2, ..., C n}, Ge = {Ge1, Ge2, ..., Ge n}, Oe={Oe1, Oe2,..., Oe n}, It={It1, It2,..., It n}, Ot={Ot1, Ot2,..., Ot n}, Hw={Ot1, Ot2,...,Ot n}, Im={Im1, Im2,..., Im n}, C i ,Ge i , Oe i 、It i , Ot i , Hw i 、Im i They represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5;

[0046] A multimodal adjustment model training step S102, wherein the multimodal adjustment model includes first, second and third branch networks, wherein the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), and the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the weight parameters of the trained first branch network and the second branch network are adjusted based on the training result of the third branch network, and the first, second and third branch networks after the weight parameters are adjusted constitute the trained multimodal adjustment model;

[0047] The coal feeder control step S103 collects the weather data Nw of the destination of the external power transmission at the current moment, the future weather data Fw and the M environmental images and inputs them into the trained multimodal adjustment model to obtain the required coal burning quantity FC in the future period, and generates a control signal of the coal feeder based on the required coal burning quantity FC in the future period, and controls the speed and coal supply quantity of the coal feeder based on the control signal.

[0048] The important inventive concept of the present invention is to form the first historical vector data set with the collected coal burning quantity C, power generation Ge and external transmission power Oe, obtain the steam input temperature It and output temperature Ot of the steam turbine to form the second historical vector data set, obtain the historical weather data Hw and environmental image of the external transmission destination to form the third historical vector data set, and train different branches of the multimodal adjustment model respectively, so that there is no need to form the data into a large data vector for training a single neural network, and adjust the weight parameters of the trained first branch network and the second branch network based on the training result of the third branch network, that is, the joint parameters of the three branch networks will be adjusted after the training is completed, which improves the training speed, and adds the image data of the destination during the training process. The image data more intuitively reflects the local climate conditions, and is integrated with multiple local images and weather data to make the trained third branch neural network more accurate, so that the subsequent control of the coal feeder is more precise, thereby saving the amount of coal burned, which not only reduces costs but also is beneficial to environmental protection. This is an important inventive concept of the present invention.

[0049] In one embodiment, the first and second branch networks are LSTM networks, and the third branch network is a CNN network. This is because the data processed by the first and second branches are mainly time-related data, and the data processed by the third branch includes image data. The CNN network can better extract image features and perform feature fusion, which is another important inventive concept of the present invention.

[0050] In one embodiment, the amount of coal required to be burned in the future period of time FC is the amount of coal required to be burned in the next 48 hours, FC = {FC1, FC2, ..., FC 48}, FC1, FC2, ..., FC 48 They respectively represent the amount of coal needed to be burned in the next 1, 2, ..., 48 hours. Frequent adjustment of the coal feeding amount of the coal feeder is not conducive to saving, but will lead to unstable power generation and user experience. Therefore, the power consumption within two days is generally predicted and converted into the amount of coal needed to be burned within the next 48 hours, and then converted into a control signal for the coal feeder, which is conducive to maintaining the overall stability of power generation. This is another inventive point of the present invention.

[0051] In one embodiment, the operation of training the third branch network based on the third historical vector data set H3 = (Hw, Im) is: using a convolutional layer to extract Im = {Im1, Im2, ..., Im n}, and compare the feature map of each image with the corresponding historical weather Hw = {Ot1, Ot2, ..., Ot n} Perform feature fusion to form a fusion vector Hm = {Hm1, Hm2, ..., Hm n}, use the fusion vector Hm to train the third branch network.

[0052] In the present invention, in order to improve the prediction accuracy, instead of just using weather data for prediction, image data of the electricity consumption area is added, and the image data is fused with the weather data to train the neural network, so that the training data can more accurately reflect the actual situation of the electricity consumption area, making the trained neural network more accurate, and the predicted electricity consumption data is more accurate after being converted into the amount of coal burned. This is another inventive point of the present invention.

[0053] In one embodiment, the operation of adjusting the weight parameters of the trained first branch network and the second branch network based on the training results of the third branch network is: adjusting the weight parameters of the trained second branch network based on the training results of the third branch network, using the first test data set to test whether the performance of the second branch network after adjusting the weight parameters is better than that of the second branch network before the weight adjustment, if so, adjusting the weight parameters of the trained first branch network based on the training results of the third branch network and adjusting the weight parameters of the trained second branch network, using the second test data set to test whether the performance of the first branch network after adjusting the weight parameters is better than that of the first branch network before the weight adjustment, if so, the weight parameter adjustment ends.

[0054] In one embodiment, the operation of adjusting the weight parameters of the trained second branch network based on the training result of the third branch network is: using the third test data to input the trained third branch network to obtain the predicted output F3 (Hm j ), each F3(Hm j ) corresponds to the label value Y3 j , where 1≤j≤m, m≥2;

[0055] The weight parameters of the second branch network after adjustment training are:

[0056]

[0057] Wherein, W2′ represents the matrix of weight parameters of the second branch network after adjustment, and W2 represents the matrix of weight parameters of the second branch network before adjustment;

[0058] In one embodiment, the operation of adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and adjusting the weight parameters of the trained second branch network is:

[0059] When the performance of the second branch network after adjusting the weight parameters in the first test data set is better than that of the second branch network before adjusting the weights, the prediction output F2((C j ,Ge j , Oe j )), each F2((C j ,Ge j , Oe j ))The corresponding label value is Y2 j ;

[0060]

[0061] Among them, W1′ represents the matrix of weight parameters of the first branch network after adjustment, and W1 represents the matrix of weight parameters of the first branch network before adjustment.

[0062] In the present invention, a specific method for jointly adjusting weight parameters between branch neural networks is proposed. Using this method, there is no need to retrain the entire neural network. Only the parameters of the corresponding branch neural network need to be adjusted according to the output of the test data set, which greatly reduces the training time of the neural network and keeps the accuracy basically the same. This is another important inventive concept of the present invention.

[0063] Figure 2 A coal feeder control device based on multi-modal thermal power energy consumption analysis of the present invention is shown, and the device comprises:

[0064] The data collection unit 201 collects the amount of coal burned C, the amount of power generated Ge, and the amount of external transmission Oe in a historical period to form a first historical vector data set H1=(C, Ge, Oe), obtains the steam input temperature It and the output temperature Ot of the steam turbine to form a second historical vector data set H2=(It, Ot), obtains the historical weather data Hw and the environmental image of the destination of the external transmission to form a third historical vector data set H3=(Hw, Im), obtains the weather data Nw of the destination of the external transmission at the current moment and the future weather data Fw, and collects M current environmental images of the destination, wherein C={C1, C2, ..., C n}, Ge = {Ge1, Ge2, ..., Ge n}, Oe={Oe1, Oe2,..., Oe n}, It={It1, It2,..., It n}, Ot={Ot1, Ot2,..., Ot n}, Hw={Ot1, Ot2,...,Ot n}, Im={Im1, Im2,..., Im n}, C i ,Ge i , Oe i 、It i , Ot i , Hw i 、Im i They represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5;

[0065] A multimodal adjustment model training unit 202, wherein the multimodal adjustment model includes first, second and third branch networks, wherein the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), and the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the weight parameters of the trained first branch network and the second branch network are adjusted based on the training result of the third branch network, and the first, second and third branch networks after the weight parameters are adjusted constitute the trained multimodal adjustment model;

[0066] The coal feeder control unit 203 collects the weather data Nw of the destination of the external power transmission at the current moment, the future weather data Fw and the M environmental images and inputs them into the trained multimodal adjustment model to obtain the required coal burning quantity FC in the future period, generates a control signal of the coal feeder based on the required coal burning quantity FC in the future period, and controls the speed and coal supply quantity of the coal feeder based on the control signal.

[0067] The important inventive concept of the present invention is to form the first historical vector data set with the collected coal burning quantity C, power generation Ge and external transmission power Oe, obtain the steam input temperature It and output temperature Ot of the steam turbine to form the second historical vector data set, obtain the historical weather data Hw and environmental image of the external transmission destination to form the third historical vector data set, and train different branches of the multimodal adjustment model respectively, so that there is no need to form the data into a large data vector for training a single neural network, and adjust the weight parameters of the trained first branch network and the second branch network based on the training result of the third branch network, that is, the joint parameters of the three branch networks will be adjusted after the training is completed, which improves the training speed, and adds the image data of the destination during the training process. The image data more intuitively reflects the local climate conditions, and is integrated with multiple local images and weather data to make the trained third branch neural network more accurate, so that the subsequent control of the coal feeder is more precise, thereby saving the amount of coal burned, which not only reduces costs but also is beneficial to environmental protection. This is an important inventive concept of the present invention.

[0068] In one embodiment, the first and second branch networks are LSTM networks, and the third branch network is a CNN network. This is because the data processed by the first and second branches are mainly time-related data, and the data processed by the third branch includes image data. The CNN network can better extract image features and perform feature fusion, which is another important inventive concept of the present invention.

[0069] In one embodiment, the amount of coal required to be burned in the future period of time FC is the amount of coal required to be burned in the next 48 hours, FC = {FC1, FC2, ..., FC 48}, FC1, FC2, ..., FC 48 They respectively represent the amount of coal needed to be burned in the next 1, 2, ..., 48 hours. Frequent adjustment of the coal feeding amount of the coal feeder is not conducive to saving, but will lead to unstable power generation and user experience. Therefore, the power consumption within two days is generally predicted and converted into the amount of coal needed to be burned within the next 48 hours, and then converted into a control signal for the coal feeder, which is conducive to maintaining the overall stability of power generation. This is another inventive point of the present invention.

[0070] In one embodiment, the operation of training the third branch network based on the third historical vector data set H3 = (Hw, Im) is: using a convolutional layer to extract Im = {Im1, Im2, ..., Im n}, and compare the feature map of each image with the corresponding historical weather Hw = {Ot1, Ot2, ..., Ot n} Perform feature fusion to form a fusion vector Hm = {Hm1, Hm2, ..., Hm n}, use the fusion vector Hm to train the third branch network.

[0071] In the present invention, in order to improve the prediction accuracy, instead of just using weather data for prediction, image data of the electricity consumption area is added, and the image data is fused with the weather data to train the neural network, so that the training data can more accurately reflect the actual situation of the electricity consumption area, making the trained neural network more accurate, and the predicted electricity consumption data is more accurate after being converted into the amount of coal burned. This is another inventive point of the present invention.

[0072] In one embodiment, the operation of adjusting the weight parameters of the trained first branch network and the second branch network based on the training results of the third branch network is: adjusting the weight parameters of the trained second branch network based on the training results of the third branch network, using the first test data set to test whether the performance of the second branch network after adjusting the weight parameters is better than that of the second branch network before the weight adjustment, if so, adjusting the weight parameters of the trained first branch network based on the training results of the third branch network and adjusting the weight parameters of the trained second branch network, using the second test data set to test whether the performance of the first branch network after adjusting the weight parameters is better than that of the first branch network before the weight adjustment, if so, the weight parameter adjustment ends.

[0073] In one embodiment, the operation of adjusting the weight parameters of the trained second branch network based on the training result of the third branch network is: using the third test data to input the trained third branch network to obtain the predicted output F3 (Hm j ), each F3(Hm j ) corresponds to the label value Y3 j , where 1≤j≤m, m≥2;

[0074] The weight parameters of the second branch network after adjustment training are:

[0075]

[0076] Wherein, W2′ represents the matrix of weight parameters of the second branch network after adjustment, and W2 represents the matrix of weight parameters of the second branch network before adjustment;

[0077] In one embodiment, the operation of adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and adjusting the weight parameters of the trained second branch network is:

[0078] When the performance of the second branch network after adjusting the weight parameters in the first test data set is better than that of the second branch network before adjusting the weights, the prediction output F2((C j ,Ge j , Oe j )), each F2((C j ,Ge j , Oe j ))The corresponding label value is Y2 j ;

[0079]

[0080] Among them, W1′ represents the matrix of weight parameters of the first branch network after adjustment, and W1 represents the matrix of weight parameters of the first branch network before adjustment.

[0081] In the present invention, a specific method for jointly adjusting weight parameters between branch neural networks is proposed. Using this method, there is no need to retrain the entire neural network. Only the parameters of the corresponding branch neural network need to be adjusted according to the output of the test data set, which greatly reduces the training time of the neural network and keeps the accuracy basically the same. This is another important inventive concept of the present invention.

[0082] In one embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored. When the computer program on the computer storage medium is executed by a processor, the above method is implemented. The computer storage medium can be a hard disk, DVD, CD, flash memory or other memory.

[0083] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0084] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the devices described in the various embodiments of the present application or certain parts of the embodiments.

[0085] Finally, it should be noted that the above embodiments are only intended to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the present invention can still be modified or replaced by equivalents. Any modification or partial replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A coal feeder control method based on multi-modal thermal power energy consumption analysis, characterized in that: The method includes: The data collection step collects the coal burning quantity C, power generation Ge and external transmission Oe of a historical period to form a first historical vector data set H1=(C, Ge, Oe), obtains the steam input temperature It and output temperature Ot of the steam turbine to form a second historical vector data set H2=(It, Ot), obtains the historical weather data Hw and environmental images of the external transmission destination to form a third historical vector data set H3=(Hw, Im), obtains the weather data Nw of the external transmission destination at the current moment and the future weather data Fw, and collects M current environmental images of the destination, wherein C={C1, C2, ..., C n }, Ge = {Ge1, Ge2, ..., Ge n }, Oe={Oe1, Oe2,..., Oe n }, It={It1, It2,..., It n }, Ot={Ot1, Ot2,..., Ot n }, Hw={Ot1, Ot2,..., Ot n }, Im={Im1, Im2,..., Im n }, C i ,Ge i , Oe i 、It i , Ot i , Hw i 、Im i They represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5; A multimodal adjustment model training step, wherein the multimodal adjustment model includes first, second and third branch networks, wherein the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), and the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the weight parameters of the trained first branch network and the second branch network are adjusted based on the training result of the third branch network, and the first, second and third branch networks after the weight parameters are adjusted constitute the trained multimodal adjustment model; A coal feeder control step, collecting the weather data Nw of the destination of the external power transmission at the current moment, the future weather data Fw and the M environmental images, inputting them into the trained multimodal adjustment model, obtaining the required coal burning quantity FC in the future period, generating a control signal of the coal feeder based on the required coal burning quantity FC in the future period, and controlling the speed and coal feeding quantity of the coal feeder based on the control signal; The first and second branch networks are LSTM networks, and the third branch network is a CNN network; The amount of coal required to be burned in the future period of time FC is the amount of coal required to be burned in the next 48 hours, FC = {FC1, FC2, ..., FC 48 }, FC1, FC2, ..., FC 48 Respectively represent the amount of coal that needs to be burned in the future 1, 2, ..., 48 hours; the operation of adjusting the weight parameters of the trained first branch network and the second branch network based on the training result of the third branch network is: adjusting the weight parameters of the trained second branch network based on the training result of the third branch network, using the first test data set to test whether the performance of the second branch network after adjusting the weight parameters is better than the second branch network before adjusting the weights, if so, adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and the weight parameters of the adjusted second branch network, using the second test data set to test whether the performance of the first branch network after adjusting the weight parameters is better than the first branch network before adjusting the weights, if so, the weight parameter adjustment ends; the operation of training the third branch network based on the third historical vector data set H3 = (Hw, Im) is: using a convolutional layer to extract Im = {Im1, Im2, ..., Im n }, and compare the feature map of each image with the corresponding historical weather Hw = {Ot1, Ot2, ..., Ot n } Perform feature fusion to form a fusion vector Hm = {Hm1, Hm2, ..., Hm n }, use the fusion vector Hm to train the third branch network; The operation of adjusting the weight parameters of the trained second branch network based on the training result of the third branch network is: using the third test data to input the trained third branch network to obtain the predicted output F3 (Hm j ), each F3(Hm j ) corresponds to the label value Y3 j , where 1≤j≤m, m≥2; The weight parameters of the second branch network after adjustment training are: Among them, W2 ′ represents the matrix of weight parameters of the second branch network after adjustment, and W2 represents the matrix of weight parameters of the second branch network before adjustment; The operation of adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and adjusting the weight parameters of the trained second branch network is: When the performance of the second branch network after adjusting the weight parameters in the first test data set is better than that of the second branch network before adjusting the weights, the prediction output F2((C j ,Ge j , Oe j )), each F2((C j ,Ge j , Oe j ))The corresponding label value is Y2 j ; Among them, W1 ′ W1 represents the matrix of weight parameters of the first branch network after adjustment, and W2 represents the matrix of weight parameters of the first branch network before adjustment.

2. A coal feeder control device based on multi-modal thermal power energy consumption analysis, characterized in that: The device includes: The data collection unit collects the amount of coal burned C, the amount of power generated Ge and the amount of external transmission Oe in a historical period to form a first historical vector data set H1=(C, Ge, Oe), obtains the steam input temperature It and the output temperature Ot of the steam turbine to form a second historical vector data set H2=(It, Ot), obtains the historical weather data Hw and the environmental image of the destination of the external transmission to form a third historical vector data set H3=(Hw, Im), obtains the weather data Nw of the destination of the external transmission at the current moment and the future weather data Fw, and collects the current M environmental images of the destination, wherein C={C1, C2, ..., C n }, Ge = {Ge1, Ge2, ..., Ge n }, Oe={Oe1, Oe2,..., Oe n }, It={It1, It2,..., It n }, Ot={Ot1, Ot2,..., Ot n }, Hw={Ot1, Ot2,..., Ot n }, Im={Im1, Im2,..., Im n }, C i ,Ge i , Oe i 、It i , Ot i , Hw i 、Im i They represent the amount of coal burned, power generation, external power transmission, steam input temperature, output temperature, weather data, and environmental image at a historical moment, 1≤i≤n, n≥5; A multimodal adjustment model training unit, wherein the multimodal adjustment model includes first, second and third branch networks, wherein the first branch network is trained based on the first historical vector data set H1=(C, Ge, Oe), the second branch network is trained based on the second historical vector H2=(It, Ot), and the third branch network is trained based on the third historical vector data set H3=(Hw, Im), and the weight parameters of the trained first branch network and the second branch network are adjusted based on the training result of the third branch network, and the first, second and third branch networks after the weight parameters are adjusted constitute the trained multimodal adjustment model; The coal feeder control unit collects the weather data Nw of the destination of the external power transmission at the current moment, the future weather data Fw and the M environmental images and inputs them into the trained multimodal adjustment model to obtain the amount of coal to be burned in the future period FC, generates a control signal of the coal feeder based on the amount of coal to be burned in the future period FC, and controls the speed and amount of coal to be fed of the coal feeder based on the control signal; the first and second branch networks are LSTM networks, and the third branch network is a CNN network; the amount of coal to be burned in the future period FC is the amount of coal to be burned in the next 48 hours, FC={FC1, FC2, ..., FC 48 }, FC1, FC2, ..., FC 48 Respectively represent the amount of coal that needs to be burned in the next 1, 2, ..., 48 hours; the operation of adjusting the weight parameters of the trained first branch network and the second branch network based on the training result of the third branch network is: adjusting the weight parameters of the trained second branch network based on the training result of the third branch network, using the first test data set to test whether the performance of the second branch network after adjusting the weight parameters is better than that of the second branch network before adjusting the weights, if yes, then adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and the weight parameters of the adjusted second branch network, using the second test data set to test whether the performance of the first branch network after adjusting the weight parameters is better than that of the first branch network before adjusting the weights, if yes, then the weight parameter adjustment ends; The operation of training the third branch network based on the third historical vector data set H3=(Hw, Im) is: using a convolutional layer to extract Im={Im1, Im2, ..., Im n }, and compare the feature map of each image with the corresponding historical weather Hw = {Ot1, Ot2, ..., Ot n } Perform feature fusion to form a fusion vector Hm = {Hm1, Hm2, ..., Hm n }, use the fusion vector Hm for training; The operation of adjusting the weight parameters of the trained second branch network based on the training result of the third branch network is: using the third test data to input the trained third branch network to obtain the predicted output F3 (Hm j ), each F3(Hm j ) corresponds to the label value Y3 j , where 1≤j≤m, m≥2; The weight parameters of the second branch network after adjustment training are: Among them, W2 ′ represents the matrix of weight parameters of the second branch network after adjustment, and W2 represents the matrix of weight parameters of the second branch network before adjustment; The operation of adjusting the weight parameters of the trained first branch network based on the training result of the third branch network and adjusting the weight parameters of the trained second branch network is: When the performance of the second branch network after adjusting the weight parameters in the first test data set is better than that of the second branch network before adjusting the weights, the prediction output F2((C j ,Ge j , Oe j )), each F2((C j ,Ge j , Oe j ))The corresponding label value is Y2 j ; Among them, W1 ′ W1 represents the matrix of weight parameters of the first branch network after adjustment, and W2 represents the matrix of weight parameters of the first branch network before adjustment.

3. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and when the computer program on the computer storage medium is executed by a processor, the method according to claim 1 is implemented.

Images