CN113408802A - Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment - Google Patents

Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment Download PDF

Info

Publication number
CN113408802A
CN113408802A CN202110689983.4A CN202110689983A CN113408802A CN 113408802 A CN113408802 A CN 113408802A CN 202110689983 A CN202110689983 A CN 202110689983A CN 113408802 A CN113408802 A CN 113408802A
Authority
CN
China
Prior art keywords
energy consumption
network
trained
characteristic
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110689983.4A
Other languages
Chinese (zh)
Other versions
CN113408802B (en
Inventor
罗洪江
何恒靖
吴昊文
杜浩东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Southern Power Grid Digital Grid Technology Guangdong Co ltd
Original Assignee
Southern Power Grid Digital Grid Research Institute Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southern Power Grid Digital Grid Research Institute Co Ltd filed Critical Southern Power Grid Digital Grid Research Institute Co Ltd
Priority to CN202110689983.4A priority Critical patent/CN113408802B/en
Publication of CN113408802A publication Critical patent/CN113408802A/en
Application granted granted Critical
Publication of CN113408802B publication Critical patent/CN113408802B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Tourism & Hospitality (AREA)
  • Manufacturing & Machinery (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Primary Health Care (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application relates to a method and a device for training an energy consumption prediction network, an energy consumption prediction method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring historical energy consumption data and actual energy consumption values of the tag moments; inputting the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and inputting the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; inputting the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; inputting the fusion energy consumption characteristics to the characteristic classification network to be trained to obtain a predicted energy consumption value; and adjusting the network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained. By adopting the method, the energy consumption monitoring efficiency can be improved.

Description

Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment
Technical Field
The present application relates to the field of computer technologies, and in particular, to a method and an apparatus for training an energy consumption prediction network, an energy consumption prediction method and apparatus, a computer device, and a storage medium.
Background
The rapid development of economy and the gradual rise of industrial energy consumption are achieved, and the development of more efficiently serving the energy to the society by adopting an energy consumption monitoring mode becomes a research hotspot at present. For industrial plants, the energy consumption is not only the total energy consumption generated in a certain period of time, but also the energy efficiency monitoring becomes very complex and expensive for the possible industrial energy consumption located at the edge side with poor connectivity, so that accurate, effective and reasonable edge side embedded lightweight industrial energy consumption monitoring is the key basis for industrial energy consumption energy saving and control, and is also the important premise for making relevant energy saving laws and policies and developing energy saving related work.
In the related technology, complicated and tedious parameter operation is often required when industrial energy consumption prediction is carried out, the number of processing parameters is large, the calculated amount is large, the prediction cannot be deployed on mobile and edge equipment, manual intervention and adjustment are used in actual energy consumption monitoring, the monitoring accuracy and usability are greatly influenced, the purposes of monitoring the energy consumption in the industrial production process and improving the energy utilization rate are not achieved, and efficient energy consumption monitoring cannot be achieved.
Disclosure of Invention
In view of the foregoing, it is desirable to provide a method, an apparatus, a computer device, and a storage medium for training an energy consumption prediction network, which can improve energy consumption monitoring efficiency.
A training method of an energy consumption prediction network, wherein the energy consumption prediction network comprises a first feature extraction network to be trained, a second feature extraction network to be trained, a feature fusion network to be trained and a feature classification network to be trained, and the method comprises the following steps:
acquiring training sample data; the training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment;
inputting the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and inputting the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic;
inputting the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained;
inputting the fusion energy consumption characteristics to the characteristic classification network to be trained to obtain a predicted energy consumption value;
adjusting network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
In one embodiment, the adjusting the network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained includes:
obtaining a network loss value of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value;
and if the network loss value is not within the preset threshold value range, adjusting the network parameters of the energy consumption prediction network according to the network loss value until the network loss value is lower than the preset threshold value, and taking the energy consumption prediction network after network parameter adjustment as the trained energy consumption prediction network.
In one embodiment, the adjusting the network parameter of the energy consumption prediction network according to the network loss value includes:
determining the gradient of each network node in the energy consumption prediction network based on the network loss value by adopting a back propagation method;
and updating the parameters of each network node according to the gradient by adopting a random gradient descending method to obtain the energy consumption prediction network after the network parameters are adjusted.
In one embodiment, the feature fusion network to be trained includes a feature adjustment layer, a feature fusion layer, and a full connection layer, and the inputting the first energy consumption feature and the second energy consumption feature into the feature fusion network to be trained to obtain a fusion energy consumption feature includes:
adjusting the first energy consumption characteristic and the second energy consumption characteristic through the characteristic adjusting layer so as to maximize the correlation between the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic;
fusing the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic through the characteristic fusion layer to obtain an initial fusion characteristic;
and carrying out full connection processing on the initial fusion characteristics through the full connection layer to obtain the fusion energy consumption characteristics.
In one embodiment, the acquiring training sample data includes:
acquiring original energy consumption data;
performing redundancy removal processing on the original energy consumption data to obtain redundancy removal energy consumption data; the characteristic redundancy degree of the redundancy-removing energy consumption data is smaller than a preset threshold value;
and carrying out normalization processing on the redundancy-removing energy consumption data to obtain the training sample data.
In one embodiment, the performing redundancy removal processing on the original energy consumption data to obtain redundancy-removed energy consumption data includes:
determining mutual information quantity corresponding to each energy consumption variable based on a mutual information feature extraction method;
sequencing the energy consumption variables based on the mutual information quantity to obtain the sequenced energy consumption variables;
determining a target energy consumption variable from the sequenced energy consumption variables; the mutual information quantity of the target energy consumption variables is greater than the energy consumption variables except the target energy consumption variables in the sequenced energy consumption variables;
and taking the characteristic data corresponding to the target energy consumption variable as the redundancy-removing energy consumption data.
A method of energy consumption prediction, the method comprising:
acquiring a trained energy consumption prediction network; the trained energy consumption prediction network is obtained by training according to the method; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network;
inputting energy consumption data to be processed into the trained first feature extraction network to obtain energy consumption convolution features, and inputting historical energy consumption data into the trained second feature extraction network to obtain energy consumption deep separation convolution features; the energy consumption data to be processed is an energy consumption data time sequence before the prediction moment;
inputting the energy consumption convolution characteristics and the energy consumption deep separation convolution characteristics into the trained characteristic fusion network to obtain fusion energy consumption characteristics; the fusion energy consumption feature is obtained by maximizing the correlation between the energy consumption convolution feature and the energy consumption deep separation convolution feature by the trained feature fusion network;
and inputting the fusion energy consumption characteristics to the trained characteristic classification network to obtain a predicted energy consumption value corresponding to the prediction moment.
An apparatus for training an energy consumption prediction network including a first feature extraction network to be trained, a second feature extraction network to be trained, a feature fusion network to be trained, and a feature classification network to be trained, the apparatus comprising:
the first acquisition module is used for acquiring training sample data; the training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment;
the first extraction module is used for inputting the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and inputting the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic;
the first fusion module is used for inputting the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained;
the first classification module is used for inputting the fusion energy consumption characteristics to the characteristic classification network to be trained to obtain a predicted energy consumption value;
a first adjusting module, configured to adjust a network parameter of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
An energy consumption prediction apparatus, the apparatus comprising:
the second acquisition module is used for acquiring the trained energy consumption prediction network; the trained energy consumption prediction network is obtained by training according to the method; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network;
the second extraction module is used for inputting the energy consumption data to be processed into the trained first feature extraction network to obtain energy consumption convolution features, and inputting the historical energy consumption data into the trained second feature extraction network to obtain energy consumption deep separation convolution features; the energy consumption data to be processed is an energy consumption data time sequence before the prediction moment;
the second fusion module is used for inputting the energy consumption convolution characteristics and the energy consumption deep separation convolution characteristics into the trained characteristic fusion network to obtain fusion energy consumption characteristics; the fusion energy consumption feature is obtained by maximizing the correlation between the energy consumption convolution feature and the energy consumption deep separation convolution feature by the trained feature fusion network;
and the second classification module is used for inputting the fusion energy consumption characteristics to the trained characteristic classification network to obtain a predicted energy consumption value corresponding to the prediction moment.
A computer device comprising a memory storing a computer program and a processor implementing the steps of the method described above when executing the computer program.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method.
The training method, the training device, the computer equipment and the storage medium of the energy consumption prediction network acquire training sample data; training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment; inputting historical energy consumption data to a first feature extraction network to be trained to obtain a first energy consumption feature, and inputting historical energy consumption data to a second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic; inputting the first energy consumption characteristic and the second energy consumption characteristic into a characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained; inputting the fusion energy consumption characteristics into a characteristic classification network to be trained to obtain a predicted energy consumption value; adjusting network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network; thus, by adopting the double-branch feature extraction network, the feature extraction processing of different scales and different convolution modes is realized on the historical energy consumption data, so that the feature extraction processing can be sufficiently performed on the input data without adopting a complex model, meanwhile, by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic when the first energy consumption characteristic and the second energy consumption characteristic are fused, can effectively reduce the feature redundancy of the fusion energy consumption features, is convenient for quickly and effectively classifying the fusion energy consumption features in the follow-up process, enables the energy consumption prediction network obtained by training to output accurate and high prediction results, and the energy consumption prediction network is light, the parameter quantity is small, the calculated quantity is small, the deployment on mobile and edge equipment is convenient, and the purposes of monitoring the energy consumption in the industrial production process and improving the energy utilization rate are realized.
Drawings
FIG. 1 is a diagram illustrating an exemplary embodiment of a method for training an energy consumption prediction network;
FIG. 2 is a schematic flow chart illustrating a method for training an energy consumption prediction network according to an embodiment;
FIG. 3 is an architecture diagram of an energy consumption prediction network, in one embodiment;
FIG. 4 is a flow diagram illustrating a method for energy consumption prediction in one embodiment;
FIG. 5 is a flowchart illustrating a method for training an energy consumption prediction network according to another embodiment;
FIG. 6 is a flow diagram illustrating a method for training an energy consumption prediction network, according to an embodiment;
FIG. 7 is a block diagram of an apparatus for training an energy consumption prediction network according to an embodiment;
FIG. 8 is a block diagram of an energy consumption prediction apparatus according to an embodiment;
FIG. 9 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The energy consumption prediction network training method provided by the application can be applied to the application environment shown in fig. 1. In practical applications, the server 110 may be implemented by a stand-alone server or a server cluster composed of a plurality of servers.
In an embodiment, as shown in fig. 2, a method for training an energy consumption prediction network is provided, where the energy consumption prediction network includes a first feature extraction network to be trained, a second feature extraction network to be trained, a feature fusion network to be trained, and a feature classification network to be trained, and as an example, the method is applied to the terminal in fig. 1, and includes the following steps:
step S210, training sample data is obtained.
The training sample data comprises historical energy consumption data and actual energy consumption values at the label moment. In practical applications, the energy consumption data may refer to energy consumption data of an industrial plant.
Wherein the historical energy consumption data is a time series of energy consumption data before the tag time. For example, assuming the tag time is 2 months and 15 days, the historical energy consumption data may be energy consumption data N days before 2 months and 15 days.
In a specific implementation, when the computer device trains the energy consumption prediction network, the computer device may obtain historical energy consumption data and an actual energy consumption value at the tag moment.
Step S220, inputting historical energy consumption data into a first characteristic extraction network to be trained to obtain a first energy consumption characteristic, and inputting historical energy consumption data into a second characteristic extraction network to be trained to obtain a second energy consumption characteristic; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a deep separation convolution characteristic.
In practical application, the first feature extraction network to be trained and the second feature extraction network to be trained are two branches of the energy consumption prediction network. To facilitate understanding by those skilled in the art, FIG. 3 illustratively provides an architectural diagram of an energy consumption prediction network; in practical applications, the first feature extraction network to be trained may be an original ResNet network. The second feature extraction network to be trained may be a base network based on a MobileNet network improvement, which improves the MobileNet v2 network into 22 convolutional layers by decomposing the normal convolution into a point convolution conv and a depth convolution conv dw by means of depth separable convolution.
The original ResNet network consists of 4 parts, the first part consists of 4 convolution layers of 3 x 3, the number of convolution kernels is M/2, the second part consists of 4 convolution layers of 3 x 3, the number of convolution kernels is 2M, the third part consists of 4 convolution layers of 3 x 3, the number of convolution kernels is 8M, the fourth part consists of 4 convolution layers of 3 x 3, and the number of convolution kernels is 32M.
The second feature extraction network to be trained is alternately connected by 5 3 × 3 deep convolutional layers and 51 × 1 dot convolutional layers, the convolutional kernels of the deep convolutional layers are respectively M, 2M, 4M and 8M, and the convolutional kernels of the dot convolutional layers are respectively 3M, 8M, 16M, 32M and 64M; then, convolution blocks alternately connected by 3 × 3 depth convolution layers of which convolution kernels are 16M and 1 × 1 point convolution layers of which convolution kernels are 256M are connected; the last convolution kernel to be concatenated is a 32M 3 x 3 depth convolution layer.
In specific implementation, the computer device inputs the historical energy consumption data to a first feature extraction network to be trained to obtain a first energy consumption feature, and inputs the historical energy consumption data to a second feature extraction network to be trained to obtain a second energy consumption feature.
Step S230, inputting the first energy consumption characteristic and the second energy consumption characteristic into a characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained.
In specific implementation, after the computer device obtains the first energy consumption characteristic and the second energy consumption characteristic again, the computer device inputs the first energy consumption characteristic and the second energy consumption characteristic into a characteristic fusion network to be trained, and a fusion energy consumption characteristic is obtained. In particular, the computer device may perform DCA feature fusion on the feature outputs of the two branches, maximizing the correlation between feature sets.
And S240, inputting the fusion energy consumption characteristics into a characteristic classification network to be trained to obtain a predicted energy consumption value.
Wherein, the feature classification network to be trained may be an SCN classifier.
In a specific implementation, the computer device may input the fusion energy consumption feature to a feature classification network to be trained, so as to perform classification processing on the fusion energy consumption feature to obtain a predicted energy consumption value.
And step S250, adjusting network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained.
The trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
In a specific implementation, after the computer device obtains the predicted energy consumption value, the computer device adjusts the network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained. And the computer equipment updates the weight and the bias of the energy consumption prediction network according to the gradient descent back propagation algorithm. The computer device can calculate the back propagation process from the output layer to the full connection layer in the energy consumption prediction network:
Figure BDA0003125811100000091
Figure BDA0003125811100000092
Figure BDA0003125811100000093
wherein j represents the number of nodes of the SCN classifier; t represents the desired output set of the network, T ═ T1,t2,...tN}；Fi,μRepresenting the feature vector after DCA fusion; fμFeature set representing the network full connectivity layer at the network update μμ＝{F1,μ,.F2,μ,...FN,μ}；βjRepresenting the weight of the SCN classifier to the output layer;
Figure BDA0003125811100000094
representing the weight of the fully connected layer to the SCN classifier; bjRepresents the bias of the fully connected layer to the SCN classifier; k (-) denotes the activation function of the network.
In the training method of the energy consumption prediction network, training sample data is obtained; training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment; inputting historical energy consumption data to a first feature extraction network to be trained to obtain a first energy consumption feature, and inputting historical energy consumption data to a second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic; inputting the first energy consumption characteristic and the second energy consumption characteristic into a characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained; inputting the fusion energy consumption characteristics into a characteristic classification network to be trained to obtain a predicted energy consumption value; adjusting network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network; thus, by adopting the double-branch feature extraction network, the feature extraction processing of different scales and different convolution modes is realized on the historical energy consumption data, so that the feature extraction processing can be sufficiently performed on the input data without adopting a complex model, meanwhile, by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic when the first energy consumption characteristic and the second energy consumption characteristic are fused, can effectively reduce the feature redundancy of the fusion energy consumption features, is convenient for quickly and effectively classifying the fusion energy consumption features in the follow-up process, enables the energy consumption prediction network obtained by training to output accurate and high prediction results, and the energy consumption prediction network is light, the parameter quantity is small, the calculated quantity is small, the deployment on mobile and edge equipment is convenient, and the purposes of monitoring the energy consumption in the industrial production process and improving the energy utilization rate are realized.
In another embodiment, adjusting network parameters of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained comprises: acquiring a network loss value of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value; and if the network loss value is not within the preset threshold value range, adjusting the network parameters of the energy consumption prediction network according to the network loss value until the network loss value is lower than the preset threshold value, and taking the energy consumption prediction network after network parameter adjustment as the trained energy consumption prediction network.
The method for adjusting the network parameters of the energy consumption prediction network according to the network loss value comprises the following steps: determining the gradient of each network node in the energy consumption prediction network by adopting a back propagation method based on the network loss value; and updating the parameters of each network node according to the gradient by adopting a random gradient descent method to obtain the energy consumption prediction network after the network parameters are adjusted.
In the specific implementation, the computer device adjusts the network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until the trained energy consumption prediction network is obtained, and the computer device can obtain the network loss value of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value; in particular, the computer device may input the predicted energy consumption value and the actual energy consumption value to a preset loss function (e.g., Softmax loss function). Then, the computer device may determine whether the network loss value is within a preset threshold range, and if the network loss value is not within the preset threshold range, the computer device adjusts a network parameter of the energy consumption prediction network according to the network loss value until the network loss value is lower than the preset threshold, and uses the energy consumption prediction network after the network parameter adjustment as a trained energy consumption prediction network. The computer equipment can determine the gradient of each network node in the energy consumption prediction network by adopting a back propagation method based on the network loss value; and updating the parameters of each network node according to the gradient by adopting a random gradient descent method to obtain the energy consumption prediction network after the network parameters are adjusted.
Specifically, the computer device may initialize the energy consumption prediction network, and the computer device may set the width hyper-parameter α of the energy consumption prediction network to 0.75, set the resolution hyper-parameter β to 0.714, and set the number M of convolution kernels to 32.
The computer then defines a Softmax loss function for the energy consumption prediction network, where the Softmax loss function may be expressed as:
Figure BDA0003125811100000111
wherein N ispIs the total number of samples, i is the node number, xiIs the input of the ith node, yiFor the output class of the ith node, σ represents all classes, θjAnd thetayiRepresenting an angle parameter;
Figure BDA0003125811100000112
wherein k belongs to [0, m-1], m is an integer and is used for controlling the size of an angle boundary, and m is more than or equal to 1; when m is 1, it is the Softmax loss function.
According to the technical scheme of the embodiment, the network loss value of the energy consumption prediction network is obtained based on the difference between the predicted energy consumption value and the actual energy consumption value; if the network loss value is not within the preset threshold range, determining the gradient of each network node in the energy consumption prediction network by adopting a back propagation method based on the network loss value; updating parameters of each network node according to the gradient by adopting a random gradient descent method to obtain an energy consumption prediction network after network parameter adjustment, and taking the energy consumption prediction network after network parameter adjustment as a trained energy consumption prediction network when the network loss value is lower than a preset threshold value; thus, the energy consumption prediction network can be accurately and effectively trained.
In another embodiment, the feature fusion network to be trained includes a feature adjustment layer, a feature fusion layer and a full connection layer, and the first energy consumption feature and the second energy consumption feature are input to the feature fusion network to be trained to obtain a fusion energy consumption feature, including: adjusting the first energy consumption characteristic and the second energy consumption characteristic through a characteristic adjusting layer so as to maximize the correlation between the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic; fusing the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic through a characteristic fusion layer to obtain an initial fusion characteristic; and carrying out full connection processing on the initial fusion characteristics through a full connection layer to obtain fusion energy consumption characteristics.
The feature fusion network to be trained comprises a feature adjusting layer, a feature fusion layer and a full connection layer.
In the specific implementation, in the process that the computer device inputs the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain the fusion energy consumption characteristic, the computer device can adjust the first energy consumption characteristic and the second energy consumption characteristic through the characteristic adjustment layer so as to maximize the correlation between the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic; fusing the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic through a characteristic fusion layer to obtain an initial fusion characteristic; and carrying out full connection processing on the initial fusion characteristics through a full connection layer to obtain fusion energy consumption characteristics.
In practical application, the computer device can perform DCA feature fusion on the first energy consumption feature and the second energy consumption feature, so as to maximize the correlation among feature sets. DCA maximizes the correlation of corresponding features in the two feature sets while maximizing the differences between the different classes. It is assumed that the collected sample data matrix comes from C separate classes, so that the n-column sample data matrix can be divided into C separate classes, the n-th classiColumns belong to class i, and the jth sample data, i.e. feature vectors, of class i are denoted as Xi,jIs epsilon.X. Defining inter-class scatter matrices, identifying the feature outputs of the two branches allows DCA feature fusion by proving that the hash matrix between classes is still a diagonal matrix. The scatter matrix formula is defined as:
Figure BDA0003125811100000121
Figure BDA0003125811100000122
wherein the content of the first and second substances,
Figure BDA0003125811100000123
and
Figure BDA0003125811100000124
respectively representing all feature setsAnd the average of the i-th class of features.
According to the technical scheme of the embodiment, the first energy consumption characteristic and the second energy consumption characteristic are adjusted through the characteristic adjusting layer, so that the correlation between the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic is maximized; fusing the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic through a characteristic fusion layer to obtain an initial fusion characteristic; performing full-connection processing on the initial fusion characteristics through a full-connection layer to obtain fusion energy consumption characteristics; in this way, the correlation of the corresponding features in the two feature sets can be maximized as much as possible.
In another embodiment, obtaining training sample data comprises: acquiring original energy consumption data; carrying out redundancy removal processing on the original energy consumption data to obtain redundancy-removed energy consumption data; the characteristic redundancy degree of the redundancy-removing energy consumption data is smaller than a preset threshold value; and carrying out normalization processing on the redundancy-removed energy consumption data to obtain training sample data.
The raw energy consumption data may refer to unprocessed industrial energy consumption data.
In the specific implementation, the computer equipment can acquire original energy consumption data in the process of acquiring training sample data by the computer equipment; then, the computer device may perform redundancy elimination on the original energy consumption data to obtain redundancy elimination energy consumption data, so that the characteristic redundancy degree of the redundancy elimination energy consumption data is smaller than a preset threshold. Then, the computer device performs normalization processing on the redundancy-removed energy consumption data to obtain normalized data, and the normalized data is used as training sample data. In practical applications, the normalized data can be expressed as:
Figure BDA0003125811100000131
wherein: a. b is two constants, and a is 0.1, and b is 0.8, which are respectively the maximum value and the minimum value of each group of factor variables; x is the number ofi，x'iRespectively before and after normalization; x is the number ofmax、xminAre respectively a sampleMaximum and minimum values in the data.
According to the technical scheme of the embodiment, original energy consumption data are obtained; performing redundancy removal processing on the original energy consumption data to obtain redundancy removal energy consumption data with characteristic redundancy degree smaller than a preset threshold value; then, carrying out normalization processing on the redundancy-removed energy consumption data to obtain training sample data; therefore, the model data processing amount can be effectively reduced, and the model processing efficiency is improved.
In another embodiment, the method for removing redundancy from the original energy consumption data includes: determining mutual information quantity corresponding to each energy consumption variable based on a mutual information feature extraction method; based on the size of each mutual information quantity, sequencing each energy consumption variable to obtain the sequenced energy consumption variable; and determining a target energy consumption variable from the sequenced energy consumption variables. And taking the characteristic data corresponding to the target energy consumption variable as redundancy-removing energy consumption data.
And the mutual information quantity of the target energy consumption variables is greater than the energy consumption variables except the target energy consumption variables in the sequenced energy consumption variables.
In a specific implementation, the original energy consumption data comprises a plurality of energy consumption variables, and in the process of performing redundancy removal processing on the original energy consumption data by the computer equipment to obtain redundancy-removed energy consumption data, the computer equipment can determine mutual information quantity corresponding to each energy consumption variable based on a mutual information feature extraction method; and sequencing the energy consumption variables based on the mutual information quantity to obtain the sequenced energy consumption variables. Then, the computer device determines a target energy consumption variable from the sorted energy consumption variables as redundancy-free energy consumption data.
For example, the computer device may select features of the normalized industrial energy consumption data X by using a mutual information-based feature extraction method, sort the features according to the size of the mutual information to remove redundant features, and select n consecutive features from the feature data X by using an mRMR increment selection method: x ═ X1,x2,…,xn)。
According to the technical scheme of the embodiment, mutual information quantity corresponding to each energy consumption variable is determined through a mutual information-based feature extraction method; based on the size of each mutual information quantity, sequencing each energy consumption variable to obtain the sequenced energy consumption variable; and determining a target energy consumption variable in the sequenced energy consumption variables, so that redundant variables can be screened from the original energy consumption data, and the data processing capacity of the model is reduced.
In one embodiment, as shown in fig. 4, an energy consumption prediction method is provided, which is exemplified by the method applied to the computer device in fig. 1, and includes the following steps:
step S410, acquiring a trained energy consumption prediction network; the trained energy consumption prediction network is obtained by training according to the method; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
Step S420, inputting the energy consumption data to be processed into a trained first feature extraction network to obtain energy consumption convolution features, and inputting historical energy consumption data into a trained second feature extraction network to obtain energy consumption deep separation convolution features; the energy consumption data to be processed is an energy consumption data time sequence before the prediction moment.
Step S430, inputting the energy consumption convolution characteristic and the energy consumption deep separation convolution characteristic into a trained characteristic fusion network to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristics are obtained by the trained characteristic fusion network through the correlation relationship between the maximum energy consumption convolution characteristics and the energy consumption deep separation convolution characteristics.
And step S440, inputting the fusion energy consumption characteristics into the trained characteristic classification network to obtain a predicted energy consumption value corresponding to the predicted moment.
In the training method of the energy consumption prediction network, the characteristic extraction processing in different scales and different convolution modes is realized by adopting the double-branch characteristic extraction network, so that the feature extraction processing can be sufficiently performed on the input data without adopting a complex model, meanwhile, by fusing the energy consumption convolution characteristic and the energy consumption deep separation convolution characteristic, the correlation relationship between the energy consumption convolution characteristic and the energy consumption deep separation convolution characteristic is maximized, can effectively reduce the feature redundancy of the fusion energy consumption features, is convenient for quickly and effectively classifying the fusion energy consumption features in the follow-up process, enables an energy consumption prediction network to output accurate and high prediction results, and the energy consumption prediction network is light, the parameter quantity is small, the calculated quantity is small, the deployment on mobile and edge equipment is convenient, and the purposes of monitoring the energy consumption in the industrial production process and improving the energy utilization rate are realized.
In another embodiment, as shown in fig. 5, there is provided a method for training an energy consumption prediction network, which is described by taking the method as an example for being applied to the computer device in fig. 1, and includes the following steps:
step S502, acquiring original energy consumption data; the raw energy consumption data includes a plurality of energy consumption variables.
Step S504, based on the mutual information feature extraction method, determining the mutual information quantity corresponding to each energy consumption variable.
And S506, sequencing the energy consumption variables based on the mutual information quantity to obtain the sequenced energy consumption variables.
Step S508, determining target energy consumption variables in the sequenced energy consumption variables; and the mutual information quantity of the target energy consumption variables is greater than the energy consumption variables except the target energy consumption variables in the sequenced energy consumption variables.
Step S510, feature data corresponding to the target energy consumption variable are obtained; and the characteristic redundancy degree of the redundancy removing energy consumption data is less than a preset threshold value.
Step S512, carrying out normalization processing on the redundancy-removing energy consumption data to obtain training sample data; the training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is a time series of energy consumption data before the tag moment.
Step S514, inputting the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and inputting the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic.
Step S516, inputting the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained.
And S518, inputting the fusion energy consumption characteristics to the characteristic classification network to be trained to obtain a predicted energy consumption value.
Step S520, obtaining a network loss value of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value.
Step S522, if the network loss value is not within the preset threshold range, determining a gradient of each network node in the energy consumption prediction network based on the network loss value by using a back propagation method.
Step 524, updating the parameters of each network node according to the gradient by adopting a random gradient descending method to obtain an energy consumption prediction network after the network parameters are adjusted, and taking the energy consumption prediction network after the network parameters are adjusted as a trained energy consumption prediction network when the network loss value is lower than a preset threshold value; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
It should be noted that, the specific limitations of the above steps can be referred to the above specific limitations of a training method for an energy consumption prediction network.
To facilitate understanding by those skilled in the art, fig. 6 provides a training flow diagram of a training method of an energy consumption prediction network; the computer equipment extracts data and divides original data into a training set and a testing set; and then, the computer equipment performs data preprocessing on the processed data and performs related performance index selection. Then, the computer equipment carries out normalization processing on the data to obtain training sample data and test sample data. The computer equipment trains the energy consumption prediction network to be trained by adopting training sample data, namely, parameter optimization, so as to obtain the trained energy consumption prediction network, then, the trained energy consumption prediction network is tested by adopting test sample data, and the model reliability of the trained energy consumption prediction network is determined.
It should be understood that, although the steps in the flowcharts of fig. 2, 4, 5 and 6 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2, 4, 5, and 6 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or in alternation with other steps or at least some of the other steps.
In one embodiment, as shown in fig. 7, there is provided a training apparatus for an energy consumption prediction network, the energy consumption prediction network including a first feature extraction network to be trained, a second feature extraction network to be trained, a feature fusion network to be trained, and a feature classification network to be trained, including:
a first obtaining module 710, configured to obtain training sample data; the training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment;
a first extraction module 720, configured to input the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and input the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic;
the first fusion module 730 is configured to input the first energy consumption feature and the second energy consumption feature to the feature fusion network to be trained to obtain a fusion energy consumption feature; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained;
the first classification module 740 is configured to input the fusion energy consumption feature to the feature classification network to be trained, so as to obtain a predicted energy consumption value;
a first adjusting module 750, configured to adjust a network parameter of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
In one embodiment, the first adjusting module 750 is specifically configured to obtain a network loss value of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value; and if the network loss value is not within the preset threshold value range, adjusting the network parameters of the energy consumption prediction network according to the network loss value until the network loss value is lower than the preset threshold value, and taking the energy consumption prediction network after network parameter adjustment as the trained energy consumption prediction network.
In one embodiment, the first adjusting module 750 is specifically configured to determine a gradient of each network node in the energy consumption prediction network based on the network loss value by using a back propagation method; and updating the parameters of each network node according to the gradient by adopting a random gradient descending method to obtain the energy consumption prediction network after the network parameters are adjusted.
In one embodiment, the feature fusion network to be trained includes a feature adjustment layer, a feature fusion layer, and a full connection layer, and the first fusion module 730 is specifically configured to adjust the first energy consumption feature and the second energy consumption feature through the feature adjustment layer, so as to maximize a correlation between the adjusted first energy consumption feature and the adjusted second energy consumption feature; fusing the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic through the characteristic fusion layer to obtain an initial fusion characteristic; and carrying out full connection processing on the initial fusion characteristics through the full connection layer to obtain the fusion energy consumption characteristics.
In one embodiment, the first obtaining module 710 is specifically configured to obtain original energy consumption data; performing redundancy removal processing on the original energy consumption data to obtain redundancy removal energy consumption data; the characteristic redundancy degree of the redundancy-removing energy consumption data is smaller than a preset threshold value; and carrying out normalization processing on the redundancy-removing energy consumption data to obtain the training sample data.
In one embodiment, the original energy consumption data includes a plurality of energy consumption variables, and the first obtaining module 710 is specifically configured to determine, based on a mutual information feature extraction method, mutual information amounts corresponding to the energy consumption variables; sequencing the energy consumption variables based on the mutual information quantity to obtain the sequenced energy consumption variables; determining a target energy consumption variable from the sequenced energy consumption variables; the mutual information quantity of the target energy consumption variables is greater than the energy consumption variables except the target energy consumption variables in the sequenced energy consumption variables; and taking the characteristic data corresponding to the target energy consumption variable as the redundancy-removing energy consumption data.
In one embodiment, as shown in fig. 8, there is provided an energy consumption prediction apparatus including:
a second obtaining module 810, configured to obtain the trained energy consumption prediction network; the trained energy consumption prediction network is obtained by training according to the method; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network;
a second extraction module 820, configured to input the energy consumption data to be processed into the trained first feature extraction network to obtain an energy consumption convolution feature, and input the historical energy consumption data into the trained second feature extraction network to obtain an energy consumption deep separation convolution feature; the energy consumption data to be processed is an energy consumption data time sequence before the prediction moment;
the second fusion module 830 is configured to input the energy consumption convolution feature and the energy consumption deep separation convolution feature to the trained feature fusion network to obtain a fusion energy consumption feature; the fusion energy consumption feature is obtained by maximizing the correlation between the energy consumption convolution feature and the energy consumption deep separation convolution feature by the trained feature fusion network;
and the second classification module 840 is configured to input the fusion energy consumption feature to the trained feature classification network, so as to obtain a predicted energy consumption value corresponding to the prediction time.
For the specific limitations of the training of the energy consumption prediction network and the energy consumption prediction device, reference may be made to the limitations of the energy consumption prediction network training and the energy consumption prediction method, which are not described herein again. The modules in the energy consumption prediction network training and energy consumption prediction device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 9. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing training and energy consumption prediction data of the energy consumption prediction network. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of energy consumption prediction network training and energy consumption prediction.
Those skilled in the art will appreciate that the architecture shown in fig. 9 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In an embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method of training, energy consumption prediction for an energy consumption prediction network as described above. Here, the steps of a method for training and predicting energy consumption of an energy consumption prediction network may be steps in a method for training and predicting energy consumption of an energy consumption prediction network according to the above embodiments.
In one embodiment, a computer readable storage medium is provided, storing a computer program that, when executed by a processor, causes the processor to perform the steps of the energy consumption prediction network training and energy consumption prediction method described above. Here, the steps of a method for training and predicting energy consumption of an energy consumption prediction network may be steps in a method for training and predicting energy consumption of an energy consumption prediction network according to the above embodiments.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical storage, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (10)

1. A training method of an energy consumption prediction network is characterized in that the energy consumption prediction network comprises a first feature extraction network to be trained, a second feature extraction network to be trained, a feature fusion network to be trained and a feature classification network to be trained, and the method comprises the following steps:
acquiring training sample data; the training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment;
inputting the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and inputting the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic;
inputting the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained;
inputting the fusion energy consumption characteristics to the characteristic classification network to be trained to obtain a predicted energy consumption value;
adjusting network parameters of the energy consumption prediction network based on the difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
2. The method of claim 1, wherein adjusting network parameters of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained comprises:
obtaining a network loss value of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value;
and if the network loss value is not within the preset threshold value range, adjusting the network parameters of the energy consumption prediction network according to the network loss value until the network loss value is lower than the preset threshold value, and taking the energy consumption prediction network after network parameter adjustment as the trained energy consumption prediction network.
3. The method of claim 2, wherein said adjusting network parameters of said energy consumption prediction network based on said network loss values comprises:
determining the gradient of each network node in the energy consumption prediction network based on the network loss value by adopting a back propagation method;
and updating the parameters of each network node according to the gradient by adopting a random gradient descending method to obtain the energy consumption prediction network after the network parameters are adjusted.
4. The method according to claim 1, wherein the feature fusion network to be trained comprises a feature adjustment layer, a feature fusion layer and a full connection layer, and the inputting the first energy consumption feature and the second energy consumption feature into the feature fusion network to be trained to obtain a fusion energy consumption feature comprises:
adjusting the first energy consumption characteristic and the second energy consumption characteristic through the characteristic adjusting layer so as to maximize the correlation between the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic;
fusing the adjusted first energy consumption characteristic and the adjusted second energy consumption characteristic through the characteristic fusion layer to obtain an initial fusion characteristic;
and carrying out full connection processing on the initial fusion characteristics through the full connection layer to obtain the fusion energy consumption characteristics.
5. The method of claim 1, wherein the obtaining training sample data comprises:
acquiring original energy consumption data;
performing redundancy removal processing on the original energy consumption data to obtain redundancy removal energy consumption data; the characteristic redundancy degree of the redundancy-removing energy consumption data is smaller than a preset threshold value;
and carrying out normalization processing on the redundancy-removing energy consumption data to obtain the training sample data.
6. The method of claim 1, wherein the raw energy consumption data comprises a plurality of energy consumption variables, and wherein the performing the de-redundancy processing on the raw energy consumption data to obtain the de-redundancy energy consumption data comprises:
determining mutual information quantity corresponding to each energy consumption variable based on a mutual information feature extraction method;
sequencing the energy consumption variables based on the mutual information quantity to obtain the sequenced energy consumption variables;
determining a target energy consumption variable from the sequenced energy consumption variables; the mutual information quantity of the target energy consumption variables is greater than the energy consumption variables except the target energy consumption variables in the sequenced energy consumption variables;
and taking the characteristic data corresponding to the target energy consumption variable as the redundancy-removing energy consumption data.
7. A method of energy consumption prediction, the method comprising:
acquiring a trained energy consumption prediction network; the trained energy consumption prediction network is trained according to the method of any one of claims 1 to 6; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network;
inputting energy consumption data to be processed into the trained first feature extraction network to obtain energy consumption convolution features, and inputting historical energy consumption data into the trained second feature extraction network to obtain energy consumption deep separation convolution features; the energy consumption data to be processed is an energy consumption data time sequence before the prediction moment;
inputting the energy consumption convolution characteristics and the energy consumption deep separation convolution characteristics into the trained characteristic fusion network to obtain fusion energy consumption characteristics; the fusion energy consumption feature is obtained by maximizing the correlation between the energy consumption convolution feature and the energy consumption deep separation convolution feature by the trained feature fusion network;
and inputting the fusion energy consumption characteristics to the trained characteristic classification network to obtain a predicted energy consumption value corresponding to the prediction moment.
8. An energy consumption prediction network training device, wherein the energy consumption prediction network comprises a first feature extraction network to be trained, a second feature extraction network to be trained, a feature fusion network to be trained and a feature classification network to be trained, the device comprises:
the first acquisition module is used for acquiring training sample data; the training sample data comprises historical energy consumption data and actual energy consumption values at the label moment; the historical energy consumption data is an energy consumption data time sequence before the tag moment;
the first extraction module is used for inputting the historical energy consumption data to the first feature extraction network to be trained to obtain a first energy consumption feature, and inputting the historical energy consumption data to the second feature extraction network to be trained to obtain a second energy consumption feature; wherein the first energy consumption characteristic is a convolution characteristic; the second energy consumption characteristic is a depth separation convolution characteristic;
the first fusion module is used for inputting the first energy consumption characteristic and the second energy consumption characteristic into the characteristic fusion network to be trained to obtain a fusion energy consumption characteristic; the fusion energy consumption characteristic is obtained by maximizing the correlation between the first energy consumption characteristic and the second energy consumption characteristic of the characteristic fusion network to be trained;
the first classification module is used for inputting the fusion energy consumption characteristics to the characteristic classification network to be trained to obtain a predicted energy consumption value;
a first adjusting module, configured to adjust a network parameter of the energy consumption prediction network based on a difference between the predicted energy consumption value and the actual energy consumption value until a trained energy consumption prediction network is obtained; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network.
9. An apparatus for energy consumption prediction, the apparatus comprising:
the second acquisition module is used for acquiring the trained energy consumption prediction network; the trained energy consumption prediction network is trained according to the method of any one of claims 1 to 6; the trained energy consumption prediction network comprises a trained first feature extraction network, a trained second feature extraction network, a trained feature fusion network and a trained feature classification network;
the second extraction module is used for inputting the energy consumption data to be processed into the trained first feature extraction network to obtain energy consumption convolution features, and inputting the historical energy consumption data into the trained second feature extraction network to obtain energy consumption deep separation convolution features; the energy consumption data to be processed is an energy consumption data time sequence before the prediction moment;
the second fusion module is used for inputting the energy consumption convolution characteristics and the energy consumption deep separation convolution characteristics into the trained characteristic fusion network to obtain fusion energy consumption characteristics; the fusion energy consumption feature is obtained by maximizing the correlation between the energy consumption convolution feature and the energy consumption deep separation convolution feature by the trained feature fusion network;
and the second classification module is used for inputting the fusion energy consumption characteristics to the trained characteristic classification network to obtain a predicted energy consumption value corresponding to the prediction moment.
10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 7.
CN202110689983.4A 2021-06-22 2021-06-22 Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment Active CN113408802B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110689983.4A CN113408802B (en) 2021-06-22 2021-06-22 Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110689983.4A CN113408802B (en) 2021-06-22 2021-06-22 Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment

Publications (2)

Publication Number Publication Date
CN113408802A true CN113408802A (en) 2021-09-17
CN113408802B CN113408802B (en) 2022-11-25

Family

ID=77682216

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110689983.4A Active CN113408802B (en) 2021-06-22 2021-06-22 Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment

Country Status (1)

Country Link
CN (1) CN113408802B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115081763A (en) * 2022-08-24 2022-09-20 山东鲁晟精工机械有限公司 Energy consumption prediction system for turning process of numerical control lathe
CN116011593A (en) * 2023-03-09 2023-04-25 支付宝(杭州)信息技术有限公司 Method and device for determining energy consumption of network model

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298374A (en) * 2019-05-09 2019-10-01 清华大学 A kind of driving locus energy consumption analysis method and apparatus based on deep learning
WO2019233421A1 (en) * 2018-06-04 2019-12-12 京东数字科技控股有限公司 Image processing method and device, electronic apparatus, and storage medium
US20200065444A1 (en) * 2012-11-06 2020-02-27 Cenergistic Llc Adjustment simulation method for energy consumption
CN110866592A (en) * 2019-10-28 2020-03-06 腾讯科技（深圳）有限公司 Model training method and device, energy efficiency prediction method and device and storage medium
CN110866528A (en) * 2019-10-28 2020-03-06 腾讯科技（深圳）有限公司 Model training method, energy consumption use efficiency prediction method, device and medium
CN110942228A (en) * 2019-10-25 2020-03-31 万翼科技有限公司 Energy consumption control method and device, computer equipment and storage medium
CN111539563A (en) * 2020-04-13 2020-08-14 珠海格力电器股份有限公司 Energy consumption safety state prediction method, device, server and storage medium
CN112434787A (en) * 2020-10-28 2021-03-02 西安交通大学 Terminal space energy consumption prediction method based on building total energy consumption, medium and equipment
CN112990591A (en) * 2021-03-26 2021-06-18 江西省能源大数据有限公司 Multi-dimensional energy consumption data analysis method based on convolutional neural network and enterprise energy consumption prediction model

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200065444A1 (en) * 2012-11-06 2020-02-27 Cenergistic Llc Adjustment simulation method for energy consumption
WO2019233421A1 (en) * 2018-06-04 2019-12-12 京东数字科技控股有限公司 Image processing method and device, electronic apparatus, and storage medium
CN110298374A (en) * 2019-05-09 2019-10-01 清华大学 A kind of driving locus energy consumption analysis method and apparatus based on deep learning
CN110942228A (en) * 2019-10-25 2020-03-31 万翼科技有限公司 Energy consumption control method and device, computer equipment and storage medium
CN110866592A (en) * 2019-10-28 2020-03-06 腾讯科技（深圳）有限公司 Model training method and device, energy efficiency prediction method and device and storage medium
CN110866528A (en) * 2019-10-28 2020-03-06 腾讯科技（深圳）有限公司 Model training method, energy consumption use efficiency prediction method, device and medium
CN111539563A (en) * 2020-04-13 2020-08-14 珠海格力电器股份有限公司 Energy consumption safety state prediction method, device, server and storage medium
CN112434787A (en) * 2020-10-28 2021-03-02 西安交通大学 Terminal space energy consumption prediction method based on building total energy consumption, medium and equipment
CN112990591A (en) * 2021-03-26 2021-06-18 江西省能源大数据有限公司 Multi-dimensional energy consumption data analysis method based on convolutional neural network and enterprise energy consumption prediction model

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
邹锋: "基于生成对抗网络的强化学习方法及其在建筑节能方面的应用", 《中国优秀硕士论文电子期刊网》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115081763A (en) * 2022-08-24 2022-09-20 山东鲁晟精工机械有限公司 Energy consumption prediction system for turning process of numerical control lathe
CN115081763B (en) * 2022-08-24 2022-11-11 山东鲁晟精工机械有限公司 Energy consumption prediction system for turning process of numerical control lathe
CN116011593A (en) * 2023-03-09 2023-04-25 支付宝(杭州)信息技术有限公司 Method and device for determining energy consumption of network model

Also Published As

Publication number Publication date
CN113408802B (en) 2022-11-25

Similar Documents

Publication Publication Date Title
Liberis et al. μnas: Constrained neural architecture search for microcontrollers
CN111931931B (en) Deep neural network training method and device for pathology full-field image
CN113408802B (en) Energy consumption prediction network training method and device, energy consumption prediction method and device, and computer equipment
CN108108762B (en) Nuclear extreme learning machine for coronary heart disease data and random forest classification method
KR20210032140A (en) Method and apparatus for performing pruning of neural network
CN110571792A (en) Analysis and evaluation method and system for operation state of power grid regulation and control system
CN112001329B (en) Method and device for predicting protein expression amount, computer device and storage medium
CN112037862B (en) Cell screening method and device based on convolutional neural network
CN109472801A (en) It is a kind of for multiple dimensioned neuromorphic detection and dividing method
JP2023502863A (en) Image incremental clustering method and apparatus, electronic device, storage medium and program product
CN115906399A (en) Improved method for predicting key process quality of product under small sample data
CN113066528B (en) Protein classification method based on active semi-supervised graph neural network
CN114881343A (en) Short-term load prediction method and device of power system based on feature selection
CN111949530B (en) Test result prediction method and device, computer equipment and storage medium
CN109886303A (en) A kind of TrAdaboost sample migration aviation image classification method based on particle group optimizing
CN112017730A (en) Cell screening method and device based on expression quantity prediction model
Liu et al. Focusformer: Focusing on what we need via architecture sampler
CN111832645A (en) Classification data feature selection method based on discrete crow difference collaborative search algorithm
CN106874927A (en) The construction method and system of a kind of random strong classifier
CN116524296A (en) Training method and device of equipment defect detection model and equipment defect detection method
CN116822702A (en) Carbon emission prediction method, apparatus, computer device, and storage medium
CN116595363A (en) Prediction method, apparatus, device, storage medium, and computer program product
Wang et al. Identification of weather phenomena based on lightweight convolutional neural networks
CN115758462A (en) Method, device, processor and computer readable storage medium for realizing sensitive data identification in trusted environment
WO2022162839A1 (en) Learning device, learning method, and recording medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20230410

Address after: Full Floor 14, Unit 3, Building 2, No. 11, Middle Spectra Road, Huangpu District, Guangzhou, Guangdong 510700

Patentee after: China Southern Power Grid Digital Grid Technology (Guangdong) Co.,Ltd.

Address before: Room 86, room 406, No.1, Yichuang street, Zhongxin Guangzhou Knowledge City, Huangpu District, Guangzhou City, Guangdong Province

Patentee before: Southern Power Grid Digital Grid Research Institute Co.,Ltd.

TR01 Transfer of patent right