CN111310905A - Neural network model training method and device and heating and ventilation system energy efficiency optimization method - Google Patents
Neural network model training method and device and heating and ventilation system energy efficiency optimization method Download PDFInfo
- Publication number
- CN111310905A CN111310905A CN202010389830.3A CN202010389830A CN111310905A CN 111310905 A CN111310905 A CN 111310905A CN 202010389830 A CN202010389830 A CN 202010389830A CN 111310905 A CN111310905 A CN 111310905A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network model
- sample
- initial
- sample pair
- 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
Links
Images
Classifications
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- 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
- 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
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The application provides a neural network model training method, a device and a heating and ventilation system energy efficiency optimization method, wherein the neural network model training method comprises the following steps: obtaining an initial sample, wherein initial parameters of the initial sample conform to monotonicity; performing data enhancement on the initial samples, and constructing to obtain a scale sample pair and a sequencing sample pair; inputting the initial sample, the scale sample pair and the sequencing sample pair to a pre-trained neural network model to obtain a corresponding prediction result, wherein the pre-trained neural network model is preset with a monotonous target and a scale target of parameters; calculating corresponding prediction error loss according to the prediction results of the initial sample, the scale sample pair and the sequencing sample pair; and updating parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the scale sample pair and the sequencing sample pair. The neural network model training method and device increase learning of the model about monotonicity and sensitive scale of an input and output curve, accelerate model convergence speed and reduce model training difficulty.
Description
Technical Field
The application relates to the technical field of machine learning, in particular to a neural network model training method and device and a heating and ventilation system energy efficiency optimization method.
Background
With the development of technologies such as cloud service, big data, AI computing and the like, enterprises and governments have built a large number of data centers. At present, the domestic data center has generally higher energy consumption, the average PUE value is basically between 2.2 and 3.0, and the electricity consumption of the domestic data center accounts for 3 percent of the electricity consumption of the whole society.
At the present stage, a lot of researches are conducted on energy efficiency optimization of a data center, for example, energy efficiency optimization of a heating and ventilation system of the data center, and energy consumption simulation software is often used for simulating and comparing energy efficiency performances of different schemes so as to assist in design decision and optimization. However, most of the neural network models applied to the energy efficiency optimization of the data center require a large number of training samples and rich sample characteristic distributions, and when the training samples do not meet the requirements, the curve distribution and the sensitivity of the neural network model with respect to input and output will be poor, so that the neural network model is difficult to train and cannot be applied online.
Disclosure of Invention
The embodiment of the application aims to provide a neural network model training method and device, so that learning of a pre-trained neural network model about monotonicity and sensitive scale of an input and output curve is increased, model convergence speed is increased, and model training difficulty is reduced.
In a first aspect, an embodiment of the present application provides a neural network model training method, including:
obtaining an initial sample, wherein initial parameters of the initial sample conform to monotonicity;
performing data enhancement on the initial samples, and constructing to obtain a scale sample pair and a sequencing sample pair;
inputting the initial sample, the scale sample pair and the sequencing sample pair to a pre-trained neural network model to obtain a corresponding prediction result, wherein the pre-trained neural network model is preset with a parameter monotone target and a scale target;
calculating corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair;
updating parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair;
and repeatedly executing the steps until the pre-trained neural network model converges.
In the implementation process, the neural network model training method according to the embodiment of the application builds a scale sample pair and a sequencing sample pair by performing data enhancement on the obtained initial sample, enriches the training samples, presets a monotonous target and a scale target of parameters of the pre-trained neural network model, and constrains an output curve of the pre-trained neural network model to be a monotonous curve, so that the pre-trained neural network model can better learn the physical characteristics of the parameters and correct the curve, and increases the learning of the pre-trained neural network model on monotonicity and sensitive scale of the input and output curves, thereby accelerating the model convergence speed, reducing the model training difficulty and reducing the model training time.
Further, the data enhancement of the initial sample, and the construction of the scale sample pair and the sequencing sample pair include:
randomly selecting to obtain a maximum parameter and a minimum parameter within the range of the value range of the initial parameter;
constructing and obtaining a scale sample pair according to the maximum parameter and the minimum parameter;
randomly selecting to obtain random parameters in the range of the value range of the initial parameters;
and constructing and obtaining a sequencing sample pair according to the random parameter and the initial parameter.
In the implementation process, the method performs data enhancement on the initial sample, and the modes of constructing the scale sample pair and sequencing the sample pair are performed in the range of the value range based on the initial parameter, so that the scale sample pair and sequencing the sample pair can be constructed more quickly, effectively and reasonably, and the training effect of the neural network model can be better guaranteed.
Further, the calculating a corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scaled sample pair, and the prediction result of the ordered sample pair includes:
calculating the mean square error loss of the initial sample according to the prediction result of the initial sample;
according to the prediction result of the scale sample pair, calculating the prediction distance between the scale samples of the scale sample pair;
calculating the distance mean square error loss of the scale sample pair according to the predicted distance between the scale samples of the scale sample pair;
calculating a prediction order between the ordered samples of the ordered sample pair according to the prediction result of the ordered sample pair;
calculating a loss of ordering of the pair of ordered samples according to a prediction order between ordered samples of the pair of ordered samples.
In the implementation process, the method calculates the corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair, so that the mean square error loss of the initial sample, the distance mean square error loss of the scale sample pair and the sequencing loss of the sequencing sample pair can be better calculated.
Further, the updating parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scaled sample pair, and the prediction error loss of the ordered sample pair includes:
fusing the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair in a weighted average and weight attenuation mode to obtain a target error loss;
and updating the parameters of the pre-trained neural network model according to the target error loss.
In the implementation process, the method fuses the prediction error losses of the initial sample, the scale sample pair and the sequencing sample pair in a weighted average and weight attenuation mode, so that the obtained target error loss can be more convenient for updating the parameters of the pre-trained neural network model.
Further, the updating parameters of the pre-trained neural network model according to the target error loss includes:
and updating parameters of the pre-trained neural network model through a gradient descent algorithm and a back propagation algorithm according to the target error loss.
In the implementation process, the method can better update the parameters of the pre-trained neural network model through a gradient descent algorithm and a back propagation algorithm.
Further, after the pre-trained neural network model converges, the method further comprises:
and cutting the obtained neural network model, and reserving a sample input channel, a neural network and a sample output channel.
In the implementation process, the method cuts the obtained neural network model, basically keeps the original performance of the obtained neural network model, reduces the number of related parameters, saves the storage space of the neural network model and reduces the calculation time of the neural network model.
In a second aspect, an embodiment of the present application provides a neural network model training apparatus, including:
the device comprises an acquisition module, a comparison module and a comparison module, wherein the acquisition module is used for acquiring an initial sample, and initial parameters of the initial sample conform to monotonicity;
the sample construction module is used for carrying out data enhancement on the initial sample, and constructing to obtain a scale sample pair and a sequencing sample pair;
the input module is used for inputting the initial sample, the scale sample pair and the sequencing sample pair to a pre-trained neural network model to obtain a corresponding prediction result, and the pre-trained neural network model is preset with a monotonous target and a scale target with parameters;
the calculation module is used for calculating corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair;
and the updating module is used for updating the parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair.
In the implementation process, the neural network model training device according to the embodiment of the application builds a scale sample pair and a sequencing sample pair by performing data enhancement on the obtained initial sample, enriches the training samples, presets a monotonous target and a scale target of parameters of the pre-trained neural network model, and constrains an output curve of the pre-trained neural network model to be a monotonous curve, so that the pre-trained neural network model can better learn the physical characteristics of the parameters and correct the curve, and increases the learning of the pre-trained neural network model on monotonicity and sensitive scale of the input and output curves, thereby accelerating the convergence speed of the model, reducing the difficulty of model training and reducing the training time of the model.
In a third aspect, an embodiment of the present application provides a heating and ventilation system energy efficiency optimization method, where the method applies a neural network model obtained by the neural network model training method.
In a fourth aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to make the electronic device execute the above neural network model training method.
In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program used in the electronic device described above.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
Fig. 1 is a schematic flowchart of a neural network model training method according to an embodiment of the present disclosure;
fig. 2 is a schematic flowchart of step S120 according to a first embodiment of the present application;
fig. 3 is a schematic flowchart of step S140 according to a first embodiment of the present application;
fig. 4 is a schematic flowchart of step S150 according to an embodiment of the present application;
fig. 5 is a block diagram of a neural network model training apparatus according to a second embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures. Meanwhile, in the description of the present application, the terms "first", "second", and the like are used only for distinguishing the description, and are not to be construed as indicating or implying relative importance.
At the present stage, a lot of researches are conducted on energy efficiency optimization of a data center, for example, energy efficiency optimization of a heating and ventilation system of the data center, and energy consumption simulation software is often used for simulating and comparing energy efficiency performances of different schemes so as to assist in design decision and optimization. However, most of the neural network models applied to the energy efficiency optimization of the data center require a large number of training samples and rich sample characteristic distributions, and when the training samples do not meet the requirements, the curve distribution and the sensitivity of the neural network model with respect to input and output will be poor, so that the neural network model is difficult to train and cannot be applied online.
In view of the above problems in the prior art, the present application provides a neural network model training method and apparatus, which increase the learning of the pre-trained neural network model with respect to monotonicity and sensitivity of the input/output curve, accelerate the convergence speed of the model, and reduce the difficulty of model training.
Example one
Referring to fig. 1, fig. 1 is a schematic flow chart of a neural network model training method provided in the embodiment of the present application. An execution subject for executing the neural network model training method described below in the embodiment of the present application may be a server.
The neural network model training method comprises the following steps:
step S110, an initial sample is obtained, and initial parameters of the initial sample conform to monotonicity.
In this embodiment, the initial samples, i.e., the initial parameters. The initial parameters conform to monotonicity, that is, the initial parameters and the prediction parameters of the pre-trained neural network model are in a monotone increasing or monotone decreasing relation.
It should be understood that so-called monotony, rather than strictly limiting complete everywhere between parameters, there may be no everywhere monotony between parameters.
In this embodiment, an example is given in which the neural network model obtained by training the neural network model training method in the embodiment of the present application is applied to energy efficiency optimization of a heating and ventilation system.
The initial parameters comprise the fan frequency of a cooling tower of the heating and ventilation system, the frequency of a cooling pump and the external temperature and humidity. Correspondingly, the prediction parameter of the pre-trained neural network model is the compressor power. The cooling tower fan frequency and the compressor power are in a monotonically decreasing relation, the cooling pump frequency and the compressor power are in a monotonically decreasing relation, and the external temperature and humidity and the compressor power are in a monotonically increasing relation.
It can be understood that the neural network model obtained by training the neural network model training method in the embodiment of the present application can be applied to energy efficiency optimization of other systems, as long as the parameters of other systems conform to monotonicity. If the neural network model obtained by training the neural network model training method in the embodiment of the application is applied to energy efficiency optimization of other systems, the initial parameters and the prediction parameters of the pre-trained neural network model change according to the parameters of other systems.
Alternatively, the initial sample may be input by the user, or may be obtained through pre-stored sample data.
And step S120, performing data enhancement on the initial samples, and constructing to obtain a scale sample pair and a sequencing sample pair.
In this embodiment, the initial sample is subjected to data enhancement, and a data enhancement sample, that is, a data enhancement parameter, can be obtained.
The scale sample pair can be constructed according to the data enhancement sample, and the sequencing sample pair can be constructed according to the data enhancement sample and the initial sample.
Step S130, inputting the initial sample, the scale sample pair and the sequencing sample pair to a pre-trained neural network model to obtain a corresponding prediction result, wherein the pre-trained neural network model is preset with a monotonous target and a scale target with parameters.
In this embodiment, the initial sample, the pair of scale samples, and the pair of ordered samples are input to the pre-trained neural network model, and the prediction result of the initial sample, the prediction result of the pair of scale samples, and the prediction result of the pair of ordered samples are obtained.
The prediction result, i.e. the prediction parameters of the pre-trained neural network model, is the predicted compressor power.
The pre-trained neural network model presets a monotonous target and a scale target of parameters, and the dependence of the model on the number of samples and the characteristic distribution is reduced.
Step S140, calculating the corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair.
In this embodiment, the prediction error loss of the initial sample, the prediction error loss of the scale sample pair, and the prediction error loss of the ordered sample pair are obtained according to the prediction result of the initial sample, the prediction result of the scale sample pair, and the prediction result of the ordered sample pair.
The prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair can be calculated in a preset prediction error loss calculation mode.
And step S150, updating parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair.
In this embodiment, the parameters of the pre-trained neural network model may be updated by a preset parameter update algorithm.
In this embodiment, the neural network model training method according to the embodiment of the present application repeatedly executes the above steps until the pre-trained neural network model converges.
According to the neural network model training method, the obtained initial samples are subjected to data enhancement, a scale sample pair and a sequencing sample pair are obtained through construction, training samples are enriched, a monotonous target and a scale target of parameters of a pre-trained neural network model are preset, an output curve of the pre-trained neural network model is restrained to be a monotonous curve, the pre-trained neural network model can better learn the physical characteristics of the parameters and correct the curve, learning of monotonicity and sensitive scale of the input and output curve of the pre-trained neural network model is increased, then the model convergence speed is accelerated, the model training difficulty is reduced, and the model training time is shortened.
In order to construct and obtain a scale sample pair and a sequencing sample pair more quickly, efficiently and reasonably, an embodiment of the present application provides a possible implementation manner, see fig. 2, where fig. 2 is a schematic flowchart of a step S120 provided in the embodiment of the present application, and a neural network model training method in the embodiment of the present application, in step S120, performs data enhancement on an initial sample, and constructs and obtains a scale sample pair and a sequencing sample pair, which may include the following steps:
step S121, randomly selecting and obtaining a maximum parameter and a minimum parameter within the range of the value range of the initial parameter;
step S122, constructing and obtaining a scale sample pair according to the maximum parameter and the minimum parameter;
s123, randomly selecting to obtain random parameters in the range of the value range of the initial parameters;
and step S124, constructing and obtaining a sequencing sample pair according to the random parameter and the initial parameter.
The maximum parameter, the minimum parameter and the random parameter are data enhancement parameters. The maximum parameter and the minimum parameter are not necessarily the maximum parameter and the minimum parameter within the range of the initial parameter, and the maximum parameter and the minimum parameter only need to satisfy that the maximum parameter is greater than the minimum parameter.
The pair of scaled samples may be < max parameter, min parameter >, and the pair of ordered samples may be < random parameter, initial parameter >.
It should be noted that the execution sequence of steps S121 to S122 and steps S123 to S124 may be that steps S121 to S122 are executed first, or steps S123 to S124 are executed first, and this is not limited herein.
In the process, the method performs data enhancement on the initial sample, and the modes of constructing the scale sample pair and sequencing the sample pair are performed in the range of the value range based on the initial parameter, so that the scale sample pair and the sequencing sample pair can be constructed more quickly, effectively and reasonably, and the training effect of the neural network model can be better guaranteed.
In order to better calculate the prediction error loss of the initial sample, the scale sample pair, and the ordered sample pair, a possible implementation manner is provided in the embodiment of the present application, referring to fig. 3, fig. 3 is a schematic flowchart of step S140 provided in the embodiment of the present application, a neural network model training method in the embodiment of the present application, in step S140, calculates the corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair, and the prediction result of the ordered sample pair, and may include the following steps:
step S141, calculating the mean square error loss of the initial sample according to the prediction result of the initial sample;
step S142, calculating the prediction distance between the scale samples of the scale sample pair according to the prediction result of the scale sample pair;
step S143, calculating the distance mean square error loss of the scale sample pair according to the predicted distance between the scale samples of the scale sample pair;
step S144, calculating the prediction order between the sequencing samples of the sequencing sample pair according to the prediction result of the sequencing sample pair;
step S145, calculating a loss of ordering of the ordered sample pairs according to the prediction order between the ordered samples of the ordered sample pairs.
The prediction distance between the scale samples of the scale sample pair, that is, the prediction difference between the scale samples of the scale sample pair, is combined with the above contents, and the prediction difference between the scale samples of the scale sample pair, that is, the prediction parameter corresponding to the maximum parameter is reduced by the value of the prediction parameter corresponding to the minimum parameter.
The prediction order between the ordered samples of the ordered sample pair is 1 or 0, and when the prediction parameter corresponding to the random parameter is greater than the prediction parameter corresponding to the initial parameter, the prediction order between the ordered samples of the ordered sample pair is 1 in combination with the above; conversely, the prediction order between the ordered samples of the ordered sample pair is 0.
It should be noted that, the execution sequence of step S141, step S142 to step S143, and step S144 to step S145 may be to execute step S141 first, or to execute step S142 to step S143 first, or to execute step S144 to step S145 first, which is not limited herein.
In the process, the method calculates the corresponding prediction error loss mode according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair, and can better calculate the mean square error loss of the initial sample, the distance mean square error loss of the scale sample pair and the sequencing loss of the sequencing sample pair.
In order to facilitate updating of parameters of a pre-trained neural network model, a possible implementation manner is provided in the embodiment of the present application, referring to fig. 4, where fig. 4 is a schematic flowchart of a step S150 provided in the embodiment of the present application, and a method for training a neural network model in the embodiment of the present application, in which step S150, parameters of a pre-trained neural network model are updated according to a prediction error loss of an initial sample, a prediction error loss of a scale sample pair, and a prediction error loss of a ranking sample pair, may include the following steps:
step S151, fusing the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair in a weighted average and weight attenuation mode to obtain a target error loss;
step S152, according to the target error loss, updating the parameters of the pre-trained neural network model.
In the process, the method fuses the prediction error losses of the initial sample, the scale sample pair and the sequencing sample pair in a weighted average and weight attenuation mode, so that the obtained target error loss can be more convenient for updating the parameters of the pre-trained neural network model.
Optionally, in step S152, the parameters of the pre-trained neural network model are updated according to the target error loss, and the parameters of the pre-trained neural network model may be updated through a gradient descent algorithm and a back propagation algorithm according to the target error loss.
In the process, the method can better update the parameters of the pre-trained neural network model through a gradient descent algorithm and a back propagation algorithm.
As an optional implementation manner, after the pre-trained neural network model converges, the neural network model training method according to the embodiment of the present application may further include the following steps:
and cutting the obtained neural network model, and reserving a sample input channel, a neural network and a sample output channel.
In the implementation process, the method cuts the obtained neural network model, basically keeps the original performance of the obtained neural network model, reduces the number of related parameters, saves the storage space of the neural network model and reduces the calculation time of the neural network model.
Example two
In order to implement a corresponding method of the above embodiments to achieve corresponding functions and technical effects, a neural network model training apparatus is provided below.
Referring to fig. 5, fig. 5 is a block diagram of a neural network model training apparatus according to an embodiment of the present disclosure.
The neural network model training device of the embodiment of the application comprises:
an obtaining module 210, configured to obtain an initial sample, where an initial parameter of the initial sample conforms to monotonicity;
the sample construction module 220 is used for performing data enhancement on the initial sample, and constructing a scale sample pair and a sequencing sample pair;
an input module 230, configured to input an initial sample, a scale sample pair, and a sequence sample pair to a pre-trained neural network model, to obtain a corresponding prediction result, where the pre-trained neural network model has preset a monotonic target and a scale target with parameters;
a calculating module 240, configured to calculate a corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scaled sample pair, and the prediction result of the ordered sample pair;
and the updating module 250 is used for updating the parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair.
The neural network model training device provided by the embodiment of the application builds a scale sample pair and a sequencing sample pair by performing data enhancement on the obtained initial sample, enriches training samples, presets a monotonous target and a scale target of parameters of a pre-trained neural network model, and restricts an output curve of the pre-trained neural network model to be a monotonous curve, so that the pre-trained neural network model can better learn the physical characteristics of the parameters and correct the curve, and increases the learning of the input and output curve monotonicity and sensitive scale of the pre-trained neural network model, thereby accelerating the convergence speed of the model, reducing the difficulty of model training and reducing the training time of the model.
As an alternative embodiment, the sample construction module 220 may be specifically configured to:
randomly selecting to obtain a maximum parameter and a minimum parameter within the range of the value range of the initial parameter;
constructing and obtaining a scale sample pair according to the maximum parameter and the minimum parameter;
randomly selecting to obtain random parameters within the range of the value range of the initial parameters;
and constructing and obtaining a sequencing sample pair according to the random parameter and the initial parameter.
As an optional implementation, the calculating module 240 may specifically be configured to:
calculating the mean square error loss of the initial sample according to the prediction result of the initial sample;
according to the prediction result of the scale sample pair, calculating the prediction distance between the scale samples of the scale sample pair;
calculating the distance mean square error loss of the scale sample pairs according to the predicted distance between the scale samples of the scale sample pairs;
calculating the prediction order between the sequencing samples of the sequencing sample pair according to the prediction result of the sequencing sample pair;
the loss of ordering of the ordered sample pairs is calculated from the prediction order between the ordered samples of the ordered sample pairs.
As an optional implementation manner, the update module 250 may specifically be configured to:
fusing the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair in a weighted average and weight attenuation mode to obtain a target error loss;
parameters of the pre-trained neural network model are updated according to the target error loss.
Optionally, when the updating module 250 updates the parameters of the pre-trained neural network model according to the target error loss, the parameters of the pre-trained neural network model may be updated through a gradient descent algorithm and a back propagation algorithm according to the target error loss.
As an optional implementation manner, the neural network model training apparatus according to the embodiment of the present application may further include:
and the model cutting module is used for cutting the obtained neural network model and reserving the sample input channel, the neural network and the sample output channel.
The neural network model training device can implement the neural network model training method of the first embodiment. The alternatives in the first embodiment are also applicable to the present embodiment, and are not described in detail here.
The rest of the embodiments of the present application may refer to the contents of the first embodiment, and in this embodiment, details are not repeated.
EXAMPLE III
The embodiment of the application provides a heating and ventilation system energy efficiency optimization method, and the method applies the neural network model obtained by the neural network model training method.
In this embodiment, the heating and ventilation system may be a heating and ventilation system in a data center, a work place, a hotel, an office building, a hospital, an airport, a train station, a subway station, or the like.
The neural network model training method in the embodiment of the present application may refer to the contents of the first embodiment, and is not described in detail in this embodiment.
Example four
An embodiment of the present application provides an electronic device, which includes a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to make the electronic device execute the above neural network model training method.
Alternatively, the electronic device may be a server.
In addition, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program used in the electronic device.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Claims (10)
1. A neural network model training method is characterized in that the neural network model is used for energy efficiency optimization of a heating and ventilation system, and the method comprises the following steps:
acquiring an initial sample, wherein initial parameters of the initial sample comprise the fan frequency of a cooling tower of a heating and ventilation system, the frequency of a cooling pump and the external temperature and humidity, and the initial parameters of the initial sample accord with monotonicity;
performing data enhancement on the initial samples, and constructing to obtain a scale sample pair and a sequencing sample pair;
inputting the initial sample, the scale sample pair and the sequencing sample pair to a pre-trained neural network model to obtain a corresponding prediction result, wherein the pre-trained neural network model is preset with a monotone target and a scale target of parameters, and the prediction result is compressor power;
calculating corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair;
updating parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair;
and repeatedly executing the steps until the pre-trained neural network model converges.
2. The neural network model training method of claim 1, wherein the performing data enhancement on the initial samples, constructing a scale sample pair and a rank sample pair comprises:
randomly selecting to obtain a maximum parameter and a minimum parameter within the range of the value range of the initial parameter;
constructing and obtaining a scale sample pair according to the maximum parameter and the minimum parameter;
randomly selecting to obtain random parameters in the range of the value range of the initial parameters;
and constructing and obtaining a sequencing sample pair according to the random parameter and the initial parameter.
3. The neural network model training method of claim 1, wherein calculating the corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scaled sample pair, and the prediction result of the ordered sample pair comprises:
calculating the mean square error loss of the initial sample according to the prediction result of the initial sample;
according to the prediction result of the scale sample pair, calculating the prediction distance between the scale samples of the scale sample pair;
calculating the distance mean square error loss of the scale sample pair according to the predicted distance between the scale samples of the scale sample pair;
calculating a prediction order between the ordered samples of the ordered sample pair according to the prediction result of the ordered sample pair;
calculating a loss of ordering of the pair of ordered samples according to a prediction order between ordered samples of the pair of ordered samples.
4. The method of claim 1, wherein updating parameters of the pre-trained neural network model according to the prediction error loss of the initial samples, the prediction error loss of the pair of scaled samples, and the prediction error loss of the pair of ordered samples comprises:
fusing the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair in a weighted average and weight attenuation mode to obtain a target error loss;
and updating the parameters of the pre-trained neural network model according to the target error loss.
5. The neural network model training method of claim 4, wherein the updating the parameters of the pre-trained neural network model according to the target error loss comprises:
and updating parameters of the pre-trained neural network model through a gradient descent algorithm and a back propagation algorithm according to the target error loss.
6. The neural network model training method of claim 1, wherein after the pre-trained neural network model converges, the method further comprises:
and cutting the obtained neural network model, and reserving a sample input channel, a neural network and a sample output channel.
7. A neural network model training apparatus, wherein the neural network model is used for energy efficiency optimization of a heating and ventilation system, the apparatus comprising:
the system comprises an acquisition module, a comparison module and a comparison module, wherein the acquisition module is used for acquiring an initial sample, the initial parameters of the initial sample comprise the fan frequency of a cooling tower of a heating and ventilation system, the frequency of a cooling pump and the external temperature and humidity, and the initial parameters of the initial sample accord with monotonicity;
the sample construction module is used for carrying out data enhancement on the initial sample, and constructing to obtain a scale sample pair and a sequencing sample pair;
the input module is used for inputting the initial sample, the scale sample pair and the sequencing sample pair to a pre-trained neural network model to obtain a corresponding prediction result, the pre-trained neural network model is preset with a monotone target and a scale target of parameters, and the prediction result is the power of a compressor;
the calculation module is used for calculating corresponding prediction error loss according to the prediction result of the initial sample, the prediction result of the scale sample pair and the prediction result of the sequencing sample pair;
and the updating module is used for updating the parameters of the pre-trained neural network model according to the prediction error loss of the initial sample, the prediction error loss of the scale sample pair and the prediction error loss of the sequencing sample pair.
8. An energy efficiency optimization method for a heating and ventilation system, which is characterized in that the method applies the neural network model obtained by the neural network model training method according to any one of claims 1 to 6.
9. An electronic device comprising a memory for storing a computer program and a processor that executes the computer program to cause the electronic device to perform the neural network model training method of any one of claims 1-6.
10. A computer-readable storage medium, characterized in that it stores a computer program for use in the electronic device of claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010389830.3A CN111310905B (en) | 2020-05-11 | 2020-05-11 | Neural network model training method and device and heating and ventilation system energy efficiency optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010389830.3A CN111310905B (en) | 2020-05-11 | 2020-05-11 | Neural network model training method and device and heating and ventilation system energy efficiency optimization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111310905A true CN111310905A (en) | 2020-06-19 |
| CN111310905B CN111310905B (en) | 2020-08-18 |
Family
ID=71152220
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010389830.3A Active CN111310905B (en) | 2020-05-11 | 2020-05-11 | Neural network model training method and device and heating and ventilation system energy efficiency optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111310905B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033801A (en) * | 2021-03-04 | 2021-06-25 | 北京百度网讯科技有限公司 | Pre-training method and device of neural network model, electronic equipment and medium |
| CN113655285A (en) * | 2021-08-30 | 2021-11-16 | 广东电网有限责任公司 | Method and device for measuring grounding resistance of power transmission line tower and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5748329A (en) * | 1996-03-08 | 1998-05-05 | Industrial Technology Research Institute | Method and apparatus for adaptive color scanning/printing data correction employing neural networks |
| US20160098987A1 (en) * | 2014-10-02 | 2016-04-07 | Microsoft Technology Licensing , LLC | Neural network-based speech processing |
| CN109036389A (en) * | 2018-08-28 | 2018-12-18 | 出门问问信息科技有限公司 | The generation method and device of a kind of pair of resisting sample |
| CN109583594A (en) * | 2018-11-16 | 2019-04-05 | 东软集团股份有限公司 | Deep learning training method, device, equipment and readable storage medium storing program for executing |
| CN111103799A (en) * | 2019-12-25 | 2020-05-05 | 创新奇智(南京)科技有限公司 | Energy efficiency optimization modeling method based on teacher-student network online exploration learning |
-
2020
- 2020-05-11 CN CN202010389830.3A patent/CN111310905B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5748329A (en) * | 1996-03-08 | 1998-05-05 | Industrial Technology Research Institute | Method and apparatus for adaptive color scanning/printing data correction employing neural networks |
| US20160098987A1 (en) * | 2014-10-02 | 2016-04-07 | Microsoft Technology Licensing , LLC | Neural network-based speech processing |
| CN109036389A (en) * | 2018-08-28 | 2018-12-18 | 出门问问信息科技有限公司 | The generation method and device of a kind of pair of resisting sample |
| CN109583594A (en) * | 2018-11-16 | 2019-04-05 | 东软集团股份有限公司 | Deep learning training method, device, equipment and readable storage medium storing program for executing |
| CN111103799A (en) * | 2019-12-25 | 2020-05-05 | 创新奇智(南京)科技有限公司 | Energy efficiency optimization modeling method based on teacher-student network online exploration learning |
Non-Patent Citations (1)
| Title |
|---|
| 石全等: "《系统决策与建模》", 31 July 2016 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033801A (en) * | 2021-03-04 | 2021-06-25 | 北京百度网讯科技有限公司 | Pre-training method and device of neural network model, electronic equipment and medium |
| CN113655285A (en) * | 2021-08-30 | 2021-11-16 | 广东电网有限责任公司 | Method and device for measuring grounding resistance of power transmission line tower and storage medium |
| CN113655285B (en) * | 2021-08-30 | 2023-07-21 | 广东电网有限责任公司 | Transmission line pole tower grounding resistance measuring method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111310905B (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111310905B (en) | Neural network model training method and device and heating and ventilation system energy efficiency optimization method | |
| Zamuda et al. | Differential evolution with self-adaptation and local search for constrained multiobjective optimization | |
| CN114662780A (en) | Carbon emission prediction method, carbon emission prediction device, electronic apparatus, and storage medium | |
| CN114492279B (en) | Parameter optimization method and system for analog integrated circuit | |
| CN107579816B (en) | Method for generating password dictionary based on recurrent neural network | |
| EP3602419A1 (en) | Neural network optimizer search | |
| JP2019197540A5 (en) | ||
| US8639643B2 (en) | Classification of a document according to a weighted search tree created by genetic algorithms | |
| CN109934330A (en) | The method of prediction model is constructed based on the drosophila optimization algorithm of diversified population | |
| JP2020035320A (en) | Reward function generating method and computer system | |
| CN107678560B (en) | Candidate result generation method and device of input method, storage medium and electronic equipment | |
| CN110675879B (en) | Audio evaluation method, system, equipment and storage medium based on big data | |
| Jin et al. | Accelerating nonlinear dc circuit simulation with reinforcement learning | |
| CN111428487B (en) | Model training method, lyric generation method, device, electronic equipment and medium | |
| CN115345106B (en) | Verilog-A model construction method, system and equipment for electronic device | |
| Wickramasinghe et al. | Building a large synthetic population from Australian census data | |
| CN112069455A (en) | Log-softmax function hardware acceleration computing method | |
| CN113051938A (en) | Machine translation model optimization method based on Transformer model | |
| CN111957053A (en) | Game player matching method and device, storage medium and electronic equipment | |
| CN112712202B (en) | Short-term wind power prediction method and device, electronic equipment and storage medium | |
| CN113254811B (en) | Interest place recommendation method, system and storage medium based on big data | |
| CN115473232B (en) | Load parameter adjusting method, system, equipment and storage medium | |
| CN117151425B (en) | RQL algorithm-based selective disassembly planning method and system | |
| KR102664600B1 (en) | Method for determining behavior of aircraft and electronic apparatus therefor | |
| CN116932714A (en) | Method and device for training generated dialogue model and realizing generated dialogue |
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 |