CN111598225A - Air conditioner cold load prediction method based on adaptive deep confidence network - Google Patents

Air conditioner cold load prediction method based on adaptive deep confidence network Download PDF

Info

Publication number
CN111598225A
CN111598225A CN202010415187.7A CN202010415187A CN111598225A CN 111598225 A CN111598225 A CN 111598225A CN 202010415187 A CN202010415187 A CN 202010415187A CN 111598225 A CN111598225 A CN 111598225A
Authority
CN
China
Prior art keywords
layer
network
training
cold load
prediction
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
CN202010415187.7A
Other languages
Chinese (zh)
Other versions
CN111598225B (en
Inventor
于军琪
冉彤
赵安军
任延欢
周昕玮
张万虎
席江涛
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.)
Xian University of Architecture and Technology
Original Assignee
Xian University of Architecture and Technology
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 Xian University of Architecture and Technology filed Critical Xian University of Architecture and Technology
Priority to CN202010415187.7A priority Critical patent/CN111598225B/en
Publication of CN111598225A publication Critical patent/CN111598225A/en
Application granted granted Critical
Publication of CN111598225B publication Critical patent/CN111598225B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Air Conditioning Control Device (AREA)

Abstract

The invention discloses an air conditioner cold load prediction method based on a self-adaptive depth confidence network, which comprises the steps of firstly collecting cold load data, adopting a Lagrange interpolation method to make up missing and abnormal energy consumption data, and carrying out normalization processing on the processed energy consumption data; processing the processed data through independent Gaussian distribution, and inputting the processed CRBM to a prediction model; training through an RBM unsupervised mechanism, inputting an RBM hidden layer of the previous layer as a visual layer of the RBM of the next layer, and adjusting network parameters; then reverse training is carried out through a supervised BP neural network, and network parameters are adjusted again; adjusting network parameters by adopting an Adam optimization algorithm; finally, selecting parameters and a structure of the prediction model, and performing structure evaluation selection on the prediction model by adopting a reconstruction error RE; and evaluating the result by adopting a root mean square relative error RMSPE and an average absolute percentage error MAPE to complete the prediction of the air conditioner cold load. The method has better prediction precision, universality and applicability.

Description

Air conditioner cold load prediction method based on adaptive deep confidence network
Technical Field
The invention belongs to the technical field of building air conditioner cold load prediction, and particularly relates to an air conditioner cold load prediction method based on a self-adaptive deep belief network.
Background
With the acceleration of urbanization, the development of industry and the increase of population, global energy demand continues to rise. Statistically, the cold load of the building operation accounts for about 40% of the global cold load, and the greenhouse gas emission accounts for 1/3 of the total emission, which indicates that the building has become the largest energy consumer. For buildings, the proportion of the heating, ventilating and air conditioning system in the total cooling load of the building is the largest, and the huge energy consumption increases the pressure of a power grid. Therefore, it is very important to manage energy for buildings.
The building cold load prediction is an important part in the building energy management process, is a key work for realizing building energy conservation, can correctly and reasonably predict the building cold load, can timely and accurately discover some abnormal conditions or potential equipment faults in the building cold load, is convenient for managers to timely take measures, and further avoids excessive waste of energy. Meanwhile, the correct and reasonable prediction of the cold load of the building can provide certain basis for reasonable energy distribution of managers, so that the energy is reasonably and effectively used. And the power generation scheme can be reasonably arranged, the balance of supply and demand of the power grid is realized, and the power system can stably operate, which is also an important part for the power system.
The artificial neural network model, the support vector machine model, the decision tree model and the hybrid model in the existing cold load prediction method are widely applied. However, these models suffer from various problems that make the prediction accuracy less than ideal. For example, the learning speed of the artificial neural network algorithm is not high enough, and local optimal and overfitting phenomena are easy to generate; the support vector machine is difficult to use in a large number of samples, and the multi-classification problem is difficult to solve; the scheme probability of the decision tree model is easily influenced by human, and the decision accuracy is reduced; the calculation amount and the calculation difficulty of the mixed model are high, and the like. The prediction algorithm only focuses on the prediction of the self algorithm on the cold load, and the complex characteristics of the building cold load cannot be fully considered, so that the prediction effect is not ideal.
Disclosure of Invention
The technical problem to be solved by the invention is to provide an air conditioner cold load prediction method based on a self-adaptive deep belief network aiming at the defects in the prior art, to consider the complex factors influencing the building cold load, to realize the effective prediction of the building cold load condition, to achieve the technical effects of energy saving and consumption reduction by controlling the influencing conditions, and to solve the problems in the prior art.
The invention adopts the following technical scheme:
a method for predicting air conditioner cold load based on a self-adaptive depth confidence network comprises the steps of firstly collecting cold load data, adopting a Lagrange interpolation method to make up missing and abnormal energy consumption data, and carrying out normalization processing on the processed energy consumption data; processing the processed data through independent Gaussian distribution, and inputting the processed CRBM to a prediction model; training through an RBM unsupervised mechanism, inputting an RBM hidden layer of a previous layer as a visual layer of a next layer of RBM, and adjusting a network parameter theta to be { w, a, b }; then reverse training is carried out through a supervised BP neural network, and the network parameter theta is adjusted again to be { w, a, b }; adjusting a network parameter theta to be { w, a, b } by adopting an Adam optimization algorithm; finally, selecting parameters and a structure of the prediction model, and performing structure evaluation selection on the prediction model by adopting a reconstruction error RE; and evaluating the result by adopting a root mean square relative error RMSPE and an average absolute percentage error MAPE to complete the prediction of the air conditioner cold load.
Specifically, the prediction model adds continuous values of independent Gaussian distribution to a linear unit to simulate real data, and an energy function E (v, h; theta) is as follows:
Figure BDA0002494719810000021
where θ ═ w, a, b, σ }, σiIs a visible layer viThe corresponding standard deviation of the gaussian noise;
the activation probability for updating the visible layer and the hidden layer is as follows:
Figure BDA0002494719810000031
Figure BDA0002494719810000032
wherein, N (mu, sigma)2) Mean μ and variance σ representing a Gaussian function2，aiOffset of visible layer, hjTo hide the binary state of cell j, wijFor the weight between the two, n is the number of visual layers predicting the cooling load, viInputting binary states of i for the visual layer, bjIs the offset of the hidden layer.
Specifically, in RBM unsupervised mechanism training, an RBM converts an input vector from a visible layer to a hidden layer through an activation function, extracts features from the hidden layer to the visible layer through training a minimized internal energy function, obtains a combined configuration energy function E (v, h; theta) between the visible layer and the hidden layer, ensures function distribution standardization through the combined distribution of the visible layer and the hidden layer, obtains conditional probability distribution through calculation according to the determined visible layer and the hidden layer, determines an optimal model through maximizing a likelihood function by using a ladder ascent method, and when a set of training samples S is given, the set S is { v } v1,v2,...vn) And then, calculating to obtain a target function as a log-likelihood function Ls under the maximization.
Further, the log-likelihood function Ls is maximized as:
Figure BDA0002494719810000033
wherein, P (v)n) For the activation probability of each input sample, N is the number of training samples.
Further, the joint distribution between the visible layer and the hidden layer is as follows:
P(v,h|θ)＝exp(-E(v,h))/Z(θ)
Figure BDA0002494719810000034
Figure BDA0002494719810000035
wherein Z (theta) is a distribution function, v is a visible layer unit, h is a hidden layer unit, and after the visible layer and the hidden layer are determined, the conditional probability distribution is calculated as follows:
Figure BDA0002494719810000041
Figure BDA0002494719810000042
wherein, aiOffset of visible layer, hjTo hide the binary state of cell j, n is the predicted amount of cold load of the visible layer, wijM is the number of hidden layers corresponding to the visible layer, v is the weight between the twoiInputting binary states of i for the visual layer, bjIs the offset of the hidden layer.
Specifically, adjusting the network parameter θ ═ { w, a, b } again specifically is:
inputting an input vector containing time, temperature, humidity and solar radiation into a network model, processing the input vector into a continuous numerical value through a Gaussian distribution CRBM (cross-correlation matrix) through unsupervised network training, and performing model training;
and then dividing the training samples into f groups of training samples, training the f groups of training samples to adjust the network parameters theta of the prediction model to be { w, a, b }, and storing the network parameters after the number of training layers of the prediction model is reached.
Specifically, adjusting a network parameter θ ═ { w, a, b } by using an Adam optimization algorithm specifically includes:
initializing the parameter vector, the first moment vector and the second moment vector, then circularly and iteratively updating each part to make the parameter theta converge, namely adding 1 to the time t, correspondingly updating each parameter of the deviation, and finally updating the parameter theta of the prediction model by using the calculated parameter value.
Specifically, a root mean square relative error RMSPE and an average absolute percentage error MAPE are used for evaluation, a difference value between a predicted value and an actual value of a prediction model is compared, a reconstruction error is adopted for the CRBM-DBN network model for evaluation, a reconstruction error is formed by a deviation between the actual value and the predicted value, the depth of the prediction model is obtained through calculation of the reconstruction error, and cold load data are predicted through training by taking the MAPE and training time as evaluation standards for the number of hidden nodes.
Further, the root mean square relative error RMSPE and the average absolute percent error MAPE are specifically:
Figure BDA0002494719810000051
Figure BDA0002494719810000052
wherein, yiAnd
Figure BDA0002494719810000053
respectively an actual load value and a predicted load value at the ith moment;
Figure BDA0002494719810000054
the average value of the real values of the air conditioner cold load is obtained, and k is the number of samples of all test sets; the reconstruction error was used for the evaluation as follows:
Figure BDA0002494719810000055
where RE is the reconstruction error, X represents the min-batch matrix, aViRepresenting the state value of the current predicted visual unit.
Furthermore, the depth of the prediction model is 2, the number of hidden nodes is 20, and the prediction step length is 1 h.
Compared with the prior art, the invention has at least the following beneficial effects:
the invention discloses an air conditioner cold load prediction method based on an adaptive deep belief network, and provides the air conditioner cold load prediction method based on the adaptive deep belief network, aiming at the problem that building cold load data often has nonlinear and dynamic characteristics to cause that the cold load data cannot be reliably and accurately predicted. The algorithm combines the advantages of supervised learning and unsupervised learning, excavates high-dimensional features, extracts feature values from bottom to top and effectively solves a series of problems caused by random initialization of traditional neural network parameters. Besides having good initial points, the problems of over-fitting and under-fitting which often occur in the neural network are effectively solved through pre-training.
Furthermore, the selected data is subjected to deficiency and abnormal data compensation processing, the problem of data reliability can be effectively solved, the input data is subjected to normalization processing to avoid prediction errors caused by different orders of magnitude of parameters of the input layer, and the prediction errors of the network model caused by data errors can be effectively solved.
Furthermore, the problem that data loss is easily caused because the RBM can only accept binary input can be effectively solved by adopting a Gauss distribution processing Boltzmann machine.
Furthermore, the deep confidence network optimized by the Adam algorithm considers the second derivative of the objective function, so that the model has strong adaptivity and good performance in processing the nonlinear problem, and the problem of slow convergence speed in the process of training the network model parameters is effectively solved.
Furthermore, the network model is evaluated by adopting the reconstruction errors, so that the size of the reconstruction errors generated by different structure selection can be clearly seen, and the parameters and the structure of the model can be better selected.
In conclusion, the method accurately predicts the building cold load, the prediction precision is higher than that of a BP neural network prediction model, the prediction precision, the universality and the applicability are better, the method is particularly suitable for large public buildings with periodically changed cold loads, and the provided cold load prediction data is more useful for energy-saving planning and energy-saving planning.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a diagram of exception data handling;
FIG. 2 is a schematic view of an RBM;
FIG. 3 is a schematic diagram of a CRBM-DBN network prediction;
FIG. 4 is a flow chart of the CRBM-DBN algorithm;
fig. 5 is a model reconstruction error map, in which (a) is a reconstruction error when the network depth is 1, (b) is a reconstruction error when the network depth is 2, (c) is a reconstruction error when the network depth is 3, and (d) is a reconstruction error when the network depth is 4;
FIG. 6 is a graph showing the prediction results, wherein (a) is a cold load prediction graph for a day of rest and a day of work, (b) is an amplification effect of 7 to 13 hours, and (c) is an amplification effect of 22 to 28 hours.
Detailed Description
The invention relates to an air conditioner cold load prediction method based on a self-adaptive deep confidence network, which comprises the following steps of:
s1, collecting cold load data (including temperature, humidity, solar load, wind speed, time and the like), performing compensation processing on missing and abnormal energy consumption data by adopting a Lagrange interpolation method, and performing normalization processing on the processed energy consumption data to be used as cold load energy consumption prediction;
s101, the model aims at a cold load prediction method of a shopping mall, and the load of the shopping mall is 08: 00-22: 00, take 15 hours. Assuming the ith data in a day is missing or abnormal, the lagrangian function is constructed using the remaining 14 known load points in the day:
Figure BDA0002494719810000071
wherein j isjPj (x) is a 14 th degree polynomial,
Figure BDA0002494719810000072
Bkas a set of coordinates for the remaining 14 time pointsAnd B isk∈ {1, 2.. i-1, i + 1.. 14}, the load value at the ith time of the day is yi＝L14(i)。
Referring to fig. 1, the missing and abnormal data are compensated by lagrange interpolation, and the original abnormal data and the result before and after the missing data are processed.
S102, considering that the orders of magnitude of each parameter in the input layer are different, in order to avoid prediction errors caused by the difference, normalization processing is carried out on the input parameters, and the specific formula is as follows:
Figure BDA0002494719810000073
wherein x' is the result of normalization, x is the input vector including temperature, humidity, solar load, time, and xminAnd xmaxAre the minimum and maximum values of the corresponding samples in the input data.
S2, the data processed in the step S1 are processed through independent Gaussian distribution, and the processed CRBM is input into a prediction model to ensure the continuity of the data;
the prediction model adds continuous values of independent Gaussian distribution in a linear unit to simulate real data, so that a traditional limited Boltzmann machine can process continuous input vectors, and an energy function E (v, h; theta) of the prediction model is as follows:
Figure BDA0002494719810000081
where θ ═ w, a, b, σ }, σiIs a visible layer viCorresponding standard deviation of Gaussian noise, generally, σ2The result is better when the value is 1.
Thus, the activation probabilities of the updated visible and hidden layers are:
Figure BDA0002494719810000082
Figure BDA0002494719810000083
wherein, N (mu, sigma)2) Mean μ and variance σ representing a Gaussian function2。
S3, training through an RBM unsupervised mechanism, inputting a previous RBM hidden layer as a next RBM visual layer, and adjusting a network parameter theta to be { w, a, b } for the first time;
referring to fig. 2, the RBM transforms the input vector from the visual layer to the hidden layer through the activation function, and performs the feature extraction by training the minimum internal energy function from the hidden layer to the visual layer, where the joint configuration energy function between the visual layer and the hidden layer is:
Figure BDA0002494719810000084
wherein v isiAnd hjIs the binary state of the visual layer input i and the hidden unit j, aiAnd bjOffset, w, of the visible layer and the hidden layer, respectivelyijM is the number of hidden layers corresponding to the visible layer, and n is the predicted cooling load number of the visible layer.
The joint distribution between the visible layer and the hidden layer is:
P(v,h|θ)＝exp(-E(v,h；θ))/Z(θ)
Figure BDA0002494719810000091
Figure BDA0002494719810000092
where Z (θ) is the partition function to ensure that the distribution is normalized, v is the visible layer element, and h is the hidden layer element to ensure that the distribution of the function is normalized.
After the visible layer and the hidden layer are determined, the conditional probability distribution is calculated as follows:
Figure BDA0002494719810000093
Figure BDA0002494719810000094
the purpose of RBM training is to determine the best model by maximizing the likelihood function using the ladder-ascending method, given a set of training samples, S ═ v1,v2,...vn) Then, the objective function is the log-likelihood function that maximizes:
Figure BDA0002494719810000095
wherein, P (v)n) And N is the number of corresponding input training samples for the activation probability of each input sample.
After a training sample set is given, adjusting a network parameter θ ═ { w, a, b } can adjust the activation probability of the corresponding adjusted hidden layer to determine the state of the hidden layer; after a hidden layer is given, the activation probability of the corresponding visual layer can be adjusted by adjusting the network parameter θ ═ { w, a, b }, so that the reconstruction of the input data is realized, and the log-likelihood function, namely the activation probability of the hidden layer and the visual layer is maximized, so that the reconstruction of the data is realized.
S4, performing reverse training through a supervised BP neural network, and adjusting a network parameter θ to { w, a, b } for the second time;
referring to fig. 3 and 4, input vectors (including time, temperature, humidity, solar radiation, etc.) are input into the network model, and are processed into continuous values through gaussian distribution CRBM by unsupervised network training, and model training is performed from bottom to top according to steps S2 and S3;
then dividing the training samples into f groups of small-batch training samples, training the f groups of training sets to adjust the network parameter theta of the model to be { w, a, b }, and after the number of training layers of the model is reached, storing the trained network parameter and carrying out the next step;
the BP algorithm reversely has supervision and fine adjustment of network parameters;
s5, after the input data are processed by gaussian distribution, and then undergo RBM unsupervised feature training and BP supervised training learning, the input data X is converted into another feature space F (Y, θ) by the model, and the network parameter θ is calculated by minimizing the error between F (Y, θ) and X as { w, a, b }, where the mean square error function is:
Figure BDA0002494719810000101
where k is the number of test set cooling loads.
Adjusting the network parameter theta to be { w, a, b } again by adopting an Adam optimization algorithm;
the Adam optimization algorithm initializes the parameter vector, the first moment vector and the second moment vector, then iteratively updates each part in a circulating way to make the parameter theta converged, namely adding 1 to the time t, corresponding to each parameter of the updating deviation, and finally updates the parameter theta of the model by using the calculated parameter value.
An Adam optimization algorithm is adopted to minimize a loss function, the Adam algorithm designs independent self-adaptive learning rates for different parameters by calculating first moment estimation and second moment estimation of a gradient, and has high calculation efficiency and low memory requirement, and the network parameter updating process is represented as follows:
Figure BDA0002494719810000102
mt＝β1×mt-1+(1-β1)×gt
Figure BDA0002494719810000103
Figure BDA0002494719810000111
Figure BDA0002494719810000112
wherein, gtIs the gradient of the mean square error function L (theta) to theta, mtIs an estimate of the first moment of the gradient, ntIs an estimate of the second moment of the gradient,
Figure BDA0002494719810000113
and
Figure BDA0002494719810000114
are respectively to mtAnd ntCorrection of deviation, exponential decay rate of moment estimation β1Is 0.9, β20.99, step size η of 0.001, and a small constant of 10 for numerical stability-8。
S6, selecting parameters and structures of the model, and performing structure evaluation selection on the model by adopting a reconstruction error RE; the results were evaluated using the root mean square relative error RMSPE and the mean absolute percent error MAPE.
The most common evaluation indexes are evaluated by using a root mean square relative error RMSPE and an average absolute percentage error MAPE, and the difference value between the predicted value and the actual value of the prediction model is compared, and the calculation formula is as follows.
Figure BDA0002494719810000115
Figure BDA0002494719810000116
Wherein, yiAnd
Figure BDA0002494719810000117
respectively an actual load value and a predicted load value at the ith moment;
Figure BDA0002494719810000118
the average value of the actual values of the air conditioner cold load is shown, k is the number of samples of all test sets, and the smaller the RMSPE and MAPE values are, the better the prediction effect is.
Simple evaluation is carried out on the CRBM-DBN (Adam) network model by adopting a reconstruction error (reconstruction error):
Figure BDA0002494719810000119
where RE is the reconstruction error, X represents the min-batch matrix, aViAnd representing the state value of the current prediction visual unit, wherein the deviation between the true value and the predicted value forms a reconstruction error, and the depth of the model can be calculated by calculating the reconstruction error.
Referring to fig. 5, the setting of the structural parameters in the deep learning model affects the training process, the training time, and the prediction result. In order to optimize training to reduce errors, the number of RBM layers is increased layer by layer in an experiment, the depth of the network is determined by calculating the reconstruction errors, when the depth of the network is 2 layers, the reconstruction errors are wholly reduced and gradually tend to be stable, the average value of the reconstruction errors is small, and therefore the accuracy is high. And when the network depth is other values, the average value of the reconstruction errors is larger and is always in a fluctuation state. Therefore, the network depth of the RBM is determined to be 2 layers, that is, 2 hidden unit layers and 1 visible unit layer are total.
And regarding the number of the hidden nodes, MAPE and training time are used as evaluation criteria, and the number is selected through experiments.
TABLE 1 implicit node number selection
Figure BDA0002494719810000121
As can be seen from table 1, when the number of hidden layer nodes is 20, the model prediction effect is the best.
Therefore, the depth of the prediction model set by the method is 2, the number of hidden nodes is 20, the prediction step length is 1h, and the cold load data is predicted through training.
The predicted daily cold load curve is generated, and after parameters and structures of a network model are obtained, the predicted daily cold load curve can be obtained, as shown in fig. 6, fig. 6(a) is a load prediction result of different algorithms, fig. 6(b) (c) is a load prediction result amplified in a break day and a working day peak load time period, and the Adam optimization algorithm takes the second derivative of an objective function into consideration, so that the model has strong adaptivity and has good performance in processing a nonlinear problem.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Predictive model result comparison analysis
The case carries out predictive analysis on the cold load data of the holidays and the workdays:
TABLE 2 working day prediction model results analysis
Figure BDA0002494719810000131
Figure BDA0002494719810000141
TABLE 3 Sunday prediction model results analysis
Figure BDA0002494719810000142
Figure BDA0002494719810000151
And (3) evaluating by using the most common evaluation indexes of root mean square relative error RMSPE and average absolute percentage error MAPE, and comparing the difference value between the predicted value and the actual value of the prediction model.
TABLE 4 model evaluation analysis
Figure BDA0002494719810000152
As can be seen from Table 4, the prediction accuracy and time complexity are greatly improved by using the method of the present invention compared with the conventional DBN and CRBM-DBN methods.
The method takes the commercial building with complex Xian as a research object and establishes a cold load prediction model on the basis of the daily cold load of the commercial building, and the building researched by the method is put into use and normally operates, so that the smoothness of data is ensured, and reasonable data is provided for prediction and verification.
In summary, the air conditioner cold load prediction method based on the adaptive deep belief network introduces a CRBM model with gaussian distribution and a problem that the convergence speed of the CRBM-DBN is slow when network model parameters are trained aiming at the problem that the RBM can only accept binary input to cause data loss, and provides the CRBM-DBN model method of adaptive learning. The experimental result shows that the method provided by the invention is better than the traditional DBN and CRBM-DBN in the aspects of prediction precision and time complexity, and is an effective air conditioner cold load prediction method.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims (10)

1. The method for predicting the cold load of the air conditioner based on the self-adaptive deep confidence network is characterized by comprising the following steps of firstly collecting cold load data, adopting a Lagrange interpolation method to make up missing and abnormal energy consumption data, and carrying out normalization processing on the processed energy consumption data; processing the processed data through independent Gaussian distribution, and inputting the processed CRBM to a prediction model; training through an RBM unsupervised mechanism, inputting an RBM hidden layer of a previous layer as a visual layer of a next layer of RBM, and adjusting a network parameter theta to be { w, a, b }; then reverse training is carried out through a supervised BP neural network, and the network parameter theta is adjusted again to be { w, a, b }; adjusting a network parameter theta to be { w, a, b } by adopting an Adam optimization algorithm; finally, selecting parameters and a structure of the prediction model, and performing structure evaluation selection on the prediction model by adopting a reconstruction error RE; and evaluating the result by adopting a root mean square relative error RMSPE and an average absolute percentage error MAPE to complete the prediction of the air conditioner cold load.
2. The method for predicting the cold load of the air conditioner based on the adaptive deep belief network as claimed in claim 1, wherein the prediction model adds continuous values of independent Gaussian distribution to linear units to simulate real data, and the energy function E (v, h; theta) is as follows:
Figure FDA0002494719800000011
where θ ═ w, a, b, σ }, σiIs a visible layer viThe corresponding standard deviation of the gaussian noise;
the activation probability for updating the visible layer and the hidden layer is as follows:
Figure FDA0002494719800000012
Figure FDA0002494719800000013
wherein, N (mu, sigma)2) Mean μ and variance σ representing a Gaussian function2，aiOffset of visible layer, hjTo be hiddenBinary state, w, of hidden cell jijFor the weight between the two, n is the number of visual layers predicting the cooling load, viInputting binary states of i for the visual layer, bjIs the offset of the hidden layer.
3. The adaptive depth-confidence-network-based air conditioner cold load prediction method of claim 1, wherein in the RBM unsupervised mechanism training, the RBM converts an input vector from a visible layer to a hidden layer through an activation function, and from the hidden layer to the visible layer, and completes the extraction of features by training a minimized internal energy function to obtain a combined configuration energy function E (v, h; theta) between the visible layer and the hidden layer, ensures the function distribution normalization through the combined distribution of the visible layer and the hidden layer, calculates a conditional probability distribution according to the determined visible layer and the hidden layer, determines an optimal model by maximizing a likelihood function through a ladder ascent method, and when a set of training samples S is given, S is { v } { (v) } v } is the set of training samples1,v2,...vn) And then, calculating to obtain a target function as a log-likelihood function Ls under the maximization.
4. The method for predicting the cold load of the air conditioner based on the adaptive deep belief network as claimed in claim 3, wherein the log likelihood function Ls is maximized as follows:
Figure FDA0002494719800000021
wherein, P (v)n) For the activation probability of each input sample, N is the number of training samples.
5. The method for predicting the cold load of the air conditioner based on the adaptive deep belief network as claimed in claim 3, wherein the joint distribution between the visual layer and the hidden layer is as follows:
P(v,h|θ)＝exp(-E(v,h))/Z(θ)
Figure FDA0002494719800000022
Figure FDA0002494719800000023
wherein Z (theta) is a distribution function, v is a visible layer unit, h is a hidden layer unit, and after the visible layer and the hidden layer are determined, the conditional probability distribution is calculated as follows:
Figure FDA0002494719800000024
Figure FDA0002494719800000025
wherein, aiOffset of visible layer, hjTo hide the binary state of cell j, n is the predicted amount of cold load of the visible layer, wijM is the number of hidden layers corresponding to the visible layer, v is the weight between the twoiInputting binary states of i for the visual layer, bjIs the offset of the hidden layer.
6. The method for predicting the cooling load of the air conditioner based on the adaptive deep belief network as claimed in claim 1, wherein the step of adjusting the network parameter θ ═ { w, a, b } is specifically as follows:
inputting an input vector containing time, temperature, humidity and solar radiation into a network model, processing the input vector into a continuous numerical value through a Gaussian distribution CRBM (cross-correlation matrix) through unsupervised network training, and performing model training;
and then dividing the training samples into f groups of training samples, training the f groups of training samples to adjust the network parameters theta of the prediction model to be { w, a, b }, and storing the network parameters after the number of training layers of the prediction model is reached.
7. The method for predicting the cold load of the air conditioner based on the adaptive deep belief network is characterized in that an Adam optimization algorithm is adopted to adjust network parameters theta ═ { w, a, b }, and specifically:
initializing the parameter vector, the first moment vector and the second moment vector, then circularly and iteratively updating each part to make the parameter theta converge, namely adding 1 to the time t, correspondingly updating each parameter of the deviation, and finally updating the parameter theta of the prediction model by using the calculated parameter value.
8. The method for predicting the air conditioning cold load based on the adaptive depth confidence network as claimed in claim 1, wherein the method is characterized in that a root mean square relative error RMSPE and an average absolute percentage error MAPE are used for evaluation, a difference value between a predicted value and an actual value of a prediction model is compared, a reconstruction error is adopted for the CRBM-DBN network model for evaluation, a deviation between the actual value and the predicted value forms the reconstruction error, the depth of the prediction model is calculated by calculating the reconstruction error, and for the number of hidden nodes, MAPE and training time are used as evaluation criteria, and cold load data are predicted by training.
9. The adaptive depth-based confidence network-based air conditioner cold load prediction method of claim 8, wherein the root mean square relative error RMSPE and the average absolute percentage error MAPE are specifically:
Figure FDA0002494719800000041
Figure FDA0002494719800000042
wherein, yiAnd
Figure FDA0002494719800000043
respectively an actual load value and a predicted load value at the ith moment;
Figure FDA0002494719800000044
the average value of the real values of the air conditioner cold load is obtained, and k is the number of samples of all test sets; using reconstruction errors for evaluation such asThe following:
Figure FDA0002494719800000045
where RE is the reconstruction error, X represents the min-batch matrix, aViRepresenting the state value of the current predicted visual unit.
10. The air conditioner cold load prediction method based on the adaptive deep belief network as claimed in claim 8, wherein the depth of the prediction model is 2, the number of hidden nodes is 20, and the prediction step length is 1 h.
CN202010415187.7A 2020-05-15 2020-05-15 Air conditioner cold load prediction method based on self-adaptive deep confidence network Active CN111598225B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010415187.7A CN111598225B (en) 2020-05-15 2020-05-15 Air conditioner cold load prediction method based on self-adaptive deep confidence network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010415187.7A CN111598225B (en) 2020-05-15 2020-05-15 Air conditioner cold load prediction method based on self-adaptive deep confidence network

Publications (2)

Publication Number Publication Date
CN111598225A true CN111598225A (en) 2020-08-28
CN111598225B CN111598225B (en) 2023-05-02

Family

ID=72189761

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010415187.7A Active CN111598225B (en) 2020-05-15 2020-05-15 Air conditioner cold load prediction method based on self-adaptive deep confidence network

Country Status (1)

Country Link
CN (1) CN111598225B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112363099A (en) * 2020-10-30 2021-02-12 天津大学 TMR current sensor temperature drift and geomagnetic field correction device and method
CN113177355A (en) * 2021-04-28 2021-07-27 南方电网科学研究院有限责任公司 Power load prediction method
CN113449467A (en) * 2021-06-21 2021-09-28 清华大学 JDAN-NFN-based online security evaluation method and device for power system
CN113776164A (en) * 2021-09-09 2021-12-10 广州珠江新城能源有限公司 Automatic control method and control system for centralized cooling system
CN114814645A (en) * 2022-06-27 2022-07-29 广东电网有限责任公司佛山供电局 On-line analysis method and related device for capacitor voltage transformer

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015172560A1 (en) * 2014-05-16 2015-11-19 华南理工大学 Central air conditioner cooling load prediction method based on bp neural network
US20160378902A1 (en) * 2015-06-25 2016-12-29 Globalfoundries Inc. Generative learning for realistic and ground rule clean hot spot synthesis
CN108549960A (en) * 2018-04-20 2018-09-18 国网重庆市电力公司永川供电分公司 A kind of 24 hours Methods of electric load forecasting
CN108665001A (en) * 2018-05-10 2018-10-16 河南工程学院 It is a kind of based on depth confidence network across subject Idle state detection method
CN110580543A (en) * 2019-08-06 2019-12-17 天津大学 Power load prediction method and system based on deep belief network
CN110783964A (en) * 2019-10-31 2020-02-11 国网河北省电力有限公司 Risk assessment method and device for static security of power grid

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015172560A1 (en) * 2014-05-16 2015-11-19 华南理工大学 Central air conditioner cooling load prediction method based on bp neural network
US20160378902A1 (en) * 2015-06-25 2016-12-29 Globalfoundries Inc. Generative learning for realistic and ground rule clean hot spot synthesis
CN108549960A (en) * 2018-04-20 2018-09-18 国网重庆市电力公司永川供电分公司 A kind of 24 hours Methods of electric load forecasting
CN108665001A (en) * 2018-05-10 2018-10-16 河南工程学院 It is a kind of based on depth confidence network across subject Idle state detection method
CN110580543A (en) * 2019-08-06 2019-12-17 天津大学 Power load prediction method and system based on deep belief network
CN110783964A (en) * 2019-10-31 2020-02-11 国网河北省电力有限公司 Risk assessment method and device for static security of power grid

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
任玮: "基于深度信念网络的网络流量预测模型", 《山西电子技术》 *
何琬等: "基于深层神经网络的电力负荷预测", 《环境与可持续发展》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112363099A (en) * 2020-10-30 2021-02-12 天津大学 TMR current sensor temperature drift and geomagnetic field correction device and method
CN112363099B (en) * 2020-10-30 2023-05-09 天津大学 TMR current sensor temperature drift and geomagnetic field correction device and method
CN113177355A (en) * 2021-04-28 2021-07-27 南方电网科学研究院有限责任公司 Power load prediction method
CN113177355B (en) * 2021-04-28 2024-01-12 南方电网科学研究院有限责任公司 Power load prediction method
CN113449467A (en) * 2021-06-21 2021-09-28 清华大学 JDAN-NFN-based online security evaluation method and device for power system
CN113776164A (en) * 2021-09-09 2021-12-10 广州珠江新城能源有限公司 Automatic control method and control system for centralized cooling system
CN113776164B (en) * 2021-09-09 2022-08-30 广州珠江新城能源有限公司 Automatic control method and control system for centralized cooling system
CN114814645A (en) * 2022-06-27 2022-07-29 广东电网有限责任公司佛山供电局 On-line analysis method and related device for capacitor voltage transformer
CN114814645B (en) * 2022-06-27 2022-09-13 广东电网有限责任公司佛山供电局 On-line analysis method and related device for capacitor voltage transformer

Also Published As

Publication number Publication date
CN111598225B (en) 2023-05-02

Similar Documents

Publication Publication Date Title
CN111598225A (en) Air conditioner cold load prediction method based on adaptive deep confidence network
CN110458443B (en) Smart home energy management method and system based on deep reinforcement learning
Jallal et al. A hybrid neuro-fuzzy inference system-based algorithm for time series forecasting applied to energy consumption prediction
CN112733462B (en) Ultra-short-term wind power plant power prediction method combining meteorological factors
CN110705760A (en) Photovoltaic power generation power prediction method based on deep belief network
CN112415924A (en) Energy-saving optimization method and system for air conditioning system
CN109146121A (en) The power predicating method stopped in the case of limited production based on PSO-BP model
Han et al. Energy saving of buildings for reducing carbon dioxide emissions using novel dendrite net integrated adaptive mean square gradient
CN111561732B (en) Heat exchange station heat supply adjusting method and system based on artificial intelligence
CN113572157A (en) User real-time autonomous energy management optimization method based on near-end policy optimization
CN105574586B (en) General-purpose aircraft boat material needing forecasting method based on MPSO-BP networks
Faiq et al. Prediction of energy consumption in campus buildings using long short-term memory
CN114298140A (en) Wind power short-term power prediction correction method considering unit classification
CN116989442A (en) Central air conditioner load prediction method and system
CN111461466A (en) Heating household valve adjusting method, system and equipment based on L STM time sequence
Georgiou et al. Implementing artificial neural networks in energy building applications—A review
CN115907195A (en) Photovoltaic power generation power prediction method, system, electronic device and medium
Ku et al. Building electric energy prediction modeling for BEMS using easily obtainable weather factors with Kriging model and data mining
Wang et al. Wind process pattern forecasting based ultra-short-term wind speed hybrid prediction
Fu et al. Predictive control of power demand peak regulation based on deep reinforcement learning
CN114611823B (en) Optimized dispatching method and system for electricity-cold-heat-gas multi-energy-demand typical park
CN117454939A (en) Wind power prediction method for optimizing BP neural network based on NSABO-Gold algorithm
Fu et al. A Sarsa-based adaptive controller for building energy conservation
CN114234392B (en) Air conditioner load fine prediction method based on improved PSO-LSTM
CN113962454A (en) LSTM energy consumption prediction method based on dual feature selection and particle swarm optimization

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