A prediction method for variable air volume air conditioning supply air volume based on deep belief neural network
Through the prediction method based on deep confidence neural network, the problem of inaccurate air supply forecast of variable air volume air conditioners is solved, and higher prediction accuracy and stability are achieved, energy consumption is reduced, and new ideas are provided for energy-saving variable air volume air conditioners research is provided. .
Patent Information
- Application Number
- CN202010477527.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-05-29
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
There are inaccurate and unstable predictions of air supply volume of variable air volume air conditioners, which affect energy conservation and indoor comfort.
Using a prediction method based on deep confidence neural network, the deep confidence neural network model is established to accurately predict air supply through the collection and processing of outdoor meteorological parameters and historical air supply data, and overcome the BP, Elman and fuzzy shallow neural network. Local convergence and overfitting problems.
It improves the accuracy and stability of air supply volume prediction, reduces building energy consumption, provides new energy-saving ideas, and helps to achieve a comfortable living environment.
Smart Images

Figure CN111594996B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of variable air volume air conditioner supply air volume prediction, and in particular to a method for predicting the variable air volume air conditioner supply air volume based on a deep belief neural network. Background Art
[0002] As the proportion of air conditioning energy consumption in building energy consumption continues to increase, variable air volume air conditioning (VAV, Variable Air Volume Air-Conditioning) has become a research object in the field of air conditioning at home and abroad due to its significant energy-saving advantages. Variable air volume air conditioning involves multiple parameters such as temperature, humidity, pressure and flow. Due to the nonlinearity and mutual coupling characteristics between these parameters, the prediction of the air supply volume of variable air volume air conditioning is inaccurate and unstable, which affects the energy saving and indoor comfort of variable air volume air conditioning. Therefore, accurately predicting the air supply volume of variable air volume air conditioning is an urgent problem to be solved.
[0003] The disclosure of the above background technology content is only used to assist in understanding the concept and technical solution of the present invention. It does not necessarily belong to the prior art of this patent application. In the absence of clear evidence that the above content has been disclosed on the filing date of this patent application, the above background technology should not be used to evaluate the novelty and creativity of the present application. Summary of the invention
[0004] In view of the above technical problems, the present invention provides a method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network, so as to achieve accurate prediction of the VAV air supply volume and improve the prediction accuracy.
[0005] To achieve the above purpose, the present invention adopts the following technical solutions:
[0006] A method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network, characterized in that it comprises the following steps:
[0007] Step 1: Determine the input parameters for training the VAV air conditioning supply air volume prediction based on the deep belief neural network, and collect data to obtain corresponding sample data, including test samples and training sample data;
[0008] Step 2: Use MATLAB software to import the input parameters of the variable air volume air conditioner supply air volume prediction based on the deep belief neural network collected in step 1 and the corresponding supply air volume as training samples into the deep belief neural network, establish a training model of the variable air volume air conditioner supply air volume prediction model based on the deep belief neural network, conduct learning training on the training samples, and continuously adjust the parameters of the model to obtain the best training model;
[0009] Step 3: Determine the variable air volume air conditioning supply air volume prediction model based on the deep belief neural network, import the test sample into the trained prediction model, and predict the corresponding supply air volume of the test sample.
[0010] Furthermore, the input parameters for the variable air volume air conditioning supply air volume prediction in step one include: the supply air volume at time t of the previous day, the supply air volume at time t two days ago, the outdoor temperature at time t of the previous day, the outdoor temperature at time t two days ago, the outdoor temperature at time t on the current day, the outdoor temperature at time t-1 on the current day, the outdoor temperature at time t-2 on the current day, the solar radiation temperature on the current day, and the atmospheric humidity on the current day.
[0011] Furthermore, in step 2, a variable air volume air conditioning air supply volume prediction model based on a deep belief neural network is established to train the training samples, and the parameters of the model are continuously adjusted to obtain the best training model. The establishment of the training model includes the following steps:
[0012] S1: Standardize the model input parameters;
[0013] S2: perform sample classification;
[0014] S3: Initialize the weights and thresholds of the deep belief neural network;
[0015] S4: Calculate the update rules for the weights and thresholds of the deep belief neural network;
[0016] S5: Set the number of neurons in the input layer, each hidden layer and the output layer;
[0017] S6: Train the DBN model;
[0018] S7: Use the trained deep belief neural network to predict the air supply volume corresponding to the test sample.
[0019] Furthermore, the sample classification in step S2 is achieved in the following way: the sample data is divided into two parts, training samples and test samples, the training samples are used for learning and training, and the test samples are used for testing; the training samples are imported into the MATLAB software in the form of a matrix, and the air supply volume at time t of the previous day, the air supply volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day and the atmospheric humidity of the current day are taken as input parameters, the air supply volume is taken as the output result, and a deep confidence neural network training model is established to carry out deep confidence neural network learning and training.
[0020] Furthermore, the variable air volume air conditioning supply air volume prediction model based on deep belief neural network has a four-layer structure; the first layer is the display layer, that is, the input layer, which is composed of outdoor meteorological parameters and the historical air volume of the variable air volume air conditioner; the second and third layers are hidden layers, and the last layer is the output layer, that is, the variable air volume air conditioner supply air volume at time t on the predicted day in the model.
[0021] Furthermore, the outdoor meteorological parameters during the data collection period include outdoor average atmospheric pressure, outdoor average wind speed, outdoor calculated dry-bulb average temperature, outdoor calculated wet-bulb average humidity, and outdoor calculated daily average temperature.
[0022] Furthermore, DBN is composed of a certain number of stacked restricted Boltzmann machines (RBMs); in the process of training RBMs, DBN solves the activation probabilities of its visible and hidden layers according to the energy function of RBMs, and then obtains the update rules of RBM training parameters.
[0023] Furthermore, the training process of the DBN model is mainly divided into two steps:
[0024] In the first step, in RBM, parameters are obtained by layer-by-layer training; the output end is connected to the BP neural network to obtain output data, completing the forward stage of the entire DBN training;
[0025] In the second step, based on the error between the actual data and the output data, the error is back-propagated from top to bottom to each layer of RBM, and each layer of RBM is fine-tuned layer by layer to make the parameters reach the global optimum and complete the training of the entire DBN.
[0026] Furthermore, in the second step of DBN model training, the actual air volume of the variable air volume air conditioner at time t is first used as the output variable of the model, and the air volume at time t the previous day, the air volume at time t the previous two days, the outdoor temperature at time t the previous day, the outdoor temperature at time t the previous two days, the outdoor temperature at time t on the current day, the outdoor temperature at time t-1 on the current day, the outdoor temperature at time t-2 on the current day, the solar radiation temperature on the current day, and the atmospheric humidity on the current day are used as the input variables of the model. The output variables and the input variables constitute the corresponding training data set S; the training data set S is then imported into the DBN model, and the RBM training parameter update rule is used to perform layer-by-layer training, first performing unsupervised pre-training, and then performing supervised reverse tuning training, until the training error reaches the set requirement and the update is stopped, and the trained optimal DBN model is obtained.
[0027] The beneficial effects of the present invention compared with the prior art are as follows:
[0028] (1) The present invention is based on deep neural network modeling, and the factors that affect the air supply volume of variable air volume air conditioners are classified and input into the algorithm, which helps to more comprehensively reflect the prediction of air supply volume. It is more advanced in theory and can overcome the interference of mutual coupling between different parameters. Secondly, the use of deep confidence neural network to establish the prediction of variable air volume air supply volume can avoid many problems such as local convergence, slow convergence speed, long training time for large amounts of data, and overfitting of BP, Elman, and fuzzy shallow neural networks, and can overcome multiple limitations in practical applications. It can better ensure the accuracy of variable air volume air supply volume prediction in actual situations and meet the requirements of practical applications.
[0029] (2) This invention introduces the theory of deep learning into the prediction field of variable air volume air conditioner supply air volume for the first time, and uses the DBN model to accurately predict its supply air volume, which can reduce building energy consumption. Accurately predicting the supply air volume of variable air volume air conditioners can provide new ideas and inspiration for the research of new energy-saving variable air volume air conditioners, and indirectly help provide humans with a comfortable living environment, which will be of great significance to the future research of variable air volume air conditioners. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 It is a flow chart of prediction of air supply volume of variable air volume air conditioner based on DBN of the present invention;
[0031] Figure 2 It is a deep belief neural network structure;
[0032] Figure 3 It is the DBN training process;
[0033] Figure 4 It is a schematic diagram comparing the predicted value of the deep belief neural network with the actual value;
[0034] Figure 5 This is a comparison chart of the prediction error of the DBN model. DETAILED DESCRIPTION
[0035] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation modes. It should be understood that the "up", "down", "left" and "right" directions mentioned in the following implementation modes of the present invention are all based on the positions of the corresponding drawings. These words used to limit the directions are only for the convenience of explanation and do not represent limitations on the specific technical solutions of the present invention. Unless otherwise specified, the same numbers in the accompanying drawings represent the same structure. Based on the principles of the present invention, other implementation modes obtained by ordinary technicians in this field without making creative work are deemed to be within the protection of the present invention.
[0036] The present invention discloses a method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network. The method applies the deep belief neural network to the prediction of the air supply volume of a VAV variable air volume air conditioner for the first time. The method specifically includes the following steps:
[0037] Step 1: Determine the input parameters for training the VAV air conditioning supply air volume prediction based on the deep belief neural network, and collect data to obtain corresponding sample data, including test samples and training sample data;
[0038] Step 2: Use MATLAB software to import the input parameters of the variable air volume air conditioner supply air volume prediction based on the deep belief neural network collected in step 1 and the corresponding supply air volume as training samples into the deep belief neural network, establish a training model of the variable air volume air conditioner supply air volume prediction model based on the deep belief neural network, conduct learning training on the training samples, and continuously adjust the parameters of the model to obtain the best training model;
[0039] Step 3: Determine the variable air volume air conditioning supply air volume prediction model based on the deep belief neural network, import the test sample into the trained prediction model, and predict the corresponding supply air volume of the test sample.
[0040] In the scheme of the present invention, the input parameters for the prediction of the air supply volume of the variable air volume air conditioner in step 1 include: the air supply volume at time t of the previous day, the air supply volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day, and the atmospheric humidity of the current day. In step 1: the mathematical relationship between each influencing factor and the air supply volume in the VAV air supply volume prediction model is as follows:
[0041] y=f(x1,x2,···,x n )i=1,2,···,n (1.1)
[0042] Where y is the air volume of VAV at time t, x i is the i-th parameter affecting the air supply volume, i.e., the input parameter for the prediction of the air supply volume of the variable air volume air conditioner in step 1, n is the number of factors affecting the air supply volume, and f is the uncertainty function relationship of DBN. Figure 1 The prediction flow chart of the air supply volume of variable air volume air conditioner based on DBN is as follows:
[0043] (1) Form a training data set S:
[0044] The actual air supply volume of VAV at time t is taken as the output variable of the model, and the air supply volume at time t of the previous day, the air supply volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day, and the atmospheric humidity of the current day are taken as the input variables of the model. The output variables and the input variables constitute the corresponding training data set S.
[0045] (2) Training DBN model:
[0046] The training data set S is imported into the DBN model, and the RBM training parameter update rule is used to perform layer-by-layer training. Unsupervised pre-training is performed first, and then supervised reverse tuning training is performed until the training error reaches the set requirement and the update is stopped. The trained optimal DBN model, that is, the deep belief neural network model, is obtained.
[0047] VAV air volume corresponding to the test data set D: Import the test samples into the trained DBN model and predict the VAV air volume y at time t.
[0048] Furthermore, field data collection was conducted on the outdoor meteorological parameters and air supply volume of a large variable air volume air conditioner of a cultural and sports center in Guangxi. The outdoor meteorological parameters and building parameters are shown in Table 1:
[0049] Table 1. Outdoor meteorological parameters and building parameters
[0050]
[0051] During the data collection process, the temperature and humidity self-recorder, black ball thermometer and HVAC electronic air volume measuring instrument FLY-1 were used to measure the outdoor meteorological parameters and VAV air volume of the Cultural and Sports Center Hall. The parameters of the measuring instruments are shown in Table 2:
[0052] Table 2. Measuring instrument models and parameters
[0053]
[0054] Taking the period from mid-June 2019 to mid-September 2019 as an example, the input parameters for the prediction of the air supply volume of the variable air volume air conditioner are collected: the three measuring instruments are set to record every 10 minutes, the recorded data are sorted, and the abnormal data are removed, and 1,000 sets of data are collected as training samples. From September 16 to 18, 2019, the three measuring instruments are set to record every hour, and 13 sets of data are collected as prediction samples. According to the measured data of VAV hourly air supply volume and outdoor hourly temperature on September 16 and 17 and the outdoor meteorological parameters on September 18, the hourly air supply volume from 8:00 to 22:00 on September 18 is predicted using the best model that has completed the training.
[0055] In the scheme of the present invention, in step 2, a variable air volume air conditioning air supply volume prediction model based on a deep belief neural network is established to learn and train the training samples, and the parameters of the model are continuously adjusted to obtain the best training model. The establishment of the training model includes the following steps:
[0056] S1: Standardize the model input parameters;
[0057] S2: perform sample classification;
[0058] S3: Initialize the weights and thresholds of the deep belief neural network;
[0059] S4: Calculate the update rules for the weights and thresholds of the deep belief neural network;
[0060] S5: Set the number of neurons in the input layer, each hidden layer and the output layer;
[0061] S6: Train the DBN model;
[0062] S7: Use the trained deep belief neural network to predict the air supply volume corresponding to the test sample.
[0063] The sample classification in step S2 is achieved in the following way: the sample data is divided into two parts: training samples and test samples, the training samples are used for learning and training, and the test samples are used for testing; the training samples are imported into the MATLAB software in the form of a matrix, and nine parameters, including the air supply volume at time t of the previous day, the air supply volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day and the atmospheric humidity of the current day, are used as input parameters, and the air supply volume is used as the output result, and a deep confidence neural network training model is established to carry out deep confidence neural network learning and training.
[0064] See also Figure 2In the step 2 of the variable air volume air conditioner air supply volume prediction method based on deep belief neural network, the present invention applies: Variable air volume air conditioner air supply volume DBN model: Deep belief neural network (DBN) is an important algorithm of deep learning, which consists of an input layer, an intermediate layer (hidden layer) and an output layer. It is a method of using unsupervised layer-by-layer training to quickly train the entire neural network layer by layer from bottom to top, so that the network has a faster training speed.
[0065] The structure of the VAV air supply volume prediction model based on the DBN algorithm is as follows: Figure 2 As shown in the figure, there are four layers in total. The first layer is the display layer, that is, the input layer, which is composed of outdoor meteorological parameters and VAV historical air supply; the second and third layers are hidden layers. The double hidden layer structure can avoid local convergence while ensuring prediction accuracy. The last layer is the output layer, which is the VAV air supply at time t in the model. DBN is composed of a certain number of restricted Boltzmann machines (RBM, Restricted BoItzmannMachine) stacked together. In the process of training RBM, DBN usually solves the activation probability of its display layer and hidden layer according to the energy function of RBM, and then obtains the update rule of RBM training parameters. RBM is a special neural network structure based on energy function. For any RBM structure, its energy function can be defined as:
[0066]
[0067] Among them, I and J are the number of units in the visible layer and hidden layer respectively, v i and h j are the i-th visible layer neuron and the j-th hidden layer neuron, respectively, i and b j Represent the bias of the i-th visible layer neuron and the j-th hidden layer neuron, respectively, and w ij is the connection v i and h j The weight between .
[0068] According to the activation probability of the hidden layer and the visible layer neurons, the connection weights and biases between the visible layer and the hidden layer neurons are recalculated, and the RBM parameters θ = {w ij ,a i ,b j}Update rule:
[0069]
[0070]
[0071]
[0072] Among them, η is the learning rate, and in the four-layer DBN structure, η is generally taken as 0.01.
[0073] During the entire DBN training process, the first RBM's visible layer generates a vector, which is then passed to the hidden layer through the RBM network. In turn, the visible layer can be reconstructed. The weights and biases between the hidden layer and the visible layer are updated according to the difference between the reconstructed layer and the visible layer until the training error reaches the set requirement and stops updating. Figure 3 As shown, the main steps are divided into two steps:
[0074] In the first step, in RBM, the parameter θ is obtained by layer-by-layer training. The output end is connected to the BP neural network to obtain the output data, completing the forward stage of the entire DBN training.
[0075] In the second step, according to the error between the actual data and the output data, the error is back-propagated from top to bottom to each layer of RBM, and each layer of RBM is fine-tuned layer by layer to make the parameter θ reach the global optimum and complete the training of the entire DBN.
[0076] See also Figure 4 , Figure 5 In the third step of the method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network, the present invention applies: using the training sample data collected in step 1, the entire DBN training is completed to obtain the optimal prediction model. Then the input parameters of the test sample are imported into the optimal prediction model to predict the corresponding air supply volume of the variable air volume air conditioner. The hourly air supply volume of the variable air volume air conditioner predicted by the deep belief neural network and the actual air supply volume are as follows: Figure 4 As shown in Figure 5 shown.
[0077] Furthermore, the performance of the deep belief neural network model in predicting VAV air flow was evaluated using the mean absolute relative error (MAPE), root mean square relative error (RMSPE) and coefficient of determination (R 2 ) to evaluate the predictive performance of the model.
[0078]
[0079]
[0080]
[0081] Among them, y k ,y p represents the actual value and the predicted value, Represents the average of the actual values and the average of the predicted values.
[0082] Generally, the smaller the MAPE value, the better the prediction performance. RMSPE is a better indicator in the prediction process. The smaller the RMSPE value, the smaller the deviation between the predicted value and the actual value, and the better the prediction performance. 2 It is an effective indicator of prediction accuracy. 2 The larger the value, the higher the linear relationship between the predicted value and the actual value. The average absolute relative error, root mean square relative error and determination coefficient of the deep confidence neural network prediction model in predicting the hourly air supply volume of the variable air volume air conditioner are calculated to be 1.555%, 0.789% and 0.9975 respectively. This shows that the deep confidence neural network prediction model has global convergence and good prediction performance, and can accurately and stably predict the hourly air supply volume of the variable air volume air conditioner.
[0083] Through the analysis of the prediction results of the variable air volume air conditioning supply air volume prediction model based on DBN, the theory of the variable air volume air conditioning supply air volume prediction model based on DBN is more advanced, and the prediction error is small after training a large amount of data. It can process massive data and has the advantages of fast convergence speed, global convergence, and stable prediction. However, in the current field of variable air volume air conditioning, the application of deep learning is relatively small. The application of the DBN model to the complex model that affects the air volume of variable air volume air conditioning has a good effect and can overcome the interference of mutual coupling between different influencing factors. Applying the theory of DBN to the field of variable air volume air conditioning can provide a more advanced intelligent algorithm for studying the air volume in complex environments, and provide greater help for the research of new energy-saving variable air volume air conditioners in the future.
[0084] Although what is considered as exemplary embodiments of the present invention has been described and described, it will be understood by those skilled in the art that various changes and substitutions may be made thereto without departing from the spirit of the present invention. In addition, appropriate modifications may be made to adapt specific circumstances to the teachings of the present invention without departing from the central concept of the present invention described herein. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but the present invention may also include all embodiments and their equivalents that fall within the scope of the present invention.
Claims
1. A prediction method for variable air volume air conditioning supply air volume based on deep belief neural network, It is characterized in that The following steps are involved: Step 1: Determine the input parameters for training the VAV air conditioning supply air volume prediction based on the deep belief neural network, and collect data to obtain corresponding sample data, including test samples and training sample data; Step 2: Use MATLAB software to import the input parameters of the variable air volume air conditioner supply air volume prediction based on the deep belief neural network collected in step 1 and the corresponding supply air volume as training samples into the deep belief neural network, establish a training model of the variable air volume air conditioner supply air volume prediction model based on the deep belief neural network, conduct learning training on the training samples, and continuously adjust the parameters of the model to obtain the best training model; Step 3: Determine a variable air volume air conditioning supply air volume prediction model based on a deep belief neural network, import the test sample into the trained prediction model, and predict the corresponding supply air volume of the test sample; The input parameters for the prediction of the variable air volume air conditioner supply air volume in step 1 include: the supply air volume at time t of the previous day, the supply air volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day, and the atmospheric humidity of the current day. The mathematical relationship between each influencing factor and the supply air volume in the VAV supply air volume prediction model is as follows: y=f(x 1 ,x 2 ,···,x n )i=1,2,···,n Where y is the air volume of VAV at time t, x i is the i-th parameter affecting the air supply volume, i.e., the input parameter for the prediction of the air supply volume of the variable air volume air conditioner in step 1, n is the number of factors affecting the air supply volume, and f is the uncertainty function relationship of DBN; The prediction of the air supply volume of variable air volume air conditioner based on DBN is calculated as follows: (1) Forming a training data set S The actual air supply volume of VAV at time t is taken as the output variable of the model, and the air supply volume at time t of the previous day, the air supply volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day, and the atmospheric humidity of the current day are taken as the input variables of the model. The output variables and the input variables constitute the corresponding training data set S; (2) Training DBN model Import the training data set S into the DBN model, and use the RBM training parameter update rule to perform layer-by-layer training. First, perform unsupervised pre-training, then perform supervised reverse tuning training until the training error reaches the set requirement and then stop updating. This will result in the best trained DBN model, i.e., the deep belief neural network model. VAV air volume corresponding to the test data set D: Import the test sample into the trained DBN model and predict the VAV air volume y at time t; In step 2, a variable air volume air conditioning air supply volume prediction model based on a deep belief neural network is established to learn and train the training samples, and the parameters of the model are continuously adjusted to obtain the best training model. The establishment of the training model includes the following steps: S1: Standardize the model input parameters; S2: perform sample classification; S3: Initialize the weights and thresholds of the deep belief neural network; S4: Calculate the update rules for the weights and thresholds of the deep belief neural network; S5: Set the number of neurons in the input layer, each hidden layer and the output layer; S6: Train the DBN model; S7: Use the trained deep belief neural network to predict the air supply volume corresponding to the test sample; The sample classification in step S2 is achieved in the following manner: the sample data is divided into two parts, a training sample and a test sample, the training sample is used for learning and training, and the test sample is used for testing; the training sample is imported into the MATLAB software in the form of a matrix, and the air supply volume at time t of the previous day, the air supply volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day, and the atmospheric humidity of the current day are taken as input parameters, and the air supply volume is taken as the output result, and a deep confidence neural network training model is established to perform deep confidence neural network learning and training; The variable air volume air conditioning air supply volume prediction model based on deep belief neural network has a total of four layers of structure; the first layer is the display layer, that is, the input layer, which is composed of outdoor meteorological parameters and the historical air supply volume of the variable air volume air conditioning; the second and third layers are hidden layers, and the last layer is the output layer, that is, the variable air volume air conditioning air supply volume at time t on the predicted day in the model; The outdoor meteorological parameters during the data collection period included outdoor average atmospheric pressure, outdoor average wind speed, outdoor calculated dry-bulb average temperature, outdoor calculated wet-bulb average humidity, and outdoor calculated daily average temperature.
2. According to claim 1, a method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network, Features: DBN is composed of a certain number of stacked restricted Boltzmann machines (RBMs); in the process of training RBMs, DBN solves the activation probabilities of its visible and hidden layers according to the energy function of RBM, and then obtains the update rules of RBM training parameters.
3. According to claim 2, a method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network, Features: The training process of the DBN model is mainly divided into two steps: In the first step, in RBM, the parameter θ is obtained by layer-by-layer training; the output end is connected to the BP neural network to obtain the output data, completing the forward stage of the entire DBN training; In the second step, according to the error between the actual data and the output data, the error is back-propagated from top to bottom to each layer of RBM, and each layer of RBM is fine-tuned layer by layer to make the parameter θ reach the global optimum and complete the training of the entire DBN.
4. According to claim 3, a method for predicting the air supply volume of a variable air volume air conditioner based on a deep belief neural network, Features: In the second step of DBN model training, the actual air volume of the variable air volume air conditioner at time t is first used as the output variable of the model, and the air volume at time t of the previous day, the air volume at time t of the previous two days, the outdoor temperature at time t of the previous day, the outdoor temperature at time t of the previous two days, the outdoor temperature at time t of the current day, the outdoor temperature at time t-1 of the current day, the outdoor temperature at time t-2 of the current day, the solar radiation temperature of the current day, and the atmospheric humidity of the current day are used as the input variables of the model. The output variables and the input variables constitute the corresponding training data set S; Then import the training data set S into the DBN model, and use the RBM training parameter update rule to perform layer-by-layer training. First, perform unsupervised pre-training, then perform supervised reverse tuning training, and stop updating when the training error reaches the set requirement, so as to obtain the trained optimal DBN model.
