JPH08210689A - System for estimating air-conditioning thermal load - Google Patents

Abstract

PURPOSE: To estimate an air-conditioning load highly accurately with a small amount of data which is easily available. CONSTITUTION: The estimate value of the air-conditioning thermal load accumulated value of next day is outputted by a method wherein informations, detected sequentially by an indoor atmosphere detecting means 28, an outdoor atmosphere detecting means 30, a time information detecting means 32 and a next day estimated weather detecting means 36, are inputted into an air-conditioning thermal load estimating means 50, taking a neural network prestudied by a studying means 52 thereinto, while a network operation is effected by the neural network. On the other hand, the amount of sunshine of next day is estimated by a photographing means 68, a picture memory means 70, a picture processing means 72, a knowledge processing means 74 and an amount of sunshine estimating means 76 to correct the estimated value of the air-conditioning thermal load accumulated value of next day, estimated by the air-conditioning thermal load estimating means 50, in accordance with the estimated amount of sunshine. A highly accurate estimation can be effected by few data in such a manner while the change of estimated value of air-conditioning thermal load accumulated value of next day due to the sudden change of weather can be coped by observing the condition of sky.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an air conditioning heat load prediction system, and more particularly to an air conditioning heat load prediction system for predicting an air conditioning load of the next day to support effective use of a heat storage type air conditioner.

[0002]

2. Description of the Related Art A heat storage type air conditioner uses inexpensive nighttime electric power to store heat in a heat storage tank at night with a heat source device such as a refrigerating machine, and stores the stored heat (cold water or hot water) in the next day. It is installed for the purpose of reducing the running cost of air conditioning by using it as a heat source for air conditioning.

[0003] In this heat storage type air conditioning equipment, a type in which the air conditioning heat load integrated value of the air-conditioned room on the next day is predicted and the amount of heat consumed by the next day is stored in the heat storage tank, and the air conditioning heat load integrated value of the next day is stored. There are two types of types that store heat as much as possible in the heat storage tank without predicting.

[0004]

However, the type of storing heat in the heat storage tank as much as possible without predicting the integrated value of the air-conditioning heat load of the next day is the amount of heat stored when the air-conditioning load is low, such as in the intermediate period. Since the heat storage tank cannot be used up, the time for keeping the heat storage tank at low temperature (cold water) or high temperature (warm water) becomes long, so that there is a disadvantage that heat loss becomes large.

On the other hand, the type that predicts the integrated value of the air-conditioning heat load on the next day is more economical than the type that does not predict the integrated value of the air-conditioning heat load on the next day, but the conventional air-conditioning heat load prediction system has a prediction accuracy. However, there are drawbacks such as low data rate, large amount of observation data, and difficult observation data. In addition, it has the drawback of not being able to cope with sudden changes in weather, which is not yet practically sufficient.

The present invention has been made in view of such a problem, and in order to support the proper operation of the heat storage type air conditioning system, it is possible to easily predict the air conditioning heat load with a small amount of data and to obtain it accurately. It is an object of the present invention to provide an air conditioning heat load prediction system capable of achieving the above and capable of coping with a sudden change in weather.

[0007]

In order to achieve the above object, the present invention relates to a heat storage type air conditioner for predicting an air conditioning heat load integrated value of a room to be air-conditioned the next day and preliminarily storing heat. Indoor environment detection means for detecting indoor environment information in the air-conditioned room,
Outside air environment detecting means for detecting outside air environment information outside the air-conditioned room, time information detecting means for detecting time information such as time and day of the week, and next day expected weather detecting means for collecting or detecting expected weather information for the next day, Thermal load detection means for detecting the actual value of the integrated air-conditioning heat load value of the air-conditioned room, detection data storage means for accumulating the data detected by each detection means, and each stored in the detection data storage means. When information is input, the neural network is learned in advance so as to output the predicted value of the air conditioning heat load integrated value of the room to be air-conditioned the next day, and the neural network is updated in consideration of the latest information from the detection means. Means, and the neural network extracted from the learning means, the indoor environment detecting means, the outside air environment detecting means, the time information detecting means, and the next day forecast Characterized in that it consists, and the air conditioning heat load prediction unit for outputting a prediction value of the next day of the air conditioning heat load integrated value of the air conditioned room by performing a network operation by sequentially inputting each information from the detection means.

[0008]

According to the present invention, the neural network in which the learning means preliminarily learns the information successively detected by the indoor environment detecting means, the outside air environment detecting means, the time information detecting means and the next-day expected weather detecting means is incorporated. The predicted value of the air-conditioning heat load integrated value of the next day is output by inputting to the air-conditioning heat load predicting means and performing a network operation by the neural network. In this way, the relationship between the indoor environment, the outside air environment, the time information and the expected next day weather required for air conditioning heat load prediction and the air conditioning heat load of the next day in the air-conditioned room is learned, and the latest information of each detection means is added. By using the updated neural network, it is possible to easily predict the air-conditioning heat load with a small amount of data that is easily available. In addition, since the neural network learns based on the information from each of the detection means, it is possible to simulate a new air-conditioned room that has no data in the past, so it is possible to easily and accurately predict the air-conditioning heat load. It can be carried out.

Further, the photographing means, the image storing means, the image processing means, the knowledge processing means, and the sunshine quantity predicting means predict the sunshine quantity of the next day, and the air conditioning heat load of the next day predicted by the air conditioning heat load predicting means. Since the predicted value of the integrated value is corrected according to the predicted sunshine amount, it is possible to make more accurate predictions, and by observing the state of the sky, the air conditioning heat of the next day due to a sudden change in the weather can be obtained. It is also possible to cope with changes in the predicted value of the load integrated value.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of an air conditioning heat load prediction system according to the present invention will be described in detail below with reference to the accompanying drawings. The present invention is an air conditioning heat load prediction system in a heat storage type air conditioning facility that predicts an air conditioning heat load integrated value of an air-conditioned room on the next day and stores heat in advance. FIG. 1 shows an air conditioning heat load prediction system according to the present invention. The schematic block diagram of the embodiment of FIG. As shown in the figure, the air conditioning heat load prediction system 10 mainly includes a detection unit 12, a detection data storage unit 14, a prediction unit 16, a first correction unit 18, and a second correction unit 20. Control target determining means 22 based on the integrated value of the air conditioning heat load of the room to be air-conditioned, which is predicted by the prediction system.
The operating condition which is the control target of the heat source device 26 is determined by the control means 24, and the control means 24 operates the heat source device 26 according to the control target.

Next, the detailed configuration of the air conditioning heat load prediction system will be described with reference to FIG. The detection means 12 is composed of an indoor environment detection means 28, an outside air environment detection means 30, a time information detection means 32, a heat load detection means 34 and a next day forecast weather detection means 36. Then, the indoor environment detecting means 28 detects the room temperature and humidity of the air-conditioned room, and outputs the detection data to the detected data storage means 14 and the air conditioning heat load predicting means 50 of the predicting means 16 described later.

The outside air environment detecting means 30 detects the integrated value of the outside air temperature, the outside air humidity and the amount of solar radiation per day, and outputs the detected data to the detected data storage means 14 and the air conditioning heat load predicting means 50 of the predicting means 16. To do. The time information detection means 32 detects the time and day of the week and outputs the detection data to the detected data storage means 14 and the air conditioning heat load prediction means 50 of the prediction means 16.

The heat load detecting means 34 detects the actual value of the integrated value of the air conditioning heat load of the air-conditioned room and outputs the detection data to the detection data storage means 14. The next-day expected weather detection means 36 collects the expected maximum temperature of the next day, the expected minimum temperature of the next day, the expected maximum humidity of the next day, and the expected minimum humidity of the next day announced by the Meteorological Agency, and the collected data is detected data storage means 14 And the air conditioning heat load prediction means 50 of the prediction means 16
Output to. It should be noted that the present invention is not limited to the collection of data from the Meteorological Agency, and a means for independently detecting the weather information of the next day may be provided.

The detection data storage means 14 comprises an indoor environment data storage means 38, an outside air environment storage means 40, a time information storage means 42, a heat load storage means 44 and a next day forecast weather storage means 46, and each storage means 38. , 40, 4
Reference numerals 2, 44 and 46 store the detection data detected by the corresponding detection means, and the detection data storage means 14
The detection data stored in is input to the learning means 52, which will be described later, of the prediction means 16, and is used as data when learning the neural network.

The predicting means 16 comprises a learning means 52 and an air conditioning heat load predicting means 50. When the information stored in the detection data storage means 14 is input, the learning means 52 preliminarily learns the neural network so as to output the predicted value of the air-conditioning heat load integrated value of the next day, and at the same time, outputs from the detection data storage means 14. The neural network can be updated by adding the latest information, and the updated latest neural network can be stored.

Here, the neural network that controls the functions of the learning means 52 and the air conditioning heat load prediction means 50 will be described. FIG. 3 shows a functional explanatory diagram of the neural network, and FIG. 4 shows the configuration of the neural network stored in the learning means 52. In the neural network, when performing network operation (forward operation), input values are x ₁ , x ₂ , ...
, X _n , internal coefficients (weights) w ₁ , w ₂ , ..., w _n
Then, the internal output value X of the accumulation processing unit 54 is The output value Y from the neuro unit 56 is: Y = 1 / (1 + exp (-AX)) where A is an arbitrary constant.

A plurality of the neuro units 56 are connected in a plurality of layers to form a neural network, and the input / output relationships of the connected neuro units 56 are as described above.
When modifying the neural network (backward calculation), the internal coefficient (weight) is used to improve the accuracy by using the back-propagation method based on the prediction accuracy of the prediction value.
To adjust.

Further, the air conditioning heat load predicting means 50 takes out the latest neural network updated by the learning means 52, and the detecting means 12 is put in the neural network.
At 21:00 (1 hour before the start of midnight power operation), outside temperature, outside humidity, room temperature, solar radiation per day, next day of the week, expected maximum temperature of the next day, expected minimum temperature of the next day,
The predicted maximum humidity of the next day and the predicted minimum humidity of the next day are input to calculate the predicted value of the integrated value of the air conditioning heat load for the next day.

Further, the learning unit 52 determines whether to newly learn the stored neural network according to the prediction accuracy of the predicted value of the integrated value of the air conditioning heat load of the next day calculated by the air conditioning heat load prediction unit 50. . When it is determined that the learning is better, the back propagation method is used to form the input layer 58, the intermediate layer 60, and the output layer 62 based on the detection data stored in the detection data storage unit 9. The internal coefficient (weight) of the input / output neural network (see FIG. 4) is corrected.

Further, as shown in FIG. 5, the construction of the neural network when the information to be learned is created by the air conditioning heat load simulation is 21 as input data.
Input the outdoor temperature, outdoor humidity, indoor temperature, amount of solar radiation per day, and the day of the next day, and simulate a prediction model for the room to be air-conditioned, and output the predicted value of the integrated air-conditioning heat load for the next day. To do. If there is no information for learning, a dynamic heat load calculation program, for example, MIC
Information to be learned can be created and learned using RO-HAPS / 1982 or the like. In this way, the detection data storage means 1 is used no matter what learning method is used.
It is possible to newly learn by each information stored in 4.

Next, the first correction means 18 will be described. The first correction means 18 independently predicts the integrated value of the air conditioning heat load of the next day based on the indoor environment information, the outside air environment information and the time information from the detecting means 12 without being influenced by the prediction of the air conditioning heat load predicting means 50. The air-conditioning heat load predicting means 50 corrects the predicted value, and a neural network is mechanically used similarly to the air-conditioning heat load predicting means 50. Then, for example, the air conditioning heat load predicting means 50 predicts the air conditioning heat load integrated value of the next day every hour after 21:00 until 8 o'clock of the next day (time when air conditioning is started). By providing the first correction means 18, for example, after the air conditioning heat load prediction means 50 predicts the air conditioning heat load integrated value at 21:00,
It can easily cope with a sudden change in the outside air environment.

Next, the second correction means 20 will be described. The second correction means 20 corrects the predicted value of the air conditioning load integrated value of the next day predicted by the air conditioning heat load prediction means 50 by predicting the sunshine amount of the next day. That is,
The second correction unit 20 includes a photographing unit 68 (for example, a video camera) that photographs the state of the sky as a sunshine amount element that influences the sunshine amount of the next day, and an image storage unit 70 that stores the image photographed by the photographing unit 68. An image processing unit 72 that takes out an image from the image storage unit 70 and processes the image so as to leave only an image related to the amount of sunlight, and an image pattern processed by the image processing unit 72 using a neural network having a learning function. And the amount of sunshine on the next day, the knowledge processing means 74 and the neural network extracted from the knowledge processing means 74 are sequentially input with the images from the image processing means 72 to perform network operation to predict the next day's sunshine. And a sunshine amount predicting means 76 for outputting the amount. In this case, the neural network of the knowledge processing means 74 may be updated with the latest data from the image processing means 72. Then, as the sunshine amount element photographed by the photographing means 68 of the second correction means 20, for example, the light of sunset and the luminosity of the sun are photographed in the evening, and the photographed image is processed by the image processing means 72. The image-processed image data is input to the sunshine amount predicting means 76, and the sunshine amount predicting means 76 outputs the next day's predicted sunshine amount using a neural network that has been learned in advance by the knowledge processing means 74. Similarly, at night, the celestial body is photographed by the photographing means 68, and the photographed image is processed by the image processing means 72. By this image processing, only the image of the moving speed of the celestial body, that is, only the celestial body is extracted, and, for example, high-speed light such as a headlight of a car is erased. Then, the image-processed image data, the past data stored in the image storage means 70, and the astronomical information at the time of photographing are input to the sunshine amount predicting means 76, and the sunshine amount predicting means 76 causes the knowledge processing means 7 to operate.
In step 4, the predicted sunshine amount for the next day is output using the neural network that has been learned in advance. In the daytime, the photographing means 68
To capture the blueness of the sky and the amount of clouds, and predict the amount of sunshine in the same way. As described above, by providing the second correcting means 20, the prediction accuracy of the predicted value of the air-conditioning heat load integrated value of the next day can be improved, and by observing the appearance of the sky, the accuracy is higher than that of the first correcting means 18. Since it is possible to grasp a sudden change in the weather, it is possible to respond to the change in the integrated value of the air conditioning heat load of the next day due to the sudden change in the weather with extremely high accuracy.

The control target determining means 22 includes the integrated value of the air conditioning heat load for the next day predicted by the predicting means 16 at 21:00, and the integrated value of the air conditioning heat load for the next day predicted by the first correcting means 18. And the control target of the air conditioner 26 is determined based on the predicted sunshine amount of the next day predicted by the second correction means 20. The control means 24 controls the control target determination means 2
An operation signal is sent to the air conditioner 26 based on the control target (operating condition) of the heat source device 26 determined in 2 to control the operation of the heat source device 26.

The heat source device 26 stores heat in the heat storage tank based on the operation signal sent from the control means 24. The operation of the embodiment of the air conditioning heat load prediction system 10 according to the present invention configured as described above is as follows. The neural network of the learning means 52 is learned in advance. That is, the detection unit 12 detects the following information. The indoor environment detecting means 28 detects the room temperature of the air-conditioned room,
In the outside air environment detecting means 30, the outside air temperature, the outside air humidity, and 1
The solar radiation amount of the day is detected, and the time information detecting means 32 detects the time and the day of the week of the next day. Further, the heat load detecting means 34 detects the actual value of the integrated air-conditioning heat load value of the air conditioning load for one day, and the next day forecast weather detecting means 36 announces the expected maximum temperature of the next day and the minimum temperature of the next day announced by the Meteorological Agency. , The maximum humidity of the next day and the minimum humidity of the next day are obtained,
It is stored in the storage means 14. Then, these data are input to the neural network of the learning means 52 of the prediction means 16 and learned so as to output the integrated value of the air conditioning heat load of the next day.

The prediction of the air-conditioning heat load integrated value of the next day is performed at 21:00, one hour before the start of the midnight power operation. That is, when the time information detecting means 32 detects 21:00, the detecting means 12 detects the following information and outputs the detected data to the predicting means 16. Indoor environment detecting means 2
At 8, the room temperature of the air-conditioned room is detected, and at the outside air environment detecting means 30, the integrated value of the outside air temperature, the outside air humidity, and the daily solar radiation amount is detected. Further, the time information detecting means 32
At, the next day of the week is detected, and the next-day expected weather detection means 36 acquires the data of the expected maximum temperature of the next day, the minimum temperature of the next day, the maximum humidity of the next day, and the minimum humidity of the next day announced by the Meteorological Agency.

The detection data detected by the detecting means 12 at 21:00 is input to the air conditioning heat load predicting means 50,
The air conditioning heat load predicting means 50 performs network calculation using the learned neural network stored in the learning means 52 and outputs the predicted value of the air conditioning heat load integrated value of the next day. In addition, the first correction means independently detects the detection means 12
The detection data (indoor environment information,
Based on the outside air environment information and time information), the predicted value of the air-conditioning heat load integrated value of the next day is output every hour until 8:00 of the next day.

Further, the second correction means 20 includes a photographing means 68.
The image of the sky in the west direction is photographed, and the image photographed by the photographing means 68 is stored in the image storage means 70. The image stored in the image storage means 70 is image-processed by the image processing means 72 and then input to the sunshine amount predicting means 76. The sunshine amount predicting means 76 is stored in the knowledge processing means 74 and has been learned. Predict the amount of sunshine on the next day using a neural network.

In the control target determining means 22, the air conditioning heat load integrated value predicted by the air conditioning heat load predicting means 50 at 21:00 is used as a base, and the air conditioning heat load integrated value of the next day predicted by the first correcting means, The control target of the heat source device 26 is determined by correcting the air-conditioning heat load integrated value of the base from the predicted sunshine amount of the next day predicted by the second correction means. Then, based on the control target determined by the control target determining unit 22, the control unit 24 sends an operation signal to the heat source device 26 to control the operation of the heat source device 26.

In this way, the relationship between the indoor environment, the outdoor air environment, the time information and the expected next day weather required for the air conditioning heat load prediction, and the air conditioning heat load of the next day in the air-conditioned room is learned, and the latest information of each detection means is learned. By using the neural network updated by taking into account, it is possible to easily predict the air conditioning heat load with a small amount of data. Further, since the neural network performs learning based on the information from each detecting means, it is possible to perform an air conditioning heat load simulation according to the air-conditioned room even for a new air-conditioned room that has no data in the past. As a result, the air conditioning heat load prediction system of the present invention can be easily introduced into a new air-conditioned room.

The first correction means 18 predicts the integrated value of the air conditioning load of the next day every hour from 22:00 to 8:00, and the air conditioning heat of the next day predicted by the air conditioning load predicting means 16 at 21:00. Since the integrated load value is corrected, a highly accurate predicted value can be obtained, and a sudden change in the outside air temperature after 21:00 can be dealt with. Further, the second correction means 20 predicts the sunshine amount on the next day, and corrects the air conditioning heat load integrated value predicted by the air conditioning heat load prediction means 50 on the next day according to the predicted sunshine amount. It is possible to make highly accurate predictions, and by observing the appearance of the sky, it is possible to respond to changes in the predicted value of the air conditioning heat load integrated value on the next day due to a sudden change in the weather.

Therefore, according to the air-conditioning system 10 of this embodiment, it is possible to predict the air-conditioning heat load with high accuracy and to cope with a sudden change in the weather with a small amount of data that is easily available. In this embodiment, the room temperature is used as input data to the simulated neural network, but a predicted value of the internal load level of the next day may be input instead of the room temperature. Although the outside air temperature is used as the input data of the neural network, the deviation of the outside air temperature may be input instead of the outside air temperature.

[0032]

As described above, according to the air conditioning heat load predicting system of the present invention, the air conditioning heat load integrated value of the next day can be accurately predicted with a small amount of data that is easily available, and the weather condition can be calculated rapidly. With such changes, it is possible to easily cope with changes in the predicted value of the air conditioning heat load integrated value of the next day. Further, since the air conditioning heat load prediction system of the present invention can learn the neural network of the learning means by the detected data, it can be easily introduced into a new air-conditioned room having no data.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of an air conditioning heat load prediction system according to the present invention.

FIG. 2 is a detailed configuration diagram of an embodiment of an air conditioning heat load prediction system according to the present invention.

FIG. 3 is a functional explanatory diagram of a neural network.

FIG. 4 is a configuration diagram of a neural network used in this embodiment.

FIG. 5 is a configuration diagram of a neural network when learning the neural network using simulation.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Air-conditioning system 12 ... Detection means 14 ... Detection data storage means 16 ... Prediction means 18 ... 1st correction means 20 ... 2nd correction means 22 ... Control target determination means 24 ... Control means 26 ... Heat source equipment 28 ... Indoor environment detection means 30 ... Outside air environment detecting means 32 ... Time information detecting means 34 ... Heat load detecting means 36 ... Next day expected weather detecting means 50 ... Air conditioning heat load predicting means 52 ... Learning means 68 ... Imaging means 70 ... Image storing means 72 ... Image processing means 74 ... Knowledge processing means 76 ... Sunlight amount prediction means

Claims (3)

[Claims] 1. A heat storage type air conditioner for predicting an air conditioning heat load integrated value of an air-conditioned room on the next day to store heat in advance, and an indoor environment detecting means for detecting indoor environment information of the air-conditioned room, An outside air environment detecting means for detecting outside air environment information outside the air-conditioned room, a time information detecting means for detecting time information such as time and day of the week, a next day forecast weather detecting means for collecting or detecting forecast weather information for the next day, and Thermal load detection means for detecting the actual value of the integrated air-conditioning heat load value of the air-conditioned room, detection data storage means for accumulating the data detected by the detection means, and information stored in the detection data storage means Input, the neural network is learned in advance so as to output the predicted value of the air-conditioning heat load integrated value of the room to be air-conditioned the next day, and the neural network is added taking into account the latest information from the detection means. A learning means for updating the network, and the neural network extracted from the learning means, sequentially inputting each information from the indoor environment detecting means, the outside air environment detecting means, the time information detecting means, and the next-day forecast weather detecting means to the network. An air conditioning heat load prediction system comprising: an air conditioning heat load predicting unit that outputs a predicted value of an air conditioning heat load integrated value of the room to be air-conditioned the next day by performing calculation. 2. A photographing means for photographing the state of the sky as a sunshine amount element that influences the sunshine amount of the next day, an image storing means for storing the image photographed by the photographing means, and an image taken out from the image storing means. Image processing means for performing image processing so as to leave only an image related to the amount of sunshine, and a relationship between the image pattern processed by the image processing means using a neural network having a learning function and the amount of sunshine on the next day in advance. The learned knowledge processing means, and the sunshine amount predicting means for outputting the predicted sunshine amount for the next day by sequentially inputting the images from the image processing means to the neural network extracted from the knowledge processing means and performing the network operation, According to the predicted sunshine amount predicted by the sunshine amount prediction means, the predicted value of the air conditioning heat load integrated value for the next day predicted by the air conditioning heat load prediction means Air conditioning heat load prediction system of claim 1, wherein the correcting. 3. The sunshine intensity factor is the amount of blue and clouds in the sky during the day, the light of sunset and the intensity of sunlight in the evening,
The air conditioning heat load prediction system according to claim 2, wherein the amount of light of a star is at night.