JP3284280B2 - Existing case effective use type inference device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an existing case effective use type inference apparatus for generating an input / output relation model of a system from past case data and inferring the conclusion of a new case.

[0002]

2. Description of the Related Art In a case where a conventional inference apparatus is applied to, for example, an air-conditioning load prediction system or the like, and a heat load several hours later is inferred from input data such as a current outside air temperature, a heat load, and a discomfort index. Is performed as follows. That is, the air-conditioning load prediction system collects data such as the outside air temperature, heat load, and discomfort index for the past one year, for example, and from these collected data, the outside air temperature, heat load, and unpleasantness at a certain time. An input / output relation model that shows the relationship between input data such as indices and output data that is heat load several hours later is generated, and when new data such as outside temperature, heat load, and discomfort index are input, The heat load after α hours is deduced based on this input / output relation model.

[0003]

As described above, in an apparatus that generates an input / output relation model from collected data and performs inference in accordance with the generated input / output relation model, accurate modeling is required. , The data collection period becomes longer. In particular, when modeling taking into account annual fluctuations as in the example applied to the air conditioning load prediction system above, the data collection period takes one year or more. In the case of a new system, During that time, there was a problem that prediction and control using the input / output relation model could not be performed. Therefore, an object of the present invention is to perform accurate prediction and control using an input / output relation model within a short period of time even in the case of a new system.

[0004]

SUMMARY OF THE INVENTION In order to solve such a problem, the present invention provides a second system which is an old system.
New system using data and input / output relation model
Input and for generating an input-output relationship model of a first system
An inference apparatus comprising a force relation model generation means and performing inference based on the generated input / output relation model,
The input / output relation model generating means includes: a model generating means for generating a model indicating a relation between a difference between an internal state and an output difference between the first and second systems in the same external state;
When new data is input to the system of
Extracting means for extracting an input / output relation model of a second system having the same or similar external state as the system, and a difference between internal states between the first and second systems with respect to the same or similar external state. And a first calculation unit based on the model generated by the model generation unit.
A second calculating means for calculating an output difference with respect to an internal state difference between the second systems; and an output value of an input / output relation model of a second system having the same or similar external state as a second value. Correction means for correcting with the value calculated by the calculation means.

[0005]

When the external conditions such as the outside air temperature and the discomfort index are the same between the first and second systems, the difference between the internal conditions such as the heat load between the first and second systems. A model is generated that shows the relationship between the output and the difference in power, which is the thermal load several hours later. When the new data to the first system is inputted, input data and the input-output relationship model its second system having a first identical or similar external conditions and system is extracted, the same or An internal state difference between the first and second systems for a similar external state is calculated, and an output for the internal state difference between the first and second systems is generated based on the generated model. Is calculated, the output value of the input / output relation model of the second system having the same or similar external state is corrected by the calculated value, and the corrected output value is compared with the output value of the first system. Become.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. FIG. 2 is a block diagram showing an example of an air conditioning load prediction system for district cooling and heating to which the present invention is applied, and the system is configured as follows. That is, in FIG. 2, reference numeral 1 denotes a storage device for storing case data such as the outside air temperature observed in the past, for example, at certain time intervals, and discomfort index at this time, and 2 denotes a data classification for classifying the case data for each month. Part 10 creates a recognition model from monthly case data,
Based on this recognition model, when new data such as outside air temperature is actually input, a state recognition automatic generation unit that recognizes the attribution to each month and performs load prediction, 20 And a reasoning unit that makes inferences based on this causal relationship model when new data is input and calculates its credibility, 30 is a load predicted value display unit, and 40 is credibility integration. Means 50 is a credibility display unit.

Here, the state recognition automatic generation unit 10 includes a recognition model generation unit 11, a recognition model learning unit 12, a recognition model storage unit 13, a recognition model situation input unit 14, an attribute recognition unit 15, and a final predicted value determination unit. 16. The recognition model generation means 11 extracts factors for recognition from past case data to generate a recognition model based on each factor, and the recognition model learning means 12 outputs the recognition model when new case data such as outside air temperature is input. Learning is performed to update the above recognition model, and these recognition models are stored in the recognition model storage unit 13. On the other hand, attribute recognition means 1
5 receives the new case data from the recognition model situation input unit 14, recognizes the attribution to each factor based on the recognition model in the storage unit 13, and sends the recognition result to the final predicted value determination unit 16. And the final predicted value determination means 16
Is to integrate the recognition result and the inference result of the inference unit 20 and output it as a final load predicted value.

The inference unit 20 includes a causality model generation unit 21, a causality model learning unit 22, a causality model storage unit 23, a causality model situation input unit 24, an inference unit 25, and a credibility calculation unit 26. Have been. The causal relationship model generating means 21 generates a model of a causal relation between past input data and a result several hours after the input of the past case data, for example, room temperature, and the causal relation model learning means 22 outputs When input, it is learned to update the causal relation model and the like, and these causal relation models are stored in the causal relation model storage unit 23. On the other hand, when the new case data is input from the causal relationship model situation input unit 24, the inference unit 25 makes an inference based on the causal relationship model in the storage unit 23 for this new case data, and determines the inference result as the final predicted value determination value. It is sent to the means 16 and integrated with the recognition result by the automatic state recognition generation device 10, and the predicted load value is displayed on the predicted load value display section 30. Further, the inference unit 25 sends the inference result to the credibility calculation unit 26 to cause the credibility calculation unit 26 to perform credibility calculation, and causes the credibility display unit 50 to display the result. When inferring new case data, the inference means 25 makes an inference based on the topology. The phase refers to a structure given to a set so that the concept of continuity can be defined, and indicates, for example, a distance between similar new case data and past case data, similarity between similar cases, and the like.

Here, the inference unit 20 is hard-wired as shown in FIG.
As shown in FIG. 5, an offline device 61 for inputting existing case data (past case data), an online input device 62 for inputting new case data, an algorithm storage unit 63, and a predetermined algorithm stored in the algorithm storage unit 63 It comprises a CPU 64 for performing an inference operation, a data storage section 65 for storing various data, and an output device 66, and is functionally constituted by each means shown in FIG.

That is, the above-described causal relation model generating means 2
1, a memory base creating unit 21a, a parameter determining unit 21b, the causal relationship model learning unit 22,
Similarity determining means 25a as inference means 25, s Nikoto example search unit 25b, importance degree determination unit 25c, case integration means 25d, the above-mentioned authenticity calculation unit 26, and an output unit 2
7.

As described above, the inference unit 20 handles the causal relationship between the input factor of past case data and the result several hours later, for example, room temperature, as described above.
X = <x1, x2,..., Xn> as an input space,
Further, when Y = <y> is assumed as the output space, events X1 (t), X2 (t),.
., Xn (t) produce output Y (t + α) after α hours, ie, {X1 (t), X2
(T),..., Xn (t), Y (t + α)｝ (t =
1,..., N), and infers data in which each input / output variable changes continuously.

The inference unit 20 first creates a case base for past case data by the causality model 21 before inference is performed. That is, an input space is first discretized and divided into a finite number of input events, and one case is generated by integrating input / output data belonging to the same input phenomenon. At this time, the condition part of the case, that is, the data such as the outside air temperature and the discomfort index,
2,..., Xn}, and the conclusion of the case, that is, the room temperature after α hours, is the centroid value Y of the output data, the number n of times the same input event has occurred, and the centroid value ΔY / Δ of its partial differential value.
X1,..., ΔY / ΔXn, that is, ｛Y, n, ΔY / Δ
X1,..., ΔY / ΔXn}.

For example, as input data, the outside air temperature X1
(° C.) and the discomfort index X2 (%) are observed, and the maximum value (max) and the minimum value (min) of these data during each time t (t = 1,..., N) in a certain period are obtained. each, X1 (max) = 30.0, X1 (min) = 20.0 X2 (max) = 80.0, X2 (min) = 70.0 , and the and the data at time t = t ₁ X1 (t ₁ ) = 25.6, X2 (t ₁ ) = 78.7, and the discretization number for discretizing this input space is “10” (division between the maximum value and the minimum value is divided into 10). Then, the discretized data is, for example, X1 = 6, X2 =
9 and is symbolized by one event {6,9}. Here, the room temperature Y (t at time t = t ₁ + α)
_{If (1} + α) is 25.0, a causal relationship of {6, 9} → 25.0 is obtained.

If the event at time t = t ₁ +1 is X1 (t ₁ ) = 25.8, X2 (t ₁ ) = 78.79 and room temperature Y (t ₁ + α + 1) = 25.5, The input events belong to the same {6, 9}, and {6, 9} → 25.25 [= {25.0 + 25.5} /
2, where n = 2] and the case data is compressed. Further, each partial differential value described above is an amount of change in output with respect to the amount of change in each input variable. In this case, since each input / output variable is continuous data, this partial differential value ΔY / ΔXi
(T) can be calculated by equation (1).

[0015]

(Equation 1)

After a case base is created for a past case (existing case), inference for a new case is performed by the inference means 25 next. First, the condition part of the new case is {Xi ^* } (i = 1, 2,..., N), and the existing case is {Xi, Y, n, ΔY / ΔXi} (i = 1, 2,
..., n). Here, the condition part of the new case is discretized and symbolized as described above at the same time as the input.

Here, when inferring a new case, first, the similarity of the existing case to the new case is determined by the similarity determining means 25a.
(This similarity corresponds to the concept of a neighborhood system in phase). The similarity of the existing case to the new case is determined by the following definition. That is, the similarity 0 is | Xi ^* −Xi | = 0 (i = 1,
2,..., N) The similarity 1 is | Xi ^* −Xi | ≦ q _Xi (i = 1,
2,..., N) Similarity 2 is | Xi ^* −Xi | ≦ q _Xi +1 (i = 1,
2,..., N) The similarity 3 is | Xi ^* −Xi | ≦ q _Xi +2 (i = 1,
2,..., N). Here, q _Xi is called a threshold,
It is a digit value determined from variance of Xi (condition part of existing case) with respect to allowable accuracy of Y (conclusion part of existing case) from existing case data.

As an example, when X1 is the outside air temperature, X2 is the discomfort index, and the room temperature after Y = α hours is considered, when an error between the room temperature Y and the predicted value is attempted within 2 degrees, an input is made. Discretization is performed so that one digit becomes twice as in the case of space. If the discretization number is, for example, "20", Y ^* (conclusion part of a new case) belonging to the same digit for each of digits 1 to 20 is collected from existing case data, clustered, and belong to the same class. The variance of the conditional parts X1 and X2 of the new case is obtained. The variance of each cluster of the condition parts X1 and X2 of the new case belonging to digit i of Y (conclusion part of the existing case) is calculated as a discrete value, and averaged by the number of clusters to obtain digit values qi _X1 and qi _X2 . . Digit values qi _X1 and qi _X2 are averaged according to the following equation. That is, q _X1 = Σqi _X1 / 20, q _X2 = Σqi _X2 / 20

Here, the closest digit value is q _X1
= 2, q _X2 = 3 (actually, q _X1 = 2.123...,
q _X2 = 3.456...). However, if Y needs to be within 1 degree, q _X1 = 1 and q _X2 = 2, and the digit value q _Xi differs depending on the required precision. That is, the distance between the existing case and new case for each variable Xi is the phase of the condition part is smaller than q _Xi is closer, the conclusion part of the new case at that time is required for the conclusion of the existing case Considered to be within precision.

Next, a similar case to the new case is searched by the similar case search means 25b. Existing cases of the optimal case are extracted as similar cases in the order of similarity to the new case. As the optimum number of cases, for example, the number of cases that makes the best inference from a simulation based on existing cases is selected.
If there are more existing cases with the same similarity than the number of optimal cases, the degree of influence of each variable Xi on Y, that is, the correlation coefficient R
_The priority is set for each variable according to the magnitude of _Xi and extracted.

Further, the importance determining means 25c determines the importance of a similar case with respect to the new case. Here, a distance is defined in the input space, and the phase between cases is considered. Here, as an example, a distance L as shown in Expression (2) is introduced.

[0022]

(Equation 2)

Here, Φi is the weight of the distance in the variable Xi. Then, the importance Wj at the time of inference of the extracted m similar cases is defined using Expression (3). That is,

[0024]

(Equation 3)

Using m similar cases extracted in this way, the optimum case number m is used to create a new case Xi ^* (i = 1, 2,...,
The inference value Y ^* for n) is calculated and integrated using equation (4) (case integration means 25d).

[0026]

(Equation 4)

Where Lij is the distance of the i-th case from the input data on the j-input variable axis, yi is the conclusion value of the i-th similar existing case, and ΔY / ΔXj is j of the i-th similar existing case. The percentage of the change that the second change gives to the conclusion value is shown.

Next, the credibility of the inference result is determined based on the calculation result of the credibility calculating means 26. That is, the credibility of the inference result is determined using the similarity of the similar case used in the inference to the new case. For example, if the similarity used in the inference is similar to the new case, the highest similarity is defined as the credibility of the inference result.
The inference result with the highest similarity “1” is determined to have credibility “1”. In this case, the credibility “0” has the highest credibility, and the credibility decreases as the number increases. The output means 27 can output the obtained inference result and its credibility, and can also output the similar case used for the inference.

As described above, the air-conditioning load prediction system performs input / output of input data such as outside air temperature and discomfort index based on past case data such as room temperature and heat load which are output data α hours later. Relational model (causal relational model)
Is generated, and when data such as the outside air temperature and the discomfort index is newly input, the room temperature and the heat load after α hours are predicted using this input / output relation model. However, in order to perform accurate load prediction in such an air conditioning load prediction system, the data collection period is one.
In the case of a new system that requires more than one year, prediction and control using the input / output relation model cannot be performed during that time.
Therefore, the existing case effective use type inference apparatus of the present invention enables accurate prediction and control using the input / output relation model within a short period of time even in the case of a new system. That is, in the new system, inference can be performed in a short period of time by effectively using the data of the old system.

FIG. 1 is a functional block diagram of the inference apparatus for inferring an output value of a new system using data of an old system. In the figure, A is the old system, B is the new system, 1A is a storage device that stores past case data of system A, 1B is a storage device that stores case data of system B for about the past 10 days, and 21A is It is a causal relation model generating means for generating an input / output relation model, that is, a causal relation model, from the past case data of the system A. Also 10
1 is an input / output relation extracting means for extracting an input / output relation model of the system A in the same state as the external state when data indicating a new external state such as an outside air temperature and a discomfort index is input to the system B; 102 is an internal state difference calculating means for calculating a difference between internal states such as room temperature and heat load in the systems A and B;
103 model generation unit 100 system A based on the generated model, the output difference calculating means for calculating a difference between the output value of the room temperature and heat load, etc. after for example α times in B, 1
Reference numeral 04 denotes an inference value determination unit that corrects the output value of the system A extracted by the input / output relationship extraction unit 101 based on the input / output relationship model using the output value of the output difference calculation unit 103 to obtain the output value of the system B.

That is, in the present inference apparatus, when the outside temperature, the discomfort index, the change in the outside temperature, the change in the heat load, and the heat load are present as the input variables, the input variables are regarded as the external states (external factors). Temperature, discomfort index, change of outside temperature, change of heat load of input variables as internal state (internal factor),
If a relational model is created between the heat load and the difference between the internal states of the systems A and B in the same external state and the difference between the heat loads after α hours, which is an output variable, the system A is recorded. An input / output relation model based on data can be effectively used.

Hereinafter, the operation of the present apparatus will be specifically described.
Each time new case data for system B is entered,
Example system A with the same external state as the external state of the data
Between the new case data of System B and the new case data of System B
The relation model of the output difference to the state difference is a model generation method.
Created by step 100. For example, in the system A, the outside air temperature at a certain time is 25 degrees, the discomfort index is 75%, the change amount of the outside air temperature is 1.0 degrees, and the change amount of the heat load is 1.5 GCa.
1 (giga calories), the heat load is 20 GCal, and the heat load after α hours (for example, 4 hours) is 25 GCal
Assume that On the other hand, in the system B, the outside air temperature at a certain time is 25 degrees, the discomfort index is 75%, the change amount of the outside air temperature is 1.0 degrees, the change amount of the heat load is 1.0 GCal, and the heat load is 15 GCal. Heat load after 4 hours is 20GC
Let it be al. At this time, the same external condition, that is, the outside air temperature of 25 degrees, the discomfort index of 75%, and the change amount of the outside air temperature of 1.0
The difference of the internal state at the time, ie, the difference ΔX
1 (1.5-1.0 = 0.5 GCal) and difference in heat load
From the difference ΔY (25−20 = 5 GCal) of the heat load after 4 hours with respect to ΔX 2 (20−15 = 5 GCal), ｛Δ
X1, ΔX2｝ → ΔY is used as the relational model of the difference between systems A and B.
If it is described as Dell, then in this case $ 0.5,
5｝ → 5 is created by the model generation means 100.

Here, the external state changes with respect to the year, but since the system follows the external state with a certain cycle, the difference between the internal states is several cycles of the system (in one cycle, the internal state is different). (It does not cover the difference space), a model of the difference between the systems A and B can be generated. In the example of such a load prediction system, the load pattern changes in a one-day cycle. Therefore, when a model of the internal state difference between the systems A and B is generated, the data collection period of the new system B is collected. Is about 10 days.

Next, in the new system B, the outside air temperature is 27 degrees, the discomfort index is 80%, the outside air temperature changes 0.5 degrees, the heat load changes 1.6 GCal, and the heat load 23.
It is assumed that new data called GCal is obtained. In this case, the input / output relation extracting unit 101 extracts case data similar to the external state, that is, the outside air temperature of 27 degrees, the discomfort index of 80%, and the change amount of the outside air temperature of 0.5 degrees from the case model of the system A. It should be noted that the input / output relation extracting means 101 performs the processing in the aforementioned paragraph 0.
017 to system B in the manner described in paragraph 0027.
Similar case data similar to the new case data
Extract from the case model. Now, it is assumed that two similar case data exist, for example. That is, the first similar case data is described as “the external air temperature at a certain time is 26.5 degrees, the discomfort index is 78%, the external air temperature is 0.4 degrees, and the heat load is 1.0 GCa.
1, heat load 25 GCaI, heat load after 4 hours is 28
GCal ”and the second similar case data is“ 27.5 degrees outside temperature, 83% discomfort index, 0.6 degrees change in outside temperature, 1.8 degrees change in heat load, 1.8 heat, 27 degree heat load
GCal, the heat load 4 hours later is 30 GCal "
And the importance of each of the first and second similar case data.
Are equivalent, for the external state of the system B,
In the system A, the causal relationship model generation means 21A
The inferred values before the correction of the heat load 4 hours after
The similarity of similar case data of
Heat load after 4 hours (28GCa)
l) and the heat load (30
GCal), i.e., 29 GCal.

Next, for the first and second similar cases, the internal state difference between the systems A and B is calculated by the internal state difference calculating means 102 (first calculating means). As a result, the first
For the similar case of the above, the difference in the variation of the heat load is -0.6
GCal (1.0-1.6 = -0.6) and the difference in heat load are obtained as 2GCal (25-23 = 2). In addition, for the second similar case, the difference in the amount of change in the heat load is 0.2 GCal (1.8-1.6 = 0.2), and the difference in the heat load is 4 GCal (27-23 = 4). Respectively. Here, the output difference calculation means 103 (second calculation means) outputs the output difference of the case of the system B with respect to the first and second similar cases, that is, 4 based on the difference model generated by the model generation unit 100. Calculate the difference in heat load after time. Specifically, for the first similar case,
Difference in load change, difference in heat load｝ = difference between {-0.6, 2}
The output difference ΔY (ΔY1) corresponding to the model of
Model similar to {0.2, 4}
Output difference ΔY (ΔY2) corresponding to the As a result, the output difference ΔY1 for the first similar case is 3GCa
1, the output difference ΔY2 with respect to the second similar case is 5 GCal
If so, the first and second similarities
Since the importance of each example is the same, the inference value determination means 1
In step 04, an average value 4GCal of these output differences is obtained and used as a correction value to correct the output value 29GCal in the system A already generated by the causal relation model generation means 21 and extracted by the input / output relation extraction means 101. As a result, the output value of the new system B, ie, 4
The heat load after time is estimated as 25 GCal (29-4 = 25).

As described above, if a relation model between the difference between the internal states of the systems A and B in the same external state and the difference in the heat load after 4 hours, which is the output variable, is created, the model is stored in the system A. The input / output relation model based on the obtained data can be effectively used, and therefore, the data collection period of the new system, System B, can be significantly reduced, and the system B can be operated in a short period of time.

[0037]

As described above, according to the present invention,
When the external condition such as the outside air temperature and the discomfort index is the same between the first and second systems, which are the new and old systems, the internal condition such as the heat load between the first and second systems is determined. In addition to generating a model showing the relationship between the difference and the difference in output, which is the heat load several hours later, when new data is input to the first system, a model having the same external state as the first system is generated. The input / output data of the second system and the input / output relation model thereof are extracted, and the difference between the internal states of the first and second systems with respect to the same external state is calculated. 1. Calculate the output difference with respect to the internal state difference between the second systems, and correct the output value of the input / output relation model of the second system having the same external state by the calculated value to obtain the first system. Since the output value of the system was used, the old system Data is data acquisition period of effective use has been the new system is significantly reduced, as a result, it is possible to run the new system in a short period of time.

[Brief description of the drawings]

FIG. 1 is a functional block diagram showing an embodiment of an existing case effective use type inference apparatus according to the present invention.

FIG. 2 is a block diagram of an air conditioning load prediction system to which the present invention is applied.

FIG. 3 is a functional block diagram of an inference unit constituting the air conditioning load prediction system.

FIG. 4 is a block diagram of the inference unit.

[Description of Signs] A Old system B New system 1A, 1B Storage device 21A Causal relation model generation means 100 Model generation means 101 Input / output relation extraction means 102 Internal state difference calculation means (first calculation means) 103 Output difference calculation Means (second arithmetic means) 104 Inference value determination means

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 9/44 330

Claims (1)

(57) [Claims] 1. The data of a second system which is an old system.
It is a new system using the data and input / output relation model.
An input / output function for generating an input / output relation model of the first system;
An existing case effective use type inference apparatus comprising a relationship model generation means and performing inference using the generated input / output relation model , wherein the input / output relation model generation means comprises a first and a second in the same external state. a model generation means for generating a model showing the relationship between the difference of the difference between the output of the internal state between systems, identical to the first system when the new data to the first system was entered or Extracting means for extracting an input / output relation model of the second system having a similar external state, and calculating an internal state difference between the first and second systems with respect to the same or similar external state. The same or similar operation means as the first operation means and the second operation means for calculating an output difference with respect to an internal state difference between the first and second systems based on the model generated by the model generation means ; <br/> Correction means for correcting an output value of an input / output relation model of the second system having a dynamic state with a value calculated by the second calculation means. .