JP2022116781A - heat source system - Google Patents

Abstract

To achieve energy saving of a heat source system.SOLUTION: In a heat source system (refrigerator system 1), a prediction execution section 66 derives prediction values of a system COP for respective condition sets through inputting the condition sets obtained by applying current values of heat loads to a condition database 72 into a machine learning model which predicts the system COP indicating the heat load with respect to entire power consumption including a heat source machine (refrigerator 10) and an auxiliary heat source machine in accordance with input of the condition set. An optimal value determination section 68 determines respective values of the condition sets corresponding to the highest prediction value of system COP among a plurality of prediction values of system COP derived with the prediction execution section 66 as optimal values.SELECTED DRAWING: Figure 1

Description

The present invention relates to a heat source system including heat source equipment.

As a heat source system including a heat source machine, for example, there is a refrigerator system including a refrigerator. Patent Literature 1 discloses a refrigerator system in which cold water cooled by a refrigerator is supplied to load equipment such as an air conditioner.

JP-A-8-35708

A refrigerator system, which is an example of a heat source system, includes, for example, a refrigerator, a cold water primary pump, a cold water secondary pump, a cooling tower, and a cooling water pump. Each of these devices is operated according to operating conditions that are artificially determined in advance. However, since this operating condition is determined artificially, the heat source system as a whole is not necessarily energy-saving.

SUMMARY OF THE INVENTION An object of the present invention is to provide a heat source system capable of saving energy.

In order to solve the above problems, the heat source system of the present invention includes a heat source device that cools or heats a heat medium sent from a heat load facility and supplies it to the heat load facility, and a heat source device that is involved in supplying the heat medium. Or a set containing a plurality of heat source auxiliary equipment, heat medium temperature difference, heat medium flow rate, heat source equipment operating condition data indicating the operating conditions of the heat source equipment, and auxiliary equipment operating condition data indicating the operating conditions of the heat source auxiliary equipment is a condition set, and a data group in which a plurality of patterns of condition sets are associated with one heat load is stored in a storage unit storing a condition database set for each of the plurality of heat loads, a heat source machine and a heat source machine auxiliary Input a set of conditions obtained by applying the current value of the heat load to the condition database to a machine learning model that predicts the system COP, which indicates the heat load for the entire power consumption including the machine, according to the input of the condition set, A prediction execution unit that derives a system COP prediction value for each input condition set, and each value of the condition set corresponding to the highest system COP prediction value among a plurality of system COP prediction values derived by the prediction execution unit and an optimum value determination unit that determines the optimum value of .

In addition, the prediction execution unit extracts a condition set that satisfies a predetermined extraction condition from among the condition sets obtained by applying the current value of the heat load to the condition database, and inputs the extracted condition sets to the machine learning model respectively. , may derive an estimate of the system COP for each input set of conditions.

In addition, when there are a plurality of corresponding condition sets indicating the condition set corresponding to the highest system COP predicted value, the optimum value determining unit determines the current value for either or both of the heat source equipment operating condition data and the auxiliary equipment operating condition data It is also possible to determine each value of the correspondence condition set that has the smallest amount of change when changing from to the value of the correspondence condition set as the optimum value.

Also, the type of machine learning model may be a gradient boosted decision tree.

Also, the heat source system may further include a model generator that performs machine learning using the condition set and the system COP as teacher data to generate a machine learning model.

Also, the model generator may periodically update the machine learning model.

According to the present invention, it is possible to save energy in the heat source system.

FIG. 1 is a schematic diagram showing the configuration of a refrigerator system, which is an example of a heat source system according to this embodiment. FIG. 2 is a block diagram showing an overview of the model generation unit and the prediction execution unit. FIG. 3 is a conceptual diagram for explaining how to properly use accumulated data. FIG. 4 is a conceptual diagram explaining the details of the machine learning model. FIG. 5 is a diagram showing an example of a condition database. FIG. 6 is a conceptual diagram for explaining the prediction execution section and the optimum value determination section. FIG. 7 is a flow chart for explaining the processing flow of the prediction execution section and the optimum value determination section. FIG. 8 is a diagram explaining the effect of the refrigerator system. FIG. 8(A) shows an example of the temporal transition of the heat load. FIG. 8B shows an example of time transition of the system COP. FIG. 9 is a diagram showing an example of obtaining a sample set.

Aspects of embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The dimensions, materials, and other specific numerical values shown in these embodiments are merely examples for facilitating understanding of the invention, and do not limit the invention unless otherwise specified. In the present specification and drawings, elements having substantially the same function and configuration are given the same reference numerals to omit redundant description, and elements that are not directly related to the present invention are omitted from the drawings. do.

FIG. 1 is a schematic diagram showing the configuration of a refrigerator system 1 as an example of a heat source system according to this embodiment. The refrigerator system 1 includes a refrigerator 10 , a cold water load facility 12 , a buffer section 14 , a cold water primary pump 16 , a cold water secondary pump 18 , a cooling tower 20 , a cooling water pump 22 and a controller 24 .

A refrigerator 10, which is an example of a heat source device, is connected to a cold water load facility 12 through a cold water flow path 30 through which cold water, which is a heat medium, flows. Cold water circulates between the refrigerator 10 and the cold water load equipment 12 through the cold water flow path 30 . The refrigerator 10 cools the cold water sent from the cold water load equipment 12 and supplies the cold water after cooling to the cold water load equipment 12 .

The cold water load equipment 12, which is an example of the heat load equipment, exchanges heat between cold water supplied from the refrigerator 10 and the object to be cooled to cool the object to be cooled. Cold water after heat exchange in the cold water load facility 12 is sent to the refrigerator 10 . The cold water load facility 12 is, for example, an air conditioner, but may be any facility in which heat is exchanged. A cold water load, which is an example of a heat load, indicates cold energy consumed by heat exchange with a cooling target.

The buffer unit 14 , the cold water primary pump 16 and the cold water secondary pump 18 are provided in the middle of the cold water flow path 30 . The buffer section 14 is, for example, an expansion tank. The buffer unit 14 functions as a buffer that temporarily stores the cold water that has been used in the cold water load equipment 12 and the cold water that has been cooled by the refrigerator 10 .

The cold water primary pump 16 draws the cold water used in the cold water load equipment 12 from the buffer unit 14 to the refrigerator 10 and sends the cold water cooled by the refrigerator 10 to the buffer unit 14 . A cold water primary pump inverter 32 is electrically connected to the cold water primary pump 16 . The cold water primary pump inverter 32 drives the cold water primary pump 16 at a rotational speed corresponding to the frequency of the electric power supplied to the cold water primary pump 16 . Hereinafter, the frequency of the cold water primary pump inverter 32 that drives the cold water primary pump 16 may be referred to as the cold water primary pump INV frequency.

The cold water secondary pump 18 draws the cold water cooled by the refrigerator 10 from the buffer unit 14 to the cold water load equipment 12 and sends the cold water used in the cold water load equipment 12 to the buffer unit 14 . A cold water secondary pump inverter 34 is electrically connected to the cold water secondary pump 18 . The cold water secondary pump inverter 34 drives the cold water secondary pump 18 at a rotation speed corresponding to the frequency of the electric power supplied to the cold water secondary pump 18 . Hereinafter, the frequency of the cold water secondary pump inverter 34 that drives the cold water secondary pump 18 may be referred to as the cold water secondary pump INV frequency.

The cooling tower 20 is connected to the refrigerator 10 through a cooling water flow path 36 through which cooling water flows. Cooling water circulates between the cooling tower 20 and the refrigerator 10 through the cooling water flow path 36 . The cooling tower 20 has a cooling tower fan 38 . The cooling tower 20 cools the cooling water by the heat of vaporization of the cooling water as the cooling tower fan 38 rotates. The cooling tower 20 supplies cooling water after cooling to the refrigerator 10 . The refrigerator 10 uses cooling water supplied from the cooling tower 20 to cool cold water. A cooling tower fan inverter 40 is electrically connected to the cooling tower fan 38 . The cooling tower fan inverter 40 drives the cooling tower fan 38 at a rotational speed corresponding to the frequency of the power supplied to the cooling tower fan 38 . Hereinafter, the frequency of the cooling tower fan inverter 40 that drives the cooling tower fan 38 may be referred to as the cooling tower fan INV frequency.

The cooling water pump 22 is provided in the middle of the cooling water flow path 36 . The cooling water pump 22 draws the cooling water cooled by the cooling tower 20 into the refrigerator 10 and sends the cooling water used by the refrigerator 10 to the cooling tower 20 . A cooling water pump inverter 42 is electrically connected to the cooling water pump 22 . The cooling water pump inverter 42 drives the cooling water pump 22 at a rotation speed corresponding to the frequency of the electric power supplied to the cooling water pump 22 . Henceforth, the frequency of the cooling water pump inverter 42 which drives the cooling water pump 22 may be called cooling water pump INV frequency.

Each of cold water primary pump 16, cold water secondary pump 18, cooling tower 20, and cooling water pump 22 is a heat source auxiliary machine involved in the supply of cold water by refrigerator 10, and is provided incidentally to refrigerator 10. .

A cold water inlet 44 and a cold water outlet 46 are formed in the refrigerator 10 . Chilled water enters refrigerator 10 through chilled water inlet 44 and exits refrigerator 10 through chilled water outlet 46 .

Hereinafter, the temperature of cold water at the cold water inlet 44 may be referred to as the cold water inlet temperature. Also, the temperature of the cold water at the cold water outlet 46 may be referred to as the cold water outlet temperature. Also, the temperature difference between the cold water inlet temperature and the cold water outlet temperature may be referred to as a heat medium temperature difference. Also, the flow rate of cold water, which is a heat medium, may be referred to as a heat medium flow rate. The heat load in the chilled water load equipment 12 corresponds to a value obtained by multiplying the heat medium temperature difference by the heat medium flow rate. Also, the heat load is equal to the refrigerating capacity of the refrigerator 10 .

Further, data indicating operating conditions of the heat source equipment such as the refrigerator 10 may be referred to as heat source equipment operating condition data. The heat source device operating condition data is, for example, the set temperature of the refrigerator 10 . The set temperature of the refrigerator 10 is specifically the target value of the chilled water outlet temperature.

Further, the data indicating the operating conditions of the heat source auxiliary equipment may be referred to as auxiliary equipment operating condition data. The accessory operating condition data is, for example, the cold water primary pump INV frequency, the cold water secondary pump INV frequency, the cooling tower fan INV frequency, and the cooling water pump INV frequency.

Control device 24 includes storage unit 50 and system control unit 52 . The storage unit 50 is composed of nonvolatile storage elements. The data stored in the storage unit 50 will be detailed later.

The system control unit 52 is a computer composed of a semiconductor integrated circuit including a central processing unit, a ROM storing programs and the like, and a RAM as a work area. The system control unit 52 functions as a command unit 60, a data acquisition unit 62, a model generation unit 64, a prediction execution unit 66, and an optimum value determination unit 68 by executing programs.

The command unit 60 controls the refrigerator 10 by transmitting a command value of the heat source device operating condition data to the refrigerator 10 . Further, the command unit 60 transmits a command value of the auxiliary machine operating condition data for each heat source machine auxiliary machine to control each heat source machine auxiliary machine.

The system control unit 52 of the present embodiment uses machine learning to determine the command value of the heat source operating condition data and the command value of the auxiliary operating condition data.

The data acquisition unit 62 can acquire the cold water inlet temperature and the cold water outlet temperature using a temperature sensor (not shown). Further, the data acquisition unit 62 can acquire the heat medium flow rate using a flow rate sensor (not shown).

At any timing, the data acquisition unit 62 acquires the chilled water inlet temperature, the chilled water outlet temperature, the heat medium flow rate, the set temperature of the refrigerator 10, the chilled water primary pump INV frequency, the chilled water secondary pump INV frequency, the cooling tower fan INV frequency, Obtain an actual measurement for each of the cooling water pump INV frequencies. The data acquisition unit 62 derives a measured value of the heat medium temperature difference from the acquired cold water inlet temperature and cold water outlet temperature. Note that the measured value of the set temperature of the refrigerator 10 is the measured value of the heat source operating condition data. Also, the measured values of the cold water primary pump INV frequency, the cold water secondary pump INV frequency, the cooling tower fan INV frequency, and the cooling water pump INV frequency are the measured values of the accessory operating condition data.

The data acquisition unit 62 stores the acquired measured values of the heat medium temperature difference, the heat medium flow rate, the heat source equipment operating condition data, and the auxiliary equipment operating condition data as accumulated data 70 in the storage unit 50 .

In addition to acquiring measured values such as heat source equipment operating condition data, the data acquiring unit 62 also acquires the power consumption of the refrigerator 10, the power consumption of the primary cold water pump 16, the power consumption of the secondary cold water pump 18, the cooling tower 20 and the power consumption of the cooling water pump 22 are obtained. The data acquisition unit 62 derives the measured value of the system COP from the acquired power consumption of each device. The system COP indicates the heat load with respect to the power consumption of the entire heat source system including the heat source equipment and the heat source auxiliary equipment. Note that COP (Coefficient Of Performance) indicates a coefficient of performance.

More specifically, the data acquisition unit 62 obtains the power consumption of the refrigerator 10, the power consumption of the primary cold water pump 16, the power consumption of the secondary cold water pump 18, the power consumption of the cooling tower 20, and the consumption of the cooling water pump 22. Each power measurement is added to derive the total power consumption measurement. The data acquisition unit 62 multiplies the measured value of the heat medium temperature difference by the measured value of the heat medium flow rate to derive the measured value of the heat load. The data acquisition unit 62 divides the actual measurement value of the heat load by the actual measurement value of the total power consumption to derive the actual measurement value of the system COP.

The data acquisition unit 62 associates the measured values of the system COP with the measured values of the heat source equipment operating condition data and stores them in the storage unit 50 as accumulated data 70 . Hereinafter, a combination of actual measurement values of the heat medium temperature difference, heat medium flow rate, heat source operating condition data, auxiliary operating condition data, and system COP at arbitrary acquisition timing may be referred to as a sample set.

The data acquisition unit 62 staggers the acquisition timing of the actual measurement values, repeatedly acquires these actual measurement values, and accumulates the acquired actual measurement values in the storage unit 50 as accumulated data 70 . In this manner, the storage unit 50 stores known combinations including the heat medium temperature difference, the heat medium flow rate, the heat source equipment operating condition data, the auxiliary equipment operating condition data, and the measured values of the system COP, and the number of times the measured values are acquired. is accumulated. In other words, in the storage unit 50, the number of sample sets corresponding to the number of acquisitions of the actual measurement values is accumulated.

The number of acquisitions of actual measurement values is, for example, 10,000 times. The acquisition period of the measured values is, for example, one year or three months in winter. It should be noted that the actual measurement value acquisition period may be part of the summer, part of the winter, and part of the middle of the year.

A condition database 72 is stored in advance in the storage unit 50 . The condition database 72 will be detailed later.

FIG. 2 is a block diagram showing an overview of the model generation unit 64 and the prediction execution unit 66. As shown in FIG. The model generator 64 processes the range enclosed by the dashed line in FIG. The prediction execution unit 66 performs processing within the range enclosed by the dashed-dotted line in FIG.

The model generation unit 64 performs machine learning using the known heat medium temperature difference, heat medium flow rate, heat source equipment operating condition data, auxiliary equipment operating condition data, and system COP accumulated as accumulated data 70 in the storage unit 50 as teacher data 80. to generate the machine learning model 82 . That is, the machine learning model 82 that has been learned is generated by the model generation unit 64 .

The machine learning model 82 generated by the model generation unit 64 derives a predicted value of the output data 86 corresponding to the input data 84 corresponding to the input data 84 that is the source data for prediction. The input data 84 are heat medium temperature difference, heat medium flow rate, heat source equipment operating condition data, and auxiliary equipment operating condition data. The output data 86 is the system COP. That is, the machine learning model 82 predicts the unknown system COP corresponding to the inputs of the heat medium temperature difference, the heat medium flow rate, the heat source equipment operating condition data, and the auxiliary equipment operating condition data. Machine learning will be detailed later.

As described above, the system COP is derived based on the total power consumption including the power consumption of the heat source equipment and the power consumption of the heat source auxiliary equipment, and the heat load. Also, the heat load is derived from the heat medium temperature difference and the heat medium flow rate. Therefore, the system COP, which is output data, has a correlation with the heat medium temperature difference, heat medium flow rate, heat source equipment operating condition data, and auxiliary equipment operating condition data, which are input data. Therefore, by using the machine learning model 82, it is possible to derive the predicted value of the system COP from the heat medium temperature difference, the heat medium flow rate, the heat source equipment operating condition data, and the auxiliary equipment operating condition data.

The prediction execution unit 66 inputs input data 84 as original data for prediction to the machine learning model 82 and derives a prediction value of the output data 86 corresponding to the input data 84 . Specifically, the prediction execution unit 66 inputs an arbitrary combination of heat medium temperature difference, heat medium flow rate, heat source equipment operating condition data, and auxiliary equipment operating condition data to the machine learning model 82 as input data 84. Then, the predicted value of the system COP for that combination is derived. The prediction execution unit 66 will be detailed later.

FIG. 3 is a conceptual diagram explaining how to use the accumulated data 70 properly. As described above, in the accumulated data 70, sample sets of heat medium temperature difference, heat medium flow rate, heat source equipment operating condition data, auxiliary equipment operating condition data, and system COP are accumulated for the number of acquisition times. In the present embodiment, a predetermined proportion of the sample set, as indicated by hatching in FIG. 3, of the entire accumulated data 70 is used as training data for actual machine learning. The predetermined percentage is, for example, 80%. As indicated by the cross-hatching in FIG. 3, the remaining sample sets are test data used to verify the accuracy of the machine learning model 82 . In an example where 80% of the sample set is used as training data, the remaining 20% of the sample set is used as test data.

Specifically, the model generating unit 64 generates a machine learning model 82 by performing machine learning using each data of the sample set classified as training data as teacher data 80 . The model generation unit 64 generates a machine learning model 82 created using the training data, from among sample sets classified as test data, the heat medium temperature difference, the heat medium flow rate, the heat source unit operating condition data, and the auxiliary unit operating conditions. Input the data to derive a predicted value for the system COP. The model generator 64 derives the prediction accuracy of the system COP based on the derived predicted value of the system COP and the known system COP of the sample set divided as test data. In this manner, the accuracy of machine learning model 82 can be verified. Note that if the prediction accuracy is not sufficient, the parameters in the machine learning may be changed and the model generation unit 64 may be made to generate the machine learning model 82 again.

FIG. 4 is a conceptual diagram explaining the details of the machine learning model 82. As shown in FIG. The type of machine learning model 82 is, for example, Gradient Boosting Decision Trees (GBDT). A gradient boosting decision tree may also be called a gradient boosting tree or gradient boosting. A practical implementation for generating a gradient boosted decision tree can use a library such as xgboost's XGBRegressor.

As shown in FIG. 4, in the gradient boosting decision tree, multiple decision trees are generated based on training data (in other words, teacher data). For example, in FIG. 4, a total of four decision trees, a first decision tree, a second decision tree, a third decision tree, and a fourth decision tree, are created. Note that the number of decision trees is not limited to four, and may be any number. Also, the number of decision trees may be changed to an arbitrary number by machine learning parameters.

In a gradient boosted decision tree, input data 84 is input to a first decision tree. Then, a first output value is output according to the first decision tree. A first residual derivation process is then performed to derive a first residual by subtracting a first output value from a predetermined target. The predetermined target is set by machine learning parameters, for example. For example, suppose the target is 11 and the first output value is 7. In this case, 11-7=4 is output as the first residual.

The first residual is then input to the second decision tree. Then, a second output value is output according to the second decision tree. Next, a second residual derivation process is performed to derive a second residual by subtracting the second output value from the first residual. For example, suppose the second output value is 1.5. In this case, 4-1.5=2.5 is output as the second residual.

The second residual is then input to the third decision tree. Then, a third output value is output according to the third decision tree. Next, a third residual derivation process is performed to derive a third residual by subtracting the third output value from the second residual. For example, suppose the third output value is 3. In this case, 2.5-3=-0.5 is output as the third residual.

The third residual is then input to the fourth decision tree. Then, a fourth output value is output according to the fourth decision tree. After that, addition processing is performed to add the first output value, the second output value, the third output value and the fourth output value. The output value of the summation process is the output data 86 output from the gradient boosting decision tree, ie the predicted value of the system COP. For example, suppose the fourth output value is -0.5. In this case, 7+1.5+3+(-0.5)=11 is output as the output data 86. FIG.

In this way, the gradient boosting decision tree repeats inputting the residual of the output value of the previous decision tree into the next decision tree, and the total value of the output values of each decision tree is finally expected value. Gradient-boosted decision trees allow highly accurate predictions of output data 86 corresponding to input data 84 to be derived.

Here, an example is described in which the type of machine learning model 82 is a gradient boosted decision tree. However, the existing machine learning model 82 may be applied without being limited to the gradient boosting decision tree.

FIG. 5 is a diagram showing an example of the condition database 72. As shown in FIG. As shown in FIG. 5, a condition set is a set of data including the items of heat medium temperature difference, heat medium flow rate, heat source operating condition data, and auxiliary operating condition data. In addition to the exemplified items, other items related to the heat source system may be added to the condition set.

In the condition database 72, a data group in which a plurality of patterns of condition sets are associated with one heat load is set for each of the plurality of heat loads.

For example, condition set “A1”, condition set “A2”, condition set “A3”, etc. are associated with heat load “A”. Also, one condition set (for example, condition set "A1") has a value different from other condition sets (for example, condition set "A2") in at least one of the items. When the values of the items of the auxiliary machine operating condition data are changed, the value of at least one of the chilled water primary pump INV frequency, the chilled water secondary pump INV frequency, the cooling tower fan INV frequency, and the cooling water pump INV frequency is changed. make it different.

Moreover, not only the heat load "A" but also other heat loads such as the heat load "B" and the heat load "C" are associated with a plurality of patterns of condition sets.

As described above, the heat load corresponds to the product of the heat medium temperature difference and the heat medium flow rate. For this reason, for example, the product of the heat medium temperature difference and the heat medium flow rate in the condition set “A1”, the product of the heat medium temperature difference and the heat medium flow rate in the condition set “A2”, and the heat medium flow rate in the condition set “A3” The product of the temperature difference and the heat medium flow rate, etc., together correspond to the heat load "A".

FIG. 6 is a conceptual diagram illustrating the prediction executing section 66 and the optimum value determining section 68. As shown in FIG. In FIG. 6, it is assumed that a learned machine learning model 82 has already been prepared.

The prediction executing unit 66 acquires the current values of the cold water inlet temperature, the cold water outlet temperature, and the heat medium flow rate, respectively, and derives the current value of the heat medium temperature difference and the current value of the heat load. Next, the prediction execution unit 66 applies the current value of the heat load to the condition database 72 and acquires a condition set of multiple patterns corresponding to the current value of the heat load. For example, if the current value of the heat load is heat load "A", the prediction execution unit 66 acquires condition sets of all patterns corresponding to heat load "A".

The prediction execution unit 66 extracts a condition set that satisfies a predetermined extraction condition from among the acquired condition sets of multiple patterns. Specifically, at least one of the items in the condition set is set as the target item for the extraction condition. Then, the prediction execution unit 66 extracts a condition set in which the value of the target item in the condition set is within a predetermined range including the current value of the target item. The predetermined range is set for each target item. For example, if the target item is the set temperature of the refrigerator 10 , the predetermined range including the current value of the target item is set to ±1° C. of the current set temperature of the refrigerator 10 . In this case, the prediction execution unit 66 extracts a condition set in which the set temperature of the refrigerator 10 is within the range of ±1° C. of the current value. In the example of FIG. 6, condition set "A2" and condition set "A3" meet the extraction conditions and are extracted, but condition set "A1" does not meet the extraction conditions and is not extracted. Although illustration is omitted, condition sets satisfying predetermined extraction conditions are also extracted for condition sets other than the condition sets "A1" to "A3". Also, the target item and the predetermined range can be arbitrarily set.

Note that the target item is not limited to one item, and may be multiple items. In this case, the prediction execution unit 66 extracts a condition set that satisfies the extraction conditions for all of the target items.

The prediction execution unit 66 inputs the extracted condition set as input data 84 to the machine learning model 82, and derives a predicted value of the system COP for each input condition set (in other words, for each extracted condition set). do.

For example, the prediction execution unit 66 inputs the condition set "A2" to the machine learning model 82 and derives the predicted value of the system COP corresponding to the condition set "A2". The prediction execution unit 66 inputs the condition set "A3" to the machine learning model 82 and derives the predicted value of the system COP corresponding to the condition set "A3". Similarly, the prediction executing unit 66 derives the predicted value of each system COP for the extracted condition set.

Here, an example of deriving the predicted value of the system COP for a condition set that satisfies a predetermined extraction condition among a plurality of patterns of condition sets corresponding to the current value of the heat load has been described. However, the prediction executing unit 66 may omit the extraction based on the predetermined extraction condition and derive the predicted value of the system COP for all condition sets corresponding to the current value of the heat load.

The optimum value determination unit 68 compares the plurality of system COP prediction values derived by the prediction execution unit 66 and selects the highest system COP prediction value. The optimum value determining unit 68 determines each value of the condition set corresponding to the selected predicted value of the highest system COP as the optimum value. For example, it is assumed that the system COP prediction value corresponding to the condition set “A2” is the highest system COP prediction value among a plurality of derived system COP prediction values. In that case, the optimum value determination unit 68 determines each value of the condition set "A2" as the optimum value for each item.

Further, for example, the predicted value of the system COP corresponding to the condition set "A2" and the predicted value of the system COP corresponding to the condition set "A3" are both the highest values. There may be multiple sets of conditions corresponding to predicted values. Hereinafter, the condition set corresponding to the highest system COP predicted value may be referred to as the corresponding condition set.

When there are a plurality of corresponding condition sets, the optimum value determination unit 68 determines the amount of change when changing one or both of the heat source equipment operating condition data and the auxiliary equipment operating condition data from the current value to the value of the corresponding condition set. Each value in the small set of corresponding conditions is determined to be the optimal value. Thereby, the optimum value determining section 68 can more reliably determine the optimum value.

For example, the optimum value determining unit 68 subtracts the value of the primary cold water pump INV frequency in the corresponding condition set from the current value of the primary cold water pump INV frequency to derive the amount of change in the primary cold water pump INV frequency. The optimum value determination unit 68 subtracts the value of the cold water secondary pump INV frequency in the corresponding condition set from the current value of the cold water secondary pump INV frequency to derive the amount of change in the cold water secondary pump INV frequency. The optimum value determining unit 68 derives the amount of change in the cooling tower fan INV frequency by subtracting the value of the cooling tower fan INV frequency in the corresponding condition set from the current value of the cooling tower fan INV frequency. The optimum value determining unit 68 derives the amount of change in the cooling water pump INV frequency by subtracting the value of the cooling water pump INV frequency in the corresponding condition set from the current value of the cooling water pump INV frequency.

The optimum value determination unit 68 adds the amount of change in the primary cold water pump INV frequency, the amount of change in the secondary cold water pump INV frequency, the amount of change in the cooling tower fan INV frequency, and the amount of change in the cooling water pump INV frequency. Then, the amount of change in the auxiliary machine operating condition data is derived. The optimum value determination unit 68 derives and compares the amount of change in the accessory operating condition data for each corresponding condition set. The optimum value determination unit 68 selects the corresponding condition set with the smallest variation in the auxiliary machine operating condition data, and determines each value of the selected corresponding condition set as the optimum value.

Here, the amount of change in the primary chilled water pump INV frequency, the amount of change in the secondary chilled water pump INV frequency, the amount of change in the cooling tower fan INV frequency, and the amount of change in the cooling water pump INV frequency are added together. , derived the amount of change in the auxiliary machine operating condition data. However, the amount of change in the auxiliary machine operating condition data is not limited to this mode. For example, add any one of the amount of change in the primary cold water pump INV frequency, the amount of change in the secondary cold water pump INV frequency, the amount of change in the cooling tower fan INV frequency, and the amount of change in the cooling water pump INV frequency. may be used as the amount of change in the auxiliary machine operating condition data, or only one of them may be used as the amount of change in the auxiliary machine operating condition data.

Further, the optimum value determining unit 68 may derive the amount of change in the heat source unit operating condition data by subtracting the value of the set temperature of the refrigerator 10 in the corresponding condition set from the current value of the set temperature of the refrigerator 10 . Then, the optimum value determination unit 68 derives and compares the amount of change in the heat source unit operating condition data for each corresponding condition set, and determines each value of the corresponding condition set with the smallest amount of change in the heat source unit operating condition data as the optimum value. may decide.

Further, the optimum value determining unit 68 may derive the total amount of change by adding the amount of change in the heat source operating condition data and the amount of changing in the auxiliary operating condition data. Then, the optimum value determination unit 68 may derive and compare the overall amount of change for each corresponding condition set, and determine each value of the corresponding condition set with the smallest overall amount of change as the optimum value.

FIG. 7 is a flow chart for explaining the processing flow of the prediction executing section 66 and the optimum value determining section 68. As shown in FIG. The predictive execution unit 66 starts the series of processes shown in FIG. 7 at an interrupt timing that comes every preset control cycle.

When the interrupt timing comes, the prediction execution unit 66 acquires the current value of the heat load (S10). Specifically, the prediction execution unit 66 acquires the current value of the cold water inlet temperature, the current value of the cold water outlet temperature, and the current value of the heat medium flow rate, and derives the current value of the heat load from them.

Next, the prediction execution unit 66 applies the current value of the heat load to the condition database 72 to acquire a condition set corresponding to the current value of the heat load (S11). Next, the prediction execution unit 66 extracts a condition set that satisfies a predetermined extraction condition from among the acquired condition sets (S12). Next, the prediction execution unit 66 inputs the extracted condition set to the machine learning model 82 and derives a predicted value of the system COP for each input condition set (S13).

Next, the optimum value determining unit 68 selects the highest system COP predicted value from the derived system COP predicted values (S14). Next, the optimum value determination unit 68 determines whether or not there are a plurality of corresponding condition sets corresponding to the highest system COP predicted value (S15).

If there are a plurality of corresponding condition sets (YES in S15), the optimum value determining unit 68 changes the current value of one or both of the heat source equipment operating condition data and the auxiliary equipment operating condition data from the current value to the value of the corresponding condition set. is derived (S16). Then, the optimum value determining unit 68 determines each value of the correspondence condition set with the smallest amount of change as the optimum value (S17), and ends the series of processes.

If there is one corresponding condition set (NO in S15), the optimum value determination unit 68 determines each value of the corresponding condition set as the optimum value (S17), and ends the series of processes.

When the optimum value for each item is determined in this way, the command unit 60 transmits the determined optimum value as a command value to each unit. For example, assume that each value of the condition set "A2" is determined to be the optimum value. In this case, the instruction unit 60 instructs the refrigerator 10 to set the temperature value of the refrigerator 10 for the condition set "A2". The instruction unit 60 instructs the cold water primary pump inverter 32 on the value of the cold water primary pump INV frequency of the condition set "A2". The instruction unit 60 instructs the cold water secondary pump inverter 34 on the value of the cold water secondary pump INV frequency of the condition set "A2". The instruction unit 60 instructs the cooling tower fan inverter 40 on the value of the cooling tower fan INV frequency of the condition set "A2". The instruction unit 60 instructs the cooling water pump inverter 42 on the value of the cooling water pump INV frequency of the condition set "A2".

FIG. 8 is a diagram for explaining the effects of the refrigerator system 1. FIG. FIG. 8(A) shows an example of the temporal transition of the heat load. FIG. 8B shows an example of time transition of the system COP. 8A and 8B have a common time axis. The dashed lines in FIGS. 8A and 8B indicate the timing T1 at which the optimum value is determined by the optimum value determining section 68 and instructed to each section. That is, before the timing T1 is before the optimum control, and after the timing T1 is after the optimum control.

As shown in FIG. 8A, the heat load after timing T1 is approximately the same as the heat load before timing T1. On the other hand, as shown in FIG. 8B, the system COP after timing T1 is significantly higher than the system COP before timing T1.

As described above, according to the refrigerator system 1, which is an example of the heat source system of the present embodiment, the power consumption of the entire system can be suppressed while maintaining the heat load, that is, the performance of the heat source. Therefore, according to this embodiment, it is possible to save energy in the heat source system.

In addition, in the heat source system of the present embodiment, machine learning is used to determine the command value of each part of the heat source system, so compared to the mode of manually determining the command value of each part, it is possible to save more energy. .

Further, in the heat source system of the present embodiment, after a condition set that satisfies a predetermined extraction condition is extracted from the condition sets obtained by applying the current value of the heat load to the condition database, the extracted condition set is converted into a machine learning model. 82. Therefore, in the heat source system of this embodiment, the processing load for deriving the predicted value of the system COP can be reduced.

FIG. 9 is a diagram showing an example of obtaining a sample set. FIG. 9 shows the time series from the first year to the fourth year of introduction of the heat source system. The hatched portions in FIG. 9 indicate the acquisition times of the sample sets. The white portions in FIG. 9 indicate the times when the optimum value determination unit 68 is performing optimum value determination control.

The data acquisition unit 62 first accumulates the sample sets in the storage unit 50 at the beginning of spring in the first year after the introduction of the heat source system. At the sample set acquisition time, the optimum value determination control by the optimum value determination unit 68 is interrupted. The data acquisition unit 62 intentionally changes the command value of each device of the heat source system at the time of sample set acquisition and accumulates the sample set. The model generation unit 64 generates the machine learning model 82 at the end of early spring of the first year of introduction using the sample set acquired in early spring of the first year of introduction.

In the middle and late spring of the first year of introduction, the optimum value determination unit 68 performs optimum value determination control.

At the beginning of the summer of the first year of introduction, the optimal value determination control is interrupted, and the data acquisition unit 62 accumulates sample sets in the storage unit 50 . The model generation unit 64 generates and updates the machine learning model 82 at the end of the beginning of the summer of the first year of introduction using the sample sets acquired in the spring and summer of the first year of introduction.

In the middle and end of the summer of the first year after introduction, and in the fall of the first year after introduction, the optimum value decision unit 68 performs control to determine the optimum value.

At the beginning of winter in the first year after the introduction, the optimum value determination control is interrupted, and the data acquisition unit 62 accumulates sample sets in the storage unit 50 . The model generation unit 64 generates and updates the machine learning model 82 at the end of the beginning of winter in the first year of introduction using the sample sets obtained in the spring, summer, and winter of the first year of introduction.

In the middle and end of winter in the first year of introduction, in the second year of introduction, and in the third year of introduction, the optimum value determination unit 68 performs control to determine the optimum value.

Then, in the fourth year after introduction, sample sets are obtained and the accumulated data 70 is updated at the same time as in the first year after introduction. Then, in the fourth year of introduction, the machine learning model 82 is updated in accordance with the update of the accumulated data 70 .

In this way, the data acquisition unit 62 periodically repeats accumulation of sample sets after the introduction of the heat source system, and periodically updates the accumulation data 70 . The model generator 64 periodically updates the machine learning model 82 using the periodically updated accumulated data 70 . Then, the optimum value determination unit 68 performs optimum value determination control using the regularly updated machine learning model 82 . Thereby, the optimum value determination unit 68 can determine an optimum value that reflects aging deterioration of each device such as the heat source machine or the heat source auxiliary machine in the heat source system.

Although the embodiments of the present invention have been described above with reference to the accompanying drawings, it goes without saying that the present invention is not limited to such embodiments. It is obvious that a person skilled in the art can conceive of various modifications or modifications within the scope described in the claims, and these also belong to the technical scope of the present invention. Understood.

For example, in the above embodiment, the refrigerator system 1 was illustrated and explained. However, the heat source system is not limited to the refrigerator system 1 including the refrigerator 10 . For example, the heat source system may be a system including a chiller or a heat pump as heat source equipment. Moreover, the heat source machine is not limited to an electric type, and may be a gas type. Further, the heat source equipment may be of a water-cooling type or an air-cooling type as a cooling method for the heat medium. Further, the heat source equipment is not limited to the configuration of cooling the heat medium and supplying it to the heat load equipment like the refrigerator 10, but may be configured to heat and supply the heat medium like a heater.

Further, in the above-described embodiment, the functions of the model generation unit 64 , the prediction execution unit 66 and the optimum value determination unit 68 are implemented by the control device 24 functioning as the command unit 60 . However, the model generation unit 64 , the prediction execution unit 66 , and the optimum value determination unit 68 may be realized by a computer separate from the control device 24 functioning as the command unit 60 . Further, each function of the model generation unit 64, the prediction execution unit 66, and the optimum value determination unit 68 may be distributed and implemented in a plurality of computers. In this case, the computer having the functions of the model generation unit 64, the prediction execution unit 66, and the optimum value determination unit 68 may acquire necessary data from the storage unit 50 of the control device 24 and perform each process.

Further, the provider having the function of the model generation unit 64 and the provider having the functions of the prediction execution unit 66 and the optimum value determination unit 68 may be different. In other words, a pre-generated machine learning model 82 may be provided, and the provided machine learning model 82 may be used to perform processing for determining the optimum command value.

1 refrigerator system 10 refrigerator 12 cold water load equipment 16 cold water primary pump 18 cold water secondary pump 20 cooling tower 22 cooling water pump 50 storage unit 64 model generation unit 66 prediction execution unit 68 optimum value determination unit 72 condition database 82 machine learning model

In order to solve the above problems, the heat source system of the present invention includes a heat source device that cools or heats a heat medium sent from a heat load facility and supplies it to the heat load facility, and a heat source device that is involved in supplying the heat medium. Or the temperature difference between the temperature of the heat medium at the inlet where the heat medium is sent into the heat source equipment and the temperature of the heat medium at the outlet where the heat medium is sent out from the heat source equipment, the temperature difference of the heat medium A set including flow rate, heat source equipment operating condition data indicating the operating conditions of the heat source equipment, and auxiliary equipment operating condition data indicating the operating conditions of the heat source equipment auxiliary equipment is defined as a condition set, and a plurality of patterns of condition sets are provided for one heat load. The associated data group includes a storage unit storing a condition database set for each of a plurality of heat loads, and a system COP indicating the heat load for the entire power consumption including the heat source machine and the heat source auxiliary machine as the condition set. A prediction execution unit that inputs a condition set obtained by applying the current heat load value to a condition database into a machine learning model that predicts according to the input, and derives a predicted value of the system COP for each input condition set. , an optimum value determination unit that determines each value of the condition set corresponding to the highest system COP prediction value among the plurality of system COP prediction values derived by the prediction execution unit as an optimum value for each heat load ; Prepare.

Further, when there are a plurality of corresponding condition sets indicating a condition set corresponding to the highest predicted value of the system COP, the optimum value determining unit selects either one of the heat source equipment operating condition data and the auxiliary equipment operating condition data for each heat load. Alternatively, each value of the corresponding condition set that has the smallest amount of change when changing from the current value to the value of the corresponding condition set for both may be determined as the optimum value.

Claims (6)

a heat source device that cools or heats a heat medium sent from a heat load facility and supplies the heat medium to the heat load facility;
one or more heat source machine auxiliaries involved in supply of the heat medium by the heat source machine;
a condition set including a temperature difference of the heat medium, a flow rate of the heat medium, heat source equipment operating condition data indicating operating conditions of the heat source equipment, and auxiliary equipment operating condition data indicating operating conditions of the heat source equipment auxiliary equipment; a storage unit storing a condition database in which data groups in which a plurality of patterns of condition sets are associated with one heat load are respectively set for a plurality of the heat loads;
A machine learning model that predicts the system COP indicating the heat load with respect to the overall power consumption including the heat source equipment and the heat source auxiliary equipment according to the input of the condition set, and the current value of the heat load is stored in the condition database. a prediction execution unit that inputs the condition sets obtained by applying and derives a predicted value of the system COP for each of the input condition sets;
an optimum value determination unit that determines each value of the condition set corresponding to the highest predicted value of the system COP among the plurality of predicted values of the system COP derived by the prediction execution unit as an optimum value;
A heat source system with The prediction execution unit extracts the condition set that satisfies a predetermined extraction condition from among the condition sets obtained by applying the current value of the heat load to the condition database, and applies the extracted condition set to the machine learning model. , respectively, to derive the predicted value of the system COP for each of the input condition sets. When there are a plurality of corresponding condition sets indicating the condition set corresponding to the highest predicted value of the system COP, the optimum value determination unit selects one or both of the heat source equipment operating condition data and the auxiliary equipment operating condition data. 3. The heat source system according to claim 1 or 2, wherein each value of said corresponding condition set having the smallest amount of change when changing from a current value to a value of said corresponding condition set is determined as an optimum value. The heat source system according to any one of claims 1 to 3, wherein the machine learning model type is a gradient boosting decision tree. The heat source system according to any one of claims 1 to 4, further comprising a model generator that performs machine learning using the condition set and the system COP as teacher data to generate the machine learning model. The heat source system according to claim 5, wherein the model generator periodically updates the machine learning model.

Images