Building management system having set offset value learning and set bias value determining system for controlling thermal environment

A set offset value learning and set bias value determining apparatus comprising:

• a set offset value learning apparatus comprising:
• a neural network having an input layer, an intermediate layer, and an output layer;
• an intermediate layer unit generation means for generating an intermediate layer unit in said intermediate layer in response to a command of changing an air-conditioning set value, set with respect to an air-conditioning control target, to a tenant set value;
• a set offset value storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to store a difference between a management set value, set with respect to said air-conditioning control target, and said tenant set value as a set offset value;
• an intrinsic number storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to store an intrinsic number indicating an ordinal number in a sequence of generation;
• and
• a membership function storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to respectively determine and store membership functions on the basis of a measured value of said air-conditioning control target, obtained when said intermediate layer unit is generated, and a measured or preset value of other predetermined parameters;
• and
• a set bias value determining apparatus comprising:
• an activation amount determination means for determining an activation amount of each said intermediate layer unit in response to a command of changing said management set value by performing a fuzz operation between each of said membership functions stored in said intermediate layer unit and a corresponding one of said membership functions of a said measured value of said air-conditioning control target, obtained when said command is issued, and said preset value of said other predetermined parameters;
• an information pack output means for outputting an information pack representing said activation amount, said set offset value, and said intrinsic number, said set offset value and said intrinsic number being stored in said intermediate layer unit;
• an air-conditioning set value calculation means for selecting said set offset value of said intermediate layer unit with a maximum activation amount, or said set offset value of one of a plurality of said intermediate layer units each having a maximum said activation amount which has a maximum said intrinsic number based on said information pack output, adding said selected set offset value, as a set bias value, to a value obtained by changing said management set value to obtain a sum, and setting said sum as a new air-conditioning set value;
• and
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal.

A set offset value learning and set bias value determining apparatus comprising:

• a set offset value learning apparatus comprising:
• a neural network having an input layer, an intermediate layer, and an output layer;
• an intermediate layer unit generation means for generating an intermediate layer unit in said intermediate layer in response to a command of changing an air-conditioning set value, set with respect to an air-conditioning control target, to a tenant set value;
• a set offset value storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to store a difference between a management set value, set with respect to said air-conditioning control target, and said tenant set value as a set offset value;
• and
• a membership function storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to respectively determine and store membership functions on the basis of a measured value of said air-conditioning control target, obtained when said intermediate layer unit is generated, and the measured or preset values of other predetermined parameters;
• and
• a set bias value determining apparatus comprising:
• a re-triggerable counter having a count value being cleared to zero in accordance with a command of changing said air-conditioning set value to said tenant set value;
• an activation amount determination means for determining an activation amount of each said intermediate layer unit in response to an incrementing operation of said re-triggerable counter by performing a fuzzy operation between each of said membership functions stored in said intermediate layer unit and a corresponding one of said membership functions of said measured value of said air-conditioning control target, obtained at the time of said incrementing operation, and said preset values of said other predetermined parameters;
• an information pack output means for outputting an information pack representing at least said activation amount and said set offset value stored in said intermediate layer unit;
• an air-conditioning set value calculation means for obtaining a set bias value by dividing a sum total of products of said activation amounts and said set offset values of the respective information packs output from said information pack output means by a sum total of said activation amounts of said information packs, adding said set bias value to said management set value, and setting a sum as a new air-conditioning set value in place of said tenant set value;
• an air-conditioning set value forcible change means for forcibly changing said air-conditioning set value to said tenant set value in accordance with a change command to said tenant set value of said air-conditioning set value;
• and
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal.

A set offset value learning and set bias value determining apparatus comprising:

• a set offset value learning apparatus comprising:
• a neural network having an input layer, an intermediate layer, and an output layer;
• an intermediate layer unit generation means for generating an intermediate layer unit in said intermediate layer in response to a command of changing an air-conditioning set value, set with respect to an air-conditioning control target, to a tenant set value;
• a set offset value storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to store a difference between a management set value, set with respect to said air-conditioning control target, and said tenant set value as a set offset value;
• and
• a membership function storage means for causing said intermediate layer unit generated by said intermediate layer unit generation means to respectively determine and store membership functions on the basis of a measured value of said air-conditioning control target, obtained when said intermediate layer unit is generated, and the measured or preset values of other predetermined parameters;
• and
• a set bias value determining apparatus comprising:
• an activation amount determination means for determining an activation amount of each said intermediate layer unit in response to a command of changing said management set value by performing a fuzzy operation between each of said membership functions stored in said intermediate layer unit and a corresponding one of said membership functions of said measured value of said air-conditioning control target, obtained when said command is issued, and said preset values of said other predetermined parameters;
• an information pack output means for outputting an information pack representing at least said activation amount and said set offset value stored in said intermediate layer unit;
• an air-conditioning set value determination means for obtaining a set bias value by dividing a sum total of products of said activation amounts and said set offset values of the respective information packs output from said information pack output means by a sum total of said activation amounts of said information packs, adding said set bias value to a value obtained by changing said management set value, and setting a new air-conditioning set value;
• and
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal.

A neural network operation apparatus comprising:

• first to Nth input layer units for receiving first to Nth parameters, at least one of which serves as a measured value, wherein said first to Nth input layer units send said first to Nth parameters as a set of fuzzy values every time data indicating dissatisfaction with the air conditioning environment is present;
• intermediate layer units arranged in correspondence with a set of lattice points of a feature space formed by a set of axes of said first to Nth parameters of said first to Nth input layer units, each unit of said intermediate layer units having a membership function centered on a corresponding lattice point of said set of lattice points for each axis of said set of axes each said membership function having a region overlapping a membership function centered on an adjacent lattice point, and said intermediate layer units receiving outputs from said first to Nth input layer units each intermediate layer unit of said intermediate layer units determines an activation amount by performing a fuzzy operation between each of said membership functions held therein and a corresponding one of said membership functions of said first to Nth parameters sent as said fuzzy values from said first to Nth input layer units, and supplies said activation amount to an output layer unit;
• said output layer unit for receiving outputs from said intermediate layer units said output layer unit determines a weighting coefficient for each intermediate layer unit of said intermediate layer units, which has supplied said activation amount, on the basis of said activation amount supplied from each said intermediate layer unit and an offset value of a thermal environment index, and sends said weighting coefficient to a corresponding intermediate layer unit of said intermediate layer units;
• said intermediate layer unit stores and learns said weighting coefficient sent from said output layer unit as a new weighting coefficient, or stores and learns a weighting coefficient obtained by a calculation based on said weighting coefficient sent from said output layer unit and a stored weighting coefficient as said new weighting coefficient;
• and
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal.

A neural network operation apparatus comprising:

• first to Nth input layer units for receiving first to Nth parameters at least one of which serves as a measured value, said first to Nth input layer units send said first to Nth parameters as fuzzy values every time data indicating dissatisfaction with the air-conditioning environment is present;
• intermediate layer units arranged in correspondence with a set of lattice points of a feature surface formed by a set of axes of said first to Nth parameters of said first to Nth input layer units, each unit of said intermediate layer units having a membership function centered on a corresponding lattice point of said set of lattice points for each axis of said set of axes each said membership function having a region overlapping a membership function centered on an adjacent lattice point and said intermediate layer units receiving outputs from said first to Nth input layer units;
• an output layer unit for receiving outputs from said intermediate layer units said output layer unit determines a weighting coefficient for each said intermediate layer unit, which has supplied said activation amount, on the basis of the activation amount supplied from said intermediate layer unit and an offset value of a thermal environment index, and sends said weighting coefficient to a corresponding intermediate layer unit of said intermediate layer units;
• each intermediate layer unit of said intermediate layer units determines an activation amount by performing a fuzzy operation between each membership function of said membership functions held therein and a corresponding one of said membership functions of said first to Nth parameters sent as said fuzzy values from said first to Nth input layer units, and supplies said activation amount to said output layer unit;
• said corresponding intermediate layer unit stores and learns said weighting coefficient sent from said output layer unit as a new weighting coefficient, or stores and learns a weighting coefficient obtained by a calculation based on said weighting coefficient sent from said output layer unit and a stored weighting coefficient as a new weighting coefficient;
• in a normal operation, said first to Nth input layer units respectively send said input first to Nth parameters, as said fuzzy values, to said intermediate layer units;
• each said intermediate layer unit determines said activation amount by performing said fuzzy operation between each of said membership functions held therein and a corresponding one of membership functions of said first to Nth parameters sent as said fuzzy values from said first to Nth input layer units, and supplies said activation amount and said stored weighting coefficient to said output layer unit;
• said output layer unit determines a sum total of products of said activation amounts and said weighting coefficient sent from said intermediate layer units, and determines and outputs said sum total as a bias value of said thermal environment index;
• and
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal.

A neural network apparatus comprising

• first to Nth input layer units for receiving first to Nth parameters at least one of which serves as a measured value;
• intermediate layer units arranged in correspondence with a set of lattice points of a feature space formed by a set of axes of said first to Nth parameters of said first to Nth input layer units, each unit of said intermediate layer units having a membership function centered on a corresponding lattice point of said set of lattice points for each axis of said set of axes, each said membership function having a region overlapping a membership function centered on an adjacent lattice point and said intermediate layer units receiving outputs from said first to Nth input layer units,
• an output layer unit for receiving outputs from said intermediate layer units
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal;
• a back propagation learning type neural network comprising:
• a set of input layer units to which first to Nth parameters are branched and input;
• a set of intermediate layer units for receiving outputs from said input layer units;
• and
• an output layer unit for receiving outputs from said set of intermediate layer units and for outputting a standard measured value of a thermal environment index.

A neural network operation apparatus comprising:

• first to Nth input layer units for receiving first to Nth parameters at least one of which serves as a measured value;
• intermediate layer units arranged in correspondence with a set of lattice points of a feature space formed by a set of axes of said first to Nth parameters of said first to Nth input layer units, each unit of said intermediate layer units having a membership function centered on a corresponding lattice point of said set of lattice points for each axis of said set of axes, each said membership function having a region overlapping a membership function centered on an adjacent lattice point, and said intermediate layer units receiving outputs from said first to Nth input layer units;
• an output layer unit for receiving outputs from said intermediate layer units;
• a tenant temperature preference setting and learning means for obtaining a temperature setting preferred by a tenant, said temperature setting entered into a tenant user terminal;
• a back propagation learning type neural network comprising:
• a set of input layer units to which first to Nth parameters are branched and input;
• a set of intermediate layer units for receiving outputs from said input layer units;
• and
• an output layer unit for receiving outputs from said set of intermediate layer units and for outputting a standard measured value of a predicted mean vote (PMV) as a thermal environment index;
• wherein a PMV bias value of said predicted mean vote (PMV) is obtained as a bias value of a thermal environment index from said neural network operation apparatus outside of said back propagation learning type neural network, said PMV bias value is added to said predicted mean vote from said back propagation learning type neural network, a sum is converted into an effective temperature by a PMV inverse operation, and said effective temperature is an output.