JP6373887B2 - Action pattern estimation device, action pattern estimation method, and action pattern estimation program - Google Patents

Description

  Embodiments described herein relate generally to a behavior pattern estimation apparatus, a behavior pattern estimation method, and a behavior pattern estimation program.

  Due to the recent liberalization of electricity, electricity retailers are required to submit demand plan values, and if the submitted demand plan values exceed the tolerance (± N%), some penalty may be imposed. . Further, when the difference between the peak value of the demand plan value and the normal value is large, the equipment cost or the cost for power procurement from the outside becomes large. This is a heavy burden for retailers and the like, and there is a certain need for a power demand adjustment method that can suppress the peak. The power demand adjustment is performed by suppressing the power demand at the peak portion or transferring the power demand to another time after performing the power demand prediction. In relation to this, there is known a method for performing demand prediction by using correlation analysis with household attributes and weather and trend analysis information. However, from this information, it is impossible to grasp what purpose the power is used for. For this reason, even if it is possible to estimate how much power demand is generated, there are cases where it is not possible to accurately estimate an action pattern in power demand or the like.

JP 2012-181805 A JP 2006-329781 A

  The problem to be solved by the present invention is to provide a behavior pattern estimation device, a behavior pattern estimation method, and a behavior pattern estimation program capable of accurately estimating a behavior pattern.

The behavior pattern estimation apparatus according to the embodiment includes an input unit and a candidate estimation unit. The input unit receives input of first vector data including a plurality of factors expressing measurement information measured by the target person acting on the device . Candidate estimation unit compares the first vector data received by the input unit, and a second vector data representing the typical behavior patterns including typical results factor predetermined said first One or more third vector data covering possible variations are generated for factors included in the vector data of 2 but not in the first vector data, and the second vector data and the generated second vector data are generated. 3 are compared with each other , a matching factor is extracted from the factors included in each vector data, and the candidate 's action pattern candidate is estimated based on the extracted factor .

The figure which shows the function structural example of the action pattern estimation apparatus which concerns on 1st Embodiment. The figure for demonstrating the relationship between a factor waveform and a typical waveform. The figure which shows the derivation example of a factor waveform. The figure which shows the data structure example of a factor weight set. The flowchart which shows an example of the production | generation process of a factor set. The flowchart which shows an example of the production | generation process of a typical action pattern (typical waveform). The flowchart which shows an example of a factor selection process. The flowchart which shows an example of a factor waveform generation process. The figure for demonstrating a mode that the initial solution of a factor waveform is produced | generated. The flowchart which shows an example of the extraction process of a factor weight set. The figure which shows the specific example of a factor weight set. The flowchart which shows an example of a candidate estimation process. The figure which shows an example of the diagonal matrix produced | generated from the information contained in context information 20B. The figure which shows an example of the matrix calculation result of a factor weight set and context information. The figure which shows an example of the enumeration result of the factor weight set output by the candidate estimation part 18. FIG. The figure which shows the hardware structural example of the action pattern estimation apparatus 1. FIG. The figure which shows the function structural example of the action pattern estimation apparatus 2 which concerns on 2nd Embodiment. The figure which shows the function structural example of the action pattern estimation apparatus 3 which concerns on 3rd Embodiment.

  Hereinafter, a behavior pattern estimation apparatus, a behavior pattern estimation method, and a behavior pattern estimation program according to an embodiment will be described with reference to the drawings. In the following embodiments, a method for estimating an action pattern in power demand will be mainly described. However, the present invention is not limited to this method. For example, the method for estimating an action pattern in a target person or a household outside the electric power field is also used. Can be applied.

(First embodiment)
FIG. 1 is a diagram illustrating a functional configuration example of the behavior pattern estimation apparatus according to the first embodiment. The behavior pattern estimation device 1 shown in FIG. 1 includes an input data storage unit 10, a factor waveform storage unit 12, a factor set generation unit 14, a factor set storage unit 16, and a candidate estimation unit 18. The input data storage unit 10, the factor waveform storage unit 12, and the factor set storage unit 16 shown in FIG. 1 may be a single storage unit or provided outside the behavior pattern estimation apparatus 1. Good.

  The input data accumulation unit 10 accumulates various input data 20 for performing the behavior pattern estimation process according to the first embodiment. The input data storage unit 10 may function as an input unit that receives input of the input data 20. The input data 20 is, for example, multidimensional vector data 20A and context information 20B, but is not limited thereto. The multidimensional vector data 20A includes a number of factors (explanatory factors) that cause behavior pattern selection. The factor can be represented by a predetermined factor type such as “deterministic factor”, “predictive factor”, “probabilistic factor”, and the like. The “deterministic factor” is, for example, the area, the scale or type of the building (wooden structure, reinforcing bar), the family structure, and the like. “Predictive factors” are, for example, the climate (weather) and the price of gasoline. “Probabilistic factors” are, for example, related to probabilistic factors such as morning-type humans, night-type humans, night meals, three meals self-cooking, frequency of going out, etc., but are not limited thereto. is not. Each factor is an item expressed by unique multidimensional vector data. The context information 20B is first vector data including a plurality of factors expressing a behavior pattern, and may include an unknown factor that is a target for estimating a typical behavior pattern. The action pattern is, for example, action contents for determining the power demand corresponding to each factor for each household, and a numerical value (for example, electric energy) of the action result in each time zone or a value obtained by normalizing the numerical value. It is a waveform pattern combined with time series. A set of factors included in each behavior pattern can be expressed as vector data.

  The factor waveform accumulating unit 12 stores factor-specific multidimensional vector data generated in advance by an arbitrary method. The factor waveform is, for example, the influence that the explanatory factor has on the power waveform or the like. For example, the reason why a household has a typical waveform (typical behavior pattern) can be explained by a combination of several explanatory factors. The typical waveform is obtained by giving a predetermined weight to each factor waveform, for example. Here, FIG. 2 is a diagram for explaining the relationship between the factor waveform and the typical waveform. For example, as shown in FIG. 2A, a typical waveform can be expressed by a combination of a factor waveform for each explanatory factor and a weight w. Examples of explanatory factors in FIG. 2A include, for example, the weather (sunny, rainy, cloudy, etc.), households (for example, the number of households), and lifestyles (for example, eating out at night, self-catering for three meals). It is not limited to this. In the example of FIG. 2, typical waveforms are, for example, predetermined weights w1 to w9, factor waveforms for each factor (weather explanatory factors A1 to A3, household factor explanatory factors B1 to B3, lifestyle explanatory factors C1. To C3) and adding the multiplied values.

  FIG. 3 is a diagram showing a derivation example of the factor waveform. For example, the degree of association between clusters of typical waveforms classified into a predetermined number and explanatory factors (for example, characteristics of target users (customers, etc., external data such as weather and day of the week)) is derived, and the explanatory factor is power The influence on the waveform used is derived as a factor waveform. Examples of explanatory factors include weather, region, day of week, number of households, presence of pets, and the like. Further, a conventional correlation analysis can be used as a method for calculating the degree of association, but is not limited to this. The factor waveform uses the degree of association between the explanatory factor and the cluster to derive what kind of waveform change is caused by the explanatory factor. For example, as shown in FIG. 3, when there is a typical waveform 1 having a relevance of 0.4 and a typical waveform 2 having a relevance of 0.7 for the “explaining factor A1,” The factor waveform of the explanatory factor A can be derived from the waveform data. Thus, from the generated factor waveform and the degree of relevance, as shown in FIG. 2B, each explanatory factor for each typical waveform (for example, waveform example 1) and the weight w in FIG. 3 described above. It is possible to obtain a simultaneous equation in which the relevance as shown is input.

  The factor set generation unit 14 generates a factor weight set (first order) from the input data (multidimensional vector data 20A) stored in the input data storage unit 10 and the factor-specific multidimensional vector data stored in the factor waveform storage unit 12. 2 vector data). Here, the factor weight set satisfies the characteristic that the result of synthesizing the factor-specific multidimensional vector data for the factor appearing in the factor weight expression is approximately equal to the waveform of the multidimensional vector data of the typical behavior pattern. There is a need. Therefore, the second vector data includes a plurality of predetermined factors that represent typical behavior patterns from which typical result factors can be derived. The factor set generation unit 14 generates second vector data by determining whether or not to adopt a plurality of factors for obtaining a typical result factor for a plurality of predetermined factors.

  FIG. 4 is a diagram illustrating an example data structure of a factor weight set. The data representation of the factor weight set in the first embodiment has a data structure as shown in FIG. 4 or a data structure that can be mutually converted between data. For example, in the example of FIG. 4, there are n factor weight expressions for one factor weight set. Also, there are n factor weight expressions for one typical waveform identifier. Further, one or more factor correspondences exist for one factor weight expression, and the factor correspondence and the factor weight have a 1: 1 relationship. In addition, there are n factor correspondences for one factor. Note that n described above is a number of 1 or more. The factor set generation unit 14 accumulates input data composed of context information 20B and actual multidimensional vector data, and generates a factor weight set from the accumulated input data and a factor waveform that has been analyzed and defined in advance. To do.

  Here, as a typical waveform derivation method, for example, past power waveforms of all customers are classified into an arbitrary number of clusters, and a typical waveform can be obtained from the power waveforms included in the classified clusters. The typical waveform can be generated by using at least one method among methods such as an average value, a median value, a mode value, an absolute value average, and a standard deviation. Alternatively, a block-type pattern waveform with high readability may be generated by discretizing a typical waveform with an arbitrary time axis and electric energy. Accordingly, it is possible to intuitively grasp at which time zone power saving is likely to be performed using the generated pattern waveform.

(Factor set generation processing)
FIG. 5 is a flowchart illustrating an example of a factor set generation process. In the example of FIG. 5, the factor set generation unit 14 uses, for example, a clustering technique (for example, a K-means method, hierarchical clustering, etc.) or the like to generate a multidimensional vector of N (N is an arbitrary numerical value) typical action pattern. A process of generating data is performed (step S100). Next, the factor set generation unit 14 performs a factor selection process for selecting a factor that explains the typical behavior pattern from a correlation analysis between the typical behavior pattern and attribute information that is a candidate factor, a questionnaire response, and the like (step) S102). Next, the factor set generation unit 14 generates a waveform of the amount of change given to the typical waveform for each selected factor, and performs a process of associating each factor with a typical action pattern (step S104). Note that the association may be objectively performed using, for example, the result of the correlation analysis described above, or may be subjectively performed using a result of hearing with a household or an individual.

  Next, the factor set generation unit 14 generates a factor weight set by using a typical behavior pattern and factor correspondence result and factor-specific multidimensional vector data derived in advance (step S106). ). Here, the weights of the factors are searched so that the result obtained by multiplying the multidimensional vector data of the factors appearing in the factor weight expression and the weights is approximately equal to the multidimensional vector data of the typical behavior pattern. Next, the factor set generation unit 14 determines whether or not a valid factor weight set has been obtained (step S108). If a valid factor weight set has not been obtained, a typical action pattern such as clustering is used. Is changed (step S110), and the process returns to step S100. If a reasonable factor weight set is obtained, this flowchart is terminated.

(Step S100: Typical Behavior Pattern Generation Processing)
FIG. 6 is a flowchart illustrating an example of processing for generating a typical behavior pattern (typical waveform). FIG. 6 is an example of the process in step S100 described above. In the example of FIG. 6, the factor set generation unit 14 performs domain-specific settings for the waveform (step S200). As the process of step S200, the factor set generation unit 14 performs a process of changing the range of one day from 4 o'clock to 28:00 or removing abnormal values on the data. In the process of removing the abnormal value, for example, the first time is not performed, and the data with the largest residual is removed from the clustering execution result by the K-means method or the like thereafter. However, the present invention is not limited to this. Absent. Moreover, the factor set production | generation part 14 performs the process according to the use of a factor pattern analysis result. For example, in the case of an application that focuses on lifestyle, the value is changed to a difference from the lowest value of the day, or in the case of an application that focuses on power consumption, the above-described processing is not performed. Although there is, it is not limited to this. Further, in the process of step S200, a guide for the number of typical waveform generations is set. For example, a guideline for the number of typical waveform generations is determined from hierarchical clustering using sampling data, or a guideline is set based on restrictions on the determination of the number of clusters. The constraints include, for example, whether or not the variance with respect to the representative waveform is close to a normal distribution, and whether or not the distance between child clusters constituting the cluster is uniform, but is not limited thereto.

  Next, the factor set generation unit 14 sets the number K of typical waveform generations (step S202), sets k = K and a = K-4 as initial values (step S204), and performs clustering of all data (step S204). Step S206). Specifically, using the K-means method, the standard number of generations of typical waveforms ± 4 is set to the number of clusters K, and a plurality of execution results are obtained. Further, the clustering is repeated a predetermined number (for example, 1000 times) or more.

  Next, the factor set generation unit 14 evaluates the clustering result (step S208) and determines whether or not the clustering result satisfies a predetermined condition (step S210). In the determination, it is determined whether or not the same constraints as in the hierarchical clustering are satisfied, and it is determined whether or not the clustering result has converged. However, the determination is not limited to this. If the condition is not satisfied, the factor set generation unit 14 increases the number of clusters by 1 (step S212), and returns to the process of step S204. If the clustering result does not converge, processing such as increasing the number of repetitions may be performed. If the clustering result still does not converge, the processing in step S200 described above is re-executed to perform processing such as removing abnormal values. May be.

  If the clustering result satisfies the conditions, K that is the closest to the standard number of typical waveform generations among the converged results is adopted (step S214), and this flowchart is terminated.

(Step S102: Factor selection process)
FIG. 7 is a flowchart illustrating an example of factor selection processing. FIG. 7 is an example of the process of step S102 described above. In the example of FIG. 7, the factor set generation unit 14 performs correlation extraction. In the extraction of correlation, correlation analysis with household attributes and weather data is performed for each typical waveform, and a correlation coefficient between the number of households having a typical waveform and weather data is a predetermined number (for example, 0. 0). 6 or more) are extracted. Further, for example, a portion where the ratio of the number of households for each attribute is significantly different from the overall average at about 10% level is extracted. In addition, about the numerical value mentioned above, it is not limited to this.

  In addition, for data that cannot be measured for correlation extraction (for example, the number of people at home and how to use home appliances), data on the results of interviews with customers, etc. are obtained. Verify whether it is correct. For example, evaluation, verification, etc. may be performed from the viewpoints of “at-home status on days when power usage is high”, “air conditioner usage status on days when power usage is high”, “difference between home appliances constituting the baseline”, etc. Yes, but not limited to this.

  Next, the factor set generation unit 14 tags the typical waveform with factors (step S302). In tagging for factor selection and typical waveforms, factors that are determined to be related to typical waveforms are tagged from correlation analysis and interview results. For example, a factor having a deviation of N times or more of the standard deviation with respect to the average can be selected and tagged, but is not limited thereto. The factor set generation unit 14 may obtain a plurality of factor pattern extraction results based on a plurality of tagging results.

  Next, the factor set generation unit 14 determines whether or not the factor selection is appropriate (step S306). Specifically, the validity is determined by “whether there is a typical waveform that could not be tagged”, “whether the combination of major factors is covered by the difference in the typical waveform”, and the like. If the factor selection is not valid, the factor set generation unit 14 increases the number of clusters given to the k-means method, for example, and regenerates a typical waveform (typical behavior pattern) as shown in FIG. Step S308) and return to Step S300. In the typical waveform generation process, if the factor selection is valid, the factor set generation unit 14 ends the present flowchart.

(Step S104: Factor waveform generation processing)
FIG. 8 is a flowchart illustrating an example of factor waveform generation processing. FIG. 8 is an example of the process of step S104 described above. In the example of FIG. 8, the factor set generation unit 14 generates an initial solution of the factor waveform (step S400). FIG. 9 is a diagram for explaining how the initial solution of the factor waveform is generated. In the generation of the initial solution, for example, an average waveform of all typical waveforms tagged with a factor and an average waveform of an untagged waveform are derived, and a difference waveform of the two average waveforms is obtained as an initial hypothetical waveform of the factor waveform. And In addition, the factor set generation unit 14 compares each typical waveform tagged with a factor with the initial hypothesis waveform, calculates a difference, and sets the calculated minimum value as the difference waveform of each typical waveform. Generate. In the example of FIG. 9A, a differential waveform from a typical waveform 1 to a typical waveform 3 is generated. Further, the time when the value becomes large is selected as a reference point in all typical waveforms in the influence time zone of the factor (for example, 10: 0 to 13:00 shown in FIG. 9A).

  Next, as illustrated in FIG. 9B, the factor set generation unit 14 converts the differential waveform of each typical waveform into a relative value with the reference point being 1. The minimum value of the difference waveforms (relative values) of all typical waveforms is used as the initial solution of the factor waveform. In the example of FIG. 9B, the initial solution of the 10:00 factor waveform is “1”, and the initial solution of the 13:00 factor waveform is “0.5”. The initial solution may be corrected in the subsequent conflict resolution process.

  Next, the factor set generation unit 14 sets initial values such as L = trial upper limit (arbitrary value), l = 0 (step S402), and checks the validity of factor selection (step S404). Next, the factor set generation unit 14 searches for the factor waveform and evaluates the factor waveform, and repeats the above processing while satisfying the condition “there is a contradiction and l <L” (step S406). Next, the factor set generation unit 14 determines whether or not there is a contradiction (step S408). Next, in the evaluation of the factor waveform, for example, assuming that the tagging results in factor selection are correct, the factor waveforms tagged to the typical waveform are stacked, and the stacked results are within the standard deviation range with respect to the average of the typical waveform. However, the present invention is not limited to this, and other evaluation criteria may be included.

  When there is a contradiction, the factor set generation unit 14 redoes the generation of the typical waveform (typical behavior pattern) as shown in FIG. 6 (step S410) and the factor selection as shown in FIG. 7 (step S410). S412), the process returns to step S404. As a solution to the contradiction, in the first embodiment, when the generated factor waveform is not valid, for example, while correcting the value of the factor waveform corresponding to a location that deviates from the typical waveform, Explore. For example, when a plurality of solutions are found by the search, the factor set generation unit 14 employs the one with the smallest error from the initial solution. In addition, if the contradiction cannot be resolved, the factor selection or the typical waveform is wrong, and the factor needs to be reviewed. When reviewing the factors, the factor set generation unit 14 adds a reverse factor, multiplies the factors, expands the factors, or performs processing such as factor division, so that a solution cannot be obtained. Do not. Here, adding the opposite factor means, for example, that if there is a factor “hot day”, a factor “normal day” or “non-hot day” is added. Also, to expand by multiplying factors, it should be expanded from factors such as “hot days”, “not hot days”, and “home” to factors such as “home on hot days” and “home on hot days”. It is. The division of the factor is, for example, to divide a “hot day” in the case of typical waveforms A, B, and C and a “hot day” in typical waveforms D, E, and F. If there is no contradiction, the factor set generation unit 14 ends this flowchart.

(Step S106: factor weight set extraction processing)
FIG. 10 is a flowchart illustrating an example of a factor weight set extraction process. FIG. 10 is an example of the process in step S106 described above. In the example of FIG. 10, the factor set generation unit 14 extracts a factor weight set (step S500). In extraction of the factor weight set, a combination of factors tagged in the factor selection is set as the factor weight set. Accordingly, if the corresponding factor waveform can be extracted by the factor waveform generation process, the factor weight set is unconditionally determined. Note that the above method cannot be applied when the main cannot be acquired, such as necessary for tagging. In such a case, factorization or analysis of variance can be used under the assumption that the factor waveform can be acquired by some method. Extract factor weight sets.

  Next, the factor set generation unit 14 evaluates the validity of the factor weight set (step S502). For example, the factor set generation unit 14 determines whether or not there is a typical waveform that cannot be explained with respect to one typical waveform in the extraction result of the factor weight set. The factor weight set extraction results are evaluated according to preset evaluation criteria such as “whether there are no factors associated with most typical waveforms”. If valid, the factor set generation unit 14 evaluates the usefulness of the factor weight set (step S506). In the evaluation of usability, for example, usability is evaluated based on a criterion such as “whether the difference in results due to factors expected in the usage scene can be explained” or the like.

  On the other hand, if it is determined that it is not valid or useful, the factor set generation unit 14 generates a typical waveform (typical behavior pattern) as shown in FIG. Each process of such factor selection (step S512) and generation of the factor waveform as shown in FIG. 8 (step S514) is performed, and the process returns to step S500. If it is determined in the process of step S508 that the factor set generation unit 14 is useful, the factor set generation unit 14 stores the factor weight set in the factor set storage unit 16 (step S516), and ends this flowchart.

  FIG. 11 is a diagram illustrating a specific example of the factor weight set. As shown in FIG. 11, the factor weight set is generated as a set of action patterns in accordance with combinations of one or a plurality of factors (factor A to factor E in the example of FIG. 11). In the example of FIG. 11, the weight is expressed by “0 (none)” and “1 (present)”, but is not limited to this, and is a predetermined decimal value between 0 and 1, for example. Etc. may be expressed.

  The candidate estimation unit 18 estimates the action pattern candidate set 22 using the context information 20B stored in the input data storage unit 10 and the factor weight set stored in the factor set storage unit 16.

(Candidate estimation processing)
FIG. 12 is a flowchart illustrating an example of the candidate estimation process. In the process of FIG. 12, when the candidate estimation unit 18 receives input of context information 20B (first vector data) (step S600), the factor weight set (second vector data) accumulated in the factor set accumulation unit 16 ) Is acquired (step S602). Next, the candidate estimation unit 18 compares the context information with the factor weight set (step S604), and is included in the factor weight set and is not included in the context information (for example, a target for estimating a behavior pattern). For one or more unknown factors, one or more vector data (third vector data) covering variations that the factor (unknown factor) can take is generated (step S606). Next, the candidate estimation unit 18 compares the generated third vector data with the factor weight set (step S608), estimates one or more action pattern candidates from the comparison result, and sets the action pattern candidate set 22 as the action pattern candidate set 22. This is output (step S610), and this flowchart is terminated.

  For example, in the process of step S608 described above, the candidate estimation unit 18 generates the third vector data and the factor weight set as matrix data corresponding to each combination of factors, and calculates the generated matrix data. By doing so, the part where the factor matches is extracted. The third vector data generates a diagonal matrix having a plurality of predetermined factors included in the factor weight set as diagonal components.

FIG. 13 is a diagram illustrating an example of a diagonal matrix generated from information included in the context information 20B. As shown in FIG. 13, in the context information 20B, when the presence or absence of the factor D is unknown in the comparison with the factor similar to the factor weight set, when the factor D is “1 (present)”, “0 Two diagonal matrices with the case of (none) ”are generated. As described above, when an unknown factor is included in the context information 20B, the candidate estimation unit 18 generates one or more vector data ( third vector data) covering the combination. In the example of FIG. 13, a plurality of diagonal matrices corresponding to each combination are generated as an example of vector data. For example, if n unknown factors exist, 2n matrices are generated.

  The candidate estimation unit 18 acquires a factor weight set as matrix data as illustrated in FIG. 13 from the factor set accumulation unit 16. Each column of the matrix data corresponds to a factor, and each row corresponds to a factor weight expression. In the element of the matrix data, the weight of each factor in the factor weight expression is stored. Here, the order and number of factors in the matrix data column must be the same as the order and number of factors in the diagonal components of the diagonal matrix.

Candidate estimation unit 18, the matrix data of the factor weights set (second vector data) is compared with the raw form diagonal matrix (third vector data), are factors included in each of the vector data matches A part is extracted, and an action pattern candidate is estimated from each extracted factor. FIG. 14 is a diagram illustrating an example of a matrix operation result of a factor weight set and context information. In the example of FIG. 14, factors are calculated. In the first embodiment, for example, in the case of calculation between the same values (for example, “1 and 1”, “0 and 0”), the result value is displayed and the values are different (“1 and 0”). , “0 and 1”), a special calculation (SC (Spcial Calculation)) process for outputting “α” is performed. That is, the candidate estimation unit 18 calculates the result of calculating the element having a value “0” and the element having a diagonal matrix value other than “0” in the factor weight set matrix as illustrated in FIG. In the factor weight set matrix, “α” is set for the calculation result of an element having a value other than “0” and an element having a diagonal matrix value “0”. When there are a plurality of diagonal matrices, the calculation is performed on each of them.

  Further, the candidate estimation unit 18 obtains a set obtained by deleting all rows including “α” as an element in the matrix data obtained from the respective calculation results. In this case, the candidate estimation unit 18 deletes the overlapping factor weight expression except for one. FIG. 15 is a diagram illustrating an example of the result of enumerating factor weight sets output by the candidate estimation unit 18. In the example shown in FIG. 15, the column in which “α” exists is deleted from the calculation results shown in FIG. 14 described above. By outputting this as the behavior pattern candidate set 22, it is possible to accurately estimate the behavior pattern candidate for the input context information 20B.

(Hardware configuration)
Next, a hardware configuration example of the behavior pattern estimation apparatus 1 will be described with reference to the drawings. FIG. 16 is a diagram illustrating a hardware configuration example of the behavior pattern estimation apparatus 1. In the example of FIG. 16, the behavior pattern estimation device 1 includes an input device 30, an output device 32, a drive device 34, an auxiliary storage device 36, a main storage device 38, a CPU (Central Processing Unit) 40, and communication. And are connected to each other by a system bus 44.

  The input device 30 includes a pointing device such as a keyboard and a mouse operated by a user, and a voice input device such as a microphone. The input device 30 activates a program execution instruction, various operation information, software, and the like from the user. The input of the information etc. is received. The output device 32 includes a display for displaying various windows and data necessary for operating the computer main body (behavior pattern estimation device 1) for performing the processing in the first embodiment. The output device 32 can display program execution progress, results, and the like by a control program of the CPU 40.

  Here, in the first embodiment, for example, the execution program installed in the computer main body is provided by the recording medium 46 or the like. The recording medium 46 can be set in the drive device 34. Based on the control signal from the CPU 40, the execution program stored in the recording medium 46 is installed from the recording medium 46 into the auxiliary storage device 36 via the drive device 34.

  The auxiliary storage device 36 is a storage means such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The auxiliary storage device 36 stores an execution program (behavior pattern estimation program) in the first embodiment, a control program provided in a computer, and the like based on a control signal from the CPU 40, and performs input / output as necessary. . The auxiliary storage device 36 can read and write necessary information from each stored information based on a control signal from the CPU 40 and the like.

  The main storage device 38 stores an execution program read from the auxiliary storage device 36 by the CPU 40. The main storage device 38 is a ROM (Read Only Memory), a RAM (Random Access Memory), or the like. The auxiliary storage device 36 and the main storage device 38 are examples of the storage unit 11 described above. Based on a control program such as an OS (Operating System) and an execution program stored in the main storage device 38, the CPU 40 performs processing of the entire computer such as various operations and data input / output with each hardware component. Control each process. Various information necessary during the execution of the program can be acquired from the auxiliary storage device 36, and the execution results and the like can also be stored.

  Specifically, the CPU 40 executes processing corresponding to the program on the main storage device 38 by executing the program installed in the auxiliary storage device 36 based on, for example, a program execution instruction obtained from the input device 30. Do. For example, the CPU 40 executes a process execution program to perform processing such as generation of the factor weight set by the factor set generation unit 14 and estimation of action pattern candidates by the candidate estimation unit 18. The processing content in the CPU 40 is not limited to the above-described content. The contents executed by the CPU 40 are stored in the auxiliary storage device 36 or the like as necessary.

  The communication device 42 communicates with an external device or the like via a communication network represented by a LAN (Local Area Network) or the Internet. The communication device 42 acquires an execution program, software, setting information, and the like from an external device or the like by connecting to a communication network or the like based on a control signal from the CPU 40. Further, the communication device 42 provides an execution result obtained by executing the program or the execution program itself in the first embodiment to an external device or the like. The communication device 42 can realize wireless communication with an external device or the like using, for example, the above-described Bluetooth (registered trademark) or Wi-Fi.

  The recording medium 46 is a computer-readable recording medium that stores an execution program and the like as described above. The recording medium 46 is, for example, a semiconductor memory such as a flash memory, or a portable recording medium such as a CD-ROM or DVD, but is not limited thereto.

  By installing an execution program (for example, an action pattern estimation program or the like) in the hardware configuration shown in FIG. 16, hardware resources and software cooperate to estimate the action pattern in the first embodiment described above. The various processes can be realized.

  As described above, according to the first embodiment, for example, in a situation where it is uncertain what action pattern is selected, an action that can be taken by giving a part of a factor that becomes a factor of action pattern selection Pattern candidates can be listed. In the first embodiment, since a plurality of factor weight expressions included in the factor weight set correspond to one typical behavior pattern, the reason why a typical behavior pattern is taken when several factors are determined is shown. It can be narrowed down by zero or more. Thus, by providing some of the factors that have been clarified, it is possible to list typical behavior patterns that can be taken by a certain target (household, etc.) and factors that are the reason for selection. If there is actual multidimensional vector data, the selected typical behavior patterns and the factors that are the reason for the selection can be listed. Therefore, it is possible to accurately estimate an action pattern of a target person such as a consumer.

  In the first embodiment, in enumerating behavior pattern candidates, it is noted that multidimensional vector data representing a behavior pattern can be approximated to one of multidimensional vector data representing a typical behavior pattern that is at most countable. Then, the reason why the typical behavior pattern is selected is calculated using a “factor weight set” which is a set of “factor weight expressions” modeled by a combination of factors and weights as factors. Therefore, for example, if demand forecasts with little deviation between the planned value and the actual value are necessary, and further deviation is expected, a power saving request can be made to the consumer, so the actual value is out of tolerance. Penalties can be avoided. In addition, according to the first embodiment, it is possible to procure cheap power by suppressing the peak, perform peak shift by power saving advice, etc., or when a retailer also serves as a power generation company, Since the investment cost can be reduced, a stable and inexpensive supply source for the demand plan value can be secured.

(Second Embodiment)
Next, a second embodiment of the behavior pattern estimation apparatus will be described with reference to the drawings. FIG. 17 is a diagram illustrating a functional configuration example of the behavior pattern estimation device 2 according to the second embodiment. In the following description, the same components as those in the behavior pattern estimation apparatus 1 according to the first embodiment described above are denoted by the same reference numerals, and detailed description thereof is omitted here. Further, the hardware configuration can be the same as that of the first embodiment described above, and a specific description thereof is omitted here. In the second embodiment, the difference from the first embodiment described above is that a behavior determining factor estimation unit 50 and a behavior modification factor estimation unit 52 are provided. Therefore, this configuration will be specifically described below.

  In the behavior pattern estimation device 2 according to the second embodiment, the behavior determination factor estimation unit 50 inputs, for example, the behavior pattern candidate set 22, and is unknown in the context information 20B described above from the enumeration results of the factor weight set. Select all the factors that existed. The selected factor is considered as a “factor determining behavior”. When all the factors that determine the behavior (behavior determining factors) are determined, the typical waveform to be selected is determined. In the above example, the factor D is an unknown factor in the context information 20B.

  The behavior modification factor estimation unit 52 estimates which factor is effective when prompting the subject to modify behavior (for example, changing a typical waveform to be selected). For example, the behavior modification factor estimation unit 52 selects, from the typical waveforms corresponding to the factor weight set enumeration results, what is desired to be selected by the target for adjusting the power demand. The behavior modification factor estimation unit 52 extracts a factor weight expression corresponding to the selected typical waveform from the list of factor weight sets included in the behavior pattern candidate set 22. The behavior modification factor estimation unit 52 extracts factors included in both the selected factor weight expression and the factors that determine the behavior extracted by the behavior determination factor estimation unit 50. This factor becomes a “factor to be acted upon (behavioral change factor)” in order to obtain a typical waveform. Selecting a factor to be acted on means that a typical waveform to be selected is acquired.

  As a result, it is possible to obtain the power demand adjustment information 54 that clarifies what may be taken in the typical waveform that the target wants to select and factors that should be worked on to select the typical waveform. Therefore, according to the second embodiment, the demand adjuster can perform desired demand suppression and peak shift based on the power demand adjustment information 54 described above.

(Third embodiment)
Next, a third embodiment of the behavior pattern estimation apparatus will be described with reference to the drawings. FIG. 18 is a diagram illustrating a functional configuration example of the behavior pattern estimation apparatus 3 according to the third embodiment. In the following description, the same components as those in the behavior pattern estimation apparatus 2 according to the second embodiment described above are denoted by the same reference numerals, and detailed description thereof is omitted here. Further, the hardware configuration can be the same as that of the first embodiment described above, and a specific description thereof is omitted here. The third embodiment differs from the second embodiment described above in that it includes a factor selection history storage unit 60 instead of the behavior modification factor estimation unit 52 and a vector data prediction unit 62 as an example of a prediction unit. Is a point. Therefore, this configuration will be specifically described below.

  In the behavior pattern estimation apparatus 3 according to the third embodiment, when the result of enumerating factor weight sets included in the behavior pattern candidate set 22 is input to the behavior determination factor estimation unit 50 as described above, the behavior determination factor estimation unit 50 selects all the factors that are unknown in the context information 20B from the list of factor weight sets. These are considered “factors that determine behavior”. Once all the factors that determine the behavior are determined, the selected typical waveform is determined.

  The vector data prediction unit 62 acquires history information (factor selection history information) in which a factor that determines a preset action is selected from the factor selection history storage unit 60. The vector data prediction unit 62 derives the probability that each factor is selected based on the acquired history information. For each factor weight expression included in the enumeration result of the factor weight set, the probability of selecting the factor weight expression is obtained by multiplying the probability of the factor that determines the action. Each factor weight expression is associated with a typical action pattern, and the sum of the selection probabilities of all factor weight expressions associated with a certain typical action pattern is the selection probability of the typical action pattern. . Therefore, the vector data prediction unit 62 outputs the power demand prediction information 64 corresponding to the selection probability of the typical behavior pattern.

According to the third embodiment described above, typical behavior patterns that a household can take are determined by giving context information that is a set of deterministic factors such as target attributes and predictable factors such as weather forecasts. Factors to be acted upon, or factors to be acted upon by the subject to select a particular typical behavior pattern. The first to third embodiments described above may be combined with some or all of at least one other embodiment.
According to the first to third embodiments described above, for example, the probability of a typical waveform for each consumer and the expected reduction amount by the power saving request / power saving advice corresponding to the explanatory factor can be grasped. Therefore, for example, the maximum DR (demand response) possible amount can be predicted. In this case, for example, the maximum DR possible amount can be acquired by calculating the sum of the expected reduction amount for each time zone. Actually, there is a DR effect for consumers who follow the power saving request. In the third embodiment, the total demand can be predicted by performing the following calculation for each time zone of a typical waveform.
“Electric energy for each time zone of each typical waveform = Σ (Energy energy of waveform x Appearance probability for each household)”
Moreover, based on said calculation result, the total electric energy for every time slot | zone can be estimated by the following calculations.
"Total energy for each time zone = Σ (Electric energy for each time zone of each typical waveform)"

  In addition, according to the third embodiment, for example, for each household, typical waveform candidates (plurality) that can be taken by the consumer, and the expected reduction amount due to the power saving request or the power saving advice corresponding to the explanation factor can be obtained. Therefore, the maximum DR possible amount and the total demand can be predicted. When the maximum DR possible amount is predicted, the sum of the above-described expected reduction amount for each time zone is the maximum DR possible amount. In practice, DR can be performed in accordance with the power saving request.

  Moreover, as a prediction of total demand, if the typical waveform which a consumer can take is confirmed to one, total demand prediction will be attained. In order to realize this, for example, a total demand forecast can be performed by collecting a plurality of explanatory factors based on a predetermined condition or by deriving new explanatory factors and weights by principal component analysis or the like.

  According to at least one embodiment described above, the behavior pattern estimation device (1, 2, 3) is accepted with an input unit that accepts input of first vector data including a plurality of factors expressing the behavior pattern. The first vector data is compared with second vector data including a plurality of predetermined factors expressing a typical behavior pattern from which a typical result factor can be derived, and the second vector data One or more third vector data covering possible variations are generated for factors included in the first vector data but not included in the first vector data, and the second vector data and the generated third vector data are generated. And a candidate estimation unit (1) for extracting matching third vector data and estimating a behavior pattern candidate from the extracted third vector data ) And, by providing, it is possible to estimate the behavior pattern accurately. In addition, according to at least one embodiment, it is possible to realize demand prediction for each household, and it is possible to predict a plurality of prediction results with probability of occurrence. In addition, the reason why the prediction result is derived, such as “a certain household eats out with a probability of 30%, in which case the prediction result is obtained”, can be presented together.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1, 2, 3 ... Behavior pattern estimation apparatus, 10 ... Input data storage part, 12 ... Factor waveform storage part, 14 ... Factor set production | generation part, 16 ... Factor set storage part, 18 ... Candidate estimation part, 20 ... Input data, DESCRIPTION OF SYMBOLS 30 ... Input device, 32 ... Output device, 34 ... Drive device, 36 ... Auxiliary storage device, 38 ... Main memory device, 40 ... CPU, 42 ... Communication device, 50 ... Behavior determination factor estimation part, 52 ... Behavior modification factor estimation , 60... Factor selection history storage unit, 62... Vector data prediction unit

Claims (10)

An input unit that receives input of first vector data including a plurality of factors expressing measurement information measured in the device by the subject acting on the device ;
Comparing the first vector data received by the input unit with the second vector data representing a typical behavior pattern including a predetermined typical result factor;
Generating one or more third vector data covering possible variations for factors included in the second vector data but not in the first vector data;
The second vector data is compared with the generated third vector data, a matching factor is extracted from the factors included in each vector data, and the subject's action is based on the extracted factor A candidate estimator for estimating pattern candidates;
An action pattern estimation device comprising: For a plurality of factors stipulated Me pre, to determine the adoption of the multiple factors for obtaining the typical outcome variables, including the factors set generation unit configured to generate the second vector data,
The behavior pattern estimation apparatus according to claim 1. The input unit accepts input of context information including a plurality of factors including an unknown factor to be estimated for a behavior pattern,
The candidate estimation unit determines a variation that the unknown factor can take, based on context information including a plurality of factors including an unknown factor to be estimated for the behavior pattern.
The behavior pattern estimation apparatus according to claim 1 or 2. The candidate estimator is
Generating the second vector data and the third vector data as matrix data corresponding to each combination of factors, calculating each generated matrix data, and extracting a portion where the factors match;
The behavior pattern estimation device according to any one of claims 1 to 3. The candidate estimator is
Generating a diagonal matrix including a plurality of predetermined factors included in the second vector data as diagonal components as the third vector data;
The behavior pattern estimation apparatus according to claim 4. Among the behavior pattern candidates obtained by the candidate estimation unit, a behavior determination factor estimation unit that estimates a factor included in the second vector data and not included in the first vector data as a behavior determination factor;
A factor weight expression of a typical behavior pattern corresponding to a behavior pattern candidate obtained by the candidate estimation unit is extracted, and based on the extracted factor weight expression and the behavior determination factor obtained by the behavior determination factor estimation unit, A behavior modification factor estimator for estimating a behavior modification factor for taking the typical behavior pattern;
The behavior pattern estimation apparatus according to any one of claims 1 to 5, further comprising: Among the behavior pattern candidates obtained by the candidate estimation unit, a behavior determination factor estimation unit that estimates a factor included in the second vector data and not included in the first vector data as a behavior determination factor;
From the factor selection history information accumulated in advance, the history information in which the factor that determines the behavior pattern is selected is acquired, and each behavior determination factor obtained by the behavior determination factor estimation unit is selected based on the acquired history information A predicting unit that predicts the probability that the typical behavior pattern is selected based on the acquired probability;
The behavior pattern estimation device according to any one of claims 1 to 5, further comprising: Wherein the action against the subject equipment is a behavior with respect to the equipment for a household in the power demand,
The behavior pattern estimation apparatus according to any one of claims 1 to 7. Computer
Receiving an input of first vector data including a plurality of factors expressing measurement information measured in the device by the subject acting on the device from the input unit;
The first vector data received by the input unit is compared with second vector data representing a typical behavior pattern including a predetermined typical result factor, and the second vector data is compared with the second vector data. One or more third vector data covering possible variations is generated for factors that are included but not included in the first vector data, and the second vector data, the generated third vector data, , Extracting a matching factor among the factors included in each vector data, and estimating the candidate 's behavior pattern candidate based on the extracted factor ,
Behavior pattern estimation method. On the computer,
Receiving an input of first vector data including a plurality of factors expressing measurement information measured in the device by the subject acting on the device from the input unit;
The first vector data received by the input unit is compared with second vector data representing a typical behavior pattern including a predetermined typical result factor, and the second vector data is compared with the second vector data. One or more third vector data covering possible variations is generated for factors that are included but not included in the first vector data, and the second vector data, the generated third vector data, , Extracting a matching factor among the factors included in each vector data, and estimating the candidate 's behavior pattern candidate based on the extracted factor ,
An action pattern estimation program for executing processing.

Images