KR102664055B1 - Industrial power control device using power data decomposition - Google Patents

Abstract

The industrial power control device of the present invention includes a plurality of loads, and is provided with a lower distribution board for controlling the voltage or power of the plurality of loads. Raw data including power data from the lower distribution board or the load. It may include a data collection unit that collects, a power prediction unit that predicts power from raw data, and a control unit that issues a voltage control command to the voltage regulator of the lower distribution panel or the voltage regulator for each individual load, and the control unit may include the raw data of the data collection unit. And it may be based on the power prediction of the power prediction unit, and the power prediction unit may predict the total power of the lower distribution board where power data for each load is collected.

Description

Industrial power control device using power data decomposition}

The present invention relates to an industrial power control device that decomposes power data by load using feature vectors, predicts the power demand of industrial companies based on this, and controls voltage.

There are existing patents that disclose defect diagnosis based on current signals of machinery being analyzed, such as motors in industrial facilities such as factories. In application number 10-2021-0064961, a bearing failure diagnosis device receives a motor 3-phase current signal, denoises the motor 3-phase current signal, and converts the denoised motor 3-phase current signal into a 1-phase current signal, 1 Steps of determining a characteristic factor based on a phase current signal and determining whether a bearing is defective based on the characteristic factor are provided. In particular, using a characteristic factor that can independently diagnose a bearing defect of a motor is disclosed.

As such, existing patents often use modulated/processed electricity-related data such as current and power of power devices, and in particular, power prediction using frequency analysis such as FFT (Fast Fourier Transform) and STFT (Short Time Fourier Transform). In most cases, the analysis is limited to the power data of a specific device, such as a motor.

The present invention is not limited to a specific device among a plurality of power devices provided in industrial facilities such as a factory, but each individual power device (load) separated from raw data collected from power data of different types of devices placed in the factory. Using power data, power demand prediction and voltage control can be performed from an overall factory perspective.

In addition, the present invention can use a method of extracting and learning preset characteristics when classifying power data for each load from the total power data collected from the power data of individual loads.

The industrial power control device of the present invention is an industrial power control device that includes a plurality of loads and is provided with a lower distribution board that controls the voltage or power of the plurality of loads, and raw data including power data from the lower distribution panel or the load. It may include a data collection unit that collects, a power prediction unit that predicts power from raw data, and a control unit that issues a voltage control command to the voltage regulator of the lower distribution panel or the voltage regulator for each individual load, and the control unit may include the raw data of the data collection unit. And it may be based on the power prediction of the power prediction unit, and the power prediction unit may predict the total power of the lower distribution board where power data for each load is collected.

The present invention not only uses current and power data of industrial companies, such as motors, to analyze specific equipment, but also makes the entire machinery equipped in industrial companies subject to power monitoring and power control.

To this end, the present invention uses a sensor capable of high-speed power data sampling, so that sufficient raw data can be provided to learn the pattern of power data for each individual load, and each load can be obtained from the collected power data from the switchboard of the factory in actual operation. Power data can be distinguished, and through this, abnormalities in each load can be detected in real time or voltage control of individual loads can be performed when predicted power demand exceeds.

The present invention can calculate the power peak reduction rate more precisely during voltage control by selecting effective sub-system control points through CVRf prediction, and controls the peak value more precisely based on this expected reduction rate, thereby limiting the upper limit of the planned power demand range. The number and frequency of exceeding the limit can be reduced, and stable supply and planning can be enabled on the power supplier side.

1 is an explanatory diagram of the arrangement of the industrial power control device of the present invention.
Figure 2 is a configuration diagram of the industrial power control device of the present invention.
Figure 3 is an explanatory diagram of feature extraction, recombination, and learning of the present invention.
Figure 4 is an explanatory diagram of a feature vector of the present invention.
Figure 5 is a flow chart of the industrial power control method of the present invention.
Figure 6 is a configuration diagram of the power data analysis step of the present invention.
Figure 7 is a configuration diagram of the power prediction and control steps of the present invention.
Figure 8 is a configuration diagram of the power peak management step of the present invention.

1 to 8, the industrial power control device and industrial power control method of the present invention will be described.

The industrial power control device of the present invention may include at least one of a server 80, a data collection unit 110, and a mapping unit 130.

Referring to FIG. 1, the present invention can be applied to all power devices that consume power provided in industrial enterprises or industrial facilities such as factories. Hereinafter, the terms factory, industry, industrial facility, etc. may be used interchangeably to mean a place where a plurality of power devices are installed. Hereinafter, power devices include resistance-type electric heaters, heat pumps, switching transformers, motors, or melting furnaces as objects of power demand prediction, and are applicable as long as they consume power and enable power data collection.

Hereinafter, industrial power devices may be used in the same sense as load (L). Power data is electricity-related data including current, voltage, and power of the corresponding load (L) and can be used interchangeably unless otherwise specified. For example, the current data of the sensor (S) that measures the real-time current of the load (L) connected to the lower switchboard (30) is classified for each load (L), the peak power of the upper switchboard (50) is managed, and whether the load (L) is broken. It can be used for identification, etc., and for this purpose, voltage drop can be performed, so current, voltage, and power data can be used interchangeably.

The data collection unit 110 may collect raw data including power data of the load L and mapping data of the mapping unit 130.

A plurality of lower distribution boards 30 corresponding to lower systems of the power system may be branched from the upper distribution board 50 corresponding to the upper system of the power system. The application of the upper system and lower system can be extended to the upper system if located upstream, and the lower system if located downstream in the power system spanning the terminal load (L) that receives power based on the power supplier. .

In FIG. 1 , as an example, each lower distribution board 30 may correspond to an individual industrial enterprise (factory), and the upper distribution board 50, from which a plurality of lower distribution boards 30 branch, may indicate a place where a transformer is installed. That is, the first sub-distribution board 30a represents the first industrial company (first factory), the second sub-distribution board 30b represents the second industrial company (second factory), and the third sub-distribution board 30c represents the third industrial company (first factory). It may represent an industrial enterprise (third factory), and a plurality of loads (L) may be connected to each lower distribution board (30).

Voltage control performed based on demand power prediction can be performed in both the upper distribution board 50 or the lower distribution board 30, but the present invention provides power data collected from the upper distribution board 50 from 111 to the individual load (L). In order to control the voltage or power for each load by distinguishing the power data of ), the lower distribution board 3 that controls the individual load (L) can be the main target of voltage control.

In Figure 1, as an example of the lower distribution board 30, the downstream loads (L1 to L6) are all shown as controllable from the lower distribution board 30, and a sensor (S) corresponding to each load (L1 to L6) is shown. Indicates that is provided. However, due to the power facility structure, power devices L located downstream of a certain lower distribution board 30 may have their voltage adjusted at individual locations rather than at the lower distribution board 30.

In FIG. 1 , the data collection unit 110 shows a first case in which all sensors S that measure power data for each power device (load L) are provided. Feature extraction by the feature extraction unit 310, label assignment by the label unit 330, recombination of features and labels by the feature combining unit 350, learning method by the learning unit 370, or classification unit 390 Depending on the power data classification method, the data collection unit 110 may be provided only in the lower distribution board 30 or the upper distribution board 50 and may include a second case in which the collected power data information is collected. At least part of the above corresponds to the first case or the second case, so it can be expanded to a mixed case of the first case and the second case.

In the first case, the server 80 can transmit and receive data with the measuring instrument of the upper distribution board 50, the measuring instrument of the lower distribution board 30, and a plurality of sensors (S1 to S6) that collect power data for each load. .

In the second case, the server 80 can transmit and receive data with the measuring instrument of the upper distribution board 50 and the measuring instrument of the lower distribution board 30, and the measuring instrument of the lower distribution board 30 includes loads located downstream ( The power data of L) may be collected and transmitted, and the power data of the lower distribution boards 30 located downstream may be collected and transmitted to the measuring instrument of the upper distribution board 30.

Except for the data collection unit 110, the remaining components may be included in the server 80. For example, the control unit 200, which can control the voltage regulator of the upper distribution board 50 or the lower distribution board 30, is located in the server 80 and is a power prediction unit ( The final voltage control command can be issued based on the prediction of 470).

The voltage regulator of the upper or lower system is a winding type that can be adjusted to a specified winding voltage by changing the tap in the live state of the main line, and may include an OLTC (On Load Tap Changer).

The voltage regulator of the upper or lower system may be a hybrid semiconductor type voltage regulator that functions as a power converter including a plurality of power semiconductor elements.

The voltage regulator of the present invention may refer to a semiconductor-type electric circuit, may be a feedback controller, and may include a proportional, integral, differential controller, or a combination thereof. .

Therefore, an embodiment of the voltage regulator of the present invention may be a three-phase power hybrid transformer, and as a semiconductor hybrid voltage regulator, voltage control can be achieved by constructing a structure that can reduce the power level and obtain optimal power density. Some of the total power can be converted into the process of alternating current-direct current-alternating current, and voltage, current, and power factor can be controlled through a power converter without using a tap changer.

Semiconductor-type voltage regulators have no concept of lifespan regarding the number of voltage changes, have very detailed control units, and can be controlled in real time by a digital signal processor (DSP), so they can be used in multi-layer systems of upper/lower systems. May be suitable for control.

The mapping unit 130 of the present invention can generate mapping data that maps or matches each load (L) and the power data of each load (L). In the first case, if all sensors (S) are provided for each load (L), each load (L) or sensor (S) is mapped to the power data for each load (L) generated in real time with the operation of the factory. It can be.

The mapping data of the mapping unit 130 can be used to classify power data when assigning a label to the label unit 330 or learning by the learning unit 370. That is, the label unit 330 can assign a label (B) used to mutually distinguish between power data for each load, and the learning unit 370 can learn to distinguish or identify patterns of power data for each load.

One of the purposes of the present invention is to control the power of the lower system (lower switchboard) when a power peak in the upper system (upper switchboard) is expected, and more specifically, all loads (L) connected to the lower switchboard 30. ) from collection to control, mapping data may be necessary for specifying and discriminating power data from power data collection to load-specific power control.

The industrial power control device of the present invention includes at least one of a control unit 200, a feature extraction unit 310, a label unit 330, a feature combining unit 350, a learning unit 370, and a classification unit 390. It can be included.

The control unit 200 can control or adjust the overall performance of components belonging to the industrial power control device shown in FIG. 2. In one embodiment, the control unit 200 uses the power prediction result by the power prediction unit 470 to adjust the voltage of the upper system for the upper distribution board 50, or to the lower distribution board 30 and the lower distribution board 30. Voltage regulation of the sub-system for connected loads (L) can be performed.

Referring to Figure 3, feature extraction and feature recombination of the present invention will be described.

The feature extractor 310 may filter at least one of the plurality of features (F) inherent in the power data.

The feature (F) may mean a unique characteristic that can be extracted from power data collected from the load (L). For example, the feature (F) is the distortion rate by order, the difference between the last time frame (or windows) and voltage or current, and the point where the differential value of the current waveform changes suddenly (or the point where the waveform is distorted). It may include at least one of a map and a normalized waveform such as min max normalization.

The waveform of general power data is one in which sinusoids and harmonics overlap, and the distortion rate may be the value of the harmonic component compared to the sinusoidal wave, which is the fundamental wave.

As such, the present invention does not analyze current and power data based only on data extraction such as frequency-amplitude through the existing Fourier transform, but frequency analysis can also be one of the features (F) of the present invention. In addition to the components of the feature (F) listed above, any element that can be used to distinguish power data for each load may be included in the feature (F).

The power data 111 input to the feature extraction unit 310 contains the plurality of features (F), and each feature (1, 2, 3) is extracted by the feature extraction units (A, B, C, D). ,4, or indicated as F1, F2, F3, F4) can be extracted (filtered). The first feature (F1) is extracted by the first feature extraction unit (A), the second feature (F2) is extracted by the second feature extraction unit (B), and so on.

The feature combiner 350 may restore power data by recombining the extracted and partially processed features (F). Processing of the extracted features F may be performed by a separate component between the feature extraction unit 310 and the feature combining unit 350, or may be performed in the feature combining unit 350 itself.

When the feature combining portion 350 is recombined, a label (B or 5) may be added by the label portion 330. The label (B or 5) may function as a name tag that distinguishes and identifies the recombined power data 113 from each other.

The recombined power data 113 may have a label 5 attached to the recombined feature (F) or be added to at least some of the features (F) to perform a discrimination function between the recombined power data 113.

When assigning a label (B), the label unit 330 can receive and utilize mapping data among the collected raw data.

Feature extraction of the feature extraction unit 310 and feature recombination of the feature combining unit 350 may be processed by embedding power data in a feature space.

Referring to FIG. 4, certain power data can be expressed as a feature vector (FV) in a feature space. Each feature (F) can form each axis of the feature space, and decomposing the feature vector (FV) into components of each axis (F) can correspond to feature (F) extraction by the feature extractor 310. FV-1 represents component decomposition of the feature vector (FV) into the first feature (1, F1), FV-2 represents component decomposition of the feature vector (FV) into the second feature (1, F2), etc. This is the method.

Component decomposition into each feature axis of the feature vector (FV) corresponding to the feature (F) extraction of the feature extraction unit 310 is not limited to simply judging by the size of the scalar quantity as shown in FIG. 4, and the distortion rate by order , the difference between the last time frame (or window) and voltage or current, a map of the point where the differential value of the current waveform changes suddenly (or the point where the waveform is distorted), and min max normalization ), etc., symbolically indicates that each feature (F) can have a scale or dimension that includes at least one of the normalized waveforms themselves.

In this way, when N features (F) are embedded in the power data, the given power data is expressed as a feature vector (FV) in a feature space consisting of N axes, is decomposed into N components, and then processed, etc. After processing, it can be recombined by the feature combination unit 350.

Recombination of the feature (F) by the feature combining unit 350 may be performed in various ways depending on the purpose. When checking and distinguishing patterns for each feature inherent in the input power data 111, the features of all feature combining units 350 can be disassembled, recombined, and assigned a label (B). If only a few of the features (F) are necessary for distinguishing power data of a certain load, the remaining features (F) can be excluded and only the necessary features can be selected.

The learning unit 370 may receive power data collected for each load (L), perform feature extraction and recombination, and learn the characteristics of the power data for each load (L).

Of course, it is also possible to predict normal or abnormal patterns for each load by learning all the power data of industrial machines, or to predict power demand fluctuations for each load and adjust the voltage. However, if the target is the entire industrial machinery, rather than specific machinery such as motors, etc., tracking and predicting all individual loads may be too time-consuming, and in reality, current data from all machines must be collected accurately in real time. It can be difficult to measure everything.

The learning of the learning unit 370 includes establishing each pattern from power data for each individual load, distinguishing power data for each load from the collected total power data, and detecting abnormal patterns for each load from the collected total power data. It may include things like:

The process of feature extraction and feature recombination by the feature extraction unit 310, label unit 335, and feature combination unit 350 may be repeatedly performed on countless amounts of power data. It may be desirable for the power data used in such iterative learning to be provided at high speed and in a large amount of data from a measuring instrument (sensor, S).

In one embodiment, the instrument may be a Hall effect current sensor with a resolution of 20 Hz to 50 khz, and may acquire raw data about current values at a sampling rate of 600 Hz to 1200 Hz.

The learning unit 370 can initially receive and learn power data patterns for pattern storage for each individual load, and learn to distinguish power data for each load from the collected total power data. At this time, the classification unit 390 can continuously record, store, and distinguish which load each component of the power data measured in real time is classified based on the learning results by the learning unit 370. That is, the classification unit 390 can classify the total power data measured in real time into power data for each load based on the learning results by the learning unit 370.

The power data classified by the learning unit 370 and the classifying unit 390 can be used to determine whether or not the load is operating normally, and can be used at the collected point (upper switchboard 50 or lower switchboard 30). When there is a need for voltage control, such as peak power management, the power data analyzed for each load can be used to find the most optimal voltage control method for each load.

The industrial power control device of the present invention may include at least one of a pattern detection unit 410, a CVRf estimation unit 430, a scheduling unit 450, and a power prediction unit 470.

The pattern recognition unit 410 can determine the normal operation pattern for each individual load or the normal pattern of the entire industrial facility including a plurality of loads, based on the learning by the learning unit 370 and the classification by the classification unit 390. It is possible to determine whether each individual load (L) is operating normally or whether the entire industrial facility including multiple loads is operating normally.

The pattern recognition unit 410 can continuously track/analyze power data for each individual load classified from the collected total power data in real time to quickly determine which load is operating abnormally, and if the malfunction pattern is repetitive, before the malfunction occurs. It may be possible to predict when it will occur and take action in advance.

In addition, industrial facilities may include resistive loads, inverter loads, air conditioner loads, electric motor loads, etc. depending on the type of load (L), and the pattern identification unit 410 can distinguish patterns of power data depending on the type of load. there is. When an abnormal pattern occurs in the total power data collected, you can first determine which type of load the pattern occurs in and then look at the data of the individual load included within each load type to quickly identify the type of load in question. Information on patterns of power data by type can also be useful from the manager's management perspective.

Even in the learning of the learning unit 370, it is difficult to predict all power data for each individual load and it is not realistically easy to provide all meters (S) for all loads, so pattern of power data of the main load or at least some of the loads is identified. Just as learning from the entire power data other than the input for By tracking the entire pattern of data and seeing a pattern different from usual, it is possible to shorten the time it takes to identify the cause of the malfunction pattern by determining which load abnormality is causing the pattern behavior.

Methods for lowering peak power through previously presented load predictions may include various methods such as demand control, price penalty method, contractual limit setting, and supply voltage control.

In addition, in managing the peak power of the upper system 50 located upstream of the plurality of lower systems 30, if the power demand of the individual loads (L) of the lower system 30 can be predicted, the entire upper system system 30 can be predicted. Accurate predictions of load peaks can be obtained.

The demand response method requires the will of the participants, so there is a limit to coverage to systematize it by applying it collectively at the switchboard 50 of the upper system level, it takes time and money, and institutional improvement may be required first.

The method of imposing price penalties and setting contractual limits cannot rule out cases where peaks occur in the upper system due to mistakes, errors, or unavoidable reasons on the part of consumers during the operation of load facilities, even if both power suppliers and users are aware of this fact. does not exist.

When attempting to reduce power through supply voltage control at the upper system substation, the number of voltage cases that can be controlled while maintaining the voltage change rate range agreed upon by the supplier is limited, and the resulting peak control coverage may be small.

In particular, it is difficult to finely adjust voltage according to power prediction using the existing tap switching voltage adjustment method, which is controlled in increments of 0.5% to 12.5%, resulting in visible changes observed by consumers in supply voltage quality. It can be.

Therefore, the present invention can adopt a semiconductor-type voltage regulator, and for example, the voltage regulator for each load can be individually controlled in 0.1% units. This fine voltage control can be linked to the analysis and scheduling of power data for each load, leading to power control without deterioration in the quality of power data for the entire upper system and for each load in the lower system.

The CVRf estimation unit 430 estimates the conservation voltage drop factor (CVRf, CVR factor, Conservation Voltage Reduction factor) for each load in real time based on the power data for each load (L) classified by the classification unit 390. You can track it.

The conservation voltage drop coefficient (CVRf) may be the power change rate relative to the voltage change rate. For example, when the voltage goes down by 1% and the power goes down by 2%, CVRf may be 2. The conservation voltage drop coefficient (CVRf) can be used to estimate the actual amount of power saved.

The CVRf estimator 430 can predict CVRf for each load of each voltage regulator or estimate the weighted average value of CVRf in real time. The CVRf weighted average may be the average of the predicted CVRf for each load multiplied by a preset weight.

The CVRf estimation unit 430 calculates the conservation voltage drop coefficient (CVRf) and the weighted average of the conservation voltage drop coefficients based on the power data for each load according to the classification of the learning unit 370 and the learning and classification unit 390. The overall reserve voltage drop coefficient upstream where power data is collected can be predicted.

The power prediction unit 470 calculates the power for each load of the lower system when there are a plurality of lower distribution boards 30 of the lower system and the plurality of lower distribution boards 30 are branched to the upper distribution board 50 of the upper system upstream. By combining the forecast values, the total future power of the upstream subsystem or the upstream parent system can be predicted.

If a problem such as peak power generation in the upper system of the power prediction unit 470 is predicted from the power data for each load of the lower system, the scheduling unit 450 determines whether the predicted power of the upper system exceeds the upper limit of the planning range. If it is predicted that it is, an algorithm for power control for each load of the lower system can be performed.

That is, based on the prediction of the CVRf estimation unit 430, the scheduling unit 450 determines how much the voltage regulator should be used for a certain load of a certain lower system when controlling the peak power of the upper switchboard 50 or the lower switchboard 30. It is possible to predict the degree of peak power control of the upper system according to voltage control.

The scheduling unit 450 considers the reserve voltage drop coefficient (CVRf) for each load calculated from the CVRf estimation unit 430 and the total reserve voltage drop coefficient upstream where power data is collected from the weighted average of the reserve voltage drop coefficients. Therefore, the adjustment amount is added to a predetermined voltage value in the order of the CVRf at each load, and the voltage adjustment range of the voltage regulator of the lower system to be adjusted can be expanded until the peak control amount required by the upper system is achieved. .

For example, if the voltage regulator is a semiconductor type, the predetermined voltage value may be 0.1%, and fine adjustment by 0.1% may be possible.

As another example, in the case of the OLTC type winding type, voltage regulators with a small voltage adjustment range can be controlled sequentially, and voltage control can be stopped when the target peak control value is achieved. In this case, one voltage regulator per regulation cycle of the voltage regulator is controlled starting from the smallest voltage regulator, and the desired voltage can be achieved without exceeding the upper limit of the power plan at the predicted peak time.

If the power prediction by the power prediction unit 470 is compared with the actual power value and the power prediction satisfies a preset reliability, the scheduling algorithm by the scheduling unit 450 can be maintained.

If the power prediction by the power prediction unit 470 is compared with the actual power value, and the power prediction deviates from the preset reliability, the scheduling unit 450 may modify, cancel, or strengthen the schedule to restore reliability. Adjustment of this schedule can be made on a load-by-load basis.

This feedback process for the difference between the power prediction value of the scheduling unit 450 and the actual power value may be applied to other components.

For example, if reliability is not satisfied despite a scheduling change due to feedback from the scheduling unit 450, the CVRf estimator 430 may adjust the CVRf weight.

Additionally, other components including the pattern recognition unit 410 may continuously update their judgment results by reflecting the difference from the predicted value in the generated peak power.

The CVRf of the CVRf estimation unit 430 is used for scheduling settings by the scheduling unit 450 or voltage control by the control unit 200 when a problem such as peak power generation by the power prediction unit 470 is predicted to occur. It can be.

Most existing CVR Factor estimation methods are limited to measurements when large temporary voltage fluctuations occur due to factors external to the customer (load group).

If the CVR Factor of all customers cannot be identified in real time at the upper system substation, the peak power reduction effect due to voltage drop cannot be predicted and evaluated, and if the CVR Factor is estimated only with the power change observed after the voltage change, after observing this Since voltage control is performed reactively according to the rule base technique, there may be difficulties in successful peak control.

Therefore, the real-time CVRf estimation method of the present invention can estimate/predict the weighted average of the CVR Factor of each load connected to the lower system in real time.

The reason why real-time CVRf estimation of the present invention is necessary is because when entering into a power supply contract with a consumer (load) for peak power control at the upper distribution board 50, equipment attachment or usage equipment details are required to receive an accurate CVR Factor in real time. This may be because it is realistically impossible to receive it.

The present invention utilizes the CVR factor calculated in real time in the sub-system switchboard 30 to determine how much to control the voltage regulator of the lower-level system located according to the predicted peak power, and determines how much to control the voltage regulator of the lower-level system based on the voltage control. A more precise prediction of the resulting peak power reduction effect may be possible. Through this, the quality of supplied power can be improved by reducing the number of times the planned power range is exceeded, managing peak power more precisely, and reducing voltage fluctuations observed at the consumer side.

Since it is known which load (L) of the sub-system needs to be controlled, it may not occur to change the individual voltages of the sub-system at once by changing the overall voltage of the voltage regulator of the upper-level system, and it may not occur due to cable voltage drop, etc. It has the advantage of not being affected by a specific sub-system with low voltage and being able to selectively control the voltage regulator or load (L) of a sub-system with good conservation voltage drop (CVR) efficiency among multiple loads.

With reference to FIGS. 5 to 8 and the contents of the industrial power control device, the industrial power control method of the present invention will be described.

The industrial power control method of the present invention may include at least one of a raw data collection step (S100), a power data analysis step (S200), and a power prediction and control step (S300).

The raw data collection step (S100) may include a power data collection step (S110) or a mapping data generation step (S130).

The power data collection step (S110) may be collecting raw data, and the raw data includes power data for each load of at least a portion of the sub-system, or collected power data of the upstream system located upstream of the sub-system. can do.

A measuring instrument (sensor, S) provided in the lower distribution panel 30 can measure power data for each load according to a set unit time.

The mapping data generation step (S130) may be collecting mapping data, and the mapping data may be divided into each load (L) and the power data of the load (L), or each load (L) and the power data of the load (L) by the mapping unit 130. It may be mutual mapping/matching of the measuring device (S) of the load (L).

The power data analysis step (S200) may decompose, learn, and classify the power data of the raw data collection step (S100) using mapping data, features (F), labels (B), recombination, etc.

The power data analysis step (S200) includes the feature extraction step (S210) of extracting the feature (F) by the feature extraction unit 310 for the input power data 111, and the label section during recombination in the recombination step (S250). Power recombined by combining the label assignment step (S230), which assigns a label (B,5) by (330), the feature (F) from the feature extraction step (S210), and the label (B) from the label assignment step (S230). A learning step of learning feature vectors of power data for each load classified or provided through a recombination step (S250), a feature extraction step (S210), a label assignment step (S230), and a recombination step (S250) that generate data 113. It may include at least one of (S270) and a classification step (S290) of classifying the entire power data into load-specific power data based on the learning in the learning step (S270) by the learning unit 370.

The power prediction and control step (S300) may include at least one of a unique pattern identification step (S310), a power increase/decrease prediction step (S330), and a power peak management step (S350).

The unique pattern identification step (S310) is to discern the pattern of unique power data for each load between normal operation and abnormal operation for each load, or to discern the pattern between normal operation and abnormal operation of the entire collected power data to determine what the pattern is. This may include discerning unique patterns that result from abnormal behavior of the load or type of load.

The power increase/decrease prediction step (S330) may be performed by the power prediction unit 470. The power prediction unit 470 may predict the peak power generation of the total power data in the collected stages 50 and 30, based on the analysis of power data for each load in the power data analysis step (S200). Conceptually, the power increase/decrease prediction step (S330) may include a unique pattern identification step (S310).

If the occurrence of a situation requiring voltage control, such as peak power generation, is predicted through the unique pattern identification step (S310) or the power increase/decrease prediction step (S330), the control unit 200 performs voltage control according to the scheduling of the scheduling unit 450. Commands can be sent to the voltage regulator.

Therefore, the power peak management step (S350) includes a CVRf estimation step (S351) of estimating the conservation voltage drop coefficient (CVRf) of the power data for each load classified or provided by the CVRf estimation unit 430 or the weighted average of the CVRf. , a power demand prediction step (S353), a scheduling step (S355), and a voltage control step for each load, in which the power demand of the lower system stage to which the load is connected by the power prediction unit 470 and the total power demand of the upper system stage are predicted. (S357), and a feedback step (S359) that compares the peak power prediction and actual power.

In the scheduling step (S355), if the peak power of the upper system is predicted to exceed the upper limit of the planning range, the scheduling unit 450 may set a schedule so that voltage control for each load of the lower system is performed before the expected peak time. .

In the load-specific voltage control step (S357), if the peak power of the upper system is predicted to exceed the upper limit of the planning range, the control unit 200 considers the conservation voltage drop coefficient (CVRf) for each load until the peak control amount is achieved. Different voltage control can be performed. At this time, the control unit 200 may schedule in consideration of the hourly voltage adjustment amount and control the voltage so as not to exceed the power plan upper limit at the predicted peak time.

30... Sub-distribution board
30a... 1st sub-distribution board
30b... 2nd sub-distribution board
30c... 3rd sub-distribution board
50... Top switchboard
80...server
110... Data collection department
111... Incoming power data
113... Recombined power data
130... Mapping unit
200... Control unit
310... Feature extraction unit
330... Label section
350... Features Joint
370... Learning Department
390... Classification department
410... Pattern recognition unit
430... CVRf estimation unit
450... Scheduling department
470... power prediction unit
B...label
L... load
S... sensor (instrument)
F... Features
FV...feature vector
FV-1,2,3,4,5,6... Feature vector components
S100... Raw data collection phase
S110... Power data collection phase
S130... Mapping data creation step
S200... power data analysis step
S210... Feature extraction step
S230... Label assignment step
S250... recombination step
S270... learning phase
S290... Sorting step
S300... power prediction and control phase
S310... Unique pattern identification step
S330... Power increase/decrease prediction step
S350... Power peak management phase
S351... CVRf estimation step
S353... Power demand forecasting step
S355... Scheduling step
S357... Load-specific voltage control steps
S359... Feedback phase

Claims (20)

In an industrial power control device that includes a plurality of loads and is provided with a lower distribution board that controls the voltage or power of the plurality of loads,
a data collection unit that collects raw data including power data from the lower distribution board or load;
a power prediction unit that predicts power from the raw data of the data collection unit;
a control unit that issues a voltage control command to the voltage regulator of the lower distribution board or a voltage regulator for each individual load;
a feature extraction unit that filters at least one of a plurality of features inherent in power data for each load; - Here, the features are unique characteristics that can be extracted from power data collected from the load, -
After the features of the power data for each load are extracted by the feature extractor, a feature combining unit that restores the power data by recombining at least some of the processed features; Including,
a label unit that assigns a label used to mutually distinguish between power data for each load; - Here, the label unit adds a label that serves an identification function to distinguish the recombined power data when recombining the feature combining unit, -
The feature extraction of the feature extraction unit and the feature recombination of the feature combining unit correspond to processing power data by embedding it in a feature space,
The power prediction unit predicts power based on power data for each load classified by the feature extraction unit, label unit, and feature combination unit,
The control unit is an industrial power control device that issues a voltage control command for each load based on the power prediction of the power prediction unit.
delete According to claim 1,
The feature includes at least one of a distortion rate by order, a difference in voltage or current by time, a map of the point where the differential value of the current waveform changes suddenly, and the normalized waveform itself. Industrial power control devices.
delete According to claim 1,
An industrial power control device in which processing of the extracted features is performed by a separate component between the feature extraction unit and the feature combining unit, or is performed in the feature combining unit itself.
delete delete According to claim 1,
The power data is expressed as a feature vector (FV) of the feature space,
Each feature forms an axis of the feature space,
Decomposing the feature vector into components of each axis corresponds to feature extraction by the feature extraction unit.
According to claim 1,
It includes a learning unit that receives power data collected for each load and learns the characteristics of the power data for each load,
The learning of the learning unit includes establishing each pattern from power data for each individual load, distinguishing power data for each load from the collected total power data, and detecting abnormal patterns for each load from the collected total power data. An industrial power control device comprising at least one.
According to claim 1,
All sensors that measure power data for each load are provided,
Power data for each load from the sensor is used for power prediction by the power prediction unit and voltage control command by the control unit,
The voltage regulator of the sub-distribution board is a semiconductor type industrial power control device capable of controlling voltage without using a tap switch.
According to claim 1,
It includes a mapping unit that generates mapping data that interconnects each load and the power data of each load,
The raw data includes power data for each load and the mapping data.
According to claim 1,
A mapping unit that is included in the raw data and generates mapping data that interconnects each load and the power data of each load, a label unit that assigns a label used to mutually distinguish between the power data for each load, and the power data for each load. It includes a learning unit that learns discrimination or pattern identification,
The mapping data is used to assign a label to the label unit and to discriminate power data for each load when learning the learning unit.
According to claim 1,
It includes a feature extraction unit that filters at least one of the plurality of features inherent in power data for each load,
It includes a feature combining unit that recombines the features extracted by the feature extracting unit,
It includes a learning unit that receives power data collected for each load and learns the characteristics of the power data for each load,
The process of feature extraction by the feature extraction unit and feature recombination by the feature combination unit is repeatedly performed on a plurality of power data,
An industrial power control device in which learning by the learning unit is performed repeatedly on a plurality of power data.
According to claim 13,
All sensors that measure power data for each load are provided,
Power data for each load is collected at high speed by the sensor,
The sensor is a current sensor using the Hall effect, and is an industrial power control device that acquires power data for current values at a sampling rate of 600 Hz to 1200 Hz.
According to claim 1,
It includes a learning unit that receives power data collected for each load and learns the characteristics of the power data for each load,
It includes a classification unit that classifies the total power data measured in real time into power data for each load based on the learning results by the learning unit,
An industrial power control device comprising a pattern recognition unit that determines the normal operation pattern for each individual load or the normal pattern of the entire industrial facility including a plurality of loads, based on the learning by the learning unit and the classification by the classification unit.
According to claim 1,
It includes a classification unit that classifies the total power data measured in real time of the lower distribution board into power data for each load,
It includes a CVRf estimation unit that estimates or tracks the Conservation Voltage Reduction factor (CVRf) for each load in real time, based on the power data for each individual load classified by the classification unit,
The conservation voltage drop coefficient (CVRf) is the power change rate relative to the voltage change rate,
The CVRf estimator is an industrial power control device that predicts CVRf for each load of each voltage regulator or estimates the weighted average value of CVRf in real time.
According to claim 1,
If there are multiple lower-level switchboards in a lower-level system and multiple lower-level switchboards are branched from the higher-level switchboards in the upper-level system located upstream,
The power prediction unit is an industrial power control device that collects power prediction values for each load of the sub-system and predicts the total future power of the upstream sub-system or the upstream upper system.
According to claim 1,
It includes a CVRf estimation unit that estimates or tracks the Conservation Voltage Reduction factor (CVRf) for each load in real time, based on power data for each load.
Based on the CVRf prediction of the CVRf estimation unit, when controlling the peak power of the lower distribution panel, predict how much voltage to control the voltage regulator for which load of which lower system, and predict the degree of peak power control of the upper system. An industrial power control device including a scheduling unit that sets.
According to clause 18,
The scheduling unit,
Considering the conservation voltage drop coefficient (CVRf) for each load calculated from the CVRf estimation unit and the total conservation voltage drop coefficient of the upstream where power data is collected from the weighted average of the conservation voltage drop coefficients,
For each load, the adjustment amount is added to a given voltage value in the order of increasing CVRf,
An industrial power control device that expands the voltage adjustment range of the voltage regulator of the lower system to be adjusted until the peak control amount required by the upper system is achieved.
According to claim 1,
It includes a power prediction unit that predicts future power of the lower distribution board,
When controlling the peak power of the lower switchboard, it includes a scheduling unit that sets a schedule to predict the degree of peak power control of the upper system,
The scheduling unit compares the power prediction by the power prediction unit with the actual generated power value, and if the power prediction is outside the preset reliability, adjusts the schedule to restore reliability,
An industrial power control device in which the schedule is adjusted for each load.

Images