JP2018037078A - Power generation monitoring system with failure detection function - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an optimal failure determination method and a device therefor, to which deep learning is applied, for improving efficiency of O&M (Operation&Maintenance) work in O&M business relating to power generation.SOLUTION: A power generation monitoring system achieves safe and stable power generation and business profit by including a failure determination device for determining the presence/absence of a failure to which machine learning is applied with time series information such as a power generation amount and weather data as input and fault occurrence information as a teacher signal.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a failure determination method and apparatus in a power generation business. More specifically, in the O & M (Operation &Maintenance; hereinafter referred to simply as “O & M”) business related to power generation, the failure information that has occurred is used as a teacher signal, and the content of the failure is determined by a deep learning technique. An invention relating to a failure determination method and apparatus for predicting and detecting the above is configured, and an optimal monitoring system is realized.

  Needless to say, the trend of energy conservation and the use of natural energy is a global development. In Europe in particular, natural energy utilization projects that take environmental measures into consideration are supported by national policies. Meanwhile, in Japan, the Agency for Natural Resources and Energy in the Ministry of Economy, Trade and Industry has begun to take measures against various energy problems. In particular, with regard to electric power and gas energy, we have set a goal as an integrated reform of the energy system, and are implementing policies to realize it. Specifically, it is a policy that achieves two goals by creating a comprehensive energy market while removing the vertical market barriers. According to the Agency for Natural Resources and Energy, first, in order to increase the industry to drive growth in Japan, the creation of innovations such as the introduction of innovative technologies and the fusion of different services; To further improve the policy, it is advocated to promote policies such as free expansion of energy selection, maximum control of tariffs, ensuring stable supply and security.

Focusing on power system reforms related to power generation, the objectives are focused on ensuring a stable supply that does not rely on planned blackouts, etc., making appropriate electricity charges, expanding electricity generation options, and expanding business opportunities.

These policy driving forces and consumer needs have expanded the flexibility of the power generation business, enabling businesses to sell electricity as well as businesses.
Although the scale of business varies, after capital investment, we can expect a certain level of profit by operating the basic cycle of “creating”, “sending” and “selling” electricity without stagnation.

  As artificial power generation, those using natural energy such as geothermal power generation, wind power generation, hydroelectric power generation, and solar power generation are expected to continue in the future. It is an infrastructure of failure and recovery measures. In particular, solar power generation has been rapidly spreading, but on the other hand, problems with solar panels and poor construction have become apparent, and the quality disparity on facilities has been exposed. In the case of large-scale facilities, repair maintenance due to panel identification at the time of failure and aging deterioration of facilities is unavoidable in order to continue stable power generation business. Against this backdrop, as a national policy, it is fully anticipated that facility maintenance will become obligatory and stricter for sound business continuity. Work burden and running cost countermeasures are serious problems.

In other words, equipment maintenance, which is necessary to convert energy supply sources into electric power, is extremely important because it greatly affects the profitability of the entire business. Continuous monitoring of power generation is an inevitable problem for power generation companies, as it becomes more difficult as equipment becomes larger.

Moreover, even if an equipment abnormality is detected, it takes a lot of time and effort to identify and cause the failure, so a specialized operator, a so-called O & M operator, usually undertakes business maintenance and provides services. Many.

In this case, O & M operators have achieved certain results by operating and managing them in cooperation with data service operators. For example, the amount of power generated by a power conditioner (PCS) at a power plant and weather data observed locally are collected on a data server managed by the data service provider via a communication network, and operation information is remotely controlled. Provided to O & M operators while checking. Further, the data service provider provides a user interface that makes it easy for the power generation company to grasp the power generation status.

Among the power generation businesses, solar power generation is a particularly high-profile item. In the monitoring system in the photovoltaic power generation business, the inventors have developed a solar cell array failure detection device for the first time in the industry and have obtained a patent. For example, Patent Documents 1 to 4 are cited.

According to the present invention, by using the tester according to the present invention, the failure position of the solar cell array can be easily and quickly detected and directly inspected from the connection box installed on the ground. For example, inspection can be performed without climbing the roof, and safety can be ensured.

  In addition, when the scale and the monitoring scale increase, the burden of monitoring increases. Therefore, an embedded fault location detector has been developed and patented so that an automatic monitoring system can be constructed. Examples thereof include Patent Documents 5 to 10.

According to the present invention, as with the portable type, measurement is performed for each PV string, and the faulty array or cable connection failure location can be identified early. Remote monitoring is also possible, which can contribute to O & M cost reduction (Fig. 1).

Furthermore, in the case where a ground fault phenomenon due to equipment failure or the like is rarely observed, the inventors have developed a useful detector for early detection of the power generation amount. For example, patent document 11 is mentioned.

JP2011-35000A JP 2011-124401 A Japanese Patent Laying-Open No. 2015-204696 JP 2012-256771 A Japanese Patent Application No. 2012-170183 JP 2013-69974 A JP 2014-33184 A Japanese Patent Laying-Open No. 2015-251581 JP-A-2015-198539 Japanese Patent Application No. 2012-84447 Japanese Patent Laying-Open No. 2015-203564 Japanese Patent Laying-Open No. 2015-014819 JP 2006-048238 A JP2015-154918A JP2015-057630A Japanese Patent Laying-Open No. 2015-184864

  These series of inventions (Patent Documents 1 to 11) and the like have greatly improved the performance of the O & M business in the photovoltaic power generation business, but still have the following problems with respect to work efficiency.

The power reduction factor in photovoltaic power generation is not only due to damage to the solar panel, but also other factors such as reduction in solar radiation, shadow effects, contamination of the solar panel, deterioration of conversion efficiency of the power conditioner (PCS), solar There are optical panel output performance degradation and cable degradation. For example, even if a slight change occurs when dead leaves, bird dung, etc. are attached to a solar panel or a cloud is applied, the output decreases due to the occurrence of uneven energy supply. Therefore, if the power service provider simply confirms the results of monitoring the output of the PCS by the data service provider, it cannot identify the cause of the output reduction or determine the failure. There is a problem that it cannot be performed efficiently (FIG. 2).

Accordingly, an object of the present invention is to increase the efficiency of O & M work in an O & M (Operation & Maintenance) business related to power generation, and to provide a perfect maintenance plan, to predict sales from the installation conditions and power generation amount prediction, failure prediction It is an object of the present invention to provide an optimal failure determination method and apparatus for applying deep learning using conditions and the like required for a power generation facility in the relationship between data and power generation amount and profit correlation.

As a result of intensive investigations to solve the above problems, the present inventor has input time-series information such as power generation amount and weather data as an input, and uses a failure occurrence information as a teacher signal to determine whether there is a failure. A desired effect was found by interposing a discriminator, and the present invention was completed.

That is, according to the present invention, first, there is a failure determination method in an O & M (Operation & Maintenance) business related to power generation, which provides failure occurrence information at a work site to a terminal of a data service provider through a communication device. Improve the ability to predict and detect the details of the failure that occurred after the data distribution step and the failure occurrence information received as a teacher signal at the terminal of the data service provider by using deep learning techniques. A failure determination method including a determination step of determining a failure based on a function is provided (hereinafter, simply referred to as “first invention”).

According to the present invention, secondly, there is a failure determination device in an O & M (Operation & Maintenance) business related to power generation, which provides failure occurrence information at a work site to a terminal of a data service provider through a communication device. Improve the ability to predict and detect the content of failure that occurred afterwards, using the deep learning method of the failure occurrence information received as a teacher signal at the terminal of the data service provider and the data service provider, There is provided a failure determination device including determination means for determining a failure based on a learning function (hereinafter, simply referred to as “second invention”).

The “failure determination device 18” in the present invention includes a “failure detection result communication device 20”, a failure detection training device (hereinafter referred to as “failure detection training device 22”) that improves the failure detection capability by learning, and a “failure detection result communication device”. As a result of deep learning performed by “20” and “failure detection training device 22”, a device that has acquired an advanced function for determining the presence or absence of a failure is used (FIGS. 3 to 5).

Here, the “failure detection result communicator 20” is the direct data of the failure occurrence information (failure location and date, etc.) when the O & M operator finds a failure during maintenance work at the power plant site. The “failure detection training device 22” has a function of transmitting to the service provider, and the data service provider receives time series information such as the amount of power generation and weather data obtained from the power generator, A machine failure occurs when time series information (characteristic data such as hours, minutes, days, and years) such as power generation and weather data given afterwards is input by machine learning using the occurrence information as a teacher signal. Thus, a failure determination apparatus capable of improving the ability to determine whether or not to perform a situation is constructed (FIG. 3).

  These series of procedures are organized as follows. (1) The O & M provider sends “failure occurrence information” to the data service provider. (2) “Failure occurrence information” is used as a teacher signal, and learning is performed using time-series information obtained from a power generation company as input. (3) Incorporation of a failure discrimination device reflecting the learning result.

As a result, the data service provider inputs a time series information such as the amount of power generation and weather data provided by the power generation company to the failure determination device constructed by the above method. It is possible to realize a photovoltaic power generation monitoring system to which a function for determining whether or not each panel is added.

As described above, in the present invention, in order to acquire a high-level failure determination function, a technique of deep learning (deep learning) is incorporated into the “fault detection training device 22”. The number of intermediate layers called “hidden layers” 30 between the “input layer” and the “output layer” is increased to form a multi-layer structure (FIG. 6). In addition, this multi-layered / multi-layered recognition is repeated randomly in several steps using this intermediate (hidden) layer, thereby extracting multiple features such as position, color, shape, texture, and overall image for more accurate discrimination. It is what makes it possible. In order to obtain a maintenance scheduling optimization function that reduces maintenance costs with high accuracy, it also provides high accuracy failure prediction using AI that combines a back-trackable rule-based engine and deep learning. Application of the algorithm is effective.

Deep learning application development has been seen in the past. For example, in the car industry, there is an invention applied to the development of a preventive safety system that prevents accidents by improving image processing ability. The said invention provides the approaching object detection system which detects an approaching object based on the captured image imaged by the imaging device (patent document 12). In addition, it is possible to improve the safety of the user by recognizing which building on the map the building included in the street view image is, and calculating the visibility and the visible range of the building. An example in which an image recognition technique using deep learning is applied to a navigation system is also disclosed (Patent Document 13). Patent Document 14 discloses an application example to a technique for detecting a lesion from a medical image. Furthermore, an application example to a behavior prediction device that estimates an emotion with respect to a user's operation status based on motion data that is an improvement in speech recognition (Patent Document 15) or a user's motion is disclosed (Patent Document). 16).

  However, as in the present invention, there is no example applied in the power generation business in order to acquire the failure discrimination function.

In particular, in the present invention, using the failure detection result communication device 20, the failure occurrence information includes the content of the failure and the time when repair is started (Ts) and the time when it is completed (Te). By notifying the operator, the failure determination device refers to time-series information such as the power generation amount and weather data separately obtained from the power generation operator, and performs failure detection training on the power generation amount and weather data at time Ts. The learning is input to the device 22 and becomes “failed” as a teacher signal. In addition, it has the feature that learning ability is improved by inputting power generation amount and weather data at time Te to the failure detection training device, and performing learning that becomes “no failure” as a teacher signal. This is a failure discriminating device 18 whose failure detection capability is improved as the number of failures detected in (1) increases.

As a result, in the present invention, for example, when the input data is an image, it is possible to determine abnormality such as a failure by converting the image data that has been image-recognized into power generation amount data. . Specific examples will be described in the following examples.

  As described above, according to the present invention, the following first and second inventions are provided.

(First invention) A failure determination method in an O & M (Operation & Maintenance) business relating to power generation, a data distribution step of providing failure occurrence information at a work site to a data service provider's terminal through a communication device, and the data Using the deep learning method, the failure occurrence information received as a teacher signal at the service provider's terminal is enhanced by the ability to predict and detect the details of the failure that occurred afterwards, and the failure is identified based on the learning function. A failure determination method including a determination step.
(Second Invention) A failure determination apparatus in an O & M (Operation & Maintenance) business relating to power generation, wherein distribution data receiving means for providing failure occurrence information at a work site to a data service provider's terminal by a communication device, The failure occurrence information received as a teacher signal at the data service provider's terminal is enhanced by the deep learning method to improve the ability to predict and detect the details of the failure that occurred afterwards. A failure discriminating device comprising discriminating means for discriminating.

  Here, in both the first invention and the second invention, the target O & M business is not particularly limited as long as it is a business related to power generation. However, the effect of the combined use with the failure detector developed by the inventor is not limited. Therefore, it is preferable to use it in a business related to solar power generation.

That is, a failure determination method characterized in that the O & M (Operation & Maintenance) business in the first invention relates to photovoltaic power generation is suitably provided, and the O & M (Operation & Maintenance) business in the second invention is A failure determination device characterized by being related to solar power generation is suitably provided.

  These series of inventions can further enhance convenience by connecting the map information and the navigation system.

For example, when the location information, power generation amount, and weather information of a solar power plant are displayed on the Google map, the data service that manages the Google map by the failure detection result communication device is the failure detection information in the field When transmitted to the business operator, the learning effect is exhibited by the present invention, the fault discrimination ability is improved, and a display notifying the occurrence of the fault is also displayed on the Google map. This display information enables a local person in charge of O & M (Operation & Maintenance) business to quickly start repairs. In addition, regional cooperation for each business site and work cooperation in remote areas in Japan and overseas are widely possible, and the monitoring system provided by the present invention can be accepted as an international business model.

Also, drone application is beneficial. For example, in a meteorological environment that continues at a high temperature, a temperature load is applied to the solar panel, and damage due to spontaneous ignition occurs. As a result, the damage is enormous and it is no longer possible for the operator to carry out extensive failure checks individually. In such a case, when the image data acquired by the drone is transmitted to the data service provider, the learning system of the present invention can be operated to provide a bird's-eye view of the location information that requires restoration of the power generation system. It provides the best performance for the response.

In this case, the applicable range is further expanded by appropriately incorporating an infrared sensor for sensing heat distribution in the drone. That is, since information on the temperature load can be obtained, it contributes not only to a reactive response but also a preventive effect.

Utilization of such drones and Google maps alone or in combination provides high performance even in daily routine management by O & M operators.

With the monitoring system with a fault detection function that can be configured by providing the present invention, the following business value and added value can be given to each business unit.
(1) Maintain high value-added power plants that can generate power safely and stabilize business profitability (total creative value)
(2) Improve profitability by improving the added value of inspection services and reducing work man-hours (creative value 1)
An O & M operator (company), using efficient failure detectors, leads periodic maintenance, for example, the solar power generation capacity is reduced due to disconnection of the internal aluminum wire It is possible to estimate the location of the optical panel and the location of disconnection between cables.
(3) Expand into the big data business by expanding the service menu and accumulating failure / inspection data (Creative Value 2)
The data service provider can collect the results obtained from the O & M business as a report and accumulate the data, thereby connecting to a new service to the power generation company such as predicting the occurrence of a failure.
(4) Reduction of power generation reduction, improvement of profitability, safety management of power plants, secondary disaster prevention and maintenance, and collateral value of power generation facilities can be maintained (Creative Value 3)
Since the power generation company can detect a failure with heat generation at an early stage, the solar power generation system can be operated safely and efficiently.

It is a figure which shows an example of the remote monitoring system of solar power generation equipment. It is a figure which shows the subject of the O & M business in a remote monitoring system. It is a figure which shows the problem solution of the O & M business in a remote monitoring system. It is a figure which shows an example of the monitoring system with a fault detection function. It is a figure which shows one example of construction of a failure discrimination device. It is a figure which shows an example of the learning function of a failure detection training device.

  Hereinafter, specific modes for carrying out the present invention will be described in detail, but the technical scope of the present invention is not limited to only the specific modes and examples disclosed below.

Hereinafter, an embodiment of the failure determination apparatus according to the present invention will be described with reference to FIG.
The failure detection result communication device 20 notifies the data service provider of the failure occurrence information, the content of the failure, the time Ts at which the repair was started, and the time Te at which the repair was completed. Here, the content of the failure, the time Ts at which the repair is started, and the time Te at which the repair is completed are collectively called failure detection data FD.
In the failure discriminating apparatus 18, a monitoring signal MD and failure detection data FD separately received from the power generation company are input as input information, and a teacher signal generation unit 27 is configured to generate a machine learning teacher signal. Here, time series information such as the amount of power generation and weather data is referred to as monitoring data MD.
The teacher signal generation unit 27 uses the time Ts that is an element of the failure detection data FD, and among the monitoring data MD stored in the monitoring data storage unit, the power generation amount at the time Ts0 that is the closest before the time Ts, Machine learning is performed by giving a teacher signal so that an output value when time series information such as weather data is inputted indicates “abnormal”. Further, machine learning is performed by giving a teacher signal so that the output value when the time series information such as the power generation amount and weather data at the nearest time Te0 after the time Te is input is “normal”.

When the monitoring data MD is input to the failure detection training device constructed as a result of generating the teacher signal and performing machine learning for the plurality of failure detection data FD in the above learning, the output obtained as a result As a result, a normal or abnormal determination result is obtained as a failure detection result.

Hereinafter, an embodiment of the fault detection training device 22 according to the present invention will be described with reference to FIG.
The failure detection training device 22 is used to input a teacher signal and perform machine learning using a neural network. It is possible to realize a fault detection trainer that improves the fault detection capability as the number of faults detected on site increases.

1 Solar radiation meter 2 Thermometer 3 PV string 4 PCS
5 Data Logger 6 Junction Box with Failure Detection Function 7 Data Center 8 Display Terminal 18 Failure Identification Device
DESCRIPTION OF SYMBOLS 19 Monitoring result communication apparatus 20 Failure detection result communication apparatus 21 Failure detection function monitoring system 22 Failure detection training device 23 Display information communication part 25 Monitoring data storage part 26 Teacher signal output part 27 Teacher signal generation part 30 Failure detection trainer (hidden) layer)

Claims (4)

A failure determination method in an O & M (Operation & Maintenance) business related to power generation, a data distribution step of providing failure occurrence information at a work site to a data service provider's terminal using a communication device, and a terminal of the data service provider Including a determination step of determining failure information based on the learning function by enhancing the ability to predict and detect the details of the failure that occurred afterwards using the deep learning technique for the failure occurrence information received as a teacher signal in Failure identification method. The failure determination method according to claim 1, wherein the O & M (Operation & Maintenance) business relates to photovoltaic power generation. A failure discriminating apparatus in O & M (Operation & Maintenance) business related to power generation, distribution data receiving means for providing failure occurrence information at a work site to a data service provider's terminal by a communication device, and the data service provider's Determining means for identifying failure based on the learning function by improving the ability to predict and detect the content of the failure that occurred afterwards using the deep learning method for failure occurrence information received as a teacher signal at the terminal A failure discrimination device. 4. The failure determination device according to claim 3, wherein the O & M (Operation & Maintenance) business relates to photovoltaic power generation.

Images