Iot-based photovoltaic power station intelligent data monitoring system

Abstract

The application discloses a photovoltaic power station intelligent data monitoring system based on an internet of things, relates to the technical field of photovoltaic operation monitoring, and improves the accuracy of photovoltaic component anomaly detection. Multidimensional data sets are extracted from historical component operation records, and a plurality of isolation detection trees are established. The multidimensional data sets are input into the isolation detection trees, and a plurality of normal numerical value combinations and abnormal numerical value combinations are generated. A component operation analysis network is established according to the normal numerical value combinations and the abnormal numerical value combinations. Real-time component operation records are input into the component operation analysis network, and prediction operation numerical value intervals of a plurality of kinds of operation data corresponding to photovoltaic components in the next operation detection period are generated. Operation monitoring is performed on the photovoltaic components according to the prediction operation numerical value intervals.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of photovoltaic operation monitoring, and particularly relates to an intelligent data monitoring system for a photovoltaic power station based on Internet of Things. BACKGROUND

[0002] With the increasing demand for clean energy worldwide, photovoltaic power generation, as a sustainable energy solution, is gradually becoming an important part of the energy field. Photovoltaic modules, as the core equipment of photovoltaic power generation systems, stable operation is crucial to ensure the efficiency and reliability of photovoltaic power generation. However, photovoltaic modules are affected by various factors during actual operation, such as environmental temperature, light intensity, humidity, etc., which may cause the performance of photovoltaic modules to decline or even fail.

[0003] The traditional photovoltaic module operation monitoring method mainly relies on manual inspection and simple sensor data collection, which is not only inefficient, but also difficult to find potential problems in photovoltaic modules in real time and accurately. In addition, due to the complexity and multidimensionality of photovoltaic module operation data, traditional data analysis methods are difficult to effectively mine the rules behind the data, and cannot accurately predict and evaluate the operation state of photovoltaic modules. Therefore, an intelligent data monitoring system for a photovoltaic power station based on Internet of Things is provided. SUMMARY

[0004] In order to solve the above technical problems, the purpose of the present application is to provide an intelligent data monitoring system for a photovoltaic power station based on Internet of Things.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme:

[0006] The intelligent data monitoring system for a photovoltaic power station based on Internet of Things comprises an Internet of Things sensing end and an edge detection end.

[0007] The Internet of Things sensing end is used to set an operation detection period, and then collect a plurality of operation data and environmental parameters of photovoltaic modules in each operation detection period through various sensors, and generate historical module operation records and real-time module operation records, and simultaneously perform operation monitoring on the photovoltaic modules according to the predicted operation value interval from the edge detection end.

[0008] The edge detection end is used to extract a multidimensional data set from the historical module operation records, and establish a plurality of isolation detection trees, input the multidimensional data set into the isolation detection trees, and then generate a plurality of normal value combinations and abnormal value combinations.

[0009] A component operation analysis network is established based on normal and abnormal value combinations. Real-time component operation records are input into the component operation analysis network to generate predicted operation value ranges for several types of photovoltaic component operation data for the next operation detection cycle, and abnormal predictions are marked for the predicted operation value ranges.

[0010] Furthermore, the process of collecting the operational data and environmental parameters includes:

[0011] The IoT sensing terminal assigns a number to each photovoltaic module in the photovoltaic power station, where the numbers are a1, a2, ..., a n n is a natural number greater than 0;

[0012] Multiple sensors of the same model are installed on each photovoltaic module, and an operation detection cycle is set for all sensors. At the beginning of each operation detection cycle, each sensor collects multiple operation data and environmental data of the photovoltaic module to which it is located.

[0013] At the end of each operational monitoring cycle, the IoT sensing device assigns a number to the data collected by each sensor and integrates the data with the same number into a component operation record.

[0014] Furthermore, the process of establishing the isolation detection tree includes:

[0015] From all historical operation records of each photovoltaic module, one type of historical environmental data is selected sequentially and combined with other historical operation data to generate a multidimensional dataset S, wherein the multidimensional dataset S = [s1, s2, ..., s...]. i h j ], s i and h j Let i and j represent the i-th type of operational data and the j-th type of environmental data, respectively, where i and j are natural numbers greater than 10.

[0016] Several isolated detection trees are established based on the total number of historical module operation records of photovoltaic modules. One type of operation data is randomly selected and combined with historical environmental data in the multidimensional dataset as a detection feature dimension combination for an isolated detection tree. Dynamic segmentation threshold combinations and positive and negative deviation unit values ​​are set for isolated detection trees with the same detection feature dimension combination. The magnitude of the positive and negative deviation unit value is ±1% of the initial value of the dynamic segmentation threshold combination.

[0017] The isolation detection tree consists of one root node and several leaf nodes.

[0018] Further, the dynamic partition threshold value combination initial value of each isolation detection tree is the intermediate value of the corresponding category environment data and operation data in the historical component operation record, that is, the same kind of historical operation data or historical environment data in the historical component operation record is sorted by size, and the intermediate value of the sorting result is selected as the dynamic partition threshold value combination initial value.

[0019] Further, the generation process of the normal value combination and the abnormal value combination includes:

[0020] Each multidimensional data set is sequentially input into each isolation detection tree, and then iterative operation is performed through the isolation forest algorithm, so that each root node of the isolation detection tree only carries one multidimensional data set, and the multidimensional data sets in each isolation detection tree are different;

[0021] In the iterative operation process, each time the isolation detection tree reaches the root node through a multidimensional data set by a dynamic partition threshold value combination, the corresponding dynamic partition threshold value combination is set to be unselectable, and then the dynamic partition threshold value combinations of other isolation detection trees are adjusted by a positive or negative deviation unit value until a multidimensional data set that has not reached the root node reaches the root node;

[0022] The abnormal threshold value of each isolation detection tree is obtained, the abnormal judgment threshold value is set, and then the size relationship between the abnormal threshold value of each isolation detection tree and the abnormal judgment threshold value is judged. If the abnormal threshold value is greater than or equal to the abnormal judgment threshold value, it is judged that the corresponding dynamic partition threshold value combination is an abnormal value combination;

[0023] If the abnormal threshold value is less than the abnormal judgment threshold value, it is judged that the corresponding dynamic partition threshold value combination is a normal value combination;

[0024] The above process of generating a normal value combination and an abnormal value combination is repeated, and then the normal value combination and the abnormal value combination between different environment data and various operation data are sequentially obtained.

[0025] Further, the establishment process of the component operation analysis network includes:

[0026] A multidimensional coordinate system is established, the abnormal value combination and the normal data combination of various environment data are mapped in the same multidimensional coordinate system, and the environment data in each abnormal value combination or normal data combination is the head node, and the rest is the tail node of the same level with the operation data. The data connection tree of the corresponding abnormal value combination or normal data combination is generated;

[0027] According to the corresponding data value of the head node of each data connection tree, each data connection tree is fitted and clustered, and a data connection network of the corresponding environment data and operation data combination is generated according to the fitting and clustering;

[0028] The fitting clustering is that according to the same environment data, the head nodes are sequentially connected according to the numerical degree of the head nodes, and the tail nodes of different data connection trees are synchronously connected according to the head node connection process.

[0029] Further, the abnormal numerical combination and the normal data combination corresponding to the data connection network under each different environment data are fitted and clustered to obtain the component operation analysis network of the corresponding photovoltaic component, and different center frame selection radii are set for various operation data and environment data in the component operation analysis network.

[0030] Further, the establishment process of the prediction operation numerical interval of the next operation detection cycle includes:

[0031] When an operation detection cycle starts, the real-time component operation record of each photovoltaic component at the end of the last operation detection cycle is input into the corresponding component operation analysis network;

[0032] Further, the component operation analysis network extracts a plurality of real-time environment data from the real-time component operation record, takes each real-time environment data as a center value, and calls the environment data within the center frame selection radius from the component operation analysis network, which is recorded as detection environment data;

[0033] Further, the detection environment data is called to generate the state detection network under the corresponding real-time environment data by clustering and fitting all data connection trees, and a plurality of prediction operation numerical intervals of various operation data are segmented from the state detection network according to the center frame selection radius of various operation data and taking the real-time operation data as a center value, if the prediction operation numerical interval contains an abnormal numerical combination interval segment, an abnormal prediction mark is set on the interval segment in the prediction operation numerical interval corresponding to the time interval, otherwise no operation is performed;

[0034] According to the number carried by the real-time component operation record, the prediction operation numerical interval of various operation data is numbered.

[0035] Further, the process of monitoring the operation of the photovoltaic component according to the prediction operation numerical interval includes:

[0036] In the next operation detection cycle, the Internet of Things sensing end calls the prediction operation numerical interval according to the number to monitor the operation of each photovoltaic component, and performs maintenance operation on the photovoltaic component in the time interval corresponding to the interval segment with the abnormal prediction mark until the corresponding operation detection cycle ends.

[0037] Compared with the prior art, the beneficial effects of the present application are:

[0038] 1、The present application can accurately identify normal value combinations and abnormal value combinations and generate predicted running value intervals for the next running detection period by establishing isolated detection trees and component running analysis networks, deeply mining and analyzing historical component running records, thereby providing strong decision support for the operation and maintenance of photovoltaic components.

[0039] 2、The present application can discover potential faults of photovoltaic components in advance, take timely measures for maintenance and replacement, avoid further expansion of faults, and reduce operation and maintenance costs and downtime losses of photovoltaic power stations by performing abnormal prediction labeling on the predicted running value intervals. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The system structure diagram of the present application. DETAILED DESCRIPTION

[0041] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific embodiments, structures, features and effects of the present application are described in detail below in combination with the drawings and preferred embodiments.

[0042] As shown in Figure 1 , the photovoltaic power station intelligent data monitoring system based on the Internet of Things includes an Internet of Things sensing end and an edge detection end.

[0043] The Internet of Things sensing end is used to set a running detection period, and then collect multiple running data and environmental parameters of photovoltaic components in each running detection period through various sensors, generate historical component running records and real-time component running records, and simultaneously perform running monitoring on photovoltaic components according to the predicted running value intervals from the edge detection end.

[0044] The edge detection end is used to extract a multidimensional data set from the historical component running records, establish a plurality of isolated detection trees, input the multidimensional data set into the isolated detection trees, and then generate a plurality of normal value combinations and abnormal value combinations.

[0045] The component running analysis network is established according to the normal value combinations and abnormal value combinations, the real-time component running records are input into the component running analysis network, and then the predicted running value intervals of the corresponding photovoltaic component running data in the next running detection period are generated, and the predicted running value intervals are labeled for abnormal prediction.

[0046] Further, the Internet of Things sensing end sets numbers for each photovoltaic component in the photovoltaic power station, wherein the numbers are a1, a2, …, a n n is a natural number greater than 0.

[0047] A plurality of same type sensors are installed on each photovoltaic module, and the types of the sensors include voltage sensors, current sensors, temperature sensors, and light intensity sensors, and all the sensors are connected to the Internet of Things sensing end, and then the Internet of Things sensing end divides the sensors on the photovoltaic module into operation sensors and environmental sensors;

[0048] A running detection cycle is set for all the sensors, and each sensor collects a plurality of running data and environmental data of the photovoltaic module where the sensor is located when a running detection cycle starts;

[0049] The running data includes voltage running values and current running values, and the environmental data includes light intensity change values and temperature change values;

[0050] The Internet of Things sensing end sets a number for the collected data of each sensor when a running detection cycle ends, and integrates the data with the same number into one component running record.

[0051] Further, the Internet of Things sensing end synchronizes the component running records of each running detection cycle to the edge detection end;

[0052] The edge detection end is provided with an anomaly detection module and a running prediction module;

[0053] The anomaly detection module is used to extract a plurality of dimensional data sets from historical component running records, and establish a plurality of isolation detection trees, input the plurality of dimensional data sets into the isolation detection trees, and then generate a plurality of normal value combinations and abnormal value combinations, and then establish a component running analysis network according to the normal value combinations and the abnormal value combinations, and the specific process includes:

[0054] From all the historical component running records of each photovoltaic module, a kind of historical environmental data is selected to generate a plurality of dimensional data sets S with other historical running data in sequence, the plurality of dimensional data sets S=[s1, s2, …, s i , h j ], s i and h j represent the i-th running data and the j-th environmental data, and i and j are natural numbers greater than 10;

[0055] A plurality of isolation detection trees are established according to the total number of the historical component running records of the photovoltaic module, and a kind of running data and the historical environmental data in the plurality of dimensional data sets are randomly selected as a detection feature dimension combination of an isolation detection tree, and a dynamic segmentation threshold combination and a positive and negative deviation unit value are set for the isolation detection trees with the same detection feature dimension combination, and the size of the positive and negative deviation unit value is ±1% of the initial value of the dynamic segmentation threshold combination;

[0056] The isolation detection tree is composed of one root node and a plurality of leaf nodes;

[0057] It should be noted that the initial value of the dynamic partition threshold combination of each isolation detection tree is the median value of the corresponding category of environmental data and running data in the historical component running record, that is, the same kind of historical running data or historical environmental data in the historical component running record is sorted by size, and the median value of the sorting result is selected as the initial value of the dynamic partition threshold combination;

[0058] Each multi-dimensional data set is sequentially input into each isolation detection tree, and then iteratively operated by the isolation forest algorithm, so that each root node of each isolation detection tree has only one multi-dimensional data set, and the multi-dimensional data sets in each isolation detection tree are different;

[0059] During the iterative operation process, each time a dynamic partition threshold combination is used to make the isolation detection tree pass through a multi-dimensional data set to reach the root node, the corresponding dynamic partition threshold combination is set to an unselectable value, and then the dynamic partition threshold combinations of other isolation detection trees are adjusted by a positive or negative deviation unit value until a multi-dimensional data set that has not reached the root node reaches the root node.

[0060] Obtain the anomaly threshold value of each isolation detection tree, and the calculation formula of the anomaly threshold value is:

[0061] ;

[0062] Wherein represents the anomaly threshold value of the kth isolation detection tree, is a correction parameter, K is the total number of the historical component running record, represents the average path length in the kth isolation detection tree, represents the path length of the xth multi-dimensional data set in the kth isolation detection tree.

[0063] Set an anomaly judgment threshold, and then judge the size relationship between the anomaly threshold value of each isolation detection tree and the anomaly judgment threshold. If the anomaly threshold value is greater than or equal to the anomaly judgment threshold, it is judged that the corresponding dynamic partition threshold combination is an abnormal value combination;

[0064] If the anomaly threshold value is less than the anomaly judgment threshold, it is judged that the corresponding dynamic partition threshold combination is a normal value combination;

[0065] Repeat the above process of generating normal value combinations and abnormal value combinations, and then sequentially obtain the normal value combinations and abnormal value combinations between different environmental data and various running data.

[0066] Further, a multi-dimensional coordinate system is established, and the abnormal value combinations and normal data combinations under various environmental data are mapped in the same multi-dimensional coordinate system, and the environmental data in each abnormal value combination or normal data combination is taken as a head node, and the rest is taken as a tail node of the same level as the running data to generate a data connection tree corresponding to the abnormal value combination or normal data combination;

[0067] According to the corresponding data values of the head nodes of each data connection tree, fitting clustering is performed on each data connection tree, and a data connection network corresponding to the combination of environmental data and running data is generated according to the fitting clustering;

[0068] The fitting clustering is to sequentially connect the head nodes according to the distribution of the same kind of environmental data in the multi-dimensional coordinate system and the degree of the head node values, and to synchronously connect the tail nodes of different data connection trees according to the head node connection process;

[0069] Further, the data connection networks corresponding to the abnormal value combinations and normal data combinations under each different environmental data are fitted and clustered to obtain a component running analysis network corresponding to the photovoltaic component, and different center frame radii are set for various running data and environmental data in the component running analysis network;

[0070] It should be noted that when each running detection period ends, the abnormality detection module updates the component running analysis network according to the component running records from the Internet of Things sensing end.

[0071] Further, the running prediction module is used to input the real-time component running records into the component running analysis network, and then generate a predicted running value interval of the corresponding running data of the photovoltaic component in the next running detection period, and mark the abnormality of the predicted running value interval, and the specific process includes:

[0072] The abnormality detection module synchronizes the component running analysis network with the running prediction module, and then when each running detection period starts, the running prediction module inputs the real-time component running records of each photovoltaic component at the end of the last running detection period into the corresponding component running analysis network;

[0073] Further, the component running analysis network extracts a plurality of real-time environmental data from the real-time component running records, takes each real-time environmental data as a center value, and retrieves the environmental data within the center frame radius from the component running analysis network, which is recorded as the detected environmental data;

[0074] Further, according to the detected environment data, the data connection tree is called, all data connection trees are clustered and fitted to generate a state detection network corresponding to real-time environment data, and according to the center frame radius of various running data, a prediction running value interval of the running data is segmented from the state detection network with the real-time running data as the center value, if the prediction running value interval has an abnormal value combination interval segment, an abnormal prediction label is set on the interval segment in the prediction running value interval corresponding to the time interval, otherwise no operation is performed;

[0075] According to the number carried by the real-time component running record, the prediction running value interval of various running data is labeled and sent to the Internet of Things sensing end;

[0076] Further, in the next running detection period, the Internet of Things sensing end calls the prediction running value interval to monitor the running of each photovoltaic component according to the number, and performs maintenance operation on the photovoltaic component in the time interval corresponding to the interval segment with abnormal prediction label until the corresponding running detection period ends.

[0077] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the present application, and any indirect modification, equivalent change and modification of the above embodiment according to the technical essence of the present application still belong to the scope of the present application.

Claims

1. An intelligent data monitoring system for a photovoltaic power station based on the Internet of Things, characterized in that, The IoT sensing end and the edge detection end are included; The IoT sensing end is used for setting a running detection period, collecting multiple running data and environmental parameters of the photovoltaic module in each running detection period through various sensors, and generating historical module running records and real-time module running records, while monitoring the running of the photovoltaic module according to a predicted running value interval from the edge detection end; The edge detection end is used for extracting a multidimensional data set from the historical module running records, and establishing a plurality of isolated detection trees, inputting the multidimensional data set into the isolated detection trees, and then generating a plurality of normal value combinations and abnormal value combinations; A module running analysis network is established according to the normal value combinations and the abnormal value combinations, and the real-time module running records are input into the module running analysis network, and then a predicted running value interval corresponding to a plurality of running data of the photovoltaic module in the next running detection period is generated, and the predicted running value interval is marked as an abnormal prediction; The establishment process of the isolated detection tree includes: From all the historical module running records of each photovoltaic module, a kind of historical environmental data is selected to generate a multidimensional data set with other historical running data; A plurality of isolated detection trees are established according to the total number of historical module running records of the photovoltaic module, and a kind of running data and the historical environmental data in the multidimensional data set are randomly selected as a detection feature dimension combination of an isolated detection tree, and a dynamic segmentation threshold combination and a positive and negative deviation unit value are set for the isolated detection trees with the same detection feature dimension combination; The isolated detection tree is composed of one root node and a plurality of leaf nodes; The generation process of the normal value combination and the abnormal value combination includes: The multidimensional data set is input into each isolated detection tree in turn, and iterative operation is performed through the isolated forest algorithm, so that the root node of each isolated detection tree only has one multidimensional data set, and the multidimensional data sets in each isolated detection tree are different; During the iterative operation process, whenever a dynamic segmentation threshold combination causes the isolated detection tree to reach the root node through a multidimensional data set, the corresponding dynamic segmentation threshold combination is set to an unselectable value, and then the dynamic segmentation threshold combinations of other isolated detection trees are adjusted through the positive and negative deviation unit values, until there is a multidimensional data set that has not reached the root node reaching the root node; The abnormal threshold values of each isolated detection tree are obtained, an abnormal judgment threshold is set, the size relationship between the abnormal threshold values of each isolated detection tree and the abnormal judgment threshold is judged, if the abnormal threshold value is greater than or equal to the abnormal judgment threshold, the corresponding dynamic segmentation threshold combination is judged as an abnormal value combination, and if the abnormal threshold value is less than the abnormal judgment threshold, the corresponding dynamic segmentation threshold combination is judged as a normal value combination; The establishment process of the module running analysis network includes: A multidimensional coordinate system is established, the abnormal value combinations and the normal data combinations under various environmental data are mapped in the same multidimensional coordinate system, and the environmental data in each abnormal value combination or normal data combination is taken as a head node, and the remaining running data are taken as tail nodes of the same level to generate a data connection tree of the corresponding abnormal value combination or normal data combination; According to the head node of each data connection tree corresponding to the data value, fitting clustering is performed on each data connection tree, and a data connection network corresponding to the combination of environmental data and operation data is generated according to the fitting clustering; The data connection network corresponding to the abnormal value combination and the normal data combination under each different environmental data is fitted and clustered to obtain a component operation analysis network corresponding to the photovoltaic component, and different center frame selection radii are set for various operation data and environmental data in the component operation analysis network. 2.The IoT-based intelligent data monitoring system for photovoltaic power stations according to claim 1, characterized in that, The collection process of the operation data and the environmental parameters includes: Each photovoltaic component in the photovoltaic power station is numbered, and the number is a1, a2, …, an, n is a natural number greater than 0; A plurality of sensors of the same type are installed on each photovoltaic component, and an operation detection period is set for all sensors. When an operation detection period starts, the sensors collect a plurality of operation data and environmental data of the photovoltaic component, the Internet of Things sensing end numbers the collected data of each sensor, and integrates the data with the same number into a component operation record. 3.The IoT-based intelligent data monitoring system for photovoltaic power stations according to claim 2, characterized in that, The same kind of historical operation data or historical environmental data in the historical component operation record is sorted by size, and the middle value of the sorting result is selected as the initial value of the dynamic segmentation threshold value combination. 4.The IoT-based photovoltaic power station intelligent data monitoring system according to claim 3, characterized in that, The establishment process of the prediction operation value interval of the next operation detection period includes: When an operation detection period starts, the real-time component operation record of each photovoltaic component at the end of the previous operation detection period is input into the corresponding component operation analysis network; The component operation analysis network extracts a plurality of real-time environmental data from the real-time component operation record, takes each real-time environmental data as a center value, and calls the environmental data within the center frame selection radius from the component operation analysis network, which is recorded as detection environmental data; According to the detection environmental data, the data connection tree is called, the clustering fitting is performed on all data connection trees to generate a state detection network corresponding to the real-time environmental data, and according to the center frame selection radius of various operation data, the real-time operation data is taken as the center value, and a plurality of prediction operation value intervals of the operation data are segmented from the state detection network. If the prediction operation value interval contains an abnormal value combination interval segment, an abnormal prediction mark is set on the interval segment in the prediction operation value interval corresponding to the time interval, otherwise no operation is performed; According to the number carried by the real-time component operation record, the prediction operation value interval of various operation data is numbered. 5.The IoT-based intelligent data monitoring system for photovoltaic power stations according to claim 4, characterized in that, The process of monitoring the operation of the photovoltaic component according to the prediction operation value interval includes: In the next operation detection period, the Internet of Things sensing end calls the prediction operation value interval to monitor the operation of each photovoltaic component according to the number, and performs maintenance operation on the photovoltaic component in the time interval corresponding to the interval segment with the abnormal prediction mark until the corresponding operation detection period ends.

Images