CN114610070B - Unmanned aerial vehicle-coordinated intelligent inspection method for wind farm - Google Patents
Unmanned aerial vehicle-coordinated intelligent inspection method for wind farm Download PDFInfo
- Publication number
- CN114610070B CN114610070B CN202210274635.5A CN202210274635A CN114610070B CN 114610070 B CN114610070 B CN 114610070B CN 202210274635 A CN202210274635 A CN 202210274635A CN 114610070 B CN114610070 B CN 114610070B
- Authority
- CN
- China
- Prior art keywords
- aerial vehicle
- unmanned aerial
- wind
- data
- wind power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000007689 inspection Methods 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000005265 energy consumption Methods 0.000 claims abstract description 33
- 238000013439 planning Methods 0.000 claims abstract description 32
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 29
- 230000002787 reinforcement Effects 0.000 claims abstract description 21
- 238000002922 simulated annealing Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 6
- 230000007613 environmental effect Effects 0.000 claims abstract description 5
- 230000009471 action Effects 0.000 claims description 35
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000000137 annealing Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000011835 investigation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 241000220317 Rosa Species 0.000 claims 1
- XOFYZVNMUHMLCC-ZPOLXVRWSA-N prednisone Chemical compound O=C1C=C[C@]2(C)[C@H]3C(=O)C[C@](C)([C@@](CC4)(O)C(=O)CO)[C@@H]4[C@@H]3CCC2=C1 XOFYZVNMUHMLCC-ZPOLXVRWSA-N 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000012423 maintenance Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
- H02J13/00002—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by monitoring
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Aviation & Aerospace Engineering (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides an unmanned plane coordinated wind power plant intelligent inspection method, and belongs to the technical field of data processing. According to the invention, the unmanned aerial vehicle carries the task load to complete the inspection work of the wind turbine generator, and the flight path of the unmanned aerial vehicle is planned through the deep reinforcement learning-simulated annealing algorithm model, so that the intelligent wind power plant inspection route planning with the lowest energy consumption is realized. The intelligent wind power station routing inspection route planning method fully considers the physical and environmental characteristics of the wind power station, realizes the intelligent wind power station routing inspection route planning with low energy consumption, has extremely strong adaptability, and can be applied to wind power stations with different geographic positions and terrains. The method not only considers the characteristic of being convenient for real-time charging in the wind power plant, but also fully considers the climate characteristics of the wind power plant, and innovatively takes the wind speed and the wind direction into consideration of unmanned aerial vehicle track planning; not only can adapt to wind power plants in different terrains and quaternary wind areas, but also can timely cope with sudden weather changes, dynamically adjusts the routing inspection route, and is very suitable for weather characteristics of the wind power plants.
Description
Technical Field
The invention belongs to the technical field of data processing, and relates to an unmanned plane-coordinated intelligent inspection method for a wind farm. According to the invention, the unmanned aerial vehicle carries the task load to complete the inspection work of the wind turbine generator, and the flight path of the unmanned aerial vehicle is planned through the deep reinforcement learning-simulated annealing algorithm model, so that the intelligent wind power plant inspection route planning with the lowest energy consumption is realized.
Background
With the rapid development of the wind power industry in China, the number of wind turbine generators is increased sharply. The natural environment at wind farms is often quite complex, potentially located in remote mountainous areas or offshore, and exposed to severe weather for long periods of time. The wind turbines are separated from each other by hundreds of meters, the turbine cabin is as high as tens of meters, and the manual inspection cost is high, the efficiency is low, the error rate is high, and the risk is high. How to reduce the operation and maintenance management cost of a large number of wind turbine generators, the intelligentization and informatization of a wind power system are promoted, so that the power generation income of a wind power plant is improved, and the method becomes a great problem to be solved urgently in the wind power industry.
With the proposal and introduction of the concept of intelligent wind power plants, wind power systems are being comprehensively optimized and upgraded in parts manufacturing, data management, operation and maintenance and other aspects. The intelligent wind farm is mainly based on measurement and control technology, communication technology, sensing technology, big data processing technology and various intelligent algorithms, and achieves the intelligentization of aspects such as fan control management, equipment state sensing, inspection maintenance and the like. The working state data of each device is obtained through various sensors in the wind turbine, the sensor data are integrated through an edge server in the wind turbine, whether the wind turbine is in a normal working state at present can be obtained, and the possible reasons of faults can be eliminated when the fan breaks down.
However, wind power systems have not achieved true intellectualization. The operation data of the fan equipment is still generally dumped and maintained by manpower; meanwhile, cracks on wind motors such as blades are difficult to confirm through sensors, and manual work is usually required to cooperate with a hanging basket or a high-power telescope for inspection. This results in the inspection system of wind power system still being time consuming and laborious and the rate of accuracy is low, is difficult to reach the operation requirement of wisdom wind power plant.
Due to the flexible and motorized nature, unmanned aerial vehicles are playing an increasingly important role in smart wind farms. The unmanned aerial vehicle carries the task load to finish the inspection work of the fan, so that the efficiency and the accuracy of the inspection work can be greatly improved, the danger of manpower inspection is reduced, and the economic benefit of power production is improved. Has great research significance and practical value.
Disclosure of Invention
The method solves the problem of how to plan the unmanned aerial vehicle inspection route by using meteorological data of the environment of the wind power place and using a deep reinforcement learning algorithm, so that the energy consumed by the planned route is minimum while traversing all wind power units. In the process, a wind turbine generator is photographed by adopting an image recognition method, so that the wind turbine generator is used for fault diagnosis of the wind turbine generator, and related data of the wind turbine generator are uploaded at the same time; time cost and labor cost in the personnel inspection process are saved, and safety of the wind turbine generator is improved. The invention realizes the intelligent inspection method of the wind power plant by the cooperation of the unmanned aerial vehicle, combines the technologies of deep reinforcement learning and the like, and finally provides theoretical basis and practical experience for the intelligent inspection field of the wind power plant by the unmanned aerial vehicle with low power consumption.
The technical scheme of the invention is as follows:
An unmanned aerial vehicle collaborative wind farm intelligent inspection method comprises an unmanned aerial vehicle route planning system based on deep reinforcement learning and simulated annealing, a wind farm fault detection based on unmanned aerial vehicles and a wind power data uploading system. The method comprises the following specific steps:
Step one: and acquiring weather forecast data of the wind farm for 4 hours in the future, and preprocessing the data.
Step two: for a plurality of wind turbine generators X= { X 1,x2,...,xn }, using a deep reinforcement learning algorithm to carry out minimum power consumption track planning and corresponding power consumption E ij on any two wind turbine generators X i、xj (i not equal to j) within the maximum cruising radius of the unmanned aerial vehicle.
Step three: and determining the starting position of the unmanned aerial vehicle according to the main wind direction of the predicted weather, and carrying out flight path planning based on a simulated annealing algorithm on the unmanned aerial vehicle according to the current weather data and the learning experience of the step two.
Step four: and 3, carrying out inspection on each unit of the wind power plant according to the track planned in the step three so as to facilitate fault investigation and data uploading.
The data preprocessing in the first step comprises the following specific steps:
step 1.1: and checking meteorological data, and if the meteorological data has a missing part, performing smoothing treatment on the missing data.
Step 1.2: for any time t, wind speed is determined according to wind direction data theta t Performing orthogonal decomposition to decompose wind speed into three wind speeds/>, which are perpendicular to each other, in three-dimensional space
Step 1.3: and carrying out data normalization processing on the wind direction and wind speed data.
In the second step, the construction steps of the deep reinforcement learning algorithm are as follows:
Step 2.1: firstly, a Markov decision process model of the unmanned aerial vehicle on the track planning of two wind turbine sets is established, five-tuple < S, A, P, R and gamma > in the process is determined, wherein S represents the current environmental state quantity of the unmanned aerial vehicle, A is the action quantity executed by the unmanned aerial vehicle, P is the transition probability among different states, R is the rewarding quantity obtained by the unmanned aerial vehicle executing the action A in the state S, and gamma is the reinforcement learning attenuation rate. The state quantity S should be able to fully represent the current state of the unmanned aerial vehicle, in the present invention, a three-dimensional coordinate system is established with the current position of the unmanned aerial vehicle, and the state quantity S includes the position coordinates Pos U = (x, y, z) where the current unmanned aerial vehicle is located, and the wind speed vector of the position where the unmanned aerial vehicle is located at the current time
Simplifying the unmanned plane motion model, wherein the executable action quantity A= < a > of the unmanned plane represents the speed vector of the unmanned planeMoving a distance in a fixed direction within the time slice tau. In order to enable the unmanned aerial vehicle to reach the target unit with the least power consumption, the following rewarding mode is designed:
Wherein, |d s′ | is the linear distance between the unmanned aerial vehicle and the target point after the action is executed, |d s | is the linear distance between the unmanned aerial vehicle and the target point before the action is executed, E ss′ is the energy consumption of the action executed at this time, and E max is the maximum energy consumption of the unmanned aerial vehicle. When the unmanned aerial vehicle approaches the target with less energy consumption, more rewards are obtained, and when the unmanned aerial vehicle reaches the target, a maximum rewards is obtained, so that the target can absorb the unmanned aerial vehicle.
For the energy consumption E ss′ required by each execution of the action of the unmanned aerial vehicle, the calculation formula E ss′＝Pu·τ.Pu is the power of the unmanned aerial vehicle, including the horizontal flight powerVertical flight power/>Resistance power/> Wherein w=mg is unmanned plane gravity; ρ is the air density; /(I)Is the total area of the rotor wing of the unmanned aerial vehicle; c D0 is the drag coefficient related to rotor geometry; and/> The relative speeds of the unmanned aerial vehicle to wind speed are respectively in the horizontal direction and the vertical direction; /(I)The speed of horizontal flight of the unmanned aerial vehicle; /(I)Hovering power for the drone.
Step 2.2: and initializing a playback experience pool D for storing data generated by the unmanned aerial vehicle in the error test process. Randomly initializing an Actor reality network mu and a Critic reality network Q, wherein parameters corresponding to the two neural networks are theta μ and theta Q respectively; and randomly initializing an Actor target network mu 'and a Critic target network Q', wherein parameters corresponding to the two neural networks are respectively theta μ′ and theta Q′, and enabling theta μ′＝θμ,θQ′＝θQ.
Step 2.3: the initial state quantity s 1 is recorded, and a random noise N conforming to a Gaussian distribution is generated.
Step 2.4: the state quantity x t at the current moment is input into an Actor reality network taking theta μ as a parameter, and random noise N t at the current moment is added. The action quantity a t＝μ(xt,θμ)+Nt is output by the Actor reality network, the action is executed, the rewards r t obtained by the action are calculated through a rewarding function, and meanwhile, the state quantity is updated to obtain x t+1.
Step 2.5: a quadruple < x t,at,rt,xt+1 > is created and stored in the playback experience pool D.
Step 2.6: randomly selecting a group of data < x j,aj,rj,xj+1 > from a playback experience pool D, inputting x j,aj into a Critic reality network to obtain Q=Q (x j,aj,θQ), inputting x j+1 into an Actor target network, calculating an action quantity a j+1＝μ′(xj+1,θμ′), inputting x j+1 and a j+1 into the Critic target network together to obtain Q (x j+1,aj+1,θQ), then taking Q 'as a label, training the Critic reality network to enable the calculated Q value to be infinitely close to the target value Q', and updating theta Q by using a gradient descent method.
Step 2.7: and updating the Actor reality network to ensure that the output action quantity is the maximum Q value calculated in the Critic reality network. Also, the gradient descent method is adopted to update theta μ, and the strategy gradient calculation method is as follows
Step 2.8: and updating the target network parameters theta [ mu ]' [ alpha ] theta μ′+(1-α)θμ,θQ′←αθQ′+(1-α)θQ.
Step 2.9: repeating the steps 2.3 to 2.8 until the loss values of the Actor target network and the Critic target network are converged, and the network parameters are unchanged. After the network converges, for two optional wind turbines x i、xj (i not equal to j), the deep reinforcement learning model gives the flight trajectory with the minimum power consumption and the power consumption E ij of the unmanned aerial vehicle.
In the third step, the specific steps of the simulated annealing algorithm are as follows:
Step 3.1: and determining the starting position of the unmanned aerial vehicle according to the current day wind direction rose diagram. If the dominant wind direction exists in the current wind power plant, setting the starting position of the unmanned aerial vehicle as a corner fan opposite to the dominant wind direction; and if the current dominant wind direction is not obvious, setting the starting position of the unmanned aerial vehicle as a fan at the central position.
Step 3.2: two neural networks of the day were trained. And (3) determining the flight path and the flight energy consumption between any two wind turbine generators x i、xj in the maximum cruising radius of the unmanned aerial vehicle according to the playback experience pool and the wind direction and the wind speed predicted by the next day weather in the second step.
Step 3.3: and starting from the initial position, sequentially selecting the lowest energy consumption track in the cruising radius of the unmanned aerial vehicle until all fans are traversed, and taking the lowest energy consumption track as an initialization path c. While initializing the start temperature T, the end temperature T 0, and the annealing speed α.
Step 3.4: by random thermal perturbation, another path c' is created in the neighborhood of c. Unlike conventional simulated annealing algorithms, this disturbance can only occur between fans located within the same cruising area.
Step 3.5: the difference deltae of unmanned energy consumption between the two paths c and c' is calculated. If Δe is less than or equal to 0, updating the path to let c=c', t++αt; otherwise, a random number rand between 0 and 1 is generated ifThe path is updated, letting c=c', t++αt.
Step 3.6: judging whether T is more than T 0 or not, if so, continuing to execute the step 3.4; otherwise, obtaining a near optimal solution of the routing inspection route planning with the lowest energy consumption, and sequentially overhauling the fan and acquiring data by the unmanned aerial vehicle according to the routing inspection route.
Step 3.7: and judging whether the routing inspection route needs to be updated or not based on the real-time meteorological data. When the real-time wind direction detected by the wind sensor in the wind power plant and the weather predicted wind direction are not in the same direction or the real-time wind speed and the weather predicted wind speed level are different by more than two stages, the wind turbine generator X' = { X 1,x2,...,xn′ } which is not traversed is counted. And (3.2) re-executing the step, and re-planning the minimum energy consumption flight path of the unmanned aerial vehicle of the remaining wind turbine. And (5) until the unmanned aerial vehicle traverses the wind turbine, ending the algorithm.
The invention has the beneficial effects that: compared with other unmanned aerial vehicle track planning methods, the method fully considers the physical and environmental characteristics of the wind power plant, realizes intelligent wind power plant routing inspection route planning with low energy consumption, has extremely strong adaptability, and can be applied to wind power plants with different geographic positions and terrains. According to the invention, the energy-saving inspection method in the intelligent wind power plant is developed to realize the planning of the inspection route of the unmanned aerial vehicle in the wind power plant, so that the characteristic of being convenient for real-time charging in the wind power plant is considered, the climate characteristic of the wind power plant is fully considered, and the wind speed and the wind direction are innovatively taken into consideration of unmanned aerial vehicle track planning; not only can adapt to wind power plants in different terrains and quaternary wind areas, but also can timely cope with sudden weather changes, dynamically adjusts the routing inspection route, and is very suitable for weather characteristics of the wind power plants.
Drawings
FIG. 1 is a view of a wind farm intelligent patrol scenario of the present invention.
FIG. 2 is a timing diagram of the intelligent patrol of the wind farm of the present invention.
FIG. 3 is a flow chart of the data preprocessing of the present invention.
Fig. 4 is a schematic diagram of the unmanned aerial vehicle track planning algorithm based on reinforcement learning.
Fig. 5 is a detailed design diagram of the reinforcement learning-based unmanned aerial vehicle track planning algorithm of the present invention.
Fig. 6 is a flow chart of a simulated annealing-based unmanned aerial vehicle track planning algorithm of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following detailed description of the specific embodiments of the present invention will be given with reference to the accompanying drawings.
FIG. 1 is a view of a wind farm intelligent patrol scenario of the present invention.
Fig. 2 is a timing chart of intelligent inspection of a wind farm, and the invention provides an unmanned plane cooperative intelligent inspection method of the wind farm, which specifically comprises the following steps:
Step one: and acquiring meteorological data of the wind farm for 4 hours in the future, and preprocessing the data.
Step two: for a plurality of wind turbine generators X= { X 1,x2,...,xn }, using a deep reinforcement learning algorithm to carry out minimum power consumption track planning on any two wind turbine generators X i、xj (i not equal to j) within the maximum cruising radius of the unmanned aerial vehicle, and calculating corresponding power consumption E ij.
Step three: and determining the starting position of the unmanned aerial vehicle according to the main wind direction of the predicted weather, and carrying out flight path planning based on a simulated annealing algorithm on the unmanned aerial vehicle according to the current weather data and the learning experience of the step two.
Step four: and 3, carrying out inspection on each unit of the wind power plant according to the planned flight path in the step three so as to facilitate fault investigation and data uploading.
Fig. 3 is a flow chart of data preprocessing in the first step of the present invention, specifically including the following steps:
step 1.1: and checking meteorological data, and if the meteorological data has a missing part, performing smoothing treatment on the missing data.
Step 1.2: for any time t, wind speed is determined according to wind direction data theta t Performing orthogonal decomposition to decompose wind speed into three wind speeds/>, which are perpendicular to each other, in three-dimensional space
Step 1.3: and carrying out normalization processing on the wind direction and wind speed data.
Before explaining the deep learning-simulated annealing algorithm of the present invention in detail, the following description is made on the problem:
Firstly, performing environment construction of simulation experiments according to physical parameters and geographic environments of a wind power plant, related parameters of an unmanned aerial vehicle and meteorological records of nearly three years, then sending the data into a neural network for training and predicting, analyzing the recognition capability of the neural network for different wind speeds and wind directions, and continuously adjusting the structure and parameters of the neural network according to experimental results so as to further improve the feasibility of the invention. The invention adopts a depth reinforcement learning algorithm based on an Actor-Critic framework as a main structure, and optimizes parameters to improve the adaptability of the algorithm. Before the unmanned aerial vehicle inspection starts every day, training the neural network according to weather forecast data of the same day. And according to the training result, calculating a near-optimal unmanned aerial vehicle energy-saving routing inspection route based on a simulated annealing algorithm. In the unmanned aerial vehicle inspection process, if the wind direction and the wind speed detected in real time have great difference with the weather forecast data, the remaining fans are subjected to track planning again according to the new weather forecast data, so that the final unmanned aerial vehicle inspection route is ensured to achieve the aim of low energy consumption.
Fig. 4 is a schematic diagram of an unmanned aerial vehicle path planning algorithm based on reinforcement learning.
Fig. 5 is a detailed design diagram of the reinforcement learning-based unmanned aerial vehicle path planning algorithm, and the process specifically includes the following steps:
Step 2.1: firstly, a Markov decision process model of the unmanned aerial vehicle on the track planning of two wind turbine sets is established, five-tuple < S, A, P, R and gamma > in the process is determined, wherein S represents the current environmental state quantity of the unmanned aerial vehicle, A is the action quantity executed by the unmanned aerial vehicle, P is the transition probability among different states, R is the rewarding quantity obtained by the unmanned aerial vehicle executing the action A in the state S, and gamma is the reinforcement learning attenuation rate. The state quantity S should be able to fully represent the current state of the unmanned aerial vehicle, in the present invention, a three-dimensional coordinate system is established with the current position of the unmanned aerial vehicle, and the state quantity S includes the position coordinates Pos U = (x, y, z) where the current unmanned aerial vehicle is located, and the wind speed vector of the position where the unmanned aerial vehicle is located at the current time
Simplifying the unmanned plane motion model, wherein the executable action quantity A= < a > of the unmanned plane represents the speed vector of the unmanned planeMoving a distance in a fixed direction within the time slice tau. In order to enable the unmanned aerial vehicle to reach the target unit with the least power consumption, the following rewarding mode is designed:
Wherein, |d s′ | is the linear distance between the unmanned aerial vehicle and the target point after the action is executed, |d s | is the linear distance between the unmanned aerial vehicle and the target point before the action is executed, E ss′ is the energy consumption of the action executed at this time, and E max is the maximum energy consumption of the unmanned aerial vehicle. When the unmanned aerial vehicle approaches the target with less energy consumption, more rewards are obtained, and when the unmanned aerial vehicle reaches the target, a maximum rewards is obtained, so that the target can absorb the unmanned aerial vehicle.
For the energy consumption E ss′ required by each execution of the action of the unmanned aerial vehicle, the calculation formula E ss′＝Pu·τ.Pu is the power of the unmanned aerial vehicle, including the horizontal flight powerVertical flight power/>Resistance power/> Wherein w=mg is unmanned plane gravity; ρ is the air density; /(I)Is the total area of the rotor wing of the unmanned aerial vehicle; c D0 is the drag coefficient related to rotor geometry; and/> The relative speeds of the unmanned aerial vehicle to wind speed are respectively in the horizontal direction and the vertical direction; /(I)The speed of horizontal flight of the unmanned aerial vehicle; /(I)Hovering power for the drone.
Step 2.2: and initializing a playback experience pool D for storing data generated by the unmanned aerial vehicle in the error test process. Randomly initializing an Actor reality network mu and a Critic reality network Q, wherein parameters corresponding to the two neural networks are theta μ and theta Q respectively; and randomly initializing an Actor target network mu 'and a Critic target network Q', wherein parameters corresponding to the two neural networks are respectively theta μ′ and theta Q′, and enabling theta μ′＝θμ,θQ′＝θQ.
Step 2.3: the initial state quantity s 1 is recorded, and a random noise N conforming to a Gaussian distribution is generated.
Step 2.4: the state quantity x t at the current moment is input into an Actor reality network taking theta μ as a parameter, and random noise N t at the current moment is added. The action quantity a t＝μ(xt,θμ)+Nt is output by the Actor reality network, the action is executed, the rewards r t obtained by the action are calculated through a rewarding function, and meanwhile, the state quantity is updated to obtain x t+1.
Step 2.5: a quadruple < x t,at,rt,xt+1 > is created and stored in the playback experience pool D.
Step 2.6: randomly selecting a group of data < x j,aj,rj,xj+1 > from a playback experience pool D, inputting x j,aj into a Critic reality network to obtain Q=Q (x j,aj,θQ), inputting x j+1 into an Actor target network, calculating an action quantity a j+1＝μ′(xj+1,θμ′), inputting x j+1 and a j+1 into the Critic target network together to obtain Q (x j+1,aj+1,θQ), then taking Q 'as a label, training the Critic reality network to enable the calculated Q value to be infinitely close to the target value Q', and updating theta Q by using a gradient descent method.
Step 2.7: and updating the Actor reality network to ensure that the output action quantity is the maximum Q value calculated in the Critic reality network. Also, the gradient descent method is adopted to update theta μ, and the strategy gradient calculation method is as follows
Step 2.8: and updating the target network parameters theta μ′←αθμ′+(1-α)θμ,θQ′←αθQ′+(1-α)θQ.
Step 2.9: repeating the steps 2.3 to 2.8 until the loss values of the Actor target network and the Critic target network are converged, and the network parameters are unchanged. After the network converges, for two optional wind turbines x i、xj (i not equal to j), the deep reinforcement learning model gives the flight trajectory with the minimum power consumption and the power consumption E ij of the unmanned aerial vehicle.
Fig. 6 is a flow chart of an unmanned aerial vehicle track planning algorithm based on simulated annealing, which specifically comprises the following steps:
step 3.1: and determining the starting position of the unmanned aerial vehicle according to the current weather direction rose diagram. If the dominant wind direction exists in the current wind power plant, setting the starting position of the unmanned aerial vehicle as a corner fan opposite to the dominant wind direction; and if the current dominant wind direction is not obvious, setting the starting position of the unmanned aerial vehicle as a fan at the central position.
Step 3.2: the neural network of the current day is trained. And (3) determining the flight path and the flight energy consumption between any two wind turbine generators x i、xj in the maximum cruising radius of the unmanned aerial vehicle according to the playback experience pool and the wind direction and the wind speed predicted by the next day weather in the second step.
Step 3.3: and starting from the initial position, sequentially selecting the lowest energy consumption track in the cruising radius of the unmanned aerial vehicle until all fans are traversed, and taking the lowest energy consumption track as an initialization path c. While initializing the start temperature T, the end temperature T 0, and the annealing speed α.
Step 3.4: by random thermal perturbation, another path c' is created in the neighborhood of c. Unlike conventional simulated annealing algorithms, this disturbance can only occur between fans located within the same cruising area.
Step 3.5: the difference deltae of unmanned energy consumption between the two paths c and c' is calculated. If Δe is less than or equal to 0, updating the path to let c=c', t++αt; otherwise, a random number rand between 0 and 1 is generated ifThe path is updated, letting c=c', t++αt.
Step 3.6: judging whether T is more than T 0 or not, if so, continuing to execute the step 3.4; otherwise, obtaining a near optimal solution of the routing inspection route planning with the lowest energy consumption, and sequentially overhauling the fan and acquiring data by the unmanned aerial vehicle according to the routing inspection route.
Step 3.7: and judging whether the routing inspection route needs to be updated or not based on the real-time meteorological data. When the real-time wind direction detected by the wind sensor in the wind power plant and the weather predicted wind direction are not in the same direction or the real-time wind speed and the weather predicted wind speed level are different by more than two stages, the wind turbine generator X' = { X 1,x2,...,xn′ } which is not traversed is counted. And (3.2) re-executing the step, and re-planning the minimum energy consumption flight path of the unmanned aerial vehicle of the remaining wind turbine. And (5) until the unmanned aerial vehicle traverses the wind turbine, ending the algorithm.
Claims (3)
1. An unmanned aerial vehicle coordinated wind farm intelligent inspection method is characterized by comprising the following steps:
Step one: acquiring weather forecast data of a wind farm for 4 hours in the future, and preprocessing the data;
Step two: for a plurality of wind turbine generators X= { X 1,x2,...,xn }, performing minimum power consumption track planning and corresponding power consumption E ij on any two wind turbine generators X i、xj within the maximum cruising radius of the unmanned aerial vehicle by using a deep reinforcement learning algorithm, wherein i is not equal to j;
step three: determining the starting position of the unmanned aerial vehicle according to the main wind direction of the predicted weather, and planning a flight path of the unmanned aerial vehicle based on a simulated annealing algorithm according to the current weather data and the learning experience of the step two;
step four: according to the flight path planned in the third step, each unit of the wind power plant is inspected so as to facilitate fault investigation and data uploading;
In the first step, the specific steps of data preprocessing are as follows:
step 1.1: checking meteorological data, and if a missing part exists, performing smoothing treatment on the missing data;
Step 1.2: for any time t, wind speed is determined according to wind direction data theta t Performing orthogonal decomposition to decompose wind speed into three wind speeds/>, which are perpendicular to each other, in three-dimensional space
Step 1.3: and carrying out normalization processing on the wind direction and wind speed data.
2. The intelligent inspection method for the wind farm coordinated with the unmanned aerial vehicle according to claim 1, wherein in the second step, the construction step of the deep reinforcement learning algorithm is as follows:
Step 2.1: firstly, establishing a Markov decision process model of the unmanned aerial vehicle on the track planning of two wind turbine units, and determining five-tuple < S, A, P, R and gamma > in the process, wherein S represents the current environmental state quantity of the unmanned aerial vehicle, A is the action quantity executed by the unmanned aerial vehicle, P is the transition probability among different states, R is the rewarding quantity obtained by the unmanned aerial vehicle executing the action A in the state S, and gamma is the reinforcement learning attenuation rate;
A three-dimensional coordinate system is established by the current position of the unmanned aerial vehicle, and the state quantity S comprises the position coordinate Pos U = (x, y, z) of the current unmanned aerial vehicle and the wind speed vector of the current unmanned aerial vehicle
Simplifying the unmanned plane motion model, wherein the executable action quantity A= < a > of the unmanned plane represents the speed vector of the unmanned planeMoving a distance in a fixed direction within a time slice tau; in order to enable the unmanned aerial vehicle to reach the target unit with the least power consumption, the following rewarding mode is designed:
R＝<r>，
wherein, |d s′ | is the linear distance from the target point after the action is executed; d s is the straight line distance from the target point before the action is executed; e max is the maximum energy consumption of the unmanned aerial vehicle; e ss′ is the energy consumption of the current execution action, the calculation formula is E ss′＝Pu·τ,Pu is the power of the unmanned aerial vehicle, including the horizontal flight power Vertical flight power/>Resistance power/> Wherein w=mg is unmanned plane gravity; ρ is the air density; /(I)Is the total area of the rotor wing of the unmanned aerial vehicle; c D0 is the drag coefficient related to rotor geometry; /(I)And/>The relative speeds of the unmanned aerial vehicle to wind speed are respectively in the horizontal direction and the vertical direction; /(I)The speed of horizontal flight of the unmanned aerial vehicle; /(I)Hovering power for the drone;
Step 2.2: initializing a playback experience pool D for storing data generated by the unmanned aerial vehicle in a trial-and-error process; randomly initializing an Actor reality network mu and a Critic reality network Q, wherein parameters corresponding to the two neural networks are theta μ and theta Q respectively; randomly initializing an Actor target network mu 'and a Critic target network Q', wherein parameters corresponding to the two neural networks are respectively theta μ′ and theta Q′, and enabling theta μ′＝θμ,θQ′＝θQ;
Step 2.3: recording initial state quantity s 1, and generating random noise N conforming to Gaussian distribution;
Step 2.4: inputting state quantity x t at the current moment into an Actor reality network taking theta μ as a parameter, and adding random noise N t at the current moment; outputting an action quantity a t＝μ(xt,θμ)+Nt by an Actor reality network, executing the action, calculating an incentive r t obtained by the action through an incentive function, and updating a state quantity to obtain x t+1;
step 2.5: creating a quadruple < x t,at,rt,xt+1 > and storing it in the playback experience pool D;
Step 2.6: randomly selecting a group of data < x j,aj,rj,xj+1 > from a playback experience pool D, inputting x j,aj into a Critic reality network to obtain Q=Q (x j,aj,θQ), inputting x j+1 into an Actor target network, calculating an action quantity a j+1＝μ′(xj+1,θμ′), inputting x j+1 and a j+1 into the Critic target network together to obtain Q (x j+1,aj+1,θQ), then taking the Q 'as a label, training the Critic reality network to enable the calculated Q value to be infinitely close to the target value Q', and updating theta Q by using a gradient descent method;
Step 2.7: updating an Actor reality network to ensure that the output action quantity is the maximum Q value calculated in the Critic reality network; also, the gradient descent method is adopted to update theta μ, and the strategy gradient calculation method is as follows
Step 2.8: updating the target network parameters theta μ′←αθμ′+(1-α)θμ,θQ′←αθQ′+(1-α)θQ;
Step 2.9: repeating the steps 2.3 to 2.8 until the loss values of the Actor target network and the Critic target network are converged, and keeping the network parameters unchanged; after the network converges, for the optional two wind turbines x i、xj, the deep reinforcement learning model gives the flight path with the minimum power consumption and the power consumption E ij of the unmanned aerial vehicle.
3. The intelligent inspection method for the wind farm coordinated with the unmanned aerial vehicle according to claim 1 or 2, wherein in the third step, the specific steps of the simulated annealing algorithm are as follows:
Step 3.1: determining the starting position of the unmanned aerial vehicle according to the current weather direction rose; if the dominant wind direction exists in the current wind power plant, setting the starting position of the unmanned aerial vehicle as a corner fan opposite to the dominant wind direction; if the current dominant wind direction is not obvious, the starting position of the unmanned aerial vehicle is set as a fan at the central position;
Step 3.2: training two neural networks on the same day; determining the flight path and the flight energy consumption between any two wind turbine generators x i、xj in the maximum cruising radius of the unmanned aerial vehicle according to the playback experience pool and the wind direction and the wind speed predicted by the next day weather in the second step;
Step 3.3: starting from the initial position, sequentially selecting the lowest energy consumption track in the cruising radius of the unmanned aerial vehicle until all fans are traversed, and taking the lowest energy consumption track as an initialization path c; simultaneously initializing an initial temperature T, a final temperature T 0 and an annealing speed alpha;
step 3.4: generating another path c' in the neighborhood of c by random thermal perturbation;
Step 3.5: calculating a difference delta E of unmanned energy consumption between the two paths c and c'; if Δe is less than or equal to 0, updating the path to let c=c', t++αt; otherwise, a random number rand between 0 and 1 is generated if Updating the path, letting c=c', t≡αt;
Step 3.6: judging whether T > T 0 is met, if so, continuing to execute the step 3.4; otherwise, obtaining a near optimal solution of the routing inspection route planning with the lowest energy consumption, and sequentially overhauling and acquiring data of the fan by the unmanned aerial vehicle according to the routing inspection route;
Step 3.7: judging whether the routing inspection route needs to be updated or not based on real-time meteorological data; when the real-time wind direction detected by the wind sensor in the wind power plant and the wind direction predicted by the weather are not in the same direction or the difference between the real-time wind speed and the wind speed level predicted by the weather is more than two stages, counting the wind turbine generator X' = { X 1,x2,...,xn′ }, re-executing the step 3.2, and re-planning the minimum energy consumption flight path of the unmanned aerial vehicle of the rest wind turbine generator; and (5) until the unmanned aerial vehicle traverses the wind turbine, ending the algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210274635.5A CN114610070B (en) | 2022-03-21 | 2022-03-21 | Unmanned aerial vehicle-coordinated intelligent inspection method for wind farm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210274635.5A CN114610070B (en) | 2022-03-21 | 2022-03-21 | Unmanned aerial vehicle-coordinated intelligent inspection method for wind farm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114610070A CN114610070A (en) | 2022-06-10 |
| CN114610070B true CN114610070B (en) | 2024-06-21 |
Family
ID=81864395
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210274635.5A Active CN114610070B (en) | 2022-03-21 | 2022-03-21 | Unmanned aerial vehicle-coordinated intelligent inspection method for wind farm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114610070B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115933787B (en) * | 2023-03-14 | 2023-05-16 | 西安英图克环境科技有限公司 | Indoor multi-terminal intelligent control system based on indoor environment monitoring |
| CN116878518B (en) * | 2023-09-06 | 2023-11-21 | 滨州市华亿电器设备有限公司 | Unmanned aerial vehicle inspection path planning method for urban power transmission line maintenance |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111256703A (en) * | 2020-05-07 | 2020-06-09 | 江苏方天电力技术有限公司 | Multi-rotor unmanned aerial vehicle inspection path planning method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110333739B (en) * | 2019-08-21 | 2020-07-31 | 哈尔滨工程大学 | AUV (autonomous Underwater vehicle) behavior planning and action control method based on reinforcement learning |
| CN112286203B (en) * | 2020-11-11 | 2021-10-15 | 大连理工大学 | Multi-agent reinforcement learning path planning method based on ant colony algorithm |
| CN113485453B (en) * | 2021-08-20 | 2024-05-10 | 中国华能集团清洁能源技术研究院有限公司 | Method and device for generating inspection flight path of marine unmanned aerial vehicle and unmanned aerial vehicle |
-
2022
- 2022-03-21 CN CN202210274635.5A patent/CN114610070B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111256703A (en) * | 2020-05-07 | 2020-06-09 | 江苏方天电力技术有限公司 | Multi-rotor unmanned aerial vehicle inspection path planning method |
Non-Patent Citations (1)
| Title |
|---|
| 海上风电场智能无人巡检方案研究;汤翔;杨源;电工技术;20211231(第020期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114610070A (en) | 2022-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114610070B (en) | Unmanned aerial vehicle-coordinated intelligent inspection method for wind farm | |
| CN110110943B (en) | Fleet energy efficiency comprehensive intelligent optimization management system and optimization method based on big data | |
| Chung et al. | Placement and routing optimization for automated inspection with unmanned aerial vehicles: A study in offshore wind farm | |
| Wang et al. | Wind power interval prediction based on improved PSO and BP neural network | |
| CN108764560A (en) | Aircraft scene trajectory predictions method based on shot and long term Memory Neural Networks | |
| CN105160444A (en) | Electrical equipment failure rate determining method and system | |
| Liu et al. | Study on UAV parallel planning system for transmission line project acceptance under the background of industry 5.0 | |
| CN112734970B (en) | Automatic inspection system and method for unmanned aerial vehicle in wind farm based on LoRaWAN positioning technology | |
| CN113705982A (en) | Scheduling decision method for vehicle-mounted machine cooperative power patrol | |
| Mohammed et al. | Design optimal PID controller for quad rotor system | |
| CN113190036A (en) | Unmanned aerial vehicle flight trajectory prediction method based on LSTM neural network | |
| CN111415010A (en) | Bayesian neural network-based wind turbine generator parameter identification method | |
| CN115840468B (en) | Autonomous line inspection method of power distribution network unmanned aerial vehicle applied to complex electromagnetic environment | |
| CN112800682A (en) | Feedback optimization fan blade fault monitoring method | |
| CN114322199A (en) | Ventilation system autonomous optimization operation regulation and control platform and method based on digital twins | |
| CN115326075A (en) | Path planning method for realizing wind field global automatic inspection based on unmanned aerial vehicle | |
| CN115185303A (en) | Unmanned aerial vehicle patrol path planning method for national parks and natural protected areas | |
| Ahmed et al. | Path planning of unmanned aerial systems for visual inspection of power transmission lines and towers | |
| Fan et al. | Temperature Prediction of Photovoltaic Panels Based on Support Vector Machine with Pigeon‐Inspired Optimization | |
| CN114548228A (en) | Unmanned aerial vehicle power grid inspection method and system based on SOC and MESH networking | |
| Zhang et al. | Energy consumption optimal design of power grid inspection trajectory for UAV mobile edge computing node | |
| CN115574826B (en) | National park unmanned aerial vehicle patrol path optimization method based on reinforcement learning | |
| Huang et al. | Study on a boat-assisted drone inspection scheme for the modern large-scale offshore wind farm | |
| CN110188939B (en) | Wind power prediction method, system, equipment and storage medium of wind power plant | |
| CN117455030A (en) | Unmanned aerial vehicle nest bionic planning method, device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |