CN-122001290-A - Pollution-degree-based photovoltaic cleaning robot path planning method and system

CN122001290ACN 122001290 ACN122001290 ACN 122001290ACN-122001290-A

Abstract

The application provides a pollution degree-based photovoltaic cleaning robot path planning method and system, and relates to the field of robot path planning. The method comprises the steps of collecting a reflectivity change event stream on the surface of a photovoltaic module and generating an asynchronous data packet containing event coordinates, time stamps and polarities. And converting the asynchronous data packet into a dynamic pollution level diagram, wherein the pollution level value and the occurrence frequency of the event in unit time are in positive correlation. And constructing a distributed dynamic pollution map by using a pulse neural network, and simulating pollution diffusion and cleaning recovery processes by neuron membrane potential evolution. And establishing an action selection competition mechanism in the impulse neural network, and generating a cleaning path instruction according to the space-time distribution characteristics of the dynamic pollution map. And executing the cleaning action and feeding back an environmental change event, and dynamically updating the dynamic pollution map and the action selection competition mechanism. The method solves the technical problems of high energy consumption, slow response to sudden pollution and dynamic environment, poor adaptability and weak anti-interference capability.

Inventors

DONG GUANGQUAN
WEN SHUJIE
WANG MIN

Assignees

山东慧匠德天智能科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. The photovoltaic cleaning robot path planning method based on the pollution degree is characterized by comprising the following steps of: Collecting a reflectivity change event stream on the surface of a photovoltaic module, and generating an asynchronous data packet containing an event coordinate, a time stamp and polarity; Converting the asynchronous data packet into a dynamic pollution level diagram, wherein the pollution level value and the occurrence frequency of events in unit time are in positive correlation; Constructing a distributed dynamic pollution map by using a pulse neural network, and simulating pollution diffusion and cleaning recovery processes by neuron membrane potential evolution; Establishing an action selection competition mechanism in a pulse neural network, and generating a cleaning path instruction according to the space-time distribution characteristics of the dynamic pollution map; And executing a cleaning action and feeding back an environment change event, and dynamically updating the dynamic pollution map and the action selection competition mechanism.
2. The pollution level-based photovoltaic cleaning robot path planning method according to claim 1, wherein the construction process of the impulse neural network specifically comprises: The method comprises the steps of constructing a pulse neural network structure comprising an event input layer, a pollution processing layer and an action output layer, wherein the event input layer consists of a plurality of event response neurons, and each neuron corresponds to a specific coordinate position of a photovoltaic module and is used for receiving an asynchronous data packet of an event camera and generating a pulse signal; the action output layer comprises competitive action neurons, basic cleaning actions are coded for each neuron, and optimal actions are selected through a pulse competition mechanism, the competition mechanism is based on the dynamic evolution of the membrane potential, and when the membrane potential of a certain action neuron reaches an issuing threshold value at first, a corresponding cleaning instruction is triggered; Integrating a special nerve morphology calculation chip, processing pulse streams in parallel, and embedding a dynamic reconfiguration interface for adapting to the space mapping of photovoltaic arrays with different sizes; and the connection weight among neurons is adjusted through pulse time, so that the accuracy of pollution simulation is optimized.
3. The method for planning a path of a photovoltaic cleaning robot based on pollution level according to claim 2, wherein the process of constructing a distributed dynamic pollution map by using a pulse neural network and simulating pollution diffusion and cleaning recovery process by neuron membrane potential evolution specifically comprises the following steps: Mapping each event response neuron in the event input layer with the space coordinate of the surface of the photovoltaic module based on the physical size of the photovoltaic module, establishing a fixed mapping relation between the physical coordinate and the virtual neuron, and forming a lookup table between the physical coordinate and the neuron; Inputting the asynchronous data packet into the impulse neural network, searching a corresponding event response neuron according to the event coordinates, calculating a membrane potential increment and driving the pollution treatment layer to update membrane potential; monitoring the reflectivity of the cleaned photovoltaic module, generating a reset event, sending the reset event to the pollution treatment layer, and updating the potential of a dynamic pollution simulation neuron of a corresponding coordinate in the pollution treatment layer; and extracting the neuron network in the pollution treatment layer to form a pollution state distribution map.
4. The method for planning a path of a photovoltaic cleaning robot based on a pollution level according to claim 3, wherein the step of establishing an action selection competition mechanism in a pulse neural network and generating a cleaning path instruction according to the space-time distribution characteristics of the dynamic pollution map specifically comprises the following steps: constructing a space-time feature coding layer, wherein the space-time feature coding layer comprises feature extraction neurons connected with a neuron cluster in the pollution treatment layer, and the feature extraction neurons are used for receiving a membrane potential pulse stream and calculating a pollution accumulation intensity feature value and a pollution diffusion trend feature value; converting the characteristic value of the accumulated pollution intensity and the characteristic value of the pollution diffusion trend into excitation pulse sequences for different action neurons in the action neuron array according to a preset action mapping rule; All the action neurons in the action neuron array perform parallel asynchronous pulse issuing competition based on the current membrane potential of the action neurons to form a cleaning path instruction.
5. The method for planning a path of a photovoltaic cleaning robot based on a pollution level according to claim 4, wherein the process of collecting the reflectance change event stream of the photovoltaic module surface and generating the asynchronous data packet including the event coordinates, the time stamp and the polarity specifically comprises: a dynamic vision sensor is deployed, the dynamic vision sensor operates in an asynchronous working mode, and each pixel unit independently responds to illumination intensity change; Triggering event acquisition to obtain a reflectance change event stream when detecting that the logarithmic change of illumination intensity of the local area exceeds a preset intensity threshold; Performing real-time filtering processing on the reflectivity change event stream, and generating a preprocessing event stream after removing global homopolar events generated by ambient light fluctuation; establishing a mapping relation between a physical coordinate system of the photovoltaic module and a pixel coordinate system of an event camera, and giving dynamic priority to pollution events and weighting the pollution events according to event polarity parameters; and sequencing the weighted preprocessing event stream according to the priority and outputting the weighted preprocessing event stream to form an asynchronous data packet sequence facing to the pollution hotspot tracking.
6. The method for planning a path of a photovoltaic cleaning robot based on a pollution level according to claim 5, wherein the process of converting the asynchronous data packet into a dynamic pollution level map specifically comprises: analyzing the asynchronous data packet, calculating pollution level and establishing pollution level mapping rules; carrying out framing treatment on the asynchronous data packet by adopting a sliding time window, and extracting the event frequency of each space coordinate; And integrating the FPGA parallel computing platform, and fusing the historical pollution accumulation value with the real-time event frequency to generate a dynamic pollution level diagram.
7. The method for planning a path of a photovoltaic cleaning robot based on a pollution level according to claim 6, wherein the process of dynamically updating the dynamic pollution map and the action selection competition mechanism specifically comprises: The method comprises the steps of monitoring illumination intensity change of a cleaning area in real time to trigger generation of a reset event, and executing membrane potential reset operation on dynamic pollution simulation neurons of corresponding coordinates of a pollution treatment layer based on polarity parameters of the reset event; based on the frequency change of pollution events before and after cleaning, adjusting the connection weight among neurons through a pulse time dependent plasticity rule; And dynamically adjusting the inhibition weight coefficient in the competition mechanism according to the execution effect of the cleaning action.
8. A pollution level-based photovoltaic cleaning robot path planning system, characterized in that the pollution level-based photovoltaic cleaning robot path planning method applied to any one of claims 1-9 comprises an event sensing module, a pulse processing module and an execution module: the event sensing module is used for collecting independent response illumination intensity changes of each pixel unit, preprocessing the response illumination intensity changes to generate an asynchronous data packet sequence, establishing a mapping relation between a physical coordinate system of the photovoltaic module and a pixel coordinate system of the sensor, and giving dynamic priority to pollution events; the pulse processing module constructs a three-level pulse neural network, and outputs a distributed dynamic pollution map and an excitation pulse sequence of an action selection competition mechanism through the asynchronous data packet sequence; the execution module selects an excitation pulse sequence of the competition mechanism based on the action, performs cleaning action and executes state feedback and environmental change event stream.
9. The contamination level based photovoltaic cleaning robot path planning system of claim 8, wherein the three-stage impulse neural network comprises an event input layer, a contamination handling layer, and an action output layer: The event input layer locates the neuron corresponding to the event through the fixed mapping relation from the physical coordinate to the virtual neuron, calculates the increment of the membrane potential and drives the pollution treatment layer to update the membrane potential; The pollution treatment layer calculates potential increment of positive polarity event and negative polarity event to generate a dynamic pollution level diagram; And the action output layer selects the optimal action through a pulse competition mechanism, and triggers a cleaning instruction when the membrane potential of a certain action neuron reaches a threshold value first.
10. The contamination level based photovoltaic cleaning robot path planning system of claim 9, wherein the construction of the event awareness module comprises: deploying a dynamic vision sensor array, wherein each pixel unit independently responds to illumination intensity changes; Establishing a dynamic mapping relation between a physical coordinate system of the photovoltaic module and a pixel coordinate system of the sensor, and carrying out coordinate conversion acceleration through an FPGA; the event filter is integrated, and a sliding time window and polarity weighting strategy is adopted.

Description

Pollution-degree-based photovoltaic cleaning robot path planning method and system Technical Field The application relates to the field of robot path planning, in particular to a photovoltaic cleaning robot path planning method and system based on pollution degree. Background With the expansion of the photovoltaic power station to large-scale, centralized and complex terrains, the problem of power generation efficiency attenuation caused by component surface pollution is increasingly outstanding, the power generation efficiency loss of components which are not cleaned in time can reach 30% at maximum, local refractory pollution such as bird droppings is more likely to cause the hot spot effect to cause permanent damage, and the cleaning robot becomes the equipment required for intelligent operation and maintenance of the power station. Most of the existing photovoltaic cleaning robots rely on traditional visual sensors such as CCD/CMOS, pollution detection is achieved through periodic full-image acquisition, so that a large amount of redundant image data needs to be processed, the energy consumption of the robots is high, the failure rate is up to more than 15% under severe environments such as high temperature, high dust and the like, pollution sensing delay is up to seconds caused by full-image transmission and processing flows, sudden pollution such as bird droppings cannot be captured timely, and cleaning timeliness is affected. Meanwhile, when the cleaning route planning is carried out, a night house adopts a static grid map or a preset route planning based on acquired data information, so that a cleaning strategy is 'one-cut', a high pollution area is difficult to focus in a targeted manner, water resources and energy consumption are wasted, and the pollution distribution dynamic change in a complex scene cannot be dealt with. Disclosure of Invention The application provides a pollution-degree-based photovoltaic cleaning robot path planning method and system, which solve the technical problems of high energy consumption, slow response to sudden pollution and dynamic environment, poor adaptability and weak anti-interference capability of the whole photovoltaic cleaning robot during use due to the dependence on full-width synchronous acquisition, static map modeling and a complex global path planning algorithm of a traditional vision sensor in the prior art. In order to achieve the above purpose, the application adopts the following technical scheme: According to the photovoltaic cleaning robot path planning method based on the pollution degree, a reflectivity change event stream of the surface of a photovoltaic module is collected, and an asynchronous data packet containing event coordinates, time stamps and polarities is generated. And converting the asynchronous data packet into a dynamic pollution level diagram, wherein the pollution level value and the occurrence frequency of the event in unit time are in positive correlation. And constructing a distributed dynamic pollution map by using a pulse neural network, and simulating pollution diffusion and cleaning recovery processes by neuron membrane potential evolution. And establishing an action selection competition mechanism in the impulse neural network, and generating a cleaning path instruction according to the space-time distribution characteristics of the dynamic pollution map. And executing the cleaning action and feeding back an environmental change event, and dynamically updating the dynamic pollution map and the action selection competition mechanism. With reference to the first aspect, in one possible implementation manner, the construction process of the impulse neural network includes constructing an impulse neural network structure including an event input layer, a pollution processing layer and an action output layer, where the event input layer is composed of a plurality of event response neurons, each neuron corresponds to a specific coordinate position of the photovoltaic module, and is used to receive an asynchronous data packet of the event camera and generate an impulse signal. The pollution treatment layer is composed of a dynamic pollution simulation neuron network, and the neurons simulate pollution accumulation and diffusion processes through membrane potential evolution. The action output layer comprises competitive action neurons, basic cleaning actions are coded for each neuron, the optimal actions are selected through a pulse competition mechanism, the competition mechanism is based on dynamic evolution of membrane potential, and when the membrane potential of a certain action neuron reaches an issuing threshold value first, a corresponding cleaning instruction is triggered. And integrating a special neuromorphic computation chip, processing pulse streams in parallel, and embedding a dynamic reconfiguration interface for adapting to the space mapping of photovoltaic arrays with different sizes. And the connection weight a