Water turbidity detection method and device for pool circulating robot

Abstract

The embodiment of the application relates to the field of water quality detection, and provides a water quality turbidity detection method and device of a swimming pool circulating robot. The method comprises the following steps: scanning a swimming pool contour of a target swimming pool to construct a three-dimensional model of the swimming pool; monitoring multi-dimensional environmental data of the target swimming pool, and predicting a pollutant distribution state of each feature region in the three-dimensional model of the swimming pool based on the multi-dimensional environmental data; generating a cruise detection strategy based on the pollutant distribution state of each feature region, and dynamically setting sampling points in each feature region according to the cruise detection strategy; detecting water quality turbidity data of each sampling point through a detection module in a cruise process of the swimming pool circulating robot; synchronously detecting water flow fluctuation data and water body monitoring images of each sampling point to obtain turbidity offset amounts of each sampling point; correcting the water quality turbidity data of each sampling point through the turbidity offset amounts; and constructing and dynamically updating a water quality turbidity distribution graph corresponding to the three-dimensional model of the swimming pool based on the water quality turbidity data, so that the accuracy and detection efficiency of water quality detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water quality detection, and in particular to a water quality turbidity detection method and device for a swimming pool circulating robot. BACKGROUND

[0002] With the increasing demand for clean swimming pool water, traditional manual cleaning or fixed circulating system for swimming pool maintenance has been difficult to meet the requirements of high efficiency, energy saving and intelligentization. At present, a swimming pool circulating robot can be used to replace traditional manual underwater cleaning operation.

[0003] In related technologies, water quality turbidity detection usually adopts a traditional fixed or single-point sampling method to monitor the content of opaque components such as suspended solids and colloidal substances in water through sensors. Water quality turbidity detection can provide a direct basis for judging water quality safety, avoiding threats to user health caused by microbial breeding or pollutant accumulation, and improving energy utilization efficiency and cleaning effect. However, the water quality in different areas of the swimming pool may be different, such as deep water area and shallow water area, water inlet and water outlet. The fixed measurement or single-point measurement in the above water quality turbidity detection method lacks representativeness and is difficult to comprehensively and accurately evaluate the turbidity of the entire swimming pool and the real situation of other water quality parameters such as residual chlorine and pH value. Taking a downstream circulating swimming pool as an example, the backwater at the bottom of the pool will cause the flow rate in the shallow water area to be slow, forming a dead water area, while the suspended solids concentration in the deep water area may be lower than that in the shallow water area due to water flow scouring. If the monitoring position is near the water outlet, the water quality turbidity in the cleaning area may be misjudged, resulting in ineffective overall cleaning strategy. When determining the overall water quality to meet the standard based on single-point data, the microorganisms in the dead water area may have exceeded the standard. Therefore, there is an urgent need to propose a technical solution to solve at least one of the above technical problems. SUMMARY

[0004] The embodiments of the present application provide a water quality turbidity detection method and device for a swimming pool circulating robot, aiming to solve the technical problem that the real water quality of the entire swimming pool cannot be comprehensively and accurately evaluated in related technologies.

[0005] In a first aspect, the embodiments of the present application provide a water quality turbidity detection method for a swimming pool circulating robot, comprising:

[0006] The swimming pool profile of the target swimming pool is scanned by a laser radar and a water pressure sensor to construct a swimming pool three-dimensional model containing water depth, water inlet position, water outlet position and water flow direction. The boundary range of each feature area in the swimming pool three-dimensional model is determined by the spatial structure, structure function and water circulation mode of the target swimming pool.

[0007] monitor multi-dimensional environment data of the target swimming pool, the multi-dimensional environment data including temperature, weather data, light conditions, objects around the swimming pool, and a location where the swimming pool is located, predict a pollutant distribution state of each feature region in the three-dimensional model of the swimming pool based on the multi-dimensional environment data, generate a cruise detection strategy based on the pollutant distribution state of each feature region, and dynamically set sampling points in each feature region according to the cruise detection strategy;

[0008] During cruise of the swimming pool circulating robot, the water quality turbidity data of each sampling point is detected by a detection module, the detection module being integrated with a multi-wavelength turbidity sensor, a residual chlorine electrode, and a pH probe; water flow fluctuation data and water body monitoring images of each sampling point are synchronously detected to obtain a turbidity offset of each sampling point; and the water quality turbidity data of each sampling point is corrected by the turbidity offset.

[0009] Based on the corrected water quality turbidity data, a water quality turbidity distribution map corresponding to the three-dimensional model of the swimming pool is constructed and dynamically updated.

[0010] In a second aspect, an embodiment of the present application provides a water quality turbidity detection device of a swimming pool circulating robot, including:

[0011] A construction module is configured to scan a swimming pool contour of a target swimming pool by a laser radar and a water pressure sensor, and construct a three-dimensional model of the swimming pool including water depth, a water inlet position, a water outlet position, and a water flow direction; a boundary range of each feature region in the three-dimensional model of the swimming pool is determined by a spatial structure, a structure function, and a water body circulation mode of the target swimming pool;

[0012] A strategy module is configured to monitor multi-dimensional environment data of the target swimming pool, the multi-dimensional environment data including temperature, weather data, light conditions, objects around the swimming pool, and a location where the swimming pool is located, predict a pollutant distribution state of each feature region in the three-dimensional model of the swimming pool based on the multi-dimensional environment data, generate a cruise detection strategy based on the pollutant distribution state of each feature region, and dynamically set sampling points in each feature region according to the cruise detection strategy;

[0013] A cruise module is configured to, during cruise of the swimming pool circulating robot, detect water quality turbidity data of each sampling point by a detection module, the detection module being integrated with a multi-wavelength turbidity sensor, a residual chlorine electrode, and a pH probe; synchronously detect water flow fluctuation data and water body monitoring images of each sampling point to obtain a turbidity offset of each sampling point; correct the water quality turbidity data of each sampling point by the turbidity offset; and based on the corrected water quality turbidity data, construct and dynamically update a water quality turbidity distribution map corresponding to the three-dimensional model of the swimming pool.

[0014] In a third aspect, the embodiments of the present application further provide an electronic device, which comprises a processor, a memory for storing a computer program, and the processor is configured to execute the computer program and implement the water turbidity detection method of the pool circulating robot according to the first aspect or any of the embodiments of the present application.

[0015] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer software program, and the computer software program is executed by a processor to implement the water turbidity detection method of the pool circulating robot according to the first aspect or any of the embodiments of the present application.

[0016] The embodiments of the present application provide a water turbidity detection method and device of a pool circulating robot. In the embodiments of the present application, a pool three-dimensional model including water depth, water inlet position, water outlet position and water flow direction is constructed by scanning the pool contour of a target pool by a laser radar and a water pressure sensor, and the boundary range of each feature area in the pool three-dimensional model is determined by the spatial structure, structural function and water circulation mode of the target pool. Further, multi-dimensional environmental data of the target pool is monitored, the multi-dimensional environmental data includes temperature, weather data, illumination condition, objects around the pool and the position of the pool, and the pollutant distribution state of each feature area in the pool three-dimensional model is predicted based on the multi-dimensional environmental data; a cruise detection strategy is generated based on the pollutant distribution state of each feature area, and the sampling points in each feature area are dynamically set according to the cruise detection strategy. In the cruise process of the pool circulating robot, the water turbidity data of each sampling point is detected by a detection module, the detection module integrates a multi-wavelength turbidity sensor, a residual chlorine electrode and a pH probe; the water flow fluctuation data and the water body monitoring image of each sampling point are synchronously detected to obtain the turbidity offset of each sampling point; and the water turbidity data of each sampling point is corrected by the turbidity offset. Finally, based on the corrected water turbidity data, a water turbidity distribution map corresponding to the pool three-dimensional model is constructed and dynamically updated.

[0017] In the embodiments of the present application, by constructing the pool three-dimensional model and the multi-dimensional environmental monitoring mode, the pollutant distribution state of each local area in the pool can be predicted, and the cruise detection strategy of the pool circulating robot is set based on the prediction result, thereby improving the adaptability of the pool detection strategy to the pool spatial structure and avoiding the occurrence of detection blind area. Further, the water turbidity data of each sampling point is detected by the cruise detection strategy, and the water turbidity data sampling result is corrected by means of water flow fluctuation monitoring and water body image monitoring, and finally the water turbidity distribution map corresponding to the pool three-dimensional model is constructed based on the corrected sampling result, thereby realizing the comprehensive detection and stereoscopic presentation of the pool, improving the accuracy and detection efficiency of the pool water quality detection, guaranteeing the cleaning ability of the pool circulating robot and reducing the energy consumption of the pool circulating robot. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of a water quality turbidity detection method of a swimming pool circulating robot provided by an embodiment of the present application is shown in FIG. 1.

[0019] Figure 2 A scene diagram of a water quality turbidity detection method of a swimming pool circulating robot provided by an embodiment of the present application is shown in FIG. 2. DETAILED DESCRIPTION

[0020] Embodiments of the present application provide a water quality turbidity detection method and device of a swimming pool circulating robot. The water quality turbidity detection method of the swimming pool circulating robot can be applied to a terminal device, which can be a mobile terminal, such as a swimming pool circulating robot or a control console of a swimming pool circulating robot. The control console can be a mobile phone, a virtual reality device, a tablet computer, a notebook computer, a desktop computer, a wearable device, or the like. The terminal device can be a server connected to a cloud service device, or a server cluster. The above connection can be realized by a hardware circuit, or by a communication module.

[0021] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. Figure 1 A flowchart of a water quality turbidity detection method of a swimming pool circulating robot provided by an embodiment of the present application is shown in FIG. 1. Figure 1 The method includes the following steps:

[0022] In step S101, a swimming pool profile of a target swimming pool is scanned by a laser radar and a water pressure sensor, and a three-dimensional model of the swimming pool containing water depth, water inlet position, water outlet position, and water flow direction is constructed.

[0023] In an embodiment of the present application, the boundary range of each feature region in the three-dimensional model of the swimming pool is determined by the spatial structure, structural function, and water circulation mode of the target swimming pool. For example, the swimming pool circulating robot can be in the swimming pool cruising scene shown in FIG. 2, and the swimming pool circulating robot can cruise and sample water quality in different regions according to the feature regions divided by the dashed lines. Figure 2

[0024] ​In step S101, first, the laser radar and water pressure sensor carried by the pool circulating robot are started, and data collection is completed by the robot cruising along the preset path on the edge and inside of the pool. The laser radar scans the horizontal and vertical profile of the target pool by emitting high-frequency laser beams, continuously receives the reflected signals to generate three-dimensional point cloud data of the pool, which contains the spatial coordinates of the pool wall and pool bottom and the shape characteristics of the pool edge, and then captures the overall shape of the pool, such as a rectangle, an irregular polygon, or a structure with arc-shaped corners. At the same time, the water pressure sensor moves with the robot and detects the water pressure values at different positions in real time. The water pressure data is converted into water depth information of each sampling point by combining the corresponding relationship between water pressure and water depth, and is matched with the spatial coordinates collected by the laser radar to establish the correlation mapping of water depth and position. In the data processing stage, the point cloud data generated by the laser radar is denoised and spliced to remove abnormal points caused by water surface reflection or pool obstacles (such as stairs and underwater lights), forming a complete three-dimensional model of the pool. By identifying regions with specific shape characteristics in the point cloud data, the positions of the water inlet and the drain are determined. The water inlet can be a convex structure with a pipe connection, and the drain is usually a recessed area on the pool bottom or wall. Combined with the relative positions of the water inlet and the drain and the initial disturbance direction of the water flow detected by the robot, the basic direction of the water flow is further determined. The water depth data obtained by the water pressure sensor is integrated into the contour model to form a three-dimensional grid containing water depth information at each position, so that the three-dimensional model of the pool can intuitively present the distribution difference between deep water areas and shallow water areas.

[0025] Further optionally, for the division of the feature areas in the three-dimensional model of the pool, the spatial structure, structural function, and water circulation mode of the target pool need to be considered to determine the boundaries. For example, a pool with a children's area and an adult's area has a clear high-low pool bottom design in terms of spatial structure. Combined with the functional requirements of the structure, the children's area is divided into a shallow water area for low-age users to play, and the adult's area is divided into a deep water area to meet the swimming needs, and the boundary between the two is determined by the starting point of the pool bottom slope change. If the pool uses a downstream circulation mode, the water inlet is set at the edge of the shallow water area, and the drain is located at the bottom of the deep water area, and the water flow direction is from the shallow water area to the deep water area. At this time, the water flow disturbance area around the water inlet is defined as an independent feature area, and its boundary is determined according to the initial diffusion range of the water flow, usually an area extending about 1.5 times the diameter of the water inlet towards the deep water area, to accurately reflect the impact of water injection on the surrounding water quality. For pools with irregular shapes, such as structures containing a circular massage area, the feature area boundary of the massage area is determined based on the physical wall range of the massage area and the influence range of the independent water flow circulation, to ensure that each feature area accurately corresponds to the actual spatial function and water flow state.

[0026] Step S102, monitoring multi-dimensional environment data of the target swimming pool, predicting the pollutant distribution state of each feature region in the three-dimensional model of the swimming pool based on the multi-dimensional environment data. In the embodiment of the present application, the multi-dimensional environment data includes temperature, weather data, light conditions, swimming pool peripheral objects, and the location of the swimming pool.

[0027] It can be understood that the multi-dimensional environment data is a key information set reflecting the influence of the environment around and inside the swimming pool on water quality, including but not limited to temperature, weather data, light conditions, swimming pool peripheral objects, and the location of the swimming pool. Among them, the temperature includes not only the real-time temperature of the swimming pool water body, but also the ambient air temperature around the swimming pool, the former directly affects the reproduction rate of microorganisms and the growth cycle of algae, and the latter is related to the change trend of water temperature. Weather data includes information such as precipitation, wind, and humidity. Precipitation may carry surrounding soil organic matter or pollutants into the swimming pool, and wind will change the flow state of the swimming pool surface, indirectly affecting the diffusion of pollutants. The light conditions need to distinguish between natural light intensity and the illumination state of artificial light sources (such as swimming pool ceiling lights). The strength and duration of natural light directly determine the growth speed of photosynthetic algae, and the stability of artificial light sources may affect the degree of optical interference during water quality detection. The swimming pool peripheral objects refer to various entities within a certain range outside the swimming pool, including vegetation (such as deciduous trees, aquatic plants), artificial facilities (such as drainage outlets, trash cans, sunshades), and building structures (such as nearby kitchens, bathrooms), which may provide sources of pollutants for the swimming pool, or change the local environment through shading, drainage, etc. The location of the swimming pool is related to the climate zone, regional environmental characteristics (such as whether it is near an industrial area, residential area), and water quality baseline conditions corresponding to its geographic coordinates, which directly affect the basic distribution type of microorganisms and algae.

[0028] The monitoring of multi-dimensional environmental data is achieved through the cooperation of sensors carried by the robot, external data networking, and image recognition. Temperature monitoring relies on contact temperature sensors and non-contact infrared temperature sensors carried by the pool circulating robot. The contact sensors insert into the water body to obtain real-time water temperature when the robot is cruising, and the infrared sensors simultaneously collect the surrounding air temperature. The combination of the two data eliminates the error of a single sensor. Weather data is mainly obtained through the robot's built-in wireless communication module to access real-time and short-term forecast weather data in the target area. Meanwhile, the robot is equipped with humidity sensors and wind sensors to verify the data on-site, avoiding deviations between remote weather data and local micro-environment of the pool. The light intensity sensor installed on the top of the robot collects real-time data, combined with the images of the sky and surrounding obstacles taken by the robot's camera, to determine whether there are sunshades, trees, and other obstacles affecting the actual illumination effect, ensuring that the light data truly reflect the light state of the pool water. The monitoring of objects around the pool relies on the robot's high-definition camera and laser radar. The camera continuously captures images around the pool, and the image recognition algorithm distinguishes between vegetation types, artificial facility categories, and building functions. The laser radar assists in measuring the distance and three-dimensional shape of objects and pool feature areas, such as identifying the crown range of deciduous trees, the location and caliber size of the drain. The determination of the pool's location is achieved by the robot's built-in GPS positioning module to obtain latitude and longitude coordinates, which are then imported into the geographic information database to match the corresponding climate zone (such as tropical and temperate zones), regional environmental attributes (such as urban residential areas and suburbs), and historical water quality monitoring data, providing regional characteristics for subsequent pollutant distribution prediction. The multi-dimensional monitoring data is integrated in real time by the robot's edge computing unit to form a structured environmental data set, laying the foundation for predicting the pollutant distribution state of each feature area in the pool three-dimensional model.

[0029] As an optional embodiment, in step S102, the method further comprises: identifying the types of microorganisms and algae contained in each feature area based on the objects around the pool and the location of the pool; predicting the pollution range and pollution trend of the types of microorganisms and algae in each feature area based on the temperature, weather data, and light conditions, to obtain the initial pollutant distribution state of each feature area; and predicting the cross-transmission of pollutants between adjacent feature areas based on the initial pollutant distribution state of each feature area, to obtain the final output of the pollutant distribution state of each feature area.

[0030] During the execution of the above steps, first, the types of microorganisms and algae in each feature area are identified based on the objects around the pool and their locations. The pool circulating robot will capture the environment around the pool through the high-definition camera mounted on it, distinguish the categories and attributes of the surrounding objects through image recognition technology, and obtain the geographic coordinates of the pool through the GPS module, matching the regional climate, environmental feature data in the geographic information database. For example, a small community outdoor pool, the robot captures a large number of willows planted around it, and through the coordinates it determines that the pool is located in the subtropical region of the south. At this time, combined with the characteristics of the environment such as the fact that willow leaves are easy to bring in humus and the subtropical region has high temperature and strong light, and matching the historical microorganism records of similar pools, it can be identified that the pool's shallow water area (close to the willow, where leaves are easy to accumulate) is prone to grow pseudomonas characterized by saprophytic growth, and the deep water area is more prone to silicious algae that adapt to moderate light due to relatively weak light and fast water flow. The whole pool has a high risk of growth of green algae due to the high temperature environment.

[0031] Subsequently, based on temperature, weather data and light conditions, the initial pollutant distribution state is predicted through a dynamic spatio-temporal graph Transformer network algorithm enhanced by physical information. During the execution of the dynamic spatio-temporal graph Transformer network algorithm, the previously identified microorganisms and algae, real-time collected temperature (such as water temperature reaching 30°C in the afternoon in summer), weather data (such as continuous light rain in recent days), light conditions (such as high light intensity at noon and short-term cloudy weather in the afternoon), and information such as water depth and water flow direction of each feature area in the pool three-dimensional model are integrated into input data. The dynamic spatio-temporal graph Transformer network algorithm will construct a dynamic spatio-temporal graph corresponding to the pool three-dimensional model, taking the shallow water area, deep water area, and the area around the water inlet as nodes in the graph. The node features integrate the reproductive characteristics of microorganisms (such as the fact that green algae reproduce faster at 30°C), the adaptability of algae to light (such as the fact that silicious algae grow slower in cloudy environments), and the water depth data of the corresponding area. The edges between nodes are dynamically adjusted according to the water flow direction and the distance between regions to reflect the possible migration correlation of pollutants. At the same time, the continuity equation and momentum equation in fluid dynamics are embedded as physical constraints in the network to ensure that the prediction conforms to the water movement law. Taking the pool in the small community as an example, under the weather condition of high temperature and short-term cloudy weather in summer, the algorithm will predict that the shallow water area will have an overall sufficient light and suitable temperature, and the coverage of green algae will expand from the pool edge to the center within the next 6 hours. The concentration of pseudomonas will gradually increase due to the continuous input of leaf humus. In the deep water area, the distribution of silicious algae is relatively uniform due to the diffusion of pollutants driven by water flow, and the trend of pollutant concentration change is relatively flat, thus forming the initial pollutant distribution state of each feature area.

[0032] Finally, the physical information graph neural network algorithm is used to predict the cross-transmission of pollutants between adjacent feature areas to obtain the final distribution state. The algorithm first integrates the initial pollutant distribution state, the water flow direction in the three-dimensional model of the pool (such as the pool in this community is downstream, the water flows from the shallow water area to the deep water area), the water circulation mode, and the temperature data related to the water flow to construct an adaptive graph structure. The nodes in the graph are the feature areas, and the node features include the initial pollutant concentration, the water flow speed (according to the downstream circulation mode, the flow speed in the shallow water area is slower and the flow speed in the deep water area is faster), and the edge weight between nodes is set according to the water flow connectivity and distance between adjacent areas. The algorithm combines hydraulic principles with graph convolution networks and uses message passing algorithms to transmit information such as pollutant concentration and water flow state between adjacent nodes to simulate the migration process of pollutants, especially the differences in transmission between different feature areas. For example, between the shallow water area and the deep water area, the green algae in the shallow water area will slowly spread to the deep water area with the water flow, but due to the faster flow speed in the deep water area, the concentration will gradually decrease during the transmission process. Between the dead water area in the corner of the pool (part of the shallow water area, with almost stagnant water flow) and the main stream area, the transmission speed of pollutants is extremely slow, and the pseudomonas in the dead water area will continue to accumulate, with only a small amount spreading to the main stream area. Through this simulation of cross-transmission, the algorithm will correct the bias in the initial pollutant distribution state that does not consider the interaction between areas, such as the initial prediction of lower green algae concentration in the deep water area, but after considering the transmission from the shallow water area, the green algae concentration prediction value on the side of the deep water area close to the shallow water area will be appropriately increased, and the final pollutant distribution state of each feature area will be formed.

[0033] It is worth noting that the dynamic spatio-temporal graph Transformer network algorithm with physical information enhancement is essentially a hybrid intelligent model that integrates dynamic spatio-temporal data structure, Transformer feature capturing ability, and domain physical laws. The core purpose is to utilize the sensitive capturing ability of data-driven models to details changes while avoiding logical biases that may occur in pure data models, such as predictions that violate objective laws, by using physical information constraints, thereby improving the accuracy and reliability of predictions. From the structural and functional disassembly, the dynamic spatio-temporal graph is the basic data carrier of the algorithm. It will first disassemble the target system to be analyzed, such as the swimming pool mentioned earlier, into individual nodes with practical significance according to the spatial dimension, such as the water inlet area, drainage area, deep water area, and shallow water area of the swimming pool. At the same time, it will divide the time dimension into continuous time steps, such as every 1 hour or every 2 hours as a time unit. Then, it will connect these nodes through edges, and the edge weights will be dynamically adjusted according to the actual correlation strength between nodes. For example, the water flow exchange between the water inlet and the adjacent shallow water area in the swimming pool is more frequent, and the edge weight between them is higher. On the other hand, the deep water area and the corner area far from the water flow have weaker correlation, and the weight is lower. Such a structure can intuitively reflect the regional correlation in space and the dynamic change in time, so that data is no longer isolated single-point information, but forms a whole network with spatio-temporal logic.

[0034] The Transformer network is the core feature extraction module of the algorithm. Unlike traditional models that can only capture simple linear correlations, Transformer can capture complex dependency relationships between different nodes at different time steps through self-attention mechanisms. For example, in the swimming pool pollutant prediction, it can analyze the influence of the microorganism concentration in the water inlet area at a certain time on the algae distribution in the shallow water area 2 hours later, the synergistic effect of temperature changes in the deep water area and light at different times of the day, and even some implicit correlations, such as the weakening of light in the shallow water area on cloudy days, which may indirectly affect the reproduction rate of microorganisms in that area, thereby affecting the spread of pollution in the downstream drainage area, and thus uncovering the spatio-temporal rules in the data that are not easily detected.

[0035] The physical information enhancement is the key difference between the dynamic spatio-temporal graph Transformer network algorithm and ordinary data-driven models. It converts the explicit physical laws and empirical knowledge in the target field into mathematical constraints or feature terms and embeds them into the training and prediction process of the model, which is equivalent to adding an objective law filter to the model. For example, in the pool pollutant prediction scenario, these physical information may include the water flow direction determining the main path of pollutant diffusion, the pollutant cannot spread rapidly against the water flow direction, the reproduction rate of certain algae will significantly increase when the temperature is between 25-30°C, and the microbial activity will decrease when the light intensity is below a certain threshold. During the prediction process, the algorithm will use these physical rules as constraints to modify the preliminary prediction results output by the Transformer. For example, if the data-driven Transformer predicts that the pollutant in a certain area spreads against the water flow direction, the model will automatically adjust the prediction result in that area based on the water flow direction, making it consistent with the actual pollutant propagation logic and avoiding errors that violate physical common sense.

[0036] In the context of the previous pool pollutant prediction scenario, the actual operation logic of the dynamic spatio-temporal graph Transformer network algorithm with physical information enhancement can be as follows: First, the characteristic areas of the pool are taken as spatial nodes, and the environmental data such as temperature, light, and water flow state at consecutive time steps are used to construct a dynamic spatio-temporal graph. Then, the Transformer module will capture the complex relationship between the concentration of microorganisms and algae and environmental factors in different areas at different times based on this graph, and preliminarily predict the pollution range and trend in each area. Finally, the preset physical rules related to the pool, such as the carrying effect of water flow on pollutants and the impact of temperature on microbial reproduction, are called to verify and optimize the preliminary prediction results. For example, if a certain area has high temperature and sufficient light, the physical law of high temperature and strong light promoting algae growth will be used to strengthen the prediction tendency of the expansion of algae pollution in that area, while unreasonable predictions of rapid algae reproduction in low-temperature areas caused by data noise will be corrected. Finally, a more realistic pollutant distribution state is output.

[0037] In the optional embodiment of the above step S102, further optionally, the pollution range covered by the microorganism types and algae and the pollution change trend in each feature region are predicted based on the temperature, weather data and illumination condition through the dynamic spatio-temporal graph Transformer network algorithm enhanced by physical information to obtain the initial pollution distribution state of each feature region, including: extracting the pollution characteristics of each feature region, the multi-dimensional environmental change characteristics in the multi-dimensional environmental data, and the spatial attribute characteristics of each feature region in the three-dimensional model of the swimming pool; constructing a dynamic spatio-temporal graph structure matched with the three-dimensional model of the swimming pool as a dynamic spatio-temporal graph Transformer network; taking each feature region as a graph node, and fusing the growth characteristics of microorganisms or algae, temperature, illumination adaptability and corresponding regional water depth data in the node characteristics of the graph node, and dynamically adjusting the edge weight between the graph nodes according to the swimming pool water flow direction and the distance between the feature regions to quantify the migration correlation of the pollutants between different feature regions; taking the continuity equation and momentum equation of fluid dynamics as priori knowledge, embedding them into the dynamic spatio-temporal graph Transformer network through a differentiable constraint layer, so that the output of the dynamic spatio-temporal graph Transformer network conforms to the physical laws of mass conservation and momentum balance of the swimming pool water body; adopting a hierarchical feature extraction strategy, capturing the local growth change characteristics of microorganisms and algae in a single feature region through a local sliding window, modeling the long-range environmental influence across feature regions through a global attention mechanism, and capturing different granularities and different dimensions of the swimming pool environment change characteristics in real time; obtaining feature interaction results by integrating multi-dimensional environmental change characteristics through a cross-attention mechanism, wherein the cross-attention mechanism is used to associate the mapping relationship between temperature and microorganism reproduction rate, illumination and algae growth cycle, and weather data and pollution diffusion intensity, and the feature interaction results are used to drive the dynamic spatio-temporal graph Transformer network to simulate the spatial coverage range of microorganism types and algae in each feature region and the change trend over time; in the constraint strength adjustment layer, a soft constraint loss function is adopted to dynamically adjust the constraint weight of the continuity equation and the momentum equation according to the water flow difference characteristics of each feature region; in the error correction mechanism, the adaptability of the model predicted pollution distribution to the water flow direction in the three-dimensional model of the swimming pool is compared to correct the data fitting deviation, and finally the initial pollution distribution state of each feature region is output.

[0038] It should be noted that, taking the pollutant prediction scenario of an outdoor swimming pool in a residential community as an example, the core logic of the above steps is to predict the initial pollutant distribution state by integrating multi-dimensional features, constraining physical laws, and using dynamic interactive modeling. This prediction not only fits the actual environmental characteristics but also conforms to objective physical laws. First, the feature extraction stage comprehensively captures key information related to pollutant distribution. For example, pollutant features focus on the saprophytic characteristics of Pseudomonas in shallow water and the low-light tolerance of diatoms in deep water. Multi-dimensional environmental change features cover water temperature in the summer afternoon, changes in sunlight between strong midday and cloudy afternoons, and recent light rain. Spatial attribute features include water depth of 0.8 meters in shallow water, water depth of 1.8 meters in deep water, and the distance between each area and the inlet. These features together constitute the basic data support for prediction, ensuring that the model can understand the pollution formation conditions from three dimensions: the characteristics of the pollutants themselves, the influence of the external environment, and spatial location differences.

[0039] Next, when constructing the dynamic spatiotemporal graph, the shallow water area, deep water area, and the area around the inlet of the pool are respectively used as nodes in the graph. The features of each node not only incorporate the microbial growth characteristics of the corresponding area (such as green algae multiplying rapidly at high temperatures), real-time temperature and light adaptability (shallow water areas have more light and green algae adaptability), but also include water depth data (shallow water areas are shallow and the water temperature rises faster). The edge weights between nodes are adjusted according to the actual water flow relationship and the distance between areas. For example, the edge weight between the inlet and the adjacent shallow water area is set higher because the water flow is directly connected and the distance is short, while the edge weight between the shallow water area and the deep water area is relatively lower because the water flow needs to cross a larger area and the flow velocity is slower. This structure can intuitively reflect the migration correlation strength of pollutants between different areas, allowing the model to initially have spatial logical cognition.

[0040] Embedding the continuity and momentum equations of fluid dynamics into the network avoids predictions that might violate physical laws due to purely data-driven approaches. For example, it prevents unreasonable model outputs such as "pollutants in shallow water areas do not diffuse with the water flow but instead appear out of nowhere in deep water areas." This ensures that the generation and migration of pollutants conform to the conservation of water mass (total volume does not increase or decrease arbitrarily) and momentum balance (movement in the direction of water flow). In a community swimming pool, this manifests as Pseudomonas aeruginosa in shallow water areas slowly diffusing towards deeper water areas with the water flow, rather than moving in the opposite direction. The hierarchical feature extraction strategy takes into account both details and the overall picture. A local sliding window can capture details of local Pseudomonas aeruginosa aggregation at the edge of shallow water areas due to leaf accumulation, while the global attention mechanism can correlate the indirect effects of increased water temperature at the inlet area on diatom growth in distant deep water areas. For instance, after the inlet water temperature rises, the warm water carried by the water flow promotes diatom activity in deep water areas. This long-range cross-regional effect can also be captured by the model, avoiding neglecting the overall environmental interaction by focusing only on the local.

[0041] The cross-attention mechanism establishes direct correlations between different environmental factors and pollutant changes, such as binding temperature and green algae reproduction rate (water temperature rises in the afternoon in summer, and green algae reproduction rate increases), light and green algae growth cycle (green algae grow more vigorously in strong light at noon, and the growth rate slows down in cloudy periods), and small rain weather and pollutant diffusion intensity (rainfall makes the water surface flow rate slightly increase, but also makes it more difficult for surrounding pollutants to enter the pool, reducing the diffusion intensity). The feature interaction results obtained by integrating these correlations can drive the model to accurately simulate the spatial coverage and time variation trend of pollutants in each region. For example, in shallow water areas, due to high temperature and sufficient light, the coverage of green algae will gradually expand from the pool edge to the center, and the expansion speed will be slower in the afternoon than at noon.

[0042] In the constraint strength adjustment layer, the weight of physical constraints will be dynamically adjusted according to the water flow differences in each region. For example, the dead water area in the corner of the pool (part of the shallow water area) has almost stagnant water flow, and the constraint weight corresponding to the continuity equation will be adjusted lower to avoid excessive constraints that prevent the model from capturing the accumulation characteristics of pollutants in the dead water area. In the main flow area (such as the water flow channel from the water inlet to the shallow water area), the constraint weight of the momentum equation will be adjusted higher to ensure that the pollutant migration conforms to the water flow dynamics. The error correction mechanism will correct the deviation by comparing the adaptability of the model's predicted pollutant distribution and the actual water flow direction in the pool. For example, if the model initially predicts that the green algae concentration in the deep water area is uniform, but considering the actual situation that the water flow in the pool flows from the shallow water area to the deep water area, it will be corrected that the green algae concentration is higher near the shallow water area and lower far away. The final output of the initial pollutant distribution state can accurately reflect the pollution situation in each region. The shallow water area has wide green algae coverage and local aggregation of Pseudomonas, and the deep water area has uniform diatom distribution and slightly more green algae near the shallow water area, which lays a precise foundation for subsequent prediction of cross-transmission between regions.

[0043] In an optional embodiment of the step S102, further optionally, the predicting, by the physical information graph neural network algorithm, the cross-transmission of pollutants between adjacent feature regions based on the initial pollutant distribution state of each feature region to obtain the final output of the pollutant distribution state of each feature region comprises: obtaining the initial pollutant distribution state of each feature region, the water depth of each feature region contained in the three-dimensional model of the swimming pool, the water flow direction and the water circulation mode, the temperature and weather data related to the water flow in the multi-dimensional environmental data; constructing a graph structure adapted to the three-dimensional model of the swimming pool as a physical information graph neural network; taking each feature region as a graph node of the physical information graph neural network, and fusing the initial pollutant distribution state of each feature region, the water flow velocity derived based on the water flow direction and the water circulation mode, and the water depth data as the feature of the graph node; the edge weight between the graph nodes is dynamically set according to the water flow connectivity, the distance and the water flow velocity of adjacent feature regions, so as to quantify the basic correlation strength of the pollutant migration between adjacent feature regions; the water flow connectivity includes the water flow channel between the deep water area and the shallow water area in the downstream swimming pool; through the physical information graph neural network, the water flow pressure difference driving the pollutant migration rule in the hydraulic principle is combined with the depth of the graph convolution network, the pollutant migration information interaction is realized by using the algorithm based on message passing; the pollutant concentration and the water flow state feature are transmitted between adjacent graph nodes through the message passing module, the pollutant migration potential between adjacent feature regions is calculated, and the migration potential between feature regions is higher between feature regions with larger water flow pressure difference; the pollutant migration amount between graph nodes is iteratively updated by the recursive method; for the feature regions formed by the spatial structure and the water circulation mode, the diffusion amount of the main flow area to the dead water area, the transport amount of the deep water area to the shallow water area along the water flow direction, and the pollutant transmission path model between feature regions are calculated, including the dead water area, the deep water area and the shallow water area; the pollutant migration rule between the dead water area and the main flow area in the target swimming pool, and between the deep water area and the shallow water area is simulated by the pollutant transmission path model, the deviation in the initial pollutant distribution state caused by not considering the cross-transmission is corrected, and the pollutant distribution state of each feature region after the cross-transmission correction is obtained.

[0044] Taking an indoor swimming pool with a downstream circulation as an example, the core of this series of steps is to accurately restore the actual propagation process of pollutants between different feature areas through the deep integration of physical laws and graph networks, making up for the lack of considering regional interaction in the initial pollutant distribution state. First, we comprehensively collect key information related to pollutant propagation, including the initial pollution conditions of each feature area. For example, the shallow water area has a higher concentration of green algae and local pseudomonas aggregation due to the accumulation of fallen leaves near the shore, while the deep water area has a more uniform distribution of diatoms due to the relatively stable water flow. At the same time, we extract the spatial and water flow attributes from the three-dimensional model of the pool, such as the difference in water depth between the shallow and deep water areas, the fixed water flow direction from the shallow to the deep water area under downstream circulation, and factors affecting water flow in multi-dimensional environmental data, such as the increase in water temperature during the summer afternoon (which changes the viscosity of the water and thus affects the flow rate), and the slight disturbance of the water surface caused by recent light rain. These data collectively form the basis for simulating pollutant propagation, ensuring that the model can adhere to the actual environmental conditions of the pool.

[0045] Next, we construct a graph structure that adapts to this downstream swimming pool, defining the shallow water area, deep water area, and dead water area in the corner of the pool as graph nodes of the physical information graph neural network. The features of each node not only include the initial pollutant concentration of the corresponding area, but also incorporate the water flow speed derived based on the water flow direction and circulation mode. The shallow water area has a slightly faster water flow speed due to its proximity to the water inlet, while the deep water area has a relatively flat flow speed. The dead water area is nearly stagnant due to its distance from the main flow channel. At the same time, we incorporate the water depth data of each area, such as the shallow water depth in the shallow water area, which makes the water flow more susceptible to external disturbances. The edge weights between nodes are dynamically adjusted according to the actual association between adjacent areas. For example, the edge weight between the shallow and deep water areas is set to a high value to reflect the strong migration association between these two areas, as there is a direct water flow channel for downstream circulation, and the distance is moderate with stable water flow. The edge weight between the dead water area and the adjacent shallow water area (part of the main flow area) is set to a low value to quantify the difference in basic association strength of pollutant migration between different areas, allowing the graph structure to truly reflect the closeness of interaction between areas.

[0046] In the model operation link, the law of water flow pressure difference driving pollutant migration in hydraulic principle is combined with graph convolution network to adapt to the water flow characteristics of the downstream swimming pool. The shallow water area continuously receives new water injected from the water inlet, and there is a certain water pressure, forming a natural pressure difference with the deep water area. This pressure difference becomes the core driving force for the migration of pollutants from the shallow water area to the deep water area. Through the message passing algorithm, adjacent nodes will continuously interact with key information, such as the concentration of green algae and water flow speed in the shallow water area, and the concentration of pollutants and water flow stability in the deep water area. The algorithm calculates the migration potential of adjacent areas based on this information. Since the pressure difference between the shallow water area and the deep water area is obvious, and the water flow direction is fixed, the migration potential between the two is much higher than that between the dead water area and the main flow area, which means that the green algae in the shallow water area is more likely to spread to the deep water area along the water flow.

[0047] Then the amount of pollutant migration between nodes is continuously updated through the recursive method. According to the characteristics of the feature area type of the swimming pool, two key propagation amounts are calculated: one is the diffusion amount of pollutants from the main flow area (main flow channel of shallow water area and deep water area) to the dead water area. Since the water flow in the dead water area is stagnant, pollutants can only enter through slow molecular diffusion, so the diffusion amount is very small. The second is the pollutant transport amount from the shallow water area to the deep water area along the downstream direction. Combined with the water flow speed and pressure difference in the shallow water area, the number of green algae carried into the deep water area per unit time along with the water flow can be calculated, and a clear pollutant propagation path model is constructed. Green algae starts from the shallow water area and gradually spreads to the deep water area along the downstream direction, with only a small amount entering the dead water area through diffusion, completely restoring the migration trajectory of pollutants in the downstream swimming pool.

[0048] Finally, the actual migration law is simulated using the propagation path model to correct the initial pollutant distribution state. For example, in the initial prediction, the green algae concentration in the deep water area is low and evenly distributed, but combined with the continuous transport process of the shallow water area to the deep water area, the green algae concentration on the side close to the shallow water area is significantly higher than that on the side far away, which is more consistent with the actual situation of pollutants gradually diffusing along the water flow under downstream circulation. At the same time, the dead water area receives a small amount of pollutants from the main flow area, so its green algae concentration is slightly higher than the initial predicted value, avoiding the problem of underestimating pollution caused by ignoring cross propagation between regions. The final pollutant distribution state of each feature area can truly reflect the overall distribution and regional interaction of pollutants in the downstream swimming pool, providing reliable decision-making basis for subsequent generation of accurate cruise detection strategies. For example, increasing the sampling point density on the side of the deep water area close to the shallow water area.

[0049] Further optionally, before obtaining the pollutant distribution state of each feature region after cross-propagation correction, a physical constraint condition adapted to fluid dynamics can also be incorporated in the unsupervised learning process. The constraint condition includes a mass conservation condition for pollutant migration matched with a continuity equation, which is used to ensure that the total amount of pollutants does not increase or decrease arbitrarily during migration. The constraint condition also includes a pollutant migration rate formula driven by water flow, which is corrected in combination with the water viscosity coefficient derived from water temperature. Thus, a constraint loss function is constructed by the constraint condition to correct the output result of the physical information graph neural network, so that the output result conforms to the actual pool water pollutant migration rule.

[0050] For example, in the pollutant propagation simulation scenario of a cell downstream open-air pool, by adding a physical constraint condition adapted to fluid dynamics in the unsupervised learning process, the model avoids relying only on data fitting to appear prediction deviation that violates the objective law, and makes the pollutant migration simulation closer to the actual water movement. First, the incorporated mass conservation condition will echo the continuity equation mentioned earlier, ensuring that the total amount of pollutants in the entire pool system does not increase or decrease arbitrarily during migration. For example, when the green algae in the shallow water area spreads to the deep water area with the water flow, the amount of green algae decreased in the shallow water area should generally match the amount of green algae increased in the deep water area (minus a small amount of natural decay), and there will be no unreasonable situation that the green algae in the shallow water area is only slightly reduced, and the green algae in the deep water area suddenly increases a lot, which is like the mass in water flow movement cannot be produced or disappeared, so that the overall logic of pollutant migration is more consistent with the actual situation.

[0051] The pollutant migration rate formula driven by water flow will be corrected in combination with the actual water temperature of the pool, because water temperature will directly affect water viscosity. When the pool water temperature rises in the summer afternoon, the water viscosity will decrease, and the speed of pollutant migration driven by water flow will be faster than when the water temperature is low. The formula will correct the water viscosity coefficient to make the migration rate calculated by the model adapt to this actual change. For example, the green algae migration rate from the shallow water area to the deep water area in the summer afternoon of the pool in the cell will be slightly faster than in the morning due to the rising water temperature, avoiding the contradictory prediction of "high water temperature but slow migration".

[0052] Subsequently, the constraint loss function converts the two physical constraints into a model-optimizable objective. When the model's predicted total amount of pollutants is not conserved or the migration rate deviates significantly from the reasonable range after water temperature correction, the loss function will produce a higher error value, driving the model to adjust parameters during training to gradually correct these deviations. The final effect is that the output of the physical information graph neural network not only matches the trends in the data, but also strictly complies with the laws of fluid dynamics, such as avoiding the error of "dead water zone pollutant total amount exceeding the input amount of the main flow zone" when simulating the accumulation of pollutants in the dead water zone, and avoiding the rigid prediction of "water temperature change has no effect on migration rate", making the subsequent pollutant cross-propagation correction result more realistic and reliable, and providing more accurate basis for the development of cruise detection strategies.

[0053] In step S103, a cruise detection strategy is generated based on the pollutant distribution state of each feature area, and sampling points in each feature area are dynamically set according to the cruise detection strategy.

[0054] In the embodiment of the present application, the cruise detection strategy is a comprehensive detection scheme based on the pollutant distribution state (such as the concentration of pollutants, whether it contains high-risk microorganisms, and the degree of algae aggregation) of each feature area and the area attribute (such as whether it is a dead water area and the degree of water flow activity), and the core purpose is to enable the pool circulation robot to complete water quality detection in an efficient and accurate manner, avoiding resource waste while ensuring that high-risk areas are not missed. When developing the strategy, the detection priority will be divided according to the pollutant distribution state, such as high-priority areas with high pollution concentration, pathogenic bacteria, or high risk of algae aggregation, medium-priority areas with medium pollution level and only containing ordinary microorganisms, and low-priority areas with low pollution concentration and no high-risk factors. The dead water area will have a higher priority in the same concentration range due to its tendency to accumulate pollutants. Then, the sampling point parameters are set according to the priority, the sampling point density is higher and the sampling frequency is more frequent in high-priority areas, and the density and frequency are appropriately reduced in low-priority areas. Finally, the cruise path and time are planned, high-priority areas are covered first, and the path connection of adjacent areas is considered, reducing the invalid movement of the robot, and the strategy is dynamically adjusted according to the actual detected water quality changes to ensure that the detection always fits the real-time pollution situation.

[0055] As an optional embodiment, in step S103, the detection priority of each feature region is divided based on the pollutant distribution state and region attribute of each feature region; the feature region with pollutant concentration higher than the preset threshold, pathogenic bacteria or algae aggregation risk is divided into a high-priority detection region; the feature region with pollutant concentration in the threshold interval, no pathogenic bacteria but ordinary microorganisms is divided into a medium-priority detection region; the feature region with pollutant concentration lower than the preset threshold, no high-risk microorganisms is divided into a low-priority detection region; the detection priority of the dead water area is increased by one level in the same pollutant concentration interval due to the easy accumulation of pollutants.

[0056] Further, the sampling point parameters of each feature region are dynamically set according to the detection priority. Specifically, for the high-priority detection region, the sampling points are arranged at a grid density of 2m x 2m, the number of sampling points is not less than the ratio of the total area of the region to 4m2, and 2-3 additional sampling points are additionally added at the places where the pollutant concentration gradient changes obviously (such as the junction of deep water area and shallow water area, downstream of water inlet), and the sampling frequency is set to 1 time per 3-5 minutes. For the medium-priority detection region, the sampling points are arranged at a grid density of 5m x 5m, the number of sampling points is the ratio of the total area of the region to 25m2, and one sampling point is added at the water flow turning place (such as the corner of the pool), and the sampling frequency is set to 1 time per 8-10 minutes. For the low-priority detection region, the sampling points are arranged at a grid density of 8m x 8m, the number of sampling points is the ratio of the total area of the region to 64m2, and only one sampling point is arranged at the center and edge of the region, and the sampling frequency is set to 1 time per 15-20 minutes. Further optionally, the robot cruising track is planned according to the principle of “high-priority region is preferentially covered, and adjacent region paths are connected”, the track needs to avoid obstacles in the pool (such as stairs, drainage grating), and sufficient stopping time is reserved when passing through each feature region sampling point to ensure that the detection module can stably collect water turbidity data and water flow fluctuation data. For example, the high-priority region sampling point stays for 10 seconds, the medium-priority stays for 8 seconds, and the low-priority stays for 5 seconds.

[0057] Finally, with the goal of high-priority feature area priority coverage and adjacent feature area path connection, the cruise path, cruise time, and cruise frequency that match the sampling point parameters of each feature area in the three-dimensional model of the swimming pool are generated to form the corresponding cruise detection strategy. Further optionally, during the robot cruise process, the corrected water turbidity data fed back by the detection module is received in real time. If the actual turbidity value of a certain feature area deviates from the predicted pollutant distribution state by more than 30%, such as a predicted turbidity of 1.0 NTU and an actual detection of 1.3 NTU, or a new high-risk type of microorganism is detected, the priority of the area is automatically increased by one level, and the sampling point density is correspondingly encrypted (such as adjusting from a medium-priority 5m×5m grid to a high-priority 2m×2m grid), the sampling frequency is increased, and the cruise path is updated to preferentially cover the adjusted high-priority area, ensuring that the sampling point setting matches the real-time pollutant distribution state.

[0058] In the embodiment of the present application, the detection module is integrated with a multi-wavelength turbidity sensor, a residual chlorine electrode and a pH probe. For example, the detection module adopts a multi-component integrated design, integrating the multi-wavelength turbidity sensor, the residual chlorine electrode and the pH probe into one body. The purpose is to enable the pool circulating robot to obtain multiple key indicators reflecting the water quality condition through only one sampling operation at each sampling point during the cruise detection process, thereby improving the detection efficiency and avoiding the detection errors that may be caused by multiple sampling. The multi-wavelength turbidity sensor is mainly responsible for collecting the water turbidity data of each sampling point. By virtue of its multi-wavelength detection capability, it can effectively eliminate the interference caused by bubbles, water color or specific algae pigments, and capture the content change of suspended particles (such as microorganisms and algae debris) in the water, thereby providing a core basis for judging the turbidity degree of the water quality. The residual chlorine electrode detects the residual chlorine content in the water sample in real time. Residual chlorine is a key substance for disinfecting the pool water. Its content is directly related to the sterilization effect of the water. Through residual chlorine detection, the disinfection state of each sampling point can be grasped in time, thereby avoiding the breeding of microorganisms due to insufficient residual chlorine or affecting the safety and comfort of the water body due to excessive residual chlorine. The pH probe is used to detect the pH value of the water sample. The pH value of the pool water affects the disinfection efficiency of residual chlorine, the existence form of pollutants and the corrosiveness of the water to the equipment. Maintaining the pH value in the appropriate range is an important link to ensure the stability of the water quality. Therefore, the pH detection data can provide a reference for subsequent judgment of the overall stability of the water quality. During actual detection, the three components work synchronously, and the turbidity, residual chlorine and pH data collected can be mutually verified and supplemented. For example, when the turbidity of a certain sampling point abnormally increases, the residual chlorine data can be combined to judge whether the microorganisms breed in large quantities due to insufficient disinfection, or the pH data can be combined to judge whether the pollutants condense due to pH imbalance, thereby providing multi-dimensional and reliable detection support for comprehensively grasping the water quality condition of each sampling point, correcting the turbidity data and subsequently constructing the water turbidity distribution map.

[0059] Further optionally, before detecting the water turbidity data of each sampling point by the detection module, the pool circulating robot can stop for 10 seconds after reaching each sampling point to wait for the water flow to stabilize. When the water sample is extracted by the micro water pump, the bubble separation device is enabled to remove the bubbles generated by the water flow disturbance by using the vortex separation principle. When the flow rate fluctuation exceeds ±0.2 m / s, secondary sampling is triggered. The cross-validation algorithm is used to compare the turbidity and conductivity change trends. If abnormal conditions such as sudden increase in turbidity without corresponding change in conductivity occur, the bubble interference is automatically marked and the backup wavelength detection channel is enabled.

[0060] Specifically, before detecting the water turbidity data of each sampling point through the detection module, preprocessing and abnormality handling operations are performed to ensure the accuracy of the detection data. After the pool circulating robot reaches each sampling point, it does not immediately start detection, but first stays for a period of time to wait for the water flow to stabilize, avoiding the influence of water instability caused by robot movement or surrounding water flow disturbance on subsequent turbidity detection results. After the water flow is stable, a micro water pump is started to extract the water sample. During the extraction process, a bubble separation device is simultaneously started. The device uses the principle of vortex separation to separate the bubbles in the water sample from the water body under the action of centrifugal force by generating rotating water flow, thereby removing the bubbles caused by water flow disturbance, preventing bubbles from adhering to the surface of the detection probe or mixing into the water sample, and causing false high turbidity detection values.

[0061] During the entire sampling and detection process, the water flow velocity of the sampling point is monitored in real time. If the flow rate fluctuation exceeds the set range, a secondary sampling mechanism is triggered to re-extract the water sample for detection to exclude the influence of rapid flow rate changes on sampling accuracy. During secondary sampling, a cross-validation algorithm is used to compare the change trends of turbidity data and conductivity data. Because under normal circumstances, if the water turbidity increases due to increased pollutants, the conductivity will usually change to some extent. If an abnormal situation occurs in which the turbidity suddenly increases but the conductivity does not change, the system will automatically determine that this abnormality is caused by bubble interference rather than real water pollution. At this time, the detection result is marked as bubble interference, and a backup wavelength detection channel in the detection module is immediately started to re-detect the water sample through different wavelengths of light to avoid the interference of bubbles on specific wavelength detection, and ultimately obtain more accurate water turbidity data to provide a reliable basis for subsequent turbidity data correction and water turbidity distribution map construction.

[0062] Step S104, simultaneously detecting water flow fluctuation data and water body monitoring images of each sampling point to obtain turbidity offset of each sampling point.

[0063] As an optional embodiment, in step S104, a fluid dynamics is used to establish a water microcirculation model in the local area where each sampling point is located based on the water flow fluctuation data; a water flow drift of each sampling point is generated based on the water microcirculation model; and a visibility difference prediction is performed on the water body monitoring image and a water body reference image corresponding to each sampling point in a preset reference image library to obtain a light source drift of each sampling point.

[0064] After synchronously detecting the water flow fluctuation data and the water body monitoring image of each sampling point, in order to accurately obtain the turbidity offset, a series of targeted operations will be gradually derived. First, for the collected water flow fluctuation data, the water flow characteristics of the local area where the sampling point is located are analyzed in combination with the principles of fluid dynamics. The water flow fluctuation data contains information such as the real-time flow rate change, flow direction deflection, and fluctuation frequency of the water body at the point. Based on these data, the water microcirculation model can present the micro flow state in the local area, such as whether there is a small range of vortex flow, stratified flow of water flow, or backflow formed due to the influence of pool walls and obstacles, thereby completely restoring the water flow environment that may cause the offset of pollutant particles.

[0065] Based on the constructed water microcirculation model, the water flow drift of each sampling point is further generated. This process extracts the dynamic change law of the water flow at the sampling point through the model, calculates the actual displacement deviation of the suspended pollutant particles (such as microorganisms and algae debris) in the water body from the preset sampling position of the detection probe under the action of the water flow. Due to the disturbance of the water flow, the actual position of the pollutant particles may differ from the sampling position of the probe, which directly leads to the inconsistency between the detected turbidity value and the actual value. The water flow drift is a quantification of this deviation, which provides a basis for subsequent correction of turbidity data.

[0066] At the same time, for the collected water body monitoring image, it will be compared with the water body reference image of the corresponding sampling point in the preset reference image library to predict the light source drift through the visibility difference. The images in the preset reference image library are the reference images of each sampling point collected under standard detection conditions (such as stable light source intensity and clean water environment), which can reflect the visibility state of the water body without interference. In the comparison process, the differences in visibility characteristics between the real-time monitoring image and the reference image are analyzed, such as the change in average gray value, the attenuation degree of image contrast, or the clarity difference of image details under a specific wavelength channel. These differences are essentially caused by changes in the light source state during detection (such as light source intensity fluctuation and slight wavelength shift), and the light source drift is a quantification of the turbidity detection deviation caused by the change of the light source. Together with the water flow drift, it constitutes the turbidity offset that needs to be corrected, ensuring the accuracy of subsequent turbidity data.

[0067] In the above embodiments, further optionally, the water microcirculation model of the local region where each sampling point is located is established based on the water flow fluctuation data by fluid dynamics, comprising: obtaining water flow fluctuation data of each sampling point, the water flow fluctuation data comprising real-time flow velocity, flow direction change amplitude and fluctuation frequency of the water body at the sampling point; determining the boundary conditions of the local region in combination with the spatial attributes of the characteristic region where the sampling point is located included in the three-dimensional model of the swimming pool, the boundary conditions comprising pool wall friction coefficient, water flow inlet and outlet position; the spatial attributes of the characteristic region comprising at least one of the following: water depth of the local region, water inlet distance, water outlet distance, whether it is a dead water area, whether it is a main flow area, and peripheral obstacle distribution. Then, based on the local flow field analysis method in fluid dynamics, the Navier-Stokes equation is adapted to the small-scale water flow characteristics of the swimming pool, the water flow fluctuation data is taken as the input boundary condition, and the spatial attribute parameters of the characteristic region are coupled to construct a water microcirculation model for reflecting vortex flow, backflow and flow velocity stratification in the local region, the model can output the instantaneous flow vector of any position in the local region; the spatial attribute parameters include the low flow velocity reference value of the dead water area and the flow velocity gradient of the main flow area.

[0068] Specifically, when constructing the water microcirculation model of the local region where each sampling point is located, the water flow fluctuation data of each sampling point will be collected first, which covers the real-time flow velocity of the water body at the point, the change range of the flow direction, and the frequency of the water flow fluctuation, providing basic data support for subsequent analysis of the local water flow state. Subsequently, the spatial attributes of the characteristic region to which the sampling point belongs recorded in the three-dimensional model of the swimming pool will be combined to determine the boundary conditions required for model construction. These spatial attributes may include the water depth of the local region, the distance to the water inlet or water outlet, whether the region belongs to a dead water area or a main flow area, and whether there are obstacles such as stairs and underwater lights in the periphery and their distribution, and the boundary conditions determined according to these attributes will specifically involve the friction coefficient of the pool wall to the water flow, the inlet and outlet position of the water flow in the local region, etc., to ensure that the model can fit the actual spatial environment of the sampling point.

[0069] Next, the local flow field analysis method in fluid dynamics is used, and the Navier-Stokes equation is adapted to the small-scale water flow characteristics of the swimming pool, the water flow fluctuation data obtained before is taken as the input boundary condition of the model, and the spatial attribute parameters of the characteristic region are coupled with the input condition. For example, the low flow velocity reference value specific to the dead water area, the flow velocity gradient existing in the main flow area, etc. Through such processing, the water microcirculation model constructed can accurately present the micro water flow conditions in the local region, including whether there is small-scale vortex flow, whether the water flow appears stratified flow phenomenon, or details such as backflow formed by obstacles, etc., and the model can finally output the instantaneous flow vector of any position in the local region, providing accurate water flow state basis for subsequent calculation of water flow drift.

[0070] In the above embodiments, further optionally, the water flow drift amount of each sampling point is generated based on the water body microcirculation model, comprising: constructing a water flow vector time sequence by extracting the instantaneous water flow vector at the sampling point through the water body microcirculation model; based on the movement law of the suspended particles in the fluid, calculating the displacement deviation of the pollutant particles in the water body from the initial position of the sampling point under the action of the water flow vector time sequence; the pollutant particles match the particle diameter and density of the identified microorganisms or algae. Further, the displacement deviation is decomposed into a main drift amount along the water flow direction and a secondary drift amount perpendicular to the water flow direction, and the main drift amount and the secondary drift amount are combined to obtain the deviation value between the actual position of the pollutant particles at the sampling point and the sampling position of the detection probe as the water flow drift amount. Wherein, in the dead water area, the water flow drift amount of the dead water area is thirty percent of the corresponding main flow area drift amount due to low water flow speed and small fluctuation; the water flow drift amount of the shallow water area needs to be superimposed with an attenuation coefficient along the depth direction due to the influence of the pool bottom friction in the shallow water area, and the attenuation coefficient decreases by a preset proportion linearly with the increase of water depth.

[0071] In the above steps, when generating the water flow drift of each sampling point based on the constructed water microcirculation model, the instantaneous water flow vectors at the sampling points are first extracted from the water microcirculation model, and these vectors are sorted in time sequence to form a water flow vector time sequence, so as to reflect the dynamic change of the water flow at the sampling point within a period of time. Considering that the diameter and density of the pollutant particles in the water body match the previously identified microorganisms or algae, according to the movement law of suspended particles in the fluid, the displacement deviation of these pollutant particles from the initial position of the sampling point (i.e. the preset sampling position of the detection probe) under the action of the water flow represented by the water flow vector time sequence is analyzed, and this deviation directly reflects the degree of deviation of the pollutant particles from the sampling position driven by the water flow. Then, the calculated displacement deviation is decomposed into two parts, one part is the main drift along the water flow direction, which is the main deviation of the pollutant particles moving along the main flow direction. The other part is the secondary drift perpendicular to the water flow direction, which is mainly caused by the secondary deviation of local eddy current, backflow and other micro water flow movements. By synthesizing the two parts of the drift, the total deviation value between the actual position of the pollutant particles at the sampling point and the sampling position of the detection probe can be obtained, and this total deviation value is the required water flow drift. At the same time, the water flow drift will be adjusted according to the characteristics of the characteristic region to which the sampling point belongs, such as the dead water area, because the water flow speed is low and the fluctuation amplitude is small, the moving range of the pollutant particles is limited, and the water flow drift of the dead water area will take a certain proportion of the corresponding main flow area water flow drift. The water flow drift of the shallow water area needs to be superimposed with a decay coefficient along the depth direction because the water depth is shallow and the water flow is more affected by the friction of the pool bottom, and the movement of the pollutant particles in the depth direction will be limited. Therefore, the water flow drift of the shallow water area needs to be superimposed with a decay coefficient along the depth direction. This decay coefficient will decrease linearly with the increase of water depth at a preset proportion, so as to better fit the actual water flow and particle movement in the shallow water area, and to ensure that the final water flow drift can accurately reflect the actual deviation in different regions.

[0072] In the above embodiment, further optionally, the water body monitoring image and the water body reference image corresponding to each sampling point in the preset reference image library are subjected to visibility difference prediction to obtain the light source drift of each sampling point. Specifically, for the sampling points of each characteristic region of the pool three-dimensional model, under standard detection conditions (including standard light source intensity, clean water body with 0 NTU turbidity, and no water flow disturbance), the water body images of each sampling point are collected by the water body monitoring camera carried by the robot (same model and same angle as the cruising detection), as the reference image. The reference image needs to be stored by characteristic region (such as dead water area sampling point reference image, deep water area sampling point reference image), and the spatial attributes (such as water depth, relative position to the multi-wavelength turbidity sensor) and standard light source parameters (such as light source wavelength, initial light intensity) of the corresponding sampling point are associated to form a preset reference image library matched with the space coordinates of the pool three-dimensional model.

[0073] Further, the water body monitoring images of each sampling point collected synchronously during the robot cruise are acquired. The color images are first converted into single-channel grayscale images through grayscale processing, and then Gaussian filtering is used to eliminate image jitter noise caused by water flow fluctuations. The monitoring images and the corresponding reference images are calibrated to the same spatial perspective with the center of the sampling point of the reference image as the anchor point through an image registration algorithm, and the contrast error caused by the slight deviation of the robot position is excluded. If there are instantaneous bubbles or robot shadows in the monitoring images, the shadowed areas are removed through image mask technology, and only the effective water area is retained for subsequent comparison. For example, the visibility features of the effective water area include but are not limited to the average gray value, the contrast (gray value standard deviation), the edge definition (edge intensity calculated by the Sobel operator), and the wavelength channel features corresponding to the multi-wavelength turbidity sensor of the detection module. For example, the gray distribution of the 450 nm blue channel and the 650 nm red channel.

[0074] Then, the deviation rate of the average gray value (ΔG = |Gmonitor-Greference| / Greference x 100%) of both, the attenuation rate of the contrast (ΔC = |Cmonitor-Creference| / Creference x 100%), and the shift distance of each wavelength channel feature peak, such as the number of pixels of the blue channel feature peak from the reference position, are calculated. Among them, the interference caused by the real change of water turbidity (such as the gray value drop caused by algae aggregation) needs to be removed in combination with the pollutant distribution state of each feature area. If the increase of the actual turbidity of a sampling point compared with the reference turbidity (0 NTU) is ≤0.3 NTU, it is determined that the visibility difference is mainly caused by light source drift.

[0075] In the intensity drift calculation process, based on the average gray value deviation rate ΔG, combined with the initial light intensity Ireference of the standard light source, the deviation between the actual intensity and the standard intensity of the light source is calculated according to the linear mapping relationship (I drift = Ireference x ΔG), that is, the intensity drift. In the wavelength drift calculation process, according to the shift distance of each wavelength channel feature peak, combined with the spectral response curve of the camera, the pixel shift is converted into the actual wavelength shift value (such as 1 pixel shift corresponding to 2 nm wavelength drift), and the average value of the wavelength shift values of each channel is taken as the wavelength drift. In the total light source drift synthesis process, the weighted sum (intensity drift weight 0.6, wavelength drift weight 0.4) is used to obtain the total light source drift of each sampling point. Among them, the wavelength drift of the shallow water area needs to be superimposed with the natural light spectrum correction coefficient (such as the correction coefficient being 1.2 in summer strong light) due to the greater influence of external natural light. The light source drift calculation result of the dead water area needs to be multiplied by a stability coefficient of 0.8 due to the poor water flow and stable light, so as to match the actual light source environment.

[0076] Step S105, correct the water quality turbidity data of each sampling point by turbidity offset, so as to eliminate the measurement error. Specifically, after obtaining the turbidity offset of each sampling point, the water flow drift obtained by the water flow fluctuation and water body microcirculation model, and the light source drift obtained by comparing the water body monitoring image with the reference image are contained, and then the original water quality turbidity data collected by the previous detection module is corrected according to the above. For example, the original turbidity data may deviate due to two conditions, one is that the water flow drives the pollutant particles to deviate from the sampling position of the detection probe, resulting in that the number of particles captured by the probe is inconsistent with the actual one; The second is that the change of light intensity or wavelength causes false fluctuation of optical detection result. In the correction process, the turbidity error caused by particle displacement will be adjusted according to the water flow drift, for example, if the pollutant particles deviate to the side of the probe due to the water flow, resulting in that the original detection value is too high, the data will be appropriately lowered according to the drift. At the same time, according to the light source drift, the deviation caused by the change of light is compensated, for example, if the light intensity is weakened, the original detection value is too low, the true turbidity level will be corrected combined with the drift. Through such bidirectional correction, the influence of various measurement interference factors can be effectively eliminated, and the water quality turbidity data of each sampling point can accurately reflect the turbidity degree of the actual water body, providing reliable basic data for subsequent construction of distribution map.

[0077] Exemplarily, the turbidity sensor (such as a multi-wavelength turbidity sensor) is essentially to calculate the number and distribution of particles by capturing the scattering and transmission signals of light of specific wavelengths by suspended particles in the water body, and then convert them into turbidity values. The two drifts are the position deviation of the detected object (suspended particles) and the state deviation of the detection tool (light source), which interfere with the accuracy of signal capture and ultimately lead to detection errors. Among them, the influence of water flow drift is due to the change of the relative position of suspended particles and the sampling area of the detection probe. The detection range of the turbidity sensor is a fixed sampling area, only the suspended particles in this area can be captured by the light and counted in the turbidity calculation. When the water flow causes the particles to drift (i.e. the deviation of the actual position of the particles and the sampling position of the probe reflected by the water flow drift), it will directly lead to the mismatch between the number of particles captured by the probe and the actual number of particles in the water body. If the water flow carries the particles away from the sampling area of the probe, the number of particles detected by the probe will be less than the true value of the water body, and the turbidity will be misjudged as being too low. If the water flow converges the particles in the surrounding area to the sampling area, such as the main flow area pushing the particles to accumulate near the probe, or the dead water area causing the particles to stay in the sampling area due to slow water flow, the particles captured by the probe will be too dense, and the turbidity will be misjudged as being too high. In addition, the water flow characteristics of different areas will also exacerbate this deviation, such as the water flow in the shallow water area being disturbed due to the friction of the pool bottom, the particle drift trajectory being unstable, and the probe being difficult to continuously capture the true particle distribution. Although the dead water area has a small drift, the particles are easy to deposit in the sampling area, which may cause the local detection value to be high for a long time. The influence of light source drift focuses on the stability of the detection light source, which directly destroys the baseline conditions of the interaction between light and particles. The multi-wavelength turbidity sensor relies on stable light source intensity, fixed wavelength parameters, and clear light path (light emission → through the water body → captured by the receiver) to ensure the accuracy of the signal. When the light source drifts (i.e. the change of the light source state reflected by the difference in visibility between the water body monitoring image and the reference image), it will interfere with the detection from multiple aspects. If the light source intensity decreases, even if the number of particles in the water body does not change, the signal scattered and absorbed by the particles when the light passes through the water body will be weakened due to the insufficient initial intensity, and the sensor will misinterpret this weak signal as a small number of particles, resulting in a low turbidity detection value; if the light source wavelength shifts (deviates from the preset detection wavelength), it will change the scattering sensitivity of light to particles of different sizes. For example, if the wavelength that was originally suitable for detecting small particles shifts, the scattering signal capture ability for small particles will decrease, even if the number of small particles in the water body increases, the sensor will not be able to accurately identify it, and thus underestimate the turbidity. In addition, the shift of the light path of the light source (such as the change of the refractive index of the water body caused by water flow fluctuation, or the change of the light path caused by the attachment of impurities on the surface of the probe due to water flow) will also cause the distortion of the light signal captured by the receiver, such as the shift of the light path making part of the scattered light unable to be received, which will also cause the sensor to misjudge that the turbidity is lower than the actual value.In short, the water flow drift affects the "source (particle) quantity" of the signal by changing "whether the particle is within the detection range", and the light source drift affects the "transmission (light) quality" of the signal by changing "whether the light stably acts on the particle", both of which ultimately cause the turbidity detection value to deviate from the true turbidity of the water body, so it is necessary to correct these two types of deviations.

[0078] Step S106, based on the corrected water turbidity data, a water turbidity distribution map corresponding to the three-dimensional model of the swimming pool is constructed and dynamically updated.

[0079] Specifically, the corrected water turbidity data of each sampling point is spatially correlated with the previously constructed three-dimensional model of the swimming pool, and the turbidity value of each sampling point is accurately corresponded to the three-dimensional coordinates of the point in the model. For the areas between the sampling points that are not directly detected, the turbidity values of these blank areas can be estimated through reasonable interpolation calculation combined with the turbidity data of the surrounding detected points and the spatial properties of the area (such as whether it is a dead water area, water flow direction), and finally integrated to form a three-dimensional water turbidity distribution map covering the entire swimming pool. The distribution map can intuitively present the turbidity difference of different areas, such as the edge of the shallow water area with high turbidity due to leaf accumulation, the main flow area of the deep water area with low turbidity, and the dead water area with significantly higher turbidity than other areas due to pollutant accumulation. At the same time, as the swimming pool circulating robot continues to detect, the newly obtained corrected turbidity data will be updated to the three-dimensional model in real time, dynamically adjusting the original turbidity distribution map. If a certain area has a lower turbidity after cleaning, or the local turbidity increases due to external factors, the distribution map will timely reflect these changes, ensuring that the operating personnel can grasp the latest status of the swimming pool water quality at any time, and providing intuitive and accurate reference for subsequent formulation of accurate water quality improvement plan (such as strengthening circulation and filtration in high turbidity area, supplementing disinfectant).

[0080] Further optionally, 10% of the sampling point data is automatically extracted for parallel sample analysis after daily detection, with a relative deviation of less than 10%, and a standard addition recovery rate test is performed once a month, adding standard suspended solids to the standard solution with known turbidity to ensure that the recovery rate is within the interval of 85% to 115%. All detection data is stored in the database after three-level audit, forming a complete traceability chain including detection time, location, water quality parameters and sensor status, providing data support for subsequent cleaning strategy optimization.

[0081] Exemplarily, after the daily detection work is completed, a certain proportion of data from all sampling points on the same day is extracted for parallel sample analysis. Through repeated detection of water samples corresponding to the same sampling point and comparison of the relative deviation of the two detection results, it is ensured that the deviation is controlled within the set standard, so as to verify the stability and accuracy of the turbidity detection data on the same day, avoid the influence of single detection error caused by instantaneous failure of the sensor or sudden fluctuation of the water flow on the overall data reliability. In addition, a standard recovery rate test is also regularly performed every month. During the test, standard suspended solids are added to the standard solution with known turbidity to simulate the state of the presence of pollutant particles in the actual swimming pool. By detecting the turbidity value of the solution after adding the standard suspended solids, the actual recovery rate of the standard suspended solids is calculated, and it is ensured that the recovery rate is maintained within a reasonable range. This process can effectively verify the detection accuracy of the detection module (such as the multi-wavelength turbidity sensor) and the reliability of the entire detection process, and timely find the potential problems in the sensor calibration deviation or the detection method. All the data that have completed detection and verification also need to go through a three-level audit process, which includes layer-by-layer checking of data collection accuracy, deviation correction rationality and result integrity. After the audit is passed, the data is uniformly stored in the database. The database will record the detection time, sampling point position, various water quality parameters (such as turbidity, residual chlorine, pH value) and the working state of the sensor (such as whether to use the standby wavelength channel, calibration time) corresponding to each data, forming a complete data chain that can be traced. These long-term accumulated data not only can reflect the change trend of the swimming pool water quality, but also can provide solid data support for subsequent optimization of cleaning strategies (such as adjusting the cleaning frequency of high-turbidity areas and optimizing the cruise path of the circulating robot).

[0082] In the embodiment of the application, by constructing a three-dimensional model of the swimming pool and a multi-dimensional environmental monitoring method, the distribution state of pollutants in each local area of the swimming pool can be predicted, and the cruise detection strategy of the swimming pool circulating robot is set based on the prediction result, thereby improving the adaptability of the swimming pool detection strategy to the swimming pool space structure and avoiding the occurrence of detection blind area. Further, the water quality turbidity data of each sampling point is detected by the cruise detection strategy, and the water quality turbidity data sampling result is corrected by means of water flow fluctuation monitoring and water body image monitoring, and finally a water quality turbidity distribution map corresponding to the three-dimensional model of the swimming pool is constructed based on the corrected sampling result, thereby realizing comprehensive detection and stereoscopic presentation of the swimming pool, improving the accuracy and efficiency of the swimming pool water quality detection, ensuring the cleaning ability of the swimming pool circulating robot, and reducing the energy consumption of the swimming pool circulating robot.

[0083] The water quality turbidity detection device of the pool circulating robot provided by the embodiment of the present application comprises: a construction module, which is configured to scan the pool contour of a target pool by a laser radar and a water pressure sensor, and to construct a pool three-dimensional model containing water depth, water inlet position, water outlet position and water flow direction; the boundary range of each feature area in the pool three-dimensional model is determined by the spatial structure, structural function and water body circulation mode of the target pool; a strategy module, which is configured to monitor multi-dimensional environmental data of the target pool, the multi-dimensional environmental data comprising temperature, weather data, illumination condition, pool peripheral objects and the position of the pool, and to predict the pollutant distribution state of each feature area in the pool three-dimensional model based on the multi-dimensional environmental data; to generate a cruise detection strategy based on the pollutant distribution state of each feature area, and to dynamically set the sampling points in each feature area according to the cruise detection strategy; and a cruise module, which is configured to detect the water quality turbidity data of each sampling point by a detection module during the cruise of the pool circulating robot, the detection module being integrated with a multi-wavelength turbidity sensor, a residual chlorine electrode and a pH probe; to synchronously detect the water flow fluctuation data and water body monitoring images of each sampling point to obtain the turbidity offset of each sampling point; to correct the water quality turbidity data of each sampling point by the turbidity offset; and to construct and dynamically update the water quality turbidity distribution map corresponding to the pool three-dimensional model based on the corrected water quality turbidity data. In some embodiments, the water quality turbidity detection device of the pool circulating robot can be applied to a terminal device. It should be noted that, for the convenience and brevity of description, the specific working process of the water quality turbidity detection device of the pool circulating robot described above can refer to the corresponding process in the foregoing embodiment of the pool circulating robot water quality turbidity detection method, which will not be described here again.

[0084] The terminal device provided by the embodiment of the present application. The terminal device comprises a processor and a memory, and the processor and the memory are connected through a bus, such as an I2C bus. Specifically, the processor is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor can be a central processing unit, and the processor can also be other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like. Those skilled in the art can understand that the structure shown in the above embodiment is only a block diagram of part of the structure related to the scheme of the embodiment of the present application, and does not constitute a limitation on the terminal device to which the scheme of the embodiment of the present application is applied. Specifically, the server can comprise more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement. The processor is used to run a computer program stored in the memory and implement any one of the water turbidity detection methods of the pool circulating robot provided by the embodiment of the present application when the computer program is executed. It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the terminal device described above can refer to the foregoing embodiment of the water turbidity detection method of the pool circulating robot, and will not be described here.

[0085] The embodiment of the present application also provides a storage medium for computer readable storage, and the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the water turbidity detection methods of the pool circulating robot provided by the specification of the embodiment of the present application.

Claims

1. A method for detecting water turbidity in a swimming pool circulation robot, characterized in that, The method includes: The outline of the target swimming pool is scanned by lidar and water pressure sensor to construct a three-dimensional model of the pool, including water depth, inlet location, outlet location, and water flow direction; the boundary range of each feature area in the three-dimensional model of the pool is determined by the spatial structure, structural function, and water circulation mode of the target pool. The system monitors multidimensional environmental data of the target swimming pool, including temperature, weather data, lighting conditions, objects around the pool, and the location of the pool. Based on the multidimensional environmental data, it predicts the distribution of pollutants in each characteristic area of ​​the pool's three-dimensional model. Based on the pollutant distribution in each characteristic area, it generates a cruise detection strategy and dynamically sets sampling points in each characteristic area according to the cruise detection strategy. During the swimming pool circulation robot's cruise, the turbidity data of the water at each sampling point is detected by a detection module. The detection module integrates a multi-wavelength turbidity sensor, a residual chlorine electrode, and a pH probe. Simultaneously, the water flow fluctuation data and water monitoring images at each sampling point are detected to obtain the turbidity offset of each sampling point. The turbidity data of each sampling point is then corrected using the turbidity offset. Based on the corrected water turbidity data, a water turbidity distribution map corresponding to the three-dimensional model of the swimming pool is constructed and dynamically updated.

2. The method for detecting water turbidity in a swimming pool circulation robot according to claim 1, characterized in that, The prediction of pollutant distribution in each feature region of the three-dimensional swimming pool model based on the multidimensional environmental data includes: Based on the objects surrounding the pool and the location of the pool, identify the types of microorganisms and algae contained in each characteristic area; Based on temperature, weather data, and illumination conditions, the Transformer network algorithm with physical information enhancement predicts the types of microorganisms and the pollution range covered by algae in each feature region, as well as the trend of pollutant changes, and obtains the initial pollutant distribution status of each feature region. By using a physical information graph neural network algorithm, combined with the initial pollutant distribution status of each feature region, the cross-propagation of pollutants between adjacent feature regions is predicted, and the final output pollutant distribution status of each feature region is obtained.

3. The method for detecting water turbidity in a swimming pool circulation robot according to claim 2, characterized in that, Based on temperature, weather data, and illumination conditions, the Transformer network algorithm with physical information enhancement predicts the types of microorganisms and the extent of pollution covered by algae in each characteristic region, as well as the trend of pollutant changes, thus obtaining the initial pollutant distribution status of each characteristic region, including: Extract pollutant features from each feature region, multidimensional environmental change features from the multidimensional environmental data, and spatial attribute features from each feature region in the three-dimensional pool model; A dynamic spatiotemporal graph structure matching the 3D model of the swimming pool is constructed as a dynamic spatiotemporal graph Transformer network. Each feature region is used as a graph node. The growth characteristics of microorganisms or algae, temperature, light adaptability and corresponding water depth data are integrated into the node features of the graph nodes. The edge weights between graph nodes are dynamically adjusted according to the direction of water flow in the pool and the distance between feature regions to quantify the migration correlation of pollutants between different feature regions. The continuity equation and momentum equation of fluid dynamics are used as prior knowledge and embedded into the dynamic spatiotemporal graph Transformer network through a differentiable constraint layer so that the output of the dynamic spatiotemporal graph Transformer network conforms to the physical laws of mass conservation and momentum balance of pool water. A hierarchical feature extraction strategy is adopted to capture the local growth and change features of microorganisms and algae in a single feature region through a local sliding window, and to model the long-range environmental impact across feature regions through a global attention mechanism, so as to capture the change features of the pool environment at different granularities and in different dimensions in real time. The cross-attention mechanism integrates multi-dimensional environmental change features to obtain feature interaction results. The cross-attention mechanism is used to associate the mapping relationship between temperature and microbial reproduction rate, light and algal growth cycle, and weather data and pollutant diffusion intensity. The feature interaction results drive the dynamic spatiotemporal graph Transformer network to simulate the spatial coverage of microbial types and algae in each feature region and their changing trends over time. In the constraint strength adjustment layer, a soft constraint loss function is adopted to dynamically adjust the constraint weights corresponding to the continuity equation and momentum equation according to the water flow difference characteristics of each feature region. In the error correction mechanism, the data fitting deviation is corrected by comparing the adaptability of the pollutant distribution predicted by the model with the water flow direction in the three-dimensional model of the pool, and finally the initial pollutant distribution state of each feature region is output.

4. The method for detecting water turbidity in a swimming pool circulation robot according to claim 3, characterized in that, The method uses a physical information graph neural network algorithm, combined with the initial pollutant distribution state of each feature region, to predict the cross-propagation of pollutants between adjacent feature regions, obtaining the final output pollutant distribution state of each feature region, including: The initial pollutant distribution status of each characteristic area, the water depth, water flow direction and water circulation mode of each characteristic area contained in the three-dimensional model of the swimming pool, and the temperature and weather data related to water flow in the multi-dimensional environmental data are obtained. A graph structure adapted to the 3D model of the swimming pool is constructed as a physical information graph neural network. Each feature region is used as a graph node in the physical information graph neural network. The graph node features are integrated with the initial pollutant distribution state of each feature region, water flow velocity, and water depth data derived based on water flow direction and water circulation mode. The edge weights between graph nodes are dynamically set according to the water flow connectivity, spacing, and water flow velocity of adjacent feature regions to quantify the basic correlation strength of pollutant migration between adjacent feature regions. Water flow connectivity includes the water flow channels between the deep and shallow water areas in a downstream swimming pool. By using a physical information graph neural network, the law of pollutant migration driven by water pressure difference in hydraulic principle is combined with the depth of graph convolutional network. A message-passing-based algorithm is used to realize the interaction of pollutant migration information. The message-passing module transmits pollutant concentration and water flow state characteristics between adjacent graph nodes, calculates the pollutant migration potential between adjacent feature regions, and the migration potential between feature regions with larger water pressure differences is higher. The pollutant migration amount between graph nodes is updated iteratively using a recursive method. For the characteristic regions formed by spatial structure and water circulation, including dead water areas, deep water areas and shallow water areas, the diffusion amount from the mainstream area to the dead water area and the transport amount from the deep water area to the shallow water area along the water flow direction are calculated to construct a pollutant propagation path model between the characteristic regions. By simulating the pollutant migration patterns between the dead water zone and the mainstream zone, as well as between the deep water zone and the shallow water zone in the target swimming pool using a pollutant propagation path model, the deviation caused by not considering cross-propagation in the initial pollutant distribution state is corrected, and the pollutant distribution state of each characteristic area after cross-propagation correction is obtained.

5. The method for detecting water turbidity in a swimming pool circulation robot according to claim 4, characterized in that, Before simulating the pollutant migration patterns between dead water zones and mainstream water zones, and between deep water zones and shallow water zones in the target swimming pool using a pollutant propagation path model, and correcting for deviations in the initial pollutant distribution state caused by not considering cross-propagation, to obtain the pollutant distribution state of each characteristic region after cross-propagation correction, the process further includes: In the unsupervised learning process, physical constraints adapted to fluid dynamics are incorporated. These constraints include: the mass conservation condition for pollutant migration is matched with the continuity equation, and the mass conservation condition is used to ensure that the total amount of pollutants does not increase or decrease out of thin air during the migration process; the constraints also include: a pollutant migration rate formula driven by water flow, which is obtained by combining the water viscosity coefficient derived from water temperature. A constraint loss function is constructed using the aforementioned constraints to correct the output of the physical information graph neural network, ensuring that the output conforms to the actual migration patterns of pollutants in swimming pool water.

6. The method for detecting water turbidity in a swimming pool circulation robot according to claim 1, characterized in that, The cruise detection strategy is generated based on the pollutant distribution status of each characteristic region, and sampling points within each characteristic region are dynamically set according to the cruise detection strategy, including: Based on the pollutant distribution and regional attributes of each characteristic area, detection priorities are assigned to each characteristic area. Areas with pollutant concentrations above a preset threshold, or those containing pathogenic microorganisms or with a high risk of algal aggregation, are classified as high-priority detection areas. Areas with pollutant concentrations within the threshold range, lacking pathogenic bacteria but containing common microorganisms, are classified as medium-priority detection areas. Areas with pollutant concentrations below the preset threshold and lacking high-risk microorganisms are classified as low-priority detection areas. Due to the tendency for pollutants to accumulate in stagnant water areas, the detection priority of stagnant water areas is increased by one level within the same pollutant concentration range. The sampling point parameters for each feature region are dynamically set according to the detection priority. With the goal of prioritizing coverage of high-priority feature regions and connecting paths between adjacent feature regions, a cruise path, cruise time, and cruise frequency are generated that match the sampling point parameters of each feature region in the 3D model of the swimming pool, so as to form a corresponding cruise detection strategy.

7. The method for detecting water turbidity in a swimming pool circulation robot according to claim 1, characterized in that, The synchronous detection of water flow fluctuation data and water monitoring images at each sampling point to obtain the turbidity shift at each sampling point includes: A microcirculation model of the water body in the local area where each sampling point is located was established using fluid dynamics based on water flow fluctuation data. The water flow drift at each sampling point is generated based on the aforementioned water microcirculation model. The visibility difference between the water monitoring image and the water reference image corresponding to each sampling point in the preset reference image library is predicted to obtain the light source drift of each sampling point.

8. The method for detecting water turbidity in a swimming pool circulation robot according to claim 7, characterized in that, The method employs fluid dynamics to establish a microcirculation model of the water body within the local area of ​​each sampling point based on water flow fluctuation data, including: Acquire water flow fluctuation data at each sampling point, including the real-time flow velocity, flow direction change amplitude, and fluctuation frequency of the water at the sampling point; combine the spatial attributes of the characteristic region where the sampling point is located in the three-dimensional model of the pool to determine the boundary conditions of the local region, including the pool wall friction coefficient and the location of the water inlet and outlet; the spatial attributes of the characteristic region include at least one of the following: water depth of the local region, distance from the inlet, distance from the outlet, whether it is a dead water zone, whether it is a mainstream zone, and the distribution of surrounding obstacles. Based on the local flow field analysis method in fluid dynamics, the Navier-Stokes equations are used to adapt to the small-scale water flow characteristics of the swimming pool. Water flow fluctuation data is used as input boundary conditions, and spatial attribute parameters of the characteristic region are coupled to construct a water microcirculation model that reflects eddies, backflows, and velocity stratification in the local area. The model can output the instantaneous water flow vector at any location in the local area. The spatial attribute parameters include the low velocity baseline value of the dead water zone and the velocity gradient of the mainstream zone.

9. The method for detecting water turbidity in a swimming pool circulation robot according to claim 8, characterized in that, The generation of water flow drift at each sampling point based on the water microcirculation model includes: The instantaneous water flow vector at the sampling point is extracted by a water microcirculation model to construct a water flow vector time series; based on the motion law of suspended particles in the fluid, the displacement deviation of pollutant particles in the water from the initial position of the sampling point is calculated under the action of the water flow vector time series; the pollutant particles are matched with the particle diameter and density of the identified microorganisms or algae. The displacement deviation is decomposed into a primary drift along the water flow direction and a secondary drift perpendicular to the water flow direction. The deviation between the actual position of the pollutant particles at the sampling point and the sampling position of the detection probe is obtained by combining the primary drift and the secondary drift, which is taken as the water flow drift.

10. A water turbidity detection device for a swimming pool circulation robot, characterized in that, The device includes: The construction module is used to scan the outline of the target swimming pool using lidar and water pressure sensors, and construct a three-dimensional model of the swimming pool that includes water depth, inlet location, outlet location, and water flow direction. The boundary range of each feature area in the three-dimensional model of the swimming pool is determined by the spatial structure, structural function, and water circulation mode of the target swimming pool. The strategy module is used to monitor multi-dimensional environmental data of the target swimming pool, including temperature, weather data, lighting conditions, objects around the pool, and the location of the pool. Based on the multi-dimensional environmental data, the module predicts the distribution of pollutants in each feature area of ​​the three-dimensional model of the pool. Based on the distribution of pollutants in each feature area, a cruise detection strategy is generated, and sampling points in each feature area are dynamically set according to the cruise detection strategy. The cruise module is used to detect water turbidity data at various sampling points during the cruise of the pool circulation robot. The detection module integrates a multi-wavelength turbidity sensor, a residual chlorine electrode, and a pH probe. It simultaneously detects water flow fluctuation data and water monitoring images at each sampling point to obtain the turbidity offset of each sampling point. The turbidity data of each sampling point is corrected using the turbidity offset. Based on the corrected water turbidity data, a water turbidity distribution map corresponding to the three-dimensional model of the pool is constructed and dynamically updated.

Images