Energy-saving industrial wastewater treatment method, equipment and system based on Internet of Things

Abstract

The invention discloses an energy-saving industrial wastewater treatment method, equipment and system based on Internet of Things, and belongs to the technical field of industrial wastewater treatment. When the industrial wastewater treatment equipment executes an industrial wastewater treatment task, visual data capable of expressing the pollutant concentration and the pollutant area in a scene and instruction data under the scene are combined, and the track of the industrial wastewater treatment equipment is controlled according to the pollutant concentration and the pollutant area; compared with a traversal mode for dosing, waste of chemicals is avoided, and the operation cost of industrial wastewater treatment equipment is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial wastewater treatment, and particularly relates to an energy-saving industrial wastewater treatment method, device and system based on the Internet of Things. Background Art

[0002] Since industrial sewage contains a large amount of harmful substances, it is particularly important to treat industrial sewage in industrial production, purify the harmful substances therein, and make it meet the water quality requirements for reuse.

[0003] The existing industrial sewage treatment mainly relies on a fixed dosing scheme, either by extensive dosing or by intelligent devices such as robots to dose the sewage in the sewage treatment chamber or sewage tank.

[0004] In the existing solutions for sewage treatment implemented by intelligent devices, the intelligent devices often traverse and dose all spaces in the sewage treatment chamber or sewage tank. The above solutions not only result in inaccurate use of the medicament, but also further increase the waste of medicament resources and the operating cost of the intelligent device. Summary of the Invention

[0005] To solve the problems of the existing technology, embodiments of the present invention provide an energy-saving industrial wastewater treatment method, device and system based on the Internet of Things, including:

[0006] On the one hand, an energy-saving industrial wastewater treatment method based on the Internet of Things is provided. The method includes:

[0007] Obtain visual data and instruction data of the industrial sewage treatment task of the industrial wastewater treatment device. The visual data is used to indicate the scene of the industrial sewage treatment task, and the instruction data is a trajectory control instruction executed on the industrial wastewater treatment device in the scene. The scene includes a sewage chamber to be treated or a sewage tank to be treated;

[0008] Construct an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data;

[0009] Train the initial industrial wastewater treatment model according to the actual visual data and actual instruction data in the scene;

[0010] Train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain an industrial wastewater treatment model for trajectory control;

[0011] Control the industrial wastewater treatment device to treat the sewage in the scene according to the industrial wastewater treatment model.

[0012] Optionally, the obtaining of the visual data and instruction data of the industrial wastewater treatment equipment includes:

[0013] Generating the visual data according to the two-dimensional RGB image and three-dimensional point cloud data of the scene;

[0014] Generating the instruction data according to the trajectory control instruction for the industrial wastewater treatment equipment within the scene.

[0015] Optionally, the constructing of the initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data includes:

[0016] Setting a visual recognition model corresponding to the visual data and a three-dimensional positioning mechanism corresponding to the visual data;

[0017] Setting a language model corresponding to the instruction data and an instruction description corresponding to the instruction data;

[0018] Aligning the visual recognition model and the language model according to the three-dimensional positioning mechanism and the instruction description to generate the industrial wastewater treatment model.

[0019] Optionally, the training of the initial industrial wastewater treatment model according to the visual data and instruction data includes:

[0020] Fusing the visual data and the instruction data into visual-language data according to visual question answering;

[0021] Training the industrial wastewater treatment model according to the visual-language data to obtain the trained industrial wastewater treatment model.

[0022] Optionally, the training of the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain an industrial wastewater treatment model for trajectory control includes:

[0023] Constructing a Generalized Voronoi Diagram (GVD) on the two-dimensional semantic map;

[0024] Extracting the nodes and edges within the Generalized Voronoi Diagram to construct the semantic map;

[0025] On the basis of the semantic map, setting trajectory control instructions and navigation result data corresponding to the trajectory control instructions;

[0026] Training the trained industrial wastewater treatment model according to the semantic map and navigation result data to obtain an industrial wastewater treatment model for trajectory control.

[0027] Optionally, the method further includes:

[0028] Based on the source domain corresponding to the scenario data, identify the scenario and the high-concentration pollution areas within the scenario, where the source domain is a set of parameters of the visual recognition model;

[0029] Based on the target domain corresponding to the control data, calculate a movement strategy for the high-concentration pollution areas; the target domain is a set of parameters of the language model;

[0030] Align the source domain and the target domain, and fine-tune the industrial wastewater treatment model.

[0031] Optionally, the identifying the scenario and the high-concentration pollution areas within the scenario based on the source domain corresponding to the scenario data includes:

[0032] Based on multiple scenario pictures and high-concentration pollution area description information, identify the scenario and the high-concentration pollution areas within the scenario, and obtain the category of the high-concentration pollution areas, the size of the high-concentration pollution areas, and the position of the high-concentration pollution areas in space. The high-concentration pollution areas include cluster-shaped pollution areas and pollution degree deposition blocks.

[0033] Optionally,

[0034] The calculating a movement strategy for the high-concentration pollution areas based on the target domain corresponding to the control data includes:

[0035] Based on multiple scenario pictures and multiple scenario description information, identify the space between the industrial wastewater treatment equipment and the high-concentration pollution areas, and obtain the movement strategy between the industrial wastewater treatment equipment and the high-concentration pollution areas through the position of the industrial wastewater treatment equipment and the position of the high-concentration pollution areas in space;

[0036] The aligning the source domain and the target domain, and fine-tuning the industrial wastewater treatment model includes:

[0037] Obtain a first adaptive weight set corresponding to the source domain respectively, and a second adaptive weight set corresponding to the target domain respectively;

[0038] Based on the first adaptive weight set and the second adaptive weight set, merge the source domain and the target domain to obtain an adjusted data set;

[0039] Based on the adjusted data set, fine-tune the industrial wastewater treatment model.

[0040] On the other hand, an energy-saving industrial wastewater treatment system based on the Internet of Things is provided. The system includes industrial wastewater treatment equipment and a processing device. The industrial wastewater treatment equipment is configured with multiple cameras and a voice module. The multiple cameras are used to obtain multiple two-dimensional RGB images and actual visual data. The voice module is used to obtain scene description information, high-concentration pollution area description information, and instruction data. The instruction data includes trajectory control instructions for the industrial wastewater treatment equipment under the scene description information and high-concentration pollution area description information. The processing device is specifically used for:

[0041] Obtain visual data and instruction data for the industrial sewage treatment task of the industrial wastewater treatment equipment. The visual data is used to indicate the scene of the industrial sewage treatment task. The instruction data is the trajectory control instruction executed on the industrial wastewater treatment equipment within the scene. The scene includes a sewage chamber to be treated or a sewage pond to be treated;

[0042] Construct an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data;

[0043] Train the initial industrial wastewater treatment model according to the actual visual data and actual instruction data within the scene;

[0044] Train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain an industrial wastewater treatment model for trajectory control;

[0045] Control the industrial wastewater treatment equipment to treat the sewage within the scene according to the industrial wastewater treatment model.

[0046] On the other hand, an energy-saving industrial wastewater treatment equipment based on the Internet of Things is also provided. The industrial wastewater treatment equipment includes:

[0047] An acquisition module, used to obtain visual data and instruction data for the industrial sewage treatment task of the industrial wastewater treatment equipment. The visual data is used to indicate the scene of the industrial sewage treatment task. The instruction data is used to indicate the trajectory control instruction executed on the industrial wastewater treatment equipment within the scene data. The scene includes a sewage chamber to be treated or a sewage pond to be treated;

[0048] A construction module, used to construct an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data;

[0049] A training module, used to train the initial industrial wastewater treatment model according to the actual visual data and actual instruction data within the scene;

[0050] The training module is further configured to train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scenario, so as to obtain an industrial wastewater treatment model for trajectory control;

[0051] An execution module, configured to control the industrial wastewater treatment equipment to treat the sewage in the scenario according to the industrial wastewater treatment model.

[0052] The present invention has at least the following beneficial effects:

[0053] Since visual data can represent pollutant concentration and pollutant area, it is possible to control the trajectory of the industrial wastewater treatment equipment according to the pollutant concentration and pollutant area by combining the visual data of the scenario and the instruction data in the scenario during the execution of the industrial sewage treatment task by the industrial wastewater treatment equipment. Compared with the method of adding medicine by traversal, it avoids the waste of medicine and reduces the operation cost of the industrial wastewater treatment equipment. Description of the Drawings

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a schematic flow chart of an energy-saving industrial wastewater treatment method based on the Internet of Things provided by an embodiment of the present invention;

[0056] Figure 2 It is a schematic flow chart of an energy-saving industrial wastewater treatment method based on the Internet of Things provided by an embodiment of the present invention;

[0057] Figure 3 It is a schematic flow chart of an energy-saving industrial wastewater treatment method based on the Internet of Things provided by an embodiment of the present invention;

[0058] Figure 4 It is a schematic diagram of an energy-saving industrial wastewater treatment system based on the Internet of Things provided by an embodiment of the present invention;

[0059] Figure 5 It is a schematic structural diagram of an energy-saving industrial wastewater treatment equipment based on the Internet of Things provided by an embodiment of the present invention. Detailed Embodiments

[0060] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0061] The knowledge domain described in the embodiments of the present invention is a set of parameters in the model, that is, all the parameters in the model. For example, the source domain is the set of all parameters in the visual recognition model; the target domain is the set of all parameters in the language model.

[0062] Refer to Figure 1 As shown, a method for energy-saving industrial wastewater treatment based on the Internet of Things is provided. The method includes:

[0063] 101. Obtain the visual data and instruction data of the industrial wastewater treatment equipment;

[0064] Among them, the visual data is used to indicate the scene of the industrial sewage treatment task, and the instruction data is the trajectory control instruction executed on the industrial wastewater treatment equipment in the scene. The scene includes the sewage chamber or sewage pool to be treated;

[0065] 102. Construct an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data;

[0066] 103. Train the initial industrial wastewater treatment model according to the visual data and instruction data in the scene;

[0067] 104. Train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain an industrial wastewater treatment model for trajectory control;

[0068] 105. Control the industrial wastewater treatment equipment to treat the sewage in the scene according to the industrial wastewater treatment model.

[0069] Optionally, obtaining the visual data and instruction data of the industrial wastewater treatment equipment includes:

[0070] Generate visual data according to the two-dimensional RGB image and three-dimensional point cloud data of the scene. This process can be specifically:

[0071] Obtain the two-dimensional RGB image and three-dimensional point cloud data;

[0072] Estimate the pixel-level depth information from one or more two-dimensional images, and "extend" the two-dimensional image information to the three-dimensional space to form a voxel grid structure;

[0073] The voxel grid is fused with the voxel grid extracted by the 3D encoder to construct visual data.

[0074] According to the trajectory control instruction for the industrial wastewater treatment equipment within the scene, instruction data is generated, including setting the cross-attention module;

[0075] By taking the processed fused visual data embedding and the optimized learnable query as inputs, and outputting the transformed query;

[0076] The transformed query is then used as a soft prompt and input into a pre-trained language model with fixed parameters;

[0077] According to the trajectory control instruction for the industrial wastewater treatment equipment within the scene, it is input into this language model to generate instruction data.

[0078] Optionally, a learnable adapter is embedded in the linear layer of the language model.

[0079] Optionally, referring to Figure 2 As shown, according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data, an initial industrial wastewater treatment model is constructed, including:

[0080] 201. Set the visual recognition model corresponding to the visual data and the three-dimensional positioning mechanism corresponding to the visual data; specifically, this three-dimensional positioning mechanism includes:

[0081] Integrate the three-dimensional position embedding into the extracted three-dimensional features. Specifically, process the three-dimensional feature X and the three-dimensional position embedding P through a specific function f to enhance its encoding ability for spatial information.

[0082] Z = f(X, P)

[0083] Among them, Z represents the feature representation integrated with three-dimensional position information, X represents the original three-dimensional feature, and P is the corresponding three-dimensional position embedding. The position embedding can be further defined as a function g, which generates the position embedding according to the three-dimensional coordinates c = (x, y, z):

[0084] P = g(c)

[0085] Enhancing the model's encoding ability for spatial information can be described by an optimization objective. For example, minimize the difference between the predicted position and the actual position. If Z′ is the output of the model based on Z (for example, the predicted spatial feature or position), and Y is the true target position or feature, then the optimization objective can be expressed as:

[0086] min L(Z′, Y) (3)

[0087] Among them, L is a loss function used to measure the difference between Z′ and Y. These position embeddings are designed to intuitively represent the relative positions and orientations of high-concentration pollution areas in three-dimensional space.

[0088] Optionally, a position identifier can also be set, which is used to guide the model to identify and describe the exact positions of specific high-concentration pollution areas in the scene.

[0089] By using these position markers as reference points during training, the model is trained to accurately capture and reflect the three-dimensional spatial relationships of high-concentration pollution areas within the scene.

[0090] 202. Set a language model corresponding to the instruction data and an instruction description corresponding to the instruction data;

[0091] 203. Align the visual recognition model and the language model according to the three-dimensional positioning mechanism and the instruction description to generate an industrial wastewater treatment model.

[0092] According to the three-dimensional positioning mechanism and the instruction description, a connection is established between the instruction description and the three-dimensional spatial position; that is, the three-dimensional spatial position output by the visual recognition model is aligned with the line of sight of the instruction description output by the language model. This process can be specifically as follows:

[0093] Freeze the 2D visual encoder and the 3D point cloud encoder;

[0094] Set up pure text generation training for the text corpus corresponding to the 3D point cloud data;

[0095] Set up a weight mixing strategy, including:

[0096] Pre-train the industrial wastewater treatment model using general vision-language data;

[0097] Fine-tune the industrial wastewater treatment model using the real data of the scene, that is, real vision-language data;

[0098] θ mix =β·θ general +(1 - β)·θ robotic (4)

[0099] Among them, β represents the mixing coefficient, and θmix represents the weight of the language model with aggregated semantics. θgeneral and θrobotic respectively represent the weights of the language model after training on the general vision dataset and the vision dataset of the industrial wastewater treatment equipment scene.

[0100] Optionally, training the industrial wastewater treatment model according to the visual data and the instruction data includes:

[0101] Fuse visual data and instruction data into visual-language data according to visual question answering;

[0102] Through general domain datasets such as VQAV2, OKVQA, GQA, OCRVQA, VizWiz, and the visual question answering dataset to be constructed in the field of industrial wastewater treatment equipment;

[0103] Through datasets such as RefCOCO3, RefCOCOg, RefCOCO+, and Flicker 30K for reference understanding and reference generation tasks.

[0104] Train the industrial wastewater treatment model according to the visual-language data to obtain the trained industrial wastewater treatment model.

[0105] Optionally, refer to Figure 3 As shown, train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain the industrial wastewater treatment model for trajectory control, including:

[0106] 301. Construct a Generalized Voronoi Diagram (GVD) on the two-dimensional semantic map;

[0107] 302. Extract the nodes and edges within the Generalized Voronoi Diagram to construct a semantic map (RVG); among them, the node types in this semantic map are divided into industrial wastewater treatment equipment nodes, neighbor nodes, ordinary nodes, and nodes to be explored. Implement the shortest path search on the RVG to plan an efficient exploration path from the industrial wastewater treatment equipment node to the exploration node. Design a special hint for each neighbor node, which combines path description and perspective description, that is, extract the relevant information of high-concentration pollution areas and their central positions along the path from the semantic map, convert it into structured text, and combine it with a predefined template to create a format suitable for interaction with the VLM.

[0108] Optionally, set an n-dimensional binary vector (where n represents the number of neighbor nodes, and the values of the vector are 0 or 1) to represent these neighbor nodes. If a certain neighbor node can lead to the exploration node, the corresponding vector value is set to 1, otherwise it is set to 0. The true value of this vector is provided by the semantic map and is supervised by the cross-entropy loss function, and its loss tensor is denoted as So as to learn the association information between neighbor nodes and exploration nodes from the semantic map.

[0109] Evaluate whether each neighbor node is located in the area that has been explored by the industrial wastewater treatment equipment before, including:

[0110] Similarly, a neighbor node is represented by an n-dimensional binary vector. If a neighbor node is located in the explored area, the corresponding value in the vector is set to 0; otherwise, it is set to 1. In this way, the learning of the model can be supervised through the historical trajectory information in the semantic map, and the cross-entropy loss function is used to optimize the model parameters. Denote the exploration efficiency loss tensor as to enable the model to identify which neighbor nodes are new exploration points, thereby improving the navigation efficiency.

[0111] Perform semantic analysis on the RGB images of the unexplored areas, including:

[0112] Capture panoramic RGB images of the surrounding environment.

[0113] Determine the rays from the industrial wastewater treatment equipment node to each neighbor node. For each ray direction, we will select the image with the smallest angular deviation from the ray as the image corresponding to the neighbor node. Let τ t ={T t-11 ,..., T t} be the set of the center line of sight (LoS). The matching process between the image and each neighbor node N i can be defined as follows:

[0114]

[0115] s.t. T k ∈τ t

[0116] where the function g(·, ·) is used to represent the angle between two rays on the map.

[0117] By adding path descriptions and long-range descriptions to the prompt, the VLM can more accurately estimate the probability of the target high-concentration pollution area appearing on each neighbor node, and construct a semantic inference vector based on this. This vector gives the semantic inference probability values for navigating to each neighbor node. The model uses the cross-entropy loss function to update the parameters and records the loss tensor as Enhance the ability to make navigation decisions based on scene information.

[0118] 303. Based on the semantic map, set trajectory control instructions and the corresponding navigation result data;

[0119] 304. According to the semantic map and the navigation result data, train the trained industrial wastewater treatment model to obtain an industrial wastewater treatment model for trajectory control.

[0120] Optionally, the method further includes:

[0121] Identify the scenario and the high-concentration pollution areas within the scenario based on the source domain corresponding to the scenario data, where the source domain is the parameter set of the visual recognition model;

[0122] Calculate the movement strategy for the high-concentration pollution areas based on the target domain corresponding to the control data; the target domain is the parameter set of the language model;

[0123] Align the source domain and the target domain, and fine-tune the industrial wastewater treatment model.

[0124] Optionally, identifying the scenario and the high-concentration pollution areas within the scenario based on the source domain corresponding to the scenario data includes:

[0125] Identify the scenario and the high-concentration pollution areas within the scenario based on multiple scenario pictures and high-concentration pollution area description information, and obtain the category of the high-concentration pollution areas, the size of the high-concentration pollution areas, and the position of the high-concentration pollution areas in space.

[0126] Optionally, calculating the movement strategy for the high-concentration pollution areas based on the target domain corresponding to the control data includes:

[0127] Identify the space between the industrial wastewater treatment equipment and the high-concentration pollution areas based on multiple scenario pictures and scenario description information, and obtain the movement strategy between the industrial wastewater treatment equipment and the high-concentration pollution areas through the position of the industrial wastewater treatment equipment and the position of the high-concentration pollution areas in space.

[0128] Optionally, aligning the source domain and the target domain, and fine-tuning the industrial wastewater treatment model includes

[0129] Obtain the first adaptive weight set corresponding to the source domain respectively, and the second adaptive weight set corresponding to the target domain respectively;

[0130] Merge the source domain and the target domain according to the first adaptive weight set and the second adaptive weight set to obtain an adjusted dataset;

[0131] For any parameter, the above process can be:

[0132] The parameter in the adjusted dataset = the parameter of the visual recognition model * the corresponding weight in the first adaptive weight set + the parameter of the language model * the corresponding weight in the second adaptive weight set.

[0133] Fine-tune the industrial wastewater treatment model according to the adjusted dataset.

[0134] In the above process, each knowledge domain has a corresponding adaptive weight. For the above adaptive weight, the above process can be: Set the adjustment weight function through the performance feedback of the industrial wastewater treatment equipment in each scenario;

[0135] Adjust the adaptive weight based on the above adjustment weight function;

[0136] Refer to Figure 4 As shown, an energy-saving industrial wastewater treatment system based on the Internet of Things is provided. The system includes industrial wastewater treatment equipment and processing equipment. The industrial wastewater treatment equipment is configured with multiple cameras and a voice module. The multiple cameras are used to obtain multiple two-dimensional RGB images and actual visual data. The voice module is used to obtain scene description information, high-concentration pollution area description information, and instruction data. The instruction data includes trajectory control instructions of the industrial wastewater treatment equipment under the scene description information and high-concentration pollution area description information; The processing equipment is specifically used for:

[0137] Obtain the visual data and instruction data of the industrial sewage treatment task of the industrial wastewater treatment equipment. The visual data is used to indicate the scene of the industrial sewage treatment task. The instruction data is the trajectory control instruction executed by the industrial wastewater treatment equipment in the scene. The scene includes a sewage chamber or sewage pond to be treated;

[0138] Construct an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data;

[0139] Train the initial industrial wastewater treatment model according to the actual visual data and actual instruction data in the scene;

[0140] Train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain an industrial wastewater treatment model for trajectory control;

[0141] Control the industrial wastewater treatment equipment to treat the sewage in the scene according to the industrial wastewater treatment model.

[0142] Optionally, the processing equipment is specifically used for:

[0143] Generate visual data according to the two-dimensional RGB image and three-dimensional point cloud data of the scene;

[0144] Generate instruction data according to the trajectory control instruction for the industrial wastewater treatment equipment in the scene.

[0145] Optionally, the processing equipment is specifically used for:

[0146] Set a visual recognition model corresponding to the visual data and a three-dimensional positioning mechanism corresponding to the visual data;

[0147] Set a language model corresponding to the instruction data and an instruction description corresponding to the instruction data;

[0148] Align the visual recognition model and the language model according to the three-dimensional positioning mechanism and instruction description to generate an industrial wastewater treatment model.

[0149] Optionally, the processing device is specifically used for:

[0150] Fuse visual data and instruction data into visual-language data according to visual question answering;

[0151] Train the industrial wastewater treatment model according to the visual-language data to obtain a trained industrial wastewater treatment model.

[0152] Optionally, the processing device is specifically used for:

[0153] Construct a Generalized Voronoi Diagram (GVD) on the two-dimensional semantic map;

[0154] Extract the nodes and edges within the Generalized Voronoi Diagram to construct a semantic map;

[0155] On the basis of the semantic map, set trajectory control instructions and the corresponding navigation result data;

[0156] Train the trained industrial wastewater treatment model according to the semantic map and the navigation result data to obtain an industrial wastewater treatment model for trajectory control.

[0157] Optionally, the processing device is specifically used for:

[0158] Identify the scene and the high-concentration pollution areas within the scene according to the source domain corresponding to the scene data, where the source domain is the parameter set of the visual recognition model;

[0159] Calculate the movement strategy for the high-concentration pollution areas according to the target domain corresponding to the control data; the target domain is the parameter set of the language model;

[0160] Align the source domain and the target domain and fine-tune the industrial wastewater treatment model.

[0161] Optionally, the processing device is specifically used for:

[0162] Identify the scene and the high-concentration pollution areas within the scene according to multiple scene pictures and the description information of the high-concentration pollution areas, and obtain the category of the high-concentration pollution areas, the size of the high-concentration pollution areas, and the position of the high-concentration pollution areas in space. The high-concentration pollution areas include cluster-shaped pollution areas and pollution degree deposition blocks.

[0163] Optionally, the processing device is specifically used for:

[0164] Based on multiple scene pictures and multiple scene description information, identify the space between the industrial wastewater treatment equipment and the high-concentration pollution area, and obtain the movement strategy between the industrial wastewater treatment equipment and the high-concentration pollution area through the position of the industrial wastewater treatment equipment and the position of the high-concentration pollution area in the space;

[0165] Optionally, the processing equipment is specifically used for:

[0166] Obtain the first adaptive weight set corresponding to the source domain and the second adaptive weight set corresponding to the target domain respectively;

[0167] Merge the source domain and the target domain according to the first adaptive weight set and the second adaptive weight set to obtain an adjusted data set;

[0168] Fine-tune the industrial wastewater treatment model according to the adjusted data set.

[0169] Refer to Figure 5 As shown, an energy-saving industrial wastewater treatment equipment based on the Internet of Things is also provided. The industrial wastewater treatment equipment includes:

[0170] An acquisition module for acquiring visual data and instruction data of the industrial sewage treatment task of the industrial wastewater treatment equipment. The visual data is used to indicate the scene of the industrial sewage treatment task, and the instruction data is used to indicate the trajectory control instruction executed by the industrial wastewater treatment equipment in the scene data. The scene includes a sewage chamber to be treated or a sewage pond to be treated;

[0171] A construction module for constructing an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data;

[0172] A training module for training the initial industrial wastewater treatment model according to the actual visual data and actual instruction data in the scene;

[0173] The training module is also used to train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scene to obtain an industrial wastewater treatment model for trajectory control;

[0174] An execution module for controlling the industrial wastewater treatment equipment to treat the sewage in the scene according to the industrial wastewater treatment model.

[0175] The above several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0176] The technical features of the above embodiments can be combined arbitrarily (as long as there is no contradiction in the combination of these technical features). For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described; these embodiments not explicitly written out should also be considered to be within the scope described in this specification.

[0177] In the foregoing, the present invention has been described in relatively specific and detailed manner through general descriptions and specific embodiments. It should be noted that, without departing from the concept of the present invention, obviously several modifications and improvements can still be made to these specific embodiments, and these all belong to the protection scope of this application. Therefore, the protection scope of the patent of this application shall be subject to the appended claims.

[0178] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An energy-saving industrial wastewater treatment method based on the Internet of Things, characterized in that, The method includes: Obtaining visual data and instruction data of an industrial sewage treatment task of an industrial sewage treatment device, where the visual data is used to indicate the scene of the industrial sewage treatment task, the instruction data is a trajectory control instruction executed on the industrial sewage treatment device within the scene, and the scene includes a sewage chamber to be treated or a sewage pond to be treated; Constructing an initial industrial sewage treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data; Training the initial industrial sewage treatment model according to the actual visual data and actual instruction data within the scene; Training the trained industrial sewage treatment model according to the semantic map and navigation result data of the scene to obtain an industrial sewage treatment model for trajectory control; Controlling the industrial sewage treatment device to treat the sewage within the scene according to the industrial sewage treatment model.

2. The method according to claim 1, characterized in that, The obtaining of the visual data and instruction data of the industrial sewage treatment device includes: Generating the visual data according to the two-dimensional RGB image and three-dimensional point cloud data of the scene; Generating the instruction data according to the trajectory control instruction for the industrial sewage treatment device within the scene.

3. The method according to claim 2, characterized in that The constructing of the initial industrial sewage treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data includes: Setting a visual recognition model corresponding to the visual data and a three-dimensional positioning mechanism corresponding to the visual data; Setting a language model corresponding to the instruction data and an instruction description corresponding to the instruction data; Aligning the visual recognition model and the language model according to the three-dimensional positioning mechanism and the instruction description to generate the industrial sewage treatment model.

4. The method according to claim 3, characterized in that, The training of the initial industrial sewage treatment model according to the visual data and instruction data includes: Fusing the visual data and the instruction data into visual-language data according to visual question answering; Training the industrial sewage treatment model according to the visual-language data to obtain the trained industrial sewage treatment model.

5. The method according to claim 4, characterized in that, The training of the trained industrial sewage treatment model according to the semantic map and navigation result data of the scene to obtain an industrial sewage treatment model for trajectory control includes: Constructing a generalized Voronoi diagram (GVD) on a two-dimensional semantic map; Extracting nodes and edges within the generalized Voronoi diagram to construct the semantic map; On the basis of the semantic map, setting a trajectory control instruction and navigation result data corresponding to the trajectory control instruction; Training the trained industrial sewage treatment model according to the semantic map and navigation result data to obtain an industrial sewage treatment model for trajectory control.

6. The method according to claim 1, wherein The method further includes: Identifying the scene and the high-concentration pollution area within the scene according to the source domain corresponding to the scene data, where the source domain is a parameter set of the visual recognition model; Calculate a movement strategy for the high-concentration pollution area according to the target domain corresponding to the control data; the target domain is the parameter set of the language model. Align the source domain and the target domain, and fine-tune the industrial wastewater treatment model.

7. The method according to claim 6, characterized in that The identifying the scenario and the high-concentration pollution area within the scenario according to the source domain corresponding to the scenario data includes: Identify the scenario and the high-concentration pollution area within the scenario based on multiple scenario pictures and high-concentration pollution area description information, and obtain the category of the high-concentration pollution area, the size of the high-concentration pollution area, and the position of the high-concentration pollution area in space. The high-concentration pollution area includes a cluster-shaped pollution area and a pollution degree deposition block.

8. The method according to claim 6, wherein The calculating a movement strategy for the high-concentration pollution area according to the target domain corresponding to the control data includes: Identify the space between the industrial wastewater treatment equipment and the high-concentration pollution area based on multiple scenario pictures and multiple scenario description information, and obtain the movement strategy between the industrial wastewater treatment equipment and the high-concentration pollution area through the position of the industrial wastewater treatment equipment and the position of the high-concentration pollution area in space. The aligning the source domain and the target domain, and fine-tuning the industrial wastewater treatment model includes: Obtain a first adaptive weight set corresponding to the source domain respectively, and a second adaptive weight set corresponding to the target domain respectively. Merge the source domain and the target domain according to the first adaptive weight set and the second adaptive weight set to obtain an adjusted data set. Fine-tune the industrial wastewater treatment model according to the adjusted data set.

9. An energy-saving industrial wastewater treatment system based on the Internet of Things, characterized in that, The system includes an industrial wastewater treatment equipment and a processing equipment. The industrial wastewater treatment equipment is configured with multiple cameras and a voice module. The multiple cameras are used to obtain multiple two-dimensional RGB images and actual visual data. The voice module is used to obtain scenario description information, high-concentration pollution area description information, and instruction data. The instruction data includes a trajectory control instruction of the industrial wastewater treatment equipment under the scenario description information and the high-concentration pollution area description information. The processing equipment is specifically used for: Obtain visual data and instruction data of the industrial sewage treatment task of the industrial wastewater treatment equipment. The visual data is used to indicate the scenario of the industrial sewage treatment task. The instruction data is a trajectory control instruction executed by the industrial wastewater treatment equipment within the scenario. The scenario includes a sewage chamber to be treated or a sewage pond to be treated. Construct an initial industrial wastewater treatment model according to the visual recognition model corresponding to the visual data and the language model corresponding to the instruction data. Train the initial industrial wastewater treatment model according to the actual visual data and actual instruction data within the scenario. Train the trained industrial wastewater treatment model according to the semantic map and navigation result data of the scenario to obtain an industrial wastewater treatment model for trajectory control. According to the industrial wastewater treatment model, the industrial wastewater treatment equipment is controlled to treat the sewage in the scene.

10. An energy-saving industrial wastewater treatment device based on the Internet of Things, characterized in that, The industrial wastewater treatment equipment includes: an acquisition module for acquiring visual data and instruction data of an industrial wastewater treatment task for an industrial wastewater treatment device, wherein the visual data is used to indicate a scenario of the industrial wastewater treatment task, and the instruction data is used to indicate a trajectory control instruction executed by the industrial wastewater treatment device within the scenario data, wherein the scenario includes a wastewater chamber to be treated or a wastewater pool to be treated; A construction module, configured to construct an initial industrial wastewater treatment model based on a visual recognition model corresponding to the visual data and a language model corresponding to the instruction data; a training module, configured to train the initial industrial wastewater treatment model based on actual visual data and actual instruction data within the scene; The training module is further used to train the trained industrial wastewater treatment model based on the semantic map of the scene and the navigation result data to obtain an industrial wastewater treatment model for trajectory control; An execution module is used to control the industrial wastewater treatment equipment to treat the sewage in the scene according to the industrial wastewater treatment model.

Images