Intelligent adaptive carbon source dosing method and system based on deep learning

Abstract

The present invention relates to the field of artificial intelligence technology, and is a deep learning-based intelligent adaptive carbon source dosing method and system, comprising: obtaining multimodal sample data, extracting multimodal fusion features from the multimodal sample data, and then using the multimodal fusion features to construct a comprehensive water quality feature state space. Furthermore, based on a preset carbon source dosing range, a carbon source dosing action space is constructed. Secondly, an initial carbon source dosing control model is constructed using the water quality feature state space and the carbon source dosing action space. The initial carbon source dosing control model is then optimized using a preset reward function and a digital twin model to obtain a target carbon source dosing control model. Finally, the target carbon source dosing control model is applied to the current multimodal data to perform actual data prediction. The present invention can improve the accuracy of carbon source dosing.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a deep learning-based intelligent adaptive carbon source addition method, system, electronic device and computer-readable storage medium. Background Art

[0002] Carbon source addition refers to the addition of organic carbon sources in sewage treatment to meet the carbon source required for microbial denitrification, which can improve the removal efficiency of total nitrogen in sewage.

[0003] Currently, common carbon source dosing schemes are often based on fixed empirical formulas or simple feedback control, which are prone to adjustment lag and difficulty adapting to dynamically changing water quality conditions, resulting in inaccurate carbon source dosing. Therefore, how to improve the accuracy of carbon source dosing has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides an intelligent adaptive carbon source addition method based on deep learning and a computer-readable storage medium, the main purpose of which is to improve the accuracy of carbon source addition.

[0005] To achieve the above objectives, the present invention provides a deep learning-based intelligent adaptive carbon source dosing method, comprising:

[0006] Obtain water quality time series sample data, hyperspectral sample data and environmental video sample data;

[0007] Extracting features from the water quality time series sample data, hyperspectral sample data, and environmental video sample data to obtain multimodal fusion features;

[0008] Based on the multimodal fusion features, constructing a water quality feature state space;

[0009] Based on the preset carbon source addition range, the carbon source addition action space is constructed;

[0010] Based on the water quality characteristic state space and the carbon source addition action space, a model is constructed to obtain an initial carbon source addition control model;

[0011] Based on a preset reward function, a preset digital twin model and the multimodal fusion feature, reinforcement learning is performed on the initial carbon source dosing control model to obtain a target carbon source dosing control model;

[0012] Obtain current water quality time series data, current hyperspectral data and current environmental video data;

[0013] Based on the target carbon source addition control model, a carbon source addition prediction is performed on the current water quality time series data, the current hyperspectral data and the current environmental video data to obtain target carbon source addition data.

[0014] Optionally, constructing a water quality feature state space based on the multimodal fusion features includes:

[0015] Based on the multimodal fusion features, feature range calculation is performed to obtain a multimodal feature range;

[0016] Based on the multimodal feature range, performing random feature combination to obtain multiple multimodal combination features;

[0017] The multiple multimodal combination features are subjected to feature aggregation to obtain a water quality feature state space.

[0018] Optionally, constructing the carbon source addition action space based on a preset carbon source addition range includes:

[0019] Performing feature extraction on the carbon source addition range to obtain a carbon source addition amount range and a carbon source addition frequency range;

[0020] Based on the carbon source dosage range and the carbon source dosage frequency range, a carbon source dosage action combination is performed to obtain a plurality of carbon source dosage combination actions;

[0021] The multiple carbon source addition combination actions are aggregated to obtain a carbon source addition action space.

[0022] Optionally, the extracting features of the water quality time series sample data, the hyperspectral sample data and the environmental video sample data to obtain multimodal fusion features includes:

[0023] Extracting dependency features from the water quality time series sample data to obtain time series local dependency features;

[0024] Extracting spectral features from the hyperspectral sample data to obtain spatial spectral features;

[0025] Extracting dynamic features from the environmental video sample data to obtain motion features;

[0026] The temporal local dependency features, spatial spectrum features and motion features are fused to obtain multimodal fusion features.

[0027] Optionally, the initial carbon source dosing control model is subjected to reinforcement learning based on a preset reward function, a preset digital twin model, and the multimodal fusion feature to obtain a target carbon source dosing control model, including:

[0028] Based on the multimodal fusion features, predicting the carbon source addition action of the initial carbon source addition control model to obtain a predicted carbon source addition action;

[0029] Based on the reward function and the digital twin model, an action evaluation is performed on the multimodal fusion feature and the predicted carbon source addition action to obtain carbon source addition evaluation data;

[0030] Based on the carbon source addition evaluation data, the initial carbon source addition control model is iteratively optimized to obtain a target carbon source addition control model.

[0031] Optionally, based on the reward function and the digital twin model, performing action evaluation on the multimodal fusion feature and the predicted carbon source addition action to obtain carbon source addition evaluation data includes:

[0032] Based on the multimodal fusion features and the predicted carbon source addition action, the digital twin model is simulated to obtain sewage treatment data;

[0033] Based on the sewage treatment data, sewage treatment performance calculation is performed to obtain target performance data;

[0034] Based on the reward function, data evaluation is performed on the target performance data to obtain carbon source addition evaluation data.

[0035] Optionally, the sewage treatment performance calculation based on the sewage treatment data to obtain target performance data includes:

[0036] Extracting features from the sewage treatment data to obtain inlet total nitrogen concentration data, effluent total nitrogen concentration data, simulated carbon source dosage, evaluation time, and evaluation times;

[0037] Based on the inlet total nitrogen concentration data, the outlet total nitrogen concentration data and the number of evaluations, the water quality total nitrogen index is estimated to obtain total nitrogen treatment performance data;

[0038] Based on the simulated carbon source dosage and evaluation time, a stability evaluation is performed to obtain stability performance data;

[0039] The total nitrogen treatment performance data and the stability performance data are integrated to obtain target performance data.

[0040] Optionally, the water quality total nitrogen index is estimated based on the inlet total nitrogen concentration data, the outlet total nitrogen concentration data and the number of evaluations to obtain total nitrogen treatment performance data, including:

[0041] Calculate the effluent total nitrogen concentration compliance rate based on a preset total nitrogen concentration threshold, the effluent total nitrogen concentration data, and the number of evaluations;

[0042] The average effluent total nitrogen concentration of the sewage treatment data is calculated using the effluent total nitrogen concentration data, the number of evaluations, and the pre-constructed average effluent total nitrogen concentration calculation formula, wherein the average effluent total nitrogen concentration calculation formula is as follows:

[0043] ;

[0044] in, represents the average effluent total nitrogen concentration, represents the number of evaluations, Indicates the The effluent total nitrogen concentration data for this assessment;

[0045] The total nitrogen removal rate of the sewage treatment data is calculated using the influent total nitrogen concentration data, the effluent total nitrogen concentration data, and a pre-established total nitrogen removal rate calculation formula, wherein the total nitrogen removal rate calculation formula is as follows:

[0046] ;

[0047] in, represents the total nitrogen removal rate, Indicates the total nitrogen concentration data of the influent, Indicates the total nitrogen concentration data of the effluent;

[0048] The data of the effluent total nitrogen concentration compliance rate, the average effluent total nitrogen concentration and the total nitrogen removal rate are integrated to obtain the total nitrogen treatment performance data.

[0049] Optionally, the stability evaluation is performed based on the simulated carbon source dosage and evaluation time to obtain stability performance data, including:

[0050] The stability performance data of the sewage treatment data is calculated using the simulated carbon source dosage, evaluation time, and a pre-built stability evaluation formula, wherein the stability evaluation formula is as follows:

[0051] ;

[0052] Wherein, S represents the stability performance data, represents the evaluation time, represents the time point of stability evaluation, Indicates time The simulated carbon source dosage.

[0053] To achieve the above objectives, the present invention also provides an intelligent adaptive carbon source dosing system based on deep learning, comprising:

[0054] A data feature extraction module is used to obtain water quality time series sample data, hyperspectral sample data and environmental video sample data, perform feature extraction on the water quality time series sample data, hyperspectral sample data and environmental video sample data, and obtain multimodal fusion features;

[0055] A target model training module is used to construct a water quality feature state space based on the multimodal fusion feature, construct a carbon source addition action space based on a preset carbon source addition range, perform model construction based on the water quality feature state space and the carbon source addition action space, obtain an initial carbon source addition control model, and perform reinforcement learning on the initial carbon source addition control model based on a preset reward function, a preset digital twin model and the multimodal fusion feature to obtain a target carbon source addition control model;

[0056] Current data acquisition module, used to obtain current water quality time series data, current hyperspectral data and current environmental video data;

[0057] The carbon source addition control module is used to perform carbon source addition prediction on the current water quality time series data, the current hyperspectral data and the current environmental video data based on the target carbon source addition control model to obtain target carbon source addition data.

[0058] In order to solve the above problem, the present invention further provides an electronic device, comprising:

[0059] a memory storing at least one instruction; and

[0060] A processor executes instructions stored in the memory to implement the above-mentioned deep learning-based intelligent adaptive carbon source addition method.

[0061] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned deep learning-based intelligent adaptive carbon source addition method.

[0062] The present invention obtains water quality time series sample data, hyperspectral sample data and environmental video sample data, and extracts multimodal fusion features from the water quality time series sample data, hyperspectral sample data and environmental video sample data to construct a comprehensive water quality feature state space, which can provide rich data support for reinforcement learning model construction. Furthermore, an action space is constructed based on a preset carbon source addition range to ensure the feasibility of the carbon source addition operation. The water quality feature state space and the carbon source addition action space are then used to construct a model to obtain an initial carbon source addition control model. The initial carbon source addition control model is optimized through reward function and digital twin model reinforcement learning to obtain a target carbon source addition control model, so that the target carbon source addition control model can adapt to complex working conditions and improve the accuracy of carbon source addition control. Finally, the optimized target carbon source addition control model is applied to the carbon source addition prediction of the current water quality time series data, the current hyperspectral data and the current environmental video data, which can achieve accurate carbon source addition, improve sewage treatment efficiency and reduce sewage treatment operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A schematic diagram of a process flow of a deep learning-based intelligent adaptive carbon source dosing method provided in one embodiment of the present invention;

[0064] Figure 2 This is a functional module diagram of an intelligent adaptive carbon source dosing system based on deep learning provided by one embodiment of the present invention;

[0065] Figure 3 A schematic structural diagram of an electronic device for implementing the deep learning-based intelligent adaptive carbon source addition method provided in one embodiment of the present invention.

[0066] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0067] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0068] The embodiments of the present application provide a method for intelligently adapting carbon source dosing based on deep learning. The execution subject of the method includes, but is not limited to, at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiments of the present application. In other words, the method for intelligently adapting carbon source dosing based on deep learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0069] Reference Figure 1 FIG2 is a flow chart of a method for intelligently adapting carbon source addition based on deep learning according to an embodiment of the present invention. In this embodiment, the method for intelligently adapting carbon source addition based on deep learning includes:

[0070] S1. Obtain water quality time series sample data, hyperspectral sample data, and environmental video sample data.

[0071] It is understandable that water quality time series sample data refers to water quality parameter data collected continuously at different time points, such as inlet flow rate, COD (Chemical Oxygen Demand) concentration, ammonia nitrogen concentration, total nitrogen concentration, dissolved oxygen concentration, water temperature, etc. It is important to know that water quality time series sample data can reflect the changing trend of water quality over time. Hyperspectral sample data refers to the spectral data of water quality parameter data. Environmental video sample data refers to video data of the sewage treatment environment. Environmental video sample data can include microbial videos and equipment operation videos, among which microbial videos can be used to observe the types, quantity and activity changes of microorganisms, and equipment operation videos can be used to observe the operating status of the equipment.

[0072] It is understandable that the embodiment of the present application uses pre-installed water quality sensors to collect data such as water inlet flow, COD concentration, ammonia nitrogen concentration, total nitrogen concentration, dissolved oxygen concentration, and water temperature according to set time intervals, thereby forming water quality time series sample data. Furthermore, hyperspectral imaging technology can be used to perform imaging processing on the water quality time series sample data to obtain hyperspectral sample data. Finally, the sewage treatment environment can be photographed by cameras and other equipment to obtain environmental video sample data.

[0073] In the embodiment of the present application, by obtaining water quality time series sample data, the trend of water quality changes over time can be obtained, which is convenient for clarifying the fluctuation of water quality parameters such as water inlet flow rate. Secondly, by obtaining hyperspectral sample data, the detailed spectral information of water quality components can be clarified, thereby facilitating the identification and quantitative analysis of pollutants in water. Finally, by obtaining environmental video sample data, the types, quantity and activity changes of microorganisms in water, as well as the status of sewage treatment equipment, can be clarified.

[0074] S2. Extract features from the water quality time series sample data, hyperspectral sample data, and environmental video sample data to obtain multimodal fusion features.

[0075] It can be understood that multimodal fusion features refer to comprehensive features that include information from multiple different data modalities, where the data modalities can be time series data, image data, video data, etc.

[0076] It can be understood that the embodiment of the present application can obtain multimodal fusion features by extracting water quality features of water quality time series sample data, spectral features of hyperspectral sample data and environmental features of environmental video sample data, and then fusing the water quality features, spectral features and environmental features.

[0077] In detail, the feature extraction of the water quality time series sample data, the hyperspectral sample data and the environmental video sample data to obtain multimodal fusion features includes:

[0078] Extracting dependency features from the water quality time series sample data to obtain time series local dependency features;

[0079] Extracting spectral features from the hyperspectral sample data to obtain spatial spectral features;

[0080] Extracting dynamic features from the environmental video sample data to obtain motion features;

[0081] The temporal local dependency features, spatial spectrum features and motion features are fused to obtain multimodal fusion features.

[0082] It can be understood that the time series local dependency feature refers to the local dependency relationship between the characteristic value at a certain moment and the characteristic values ​​at several adjacent moments before and after the moment in the water quality time series sample data. For example, in the water quality time series sample data, there are COD concentration data at time point a, COD concentration data at time point b, COD concentration data at time point c, COD concentration data at time point d, COD concentration data at time point e, COD concentration data at time point f and COD concentration data at the current time point. The Pearson correlation coefficient between the COD concentration data at each time point and the COD concentration data at the current time point is calculated. It can be obtained that the Pearson correlation coefficient between the COD concentration data at time point d, the COD concentration data at time point e, the COD concentration data at time point f and the COD concentration data at the current time point is greater than 0.5, then it is determined that the COD concentration data at time point d, the COD concentration data at time point e, and the COD concentration data at time point f have a significant impact on the COD concentration data at the current time point. At this time, it can be determined that the correlation between the COD concentration data at time point d, the COD concentration data at time point e, the COD concentration data at time point f and the COD concentration data at the current time point can constitute a time series local dependence feature.

[0083] It can be understood that spatial-spectral features are features that combine spatial and spectral information. Spatial information refers to the coordinate position of each pixel in the hyperspectral sample data, the spatial relationship between each pixel and its adjacent pixels, the spatial structure of the local area, and the spatial distribution structure of different bands. Motion features are features used to describe the motion state of microorganisms in the video, where the motion state can be movement speed, direction, movement trajectory, etc.

[0084] It is understandable that by mining the interdependence between water quality parameters at different time nodes from the water quality time series sample data, the time series local dependence features can be obtained. Furthermore, by searching for reflectance peaks, valleys and other features of specific bands that can characterize the materiality from the hyperspectral sample data, and then integrating the reflectance peaks, valleys and other features, the spatial spectral features can be obtained. Secondly, by analyzing the position, speed, acceleration and other information of micro-objects in the environmental video sample data, the movement trajectory and behavior pattern of microorganisms can be captured, and the movement trajectory and behavior pattern of microorganisms can be represented in the form of features to obtain motion features. Finally, the above-mentioned time series local dependence features, spatial spectral features and motion features are feature spliced ​​according to a pre-set feature splicing strategy to obtain multimodal fusion features, wherein the feature splicing strategy can be the feature tail of the time series local dependence feature and the feature head of the spatial spectral feature, or the feature tail of the spatial spectral feature and the feature head of the motion feature.

[0085] In an embodiment of the present application, by extracting dependency features from the water quality time series sample data to obtain time series local dependency features, the intrinsic correlation and change patterns of water quality parameters in the time dimension can be clarified, and key time series information can be provided for water quality status assessment. Furthermore, spectral feature extraction is performed on the hyperspectral sample data to obtain spatial spectral features, which can accurately capture the spectral characteristics of water quality components and the spatial distribution of different bands of their hyperspectral spectrum, thereby enhancing a detailed understanding of the water quality status. At the same time, dynamic feature extraction is performed on the environmental video sample data to obtain motion features, which can clarify the activity status and dynamic changes of microorganisms in the water environment, adding a dynamic dimension to water quality assessment. Finally, the time series local dependency features, spatial spectral features, and motion features are fused to obtain multimodal fusion features, which achieves a comprehensive description of water quality from multiple perspectives of time series, spectrum, and dynamics, making the multimodal fusion features richer and more comprehensive. In addition, the multimodal fusion features can also accurately reflect the actual water quality, provide a data basis for the formulation of carbon source addition control strategies, and thus improve the intelligence level of the sewage treatment process.

[0086] S3. Constructing a water quality feature state space based on the multimodal fusion features.

[0087] It is understandable that the water quality characteristic state space is used to characterize various possible states of water quality. What needs to be known is that in the water quality characteristic state space, there are multiple dimensions and multiple coordinate points. Each dimension represents a specific water quality characteristic, and each coordinate point corresponds to a water quality state. For example, the water quality characteristic state space integrates the COD concentration trend characteristics, hyperspectral principal component characteristics and microbial activity characteristics, then the water quality characteristic state space has three dimensions and multiple coordinate points, and different coordinate points represent different combinations of COD concentration trend characteristics, hyperspectral principal component characteristics and microbial activity characteristics.

[0088] It is understandable that the embodiment of the present application can clarify the range of features based on multimodal fusion features, and randomly combine features within this range to obtain combined features. Furthermore, all the obtained combined features are embedded in a preset empty set to obtain the water quality feature state space.

[0089] In detail, the water quality feature state space is constructed based on the multimodal fusion features, including:

[0090] Based on the multimodal fusion features, feature range calculation is performed to obtain a multimodal feature range;

[0091] Based on the multimodal feature range, performing random feature combination to obtain multiple multimodal combination features;

[0092] The multiple multimodal combination features are subjected to feature aggregation to obtain a water quality feature state space.

[0093] It is understood that the multimodal feature range refers to the value range of each feature component in the multimodal fusion feature. For example, after calculating the feature range, it can be obtained that the change rate of the water quality parameter range is [-0.5, 0.5] mg / L·min, and the intensity range of the spectral feature is [0.2, 0.8]. Multimodal combination features refer to the feature set obtained by randomly combining features. It is important to know that each multimodal combination feature is a possible variant of the multimodal fusion feature. For example, the feature composed of a water quality parameter change rate of 0.3 mg / L·min and a spectral feature intensity of 0.6 can be a multimodal combination feature.

[0094] It can be understood that the embodiment of the present application can obtain the value range of each feature component, that is, the multimodal feature range, by querying the maximum and minimum values ​​of each feature component in the multimodal fusion feature. Secondly, within the multimodal feature range, the values ​​of different feature components are randomly selected and combined to generate a new multimodal combination feature. For example, a change rate value of 0.3 mg / L·min is randomly selected within the water quality parameter change rate range of [-0.5, 0.5] mg / L·min, and an intensity value of 0.6 is randomly selected within the spectral feature intensity range of [0.2, 0.8] to form a multimodal combination feature (water quality parameter change rate 0.3 mg / L·min, spectral feature intensity 0.6). Finally, all combined multimodal combination features are embedded in a pre-set empty set to obtain the water quality feature state space.

[0095] It can be understood that in the embodiments of the present application, the feature range is calculated based on the multimodal fusion features, and the value range of each feature is clarified, which lays the foundation for feature combination. Secondly, random combination of features within the multimodal feature range can effectively increase the diversity of multimodal combination features. Finally, by performing feature aggregation on multiple multimodal combination features, a comprehensive and rich water quality feature state space can be constructed, which provides sufficient and diverse data support for water quality analysis, thereby improving the model's ability to accurately characterize and predict water quality status.

[0096] S4. Based on the preset carbon source addition range, a carbon source addition action space is constructed.

[0097] It is understandable that the preset carbon source addition range refers to the range of carbon source addition amounts predetermined according to the sewage treatment process and sewage treatment experience. For example, according to the sewage treatment process and sewage treatment experience, the carbon source addition range can be determined to be 0-100 kg per hour. The carbon source addition action space refers to the set of all possible actions of the carbon source addition operation, wherein each carbon source addition operation corresponds to a specific combination of carbon source addition amount and carbon source addition frequency. For example, when the carbon source addition range is between 0-100 kg per hour, the carbon source addition action space can include adding once every 5 minutes, 5 kg each time, adding once every 5 minutes, 6 kg each time, adding once every 5 minutes, 7 kg each time, adding once every 30 minutes, 10 kg each time, adding once every 45 minutes, 10 kg each time, etc. carbon source addition operations.

[0098] It is understandable that the embodiment of the present application can divide the carbon source addition range into addition levels according to the pre-set step size and frequency, and then form a series of discrete addition actions according to the divided addition levels, so as to obtain the carbon source addition action space. For example, when the carbon source addition range is 0-100 kilograms per hour, the carbon source addition range can be divided into multiple addition levels with a step size of 10 kilograms and a frequency of addition every 10 minutes. Furthermore, a series of discrete addition actions can be formed according to the addition levels. Finally, this series of addition actions are combined into the carbon source addition action space.

[0099] In detail, the carbon source addition action space is constructed based on the preset carbon source addition range, including:

[0100] Performing feature extraction on the carbon source addition range to obtain a carbon source addition amount range and a carbon source addition frequency range;

[0101] Based on the carbon source dosage range and the carbon source dosage frequency range, a carbon source dosage action combination is performed to obtain a plurality of carbon source dosage combination actions;

[0102] The multiple carbon source addition combination actions are aggregated to obtain a carbon source addition action space.

[0103] It is understood that the carbon source addition amount range refers to a specific numerical interval of the carbon source addition amount, for example, 10 kg to 100 kg. The carbon source addition frequency range refers to a specific numerical interval of the carbon source addition frequency, for example, once every 5 to 60 minutes. The carbon source addition combination action refers to a carbon source addition action composed of different carbon source addition amounts and carbon source addition frequencies. For example, the carbon source addition combination action can be an addition amount of 50 kg / hour and a frequency of 15 minutes / time, and the carbon source addition combination action can also be an addition amount of 80 kg / hour and a frequency of 30 minutes / time.

[0104] It is understandable that by analyzing the carbon source addition range, clarifying the two types of data, carbon source addition amount and carbon source addition frequency, and then querying the maximum and minimum values ​​of the carbon source addition amount, the carbon source addition amount range can be obtained, and by querying the maximum and minimum values ​​of the carbon source addition frequency, the carbon source addition frequency range can be obtained. Further, within the carbon source addition amount range and the carbon source addition frequency range, different carbon source addition amounts and carbon source addition frequencies are combined to obtain multiple carbon source addition combination actions. For example, by combining By combining different carbon source addition amounts and carbon source addition frequencies, we can obtain multiple carbon source addition combination actions such as carbon source addition combination action 1 and carbon source addition combination action 2, among which, carbon source addition combination action 1 is a carbon source addition amount of 50 kg / hour and a carbon source addition frequency of 15 minutes / time, and carbon source addition combination action 2 is a carbon source addition amount of 80 kg / hour and a carbon source addition frequency of 30 minutes / time. Finally, all the obtained carbon source addition combination actions are embedded into the pre-set empty set to obtain the carbon source addition action space.

[0105] It is understandable that in the embodiments of the present application, by extracting features of the carbon source addition range and clarifying the carbon source addition amount range and the carbon source addition frequency range, a clear parameter range can be provided for the carbon source addition action combination. Secondly, the carbon source addition action combination is performed within the carbon source addition amount range and the carbon source addition frequency range to generate multiple different carbon source addition combination actions, thereby increasing the diversity and flexibility of the operation. Finally, by aggregating multiple carbon source addition combination actions, a comprehensive carbon source addition action space is constructed, which can provide rich training samples and operation options for the reinforcement learning model, thereby helping the reinforcement learning model to more accurately select the optimal carbon source addition strategy, thereby improving the accuracy of carbon source addition in the sewage treatment process.

[0106] S5. Based on the water quality characteristic state space and the carbon source addition action space, a model is constructed to obtain an initial carbon source addition control model.

[0107] It is understandable that the preset reward function refers to a function that is pre-designed according to the goals and requirements of sewage treatment to measure the effect of carbon source addition control. It is important to know that the input of the reward function can be the water quality characteristic state in the water quality characteristic state space and the carbon source addition action in the carbon source addition action space. The output of the reward function is the reward value, where the reward value is used to guide the learning of the initial carbon source addition control model. The initial carbon source addition control model refers to a model that can select a carbon source addition action according to a pre-set strategy when a water quality characteristic state is given. It is important to know that the performance of the initial carbon source addition control model is not optimal and requires further reinforcement learning and optimization.

[0108] It can be understood that the embodiment of the present application designs an initialization strategy network, and the number of neurons in the network input layer of the initialization strategy network must match the dimension of the water quality characteristic state space, and the number of neurons in the output layer must correspond to the number of actions in the carbon source addition action space, so as to obtain an initial carbon source addition control model.

[0109] S6. Based on the preset reward function, the preset digital twin model and the multimodal fusion feature, the initial carbon source addition control model is reinforced learned to obtain the target carbon source addition control model.

[0110] It is understood that the preset reward function refers to a function designed in advance based on the goals and requirements of sewage treatment to measure the effectiveness of carbon source addition control. It is important to note that the input of the reward function can be the water quality characteristic state in the water quality characteristic state space and the carbon source addition action in the carbon source addition action space. The output of the reward function is the reward value, where the reward value is used to guide the learning of the initial carbon source addition control model. The preset digital twin model refers to a pre-built virtual model corresponding to the actual sewage treatment plant. Specifically, the digital twin model can simulate various physical, chemical, and biological reactions in the sewage treatment process.

[0111] It can be understood that the embodiment of the present application uses the initial carbon source addition control model to preliminarily predict the carbon source addition action corresponding to the multimodal fusion feature, and puts the multimodal fusion feature and its corresponding carbon source addition action into the digital twin model for simulation operation, and uses the reward function to evaluate the rationality of the carbon source addition action during this simulation operation to obtain evaluation data. Furthermore, based on the evaluation data, the parameters of the initial carbon source addition control model are adjusted to obtain the target carbon source addition control model.

[0112] In detail, based on the preset reward function, the preset digital twin model and the multimodal fusion feature, the initial carbon source dosing control model is reinforced learned to obtain the target carbon source dosing control model, including:

[0113] Based on the multimodal fusion features, predicting the carbon source addition action of the initial carbon source addition control model to obtain a predicted carbon source addition action;

[0114] Based on the reward function and the digital twin model, an action evaluation is performed on the multimodal fusion feature and the predicted carbon source addition action to obtain carbon source addition evaluation data;

[0115] Based on the carbon source addition evaluation data, the initial carbon source addition control model is iteratively optimized to obtain a target carbon source addition control model.

[0116] It is understood that the predicted carbon source addition action refers to the carbon source addition amount and frequency selected by the initial carbon source addition control model based on multimodal fusion features and its own action selection strategy. The carbon source addition evaluation data refers to the rationality evaluation data of the predicted carbon source addition action.

[0117] It can be understood that the embodiment of the present application inputs the multimodal fusion feature into the initial carbon source addition control model, and the initial carbon source addition control model selects the carbon source addition action corresponding to the multimodal fusion feature from the carbon source addition action space as the predicted carbon source addition action based on a pre-set action selection strategy. Secondly, the multimodal fusion feature and the predicted carbon source addition action are input into a pre-built digital twin model, so that the digital twin model performs simulation operation according to the multimodal fusion feature and the predicted carbon source addition action. Then, according to the reward function, the data obtained from the operation of the digital twin model is evaluated to obtain carbon source addition evaluation data. Finally, according to the carbon source addition evaluation data, the action selection strategy pre-set in the initial carbon source addition control model is adjusted to achieve iterative optimization of the initial carbon source addition control model, thereby obtaining the target carbon source addition control model.

[0118] In detail, based on the reward function and the digital twin model, the multimodal fusion feature and the predicted carbon source addition action are evaluated to obtain carbon source addition evaluation data, including:

[0119] Based on the multimodal fusion features and the predicted carbon source addition action, the digital twin model is simulated to obtain sewage treatment data;

[0120] Based on the sewage treatment data, sewage treatment performance calculation is performed to obtain target performance data;

[0121] Based on the reward function, data evaluation is performed on the target performance data to obtain carbon source addition evaluation data.

[0122] It is understood that sewage treatment data refers to data obtained after the digital twin model simulation runs. Sewage treatment data includes effluent water quality indicators, carbon source consumption, operating costs, etc. Among them, effluent water quality indicators can be total nitrogen concentration, COD concentration, etc. Target performance data refers to key performance indicators of sewage treatment, such as total nitrogen removal rate and effluent total nitrogen compliance rate.

[0123] It can be understood that the embodiment of the present application runs the multimodal fusion features and predicts the carbon source addition action through the digital twin model, which can simulate the sewage treatment situation of the sewage treatment plant under the multimodal fusion features and predicted carbon source addition action, and at the same time records the effluent water quality indicators, carbon source consumption, operating costs and other processing data generated by the digital twin model when running the multimodal fusion features and predicting the carbon source addition action, and obtains sewage treatment data. Furthermore, according to the pre-constructed performance calculation formula and the above-mentioned sewage treatment data, the sewage treatment performance data of the digital twin model under the multimodal fusion features and predicted carbon source addition action, that is, the target performance data, can be calculated. Finally, the target performance data is substituted into the pre-set reward data to obtain the sewage treatment capacity evaluation data of the digital twin model under the multimodal fusion features and predicted carbon source addition action, that is, the carbon source addition evaluation data.

[0124] Specifically, the sewage treatment performance calculation is performed based on the sewage treatment data to obtain target performance data, including:

[0125] Extracting features from the sewage treatment data to obtain inlet total nitrogen concentration data, effluent total nitrogen concentration data, simulated carbon source dosage, evaluation time, and evaluation times;

[0126] Based on the inlet total nitrogen concentration data, the outlet total nitrogen concentration data and the number of evaluations, the water quality total nitrogen index is estimated to obtain total nitrogen treatment performance data;

[0127] Based on the simulated carbon source dosage and evaluation time, a stability evaluation is performed to obtain stability performance data;

[0128] The total nitrogen treatment performance data and the stability performance data are integrated to obtain target performance data.

[0129] It is understandable that the inlet total nitrogen concentration data refers to the total nitrogen concentration measurement value at the water inlet in the digital twin model. It is important to know that the inlet total nitrogen concentration data represents the total nitrogen content of the sewage to be treated. The effluent total nitrogen concentration data is the total nitrogen concentration measurement value at the water outlet in the digital twin model. It is important to know that the effluent total nitrogen concentration data represents the total nitrogen content in the water quality after carbon source addition treatment. The simulated carbon source dosage refers to the carbon source dosage used in the simulation operation of the digital twin model. The evaluation time refers to the time period covered by each performance evaluation. For example, in a certain performance evaluation, the evaluation time is 24 hours. The number of evaluations refers to the frequency of evaluating the operation results of the digital twin model during the performance evaluation process. For example, in a certain performance evaluation, the number of evaluations is 100 times.

[0130] It is understood that total nitrogen treatment performance data refers to data that characterizes the digital twin model's effect on total nitrogen treatment, such as total nitrogen removal rate, effluent total nitrogen compliance rate, etc. Stability performance data refers to data that describes the operational stability of the digital twin model.

[0131] It is understandable that the embodiment of the present application can determine whether the effluent total nitrogen concentration data meets the standard during each performance evaluation based on a pre-set total nitrogen compliance threshold, and then calculate the effluent total nitrogen concentration compliance rate of the digital twin model under the multimodal fusion feature and the predicted carbon source addition action according to the number of evaluations of the digital twin model. Then, the pre-constructed average effluent total nitrogen concentration calculation formula can be used to calculate the average effluent total nitrogen concentration corresponding to the above-mentioned sewage treatment data, and the pre-constructed total nitrogen removal rate calculation formula can be used to calculate the total nitrogen removal rate corresponding to the sewage treatment data. Then, the above-mentioned effluent total nitrogen concentration compliance rate, average effluent total nitrogen concentration and total nitrogen removal rate are embedded and merged to obtain total nitrogen treatment performance data. For example, when the effluent total nitrogen concentration compliance rate is 85%, the average effluent total nitrogen concentration is 12 mg / L, and the total nitrogen removal rate is 60%, the effluent total nitrogen concentration compliance rate, the average effluent total nitrogen concentration and the total nitrogen removal rate are filled into the pre-constructed empty set to obtain the total nitrogen treatment performance data {85%, 12, 60%}.

[0132] Furthermore, the stability performance data corresponding to the sewage treatment data can be calculated using a pre-built stability evaluation formula. Finally, the above-mentioned total nitrogen treatment performance data and the stability performance data are embedded and merged to obtain the target performance data.

[0133] In detail, the water quality total nitrogen index is estimated based on the inlet total nitrogen concentration data, the outlet total nitrogen concentration data and the number of evaluations to obtain the total nitrogen treatment performance data, including:

[0134] Calculate the effluent total nitrogen concentration compliance rate based on a preset total nitrogen concentration threshold, the effluent total nitrogen concentration data, and the number of evaluations;

[0135] The average effluent total nitrogen concentration of the sewage treatment data is calculated using the effluent total nitrogen concentration data, the number of evaluations, and the pre-constructed average effluent total nitrogen concentration calculation formula, wherein the average effluent total nitrogen concentration calculation formula is as follows:

[0136] ;

[0137] in, represents the average effluent total nitrogen concentration, represents the number of evaluations, Indicates the The effluent total nitrogen concentration data for this assessment;

[0138] The total nitrogen removal rate of the sewage treatment data is calculated using the influent total nitrogen concentration data, the effluent total nitrogen concentration data, and a pre-established total nitrogen removal rate calculation formula, wherein the total nitrogen removal rate calculation formula is as follows:

[0139] ;

[0140] in, represents the total nitrogen removal rate, Indicates the total nitrogen concentration data of the influent, Indicates the total nitrogen concentration data of the effluent;

[0141] The data of the effluent total nitrogen concentration compliance rate, the average effluent total nitrogen concentration and the total nitrogen removal rate are integrated to obtain the total nitrogen treatment performance data.

[0142] In addition, the stability evaluation is performed based on the simulated carbon source dosage and evaluation time to obtain stability performance data, including:

[0143] The stability performance data of the sewage treatment data is calculated using the simulated carbon source dosage, evaluation time, and a pre-built stability evaluation formula, wherein the stability evaluation formula is as follows:

[0144] ;

[0145] Wherein, S represents the stability performance data, represents the evaluation time, represents the time point of stability evaluation, Indicates time The simulated carbon source dosage.

[0146] It is understandable that by predicting the initial carbon source addition control model based on multimodal fusion features, accurate carbon source addition actions can be obtained, and the predicted actions can be evaluated using reward functions and digital twin models to generate evaluation data. The evaluation data is then used to optimize the initial carbon source addition control model, which can improve the accuracy, economy and stability of carbon source addition, thereby ensuring the effect of sewage treatment.

[0147] S7. Obtain current water quality time series data, current hyperspectral data, and current environmental video data.

[0148] It is understood that current water quality time series data refers to the latest water quality parameter data collected at the current time point. It is important to note that current water quality time series data also includes data such as influent flow rate, COD concentration, and ammonia nitrogen concentration. Current hyperspectral data refers to the latest hyperspectral water quality data obtained at the current time point. Current environmental video data refers to video data of the sewage treatment environment captured at the current time point. Current environmental video data can include videos of microbial activity and equipment operation captured at the current time point.

[0149] It is understandable that the embodiments of the present application can use the above-mentioned water quality sensor to collect data such as the water inlet flow, COD concentration, ammonia nitrogen concentration, total nitrogen concentration, dissolved oxygen concentration, water temperature, etc. at the current time point, thereby obtaining the current water quality time series data. Secondly, the current water quality time series data is imaged using hyperspectral imaging technology to obtain the current hyperspectral data. Finally, at the current time point, the sewage treatment environment is photographed by cameras and other equipment to obtain the current environment video data.

[0150] S8. Based on the target carbon source addition control model, a carbon source addition prediction is performed on the current water quality time series data, the current hyperspectral data, and the current environmental video data to obtain target carbon source addition data.

[0151] It is understandable that the target carbon source addition data refers to the carbon source addition amount and carbon source addition frequency data that are adapted to the current water quality time series data, the current hyperspectral data, and the current environmental video data.

[0152] It can be understood that the embodiment of the present application obtains current water quality characteristics, current spectral characteristics and current environmental characteristics by performing feature extraction on current water quality time series data, current hyperspectral data and current environmental video data, and then performs feature fusion on the current water quality characteristics, current spectral characteristics and current environmental characteristics to obtain current fused characteristics. Furthermore, the current fused characteristics are input into the target carbon source addition control model. The target carbon source addition control model can select the optimal carbon source addition amount and carbon source addition frequency from the carbon source addition action space according to the optimized action selection strategy, and integrate the optimal carbon source addition amount and carbon source addition frequency into the target carbon source addition data output.

[0153] The present invention obtains water quality time series sample data, hyperspectral sample data and environmental video sample data, and extracts multimodal fusion features from the water quality time series sample data, hyperspectral sample data and environmental video sample data to construct a comprehensive water quality feature state space, which can provide rich data support for reinforcement learning model construction. Furthermore, an action space is constructed based on a preset carbon source addition range to ensure the feasibility of the carbon source addition operation. The water quality feature state space and the carbon source addition action space are then used to construct a model to obtain an initial carbon source addition control model. The initial carbon source addition control model is optimized through reward function and digital twin model reinforcement learning to obtain a target carbon source addition control model, so that the target carbon source addition control model can adapt to complex working conditions and improve the accuracy of carbon source addition control. Finally, the optimized target carbon source addition control model is applied to the carbon source addition prediction of the current water quality time series data, the current hyperspectral data and the current environmental video data, which can achieve accurate carbon source addition, improve sewage treatment efficiency and reduce sewage treatment operating costs.

[0154] like Figure 2 , which is a functional module diagram of an intelligent adaptive carbon source dosing system based on deep learning provided by one embodiment of the present invention.

[0155] The deep learning-based intelligent adaptive carbon source dosing system 100 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the deep learning-based intelligent adaptive carbon source dosing system 100 can include a data feature extraction module 101, a target model training module 102, a current data acquisition module 103, and a carbon source dosing control module 104. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the electronic device's memory.

[0156] The data feature extraction module 101 is used to obtain water quality time series sample data, hyperspectral sample data and environmental video sample data, perform feature extraction on the water quality time series sample data, hyperspectral sample data and environmental video sample data, and obtain multimodal fusion features;

[0157] The target model training module 102 is used to construct a water quality feature state space based on the multimodal fusion feature, construct a carbon source addition action space based on a preset carbon source addition range, perform model construction based on the water quality feature state space and the carbon source addition action space, obtain an initial carbon source addition control model, and perform reinforcement learning on the initial carbon source addition control model based on a preset reward function, a preset digital twin model and the multimodal fusion feature to obtain a target carbon source addition control model;

[0158] The current data acquisition module 103 is used to acquire current water quality time series data, current hyperspectral data and current environmental video data;

[0159] The carbon source addition control module 104 is used to perform carbon source addition prediction on the current water quality time series data, the current hyperspectral data and the current environmental video data based on the target carbon source addition control model to obtain target carbon source addition data.

[0160] In detail, each module in the intelligent adaptive carbon source dosing system 100 based on deep learning in the embodiment of the present invention adopts the same method as above when in use. Figure 1 The technical means are the same as the intelligent adaptive carbon source addition method based on deep learning described in , and can produce the same technical effects, so I will not go into details here.

[0161] like Figure 3 , which is a structural diagram of an electronic device for implementing an intelligent adaptive carbon source dosing method based on deep learning provided by one embodiment of the present invention.

[0162] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as an intelligent adaptive carbon source dosing method program based on deep learning.

[0163] The memory 11 includes at least one type of readable storage medium, including flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 1. Furthermore, the memory 11 includes both the internal storage unit of the electronic device 1 and an external storage device. The memory 11 can be used not only to store application software installed in the electronic device 1 and various types of data, such as the code of the deep learning-based intelligent adaptive carbon source dosing method program, but also to temporarily store data that has been output or is about to be output.

[0164] In some embodiments, the processor 10 may be composed of an integrated circuit, such as a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting the various components of the entire electronic device using various interfaces and lines, and executing programs or modules stored in the memory 11 (such as a program for an intelligent adaptive carbon source dosing method based on deep learning, etc.), as well as calling data stored in the memory 11, to perform various functions of the electronic device 1 and process data.

[0165] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0166] Figure 3 Only the electronic device with components is shown, and it can be understood by those skilled in the art that Figure 3 The structure shown does not constitute a limitation on the electronic device 1 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0167] For example, although not shown, the electronic device 1 may further include a power source (e.g., a battery) to power various components. Preferably, the power source may be logically connected to the at least one processor 10 via a power management system, thereby enabling functions such as charge management, discharge management, and power consumption management through the power management system. The power source may further include any components such as one or more DC or AC power sources, a recharging system, a power failure detection circuit, a power converter or inverter, and a power status indicator. The electronic device 1 may further include various sensors, Bluetooth modules, Wi-Fi modules, etc., which are not further described here.

[0168] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0169] Optionally, the electronic device 1 may further include a user interface, which may be a display or an input unit (such as a keyboard). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display may also be appropriately referred to as a display screen or a display unit, and is used to display information processed by the electronic device 1 and to display a visual user interface.

[0170] The program of the intelligent adaptive carbon source dosing method based on deep learning stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0171] Obtain water quality time series sample data, hyperspectral sample data and environmental video sample data;

[0172] Extracting features from the water quality time series sample data, hyperspectral sample data, and environmental video sample data to obtain multimodal fusion features;

[0173] Based on the multimodal fusion features, constructing a water quality feature state space;

[0174] Based on the preset carbon source addition range, the carbon source addition action space is constructed;

[0175] Based on the water quality characteristic state space and the carbon source addition action space, a model is constructed to obtain an initial carbon source addition control model;

[0176] Based on a preset reward function, a preset digital twin model and the multimodal fusion feature, reinforcement learning is performed on the initial carbon source dosing control model to obtain a target carbon source dosing control model;

[0177] Obtain current water quality time series data, current hyperspectral data and current environmental video data;

[0178] Based on the target carbon source addition control model, a carbon source addition prediction is performed on the current water quality time series data, the current hyperspectral data and the current environmental video data to obtain target carbon source addition data.

[0179] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0180] Furthermore, if the modules / units integrated into the electronic device 1 are implemented as software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. The computer-readable storage medium may be volatile or non-volatile. For example, the computer-readable medium may include any entity or system capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0181] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed by a processor of an electronic device, the computer program can implement:

[0182] Obtain water quality time series sample data, hyperspectral sample data and environmental video sample data;

[0183] Extracting features from the water quality time series sample data, hyperspectral sample data, and environmental video sample data to obtain multimodal fusion features;

[0184] Based on the multimodal fusion features, constructing a water quality feature state space;

[0185] Based on the preset carbon source addition range, the carbon source addition action space is constructed;

[0186] Based on the water quality characteristic state space and the carbon source addition action space, a model is constructed to obtain an initial carbon source addition control model;

[0187] Based on a preset reward function, a preset digital twin model and the multimodal fusion feature, reinforcement learning is performed on the initial carbon source dosing control model to obtain a target carbon source dosing control model;

[0188] Obtain current water quality time series data, current hyperspectral data and current environmental video data;

[0189] Based on the target carbon source addition control model, a carbon source addition prediction is performed on the current water quality time series data, the current hyperspectral data and the current environmental video data to obtain target carbon source addition data.

[0190] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only exemplary, and actual implementations may have other division methods.

[0191] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0192] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0193] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart adaptive carbon source dosing method based on deep learning, characterized in that: The method comprises: Obtain water quality time series sample data, hyperspectral sample data and environmental video sample data; Extracting features from the water quality time series sample data, hyperspectral sample data, and environmental video sample data to obtain multimodal fusion features; Based on the multimodal fusion features, constructing a water quality feature state space; Based on the preset carbon source addition range, the carbon source addition action space is constructed; Based on the water quality characteristic state space and the carbon source addition action space, a model is constructed to obtain an initial carbon source addition control model; Based on a preset reward function, a preset digital twin model and the multimodal fusion feature, reinforcement learning is performed on the initial carbon source dosing control model to obtain a target carbon source dosing control model; Obtain current water quality time series data, current hyperspectral data and current environmental video data; Based on the target carbon source addition control model, a carbon source addition prediction is performed on the current water quality time series data, the current hyperspectral data and the current environmental video data to obtain target carbon source addition data.

2. The deep learning-based intelligent adaptive carbon source dosing method according to claim 1, characterized in that: The step of constructing a water quality feature state space based on the multimodal fusion features includes: Based on the multimodal fusion features, feature range calculation is performed to obtain a multimodal feature range; Based on the multimodal feature range, performing random feature combination to obtain multiple multimodal combination features; The multiple multimodal combination features are subjected to feature aggregation to obtain a water quality feature state space.

3. The intelligent adaptive carbon source dosing method based on deep learning according to claim 1, characterized in that: The carbon source addition action space is constructed based on the preset carbon source addition range, including: Performing feature extraction on the carbon source addition range to obtain a carbon source addition amount range and a carbon source addition frequency range; Based on the carbon source dosage range and the carbon source dosage frequency range, a carbon source dosage action combination is performed to obtain a plurality of carbon source dosage combination actions; The multiple carbon source addition combination actions are aggregated to obtain a carbon source addition action space.

4. The intelligent adaptive carbon source dosing method based on deep learning according to claim 1, characterized in that: The feature extraction of the water quality time series sample data, the hyperspectral sample data and the environmental video sample data to obtain multimodal fusion features includes: Extracting dependency features from the water quality time series sample data to obtain time series local dependency features; Extracting spectral features from the hyperspectral sample data to obtain spatial spectral features; Extracting dynamic features from the environmental video sample data to obtain motion features; The temporal local dependency features, spatial spectrum features and motion features are fused to obtain multimodal fusion features.

5. The deep learning-based intelligent adaptive carbon source dosing method according to claim 1, characterized in that: The method of performing reinforcement learning on the initial carbon source dosing control model based on the preset reward function, the preset digital twin model and the multimodal fusion feature to obtain a target carbon source dosing control model includes: Based on the multimodal fusion features, predicting the carbon source addition action of the initial carbon source addition control model to obtain a predicted carbon source addition action; Based on the reward function and the digital twin model, an action evaluation is performed on the multimodal fusion feature and the predicted carbon source addition action to obtain carbon source addition evaluation data; Based on the carbon source addition evaluation data, the initial carbon source addition control model is iteratively optimized to obtain a target carbon source addition control model.

6. The intelligent adaptive carbon source dosing method based on deep learning according to claim 5, characterized in that: Based on the reward function and the digital twin model, the multimodal fusion feature and the predicted carbon source addition action are evaluated to obtain carbon source addition evaluation data, including: Based on the multimodal fusion features and the predicted carbon source addition action, the digital twin model is simulated to obtain sewage treatment data; Based on the sewage treatment data, sewage treatment performance calculation is performed to obtain target performance data; Based on the reward function, data evaluation is performed on the target performance data to obtain carbon source addition evaluation data.

7. The deep learning-based intelligent adaptive carbon source dosing method according to claim 6, characterized in that: The sewage treatment performance calculation is performed based on the sewage treatment data to obtain target performance data, including: Extracting features from the sewage treatment data to obtain inlet total nitrogen concentration data, effluent total nitrogen concentration data, simulated carbon source dosage, evaluation time, and evaluation times; Based on the inlet total nitrogen concentration data, the outlet total nitrogen concentration data and the number of evaluations, the water quality total nitrogen index is estimated to obtain total nitrogen treatment performance data; Based on the simulated carbon source dosage and evaluation time, a stability evaluation is performed to obtain stability performance data; The total nitrogen treatment performance data and the stability performance data are integrated to obtain target performance data.

8. The deep learning-based intelligent adaptive carbon source dosing method according to claim 7, characterized in that: The water quality total nitrogen index is estimated based on the inlet total nitrogen concentration data, the outlet total nitrogen concentration data and the number of evaluations to obtain total nitrogen treatment performance data, including: Calculate the effluent total nitrogen concentration compliance rate based on a preset total nitrogen concentration threshold, the effluent total nitrogen concentration data, and the number of evaluations; The average effluent total nitrogen concentration of the sewage treatment data is calculated using the effluent total nitrogen concentration data, the number of evaluations, and the pre-constructed average effluent total nitrogen concentration calculation formula, wherein the average effluent total nitrogen concentration calculation formula is as follows: ； in, represents the average effluent total nitrogen concentration, represents the number of evaluations, Indicates the The effluent total nitrogen concentration data for this assessment; The total nitrogen removal rate of the sewage treatment data is calculated using the influent total nitrogen concentration data, the effluent total nitrogen concentration data, and a pre-established total nitrogen removal rate calculation formula, wherein the total nitrogen removal rate calculation formula is as follows: ； in, represents the total nitrogen removal rate, Indicates the total nitrogen concentration data of the influent, Indicates the total nitrogen concentration data of the effluent; The data of the effluent total nitrogen concentration compliance rate, the average effluent total nitrogen concentration and the total nitrogen removal rate are integrated to obtain the total nitrogen treatment performance data.

9. The intelligent adaptive carbon source dosing method based on deep learning according to claim 7, characterized in that: The stability evaluation is performed based on the simulated carbon source dosage and evaluation time to obtain stability performance data, including: The stability performance data of the sewage treatment data is calculated using the simulated carbon source dosage, evaluation time, and a pre-built stability evaluation formula, wherein the stability evaluation formula is as follows: ； Wherein, S represents the stability performance data, represents the evaluation time, represents the time point of stability evaluation, Indicates time The simulated carbon source dosage.

10. A deep learning-based intelligent adaptive carbon source dosing system, characterized in that: The system comprises: A data feature extraction module is used to obtain water quality time series sample data, hyperspectral sample data and environmental video sample data, perform feature extraction on the water quality time series sample data, hyperspectral sample data and environmental video sample data, and obtain multimodal fusion features; A target model training module is used to construct a water quality feature state space based on the multimodal fusion feature, construct a carbon source addition action space based on a preset carbon source addition range, perform model construction based on the water quality feature state space and the carbon source addition action space, obtain an initial carbon source addition control model, and perform reinforcement learning on the initial carbon source addition control model based on a preset reward function, a preset digital twin model and the multimodal fusion feature to obtain a target carbon source addition control model; Current data acquisition module, used to obtain current water quality time series data, current hyperspectral data and current environmental video data; The carbon source addition control module is used to perform carbon source addition prediction on the current water quality time series data, the current hyperspectral data and the current environmental video data based on the target carbon source addition control model to obtain target carbon source addition data.

Images