Water conservancy system control method and device based on voice interaction

Abstract

The invention discloses a water conservancy system control method and device based on voice interaction, and belongs to the technical field of intelligent control. The method comprises the steps of performing wake-up word detection based on a wake-up threshold dynamic generation mechanism, and performing identity recognition and permission verification on environment audio data by using a voiceprint matching model when a preset wake-up word is detected, so as to generate an identity authentication result; if the identity authentication result meets the requirement, performing voice text recognition on the environment audio data by using a voice text recognition model, performing intention recognition on the obtained voice recognition text to obtain an intention recognition result, and generating a system control instruction; and performing cooperative control decision generation and decision optimization by using a cooperative control model constructed based on a QMIX-ITOC algorithm according to the system control instruction, and controlling execution equipment of the water conservancy system according to the obtained cooperative control decision. The problems that in the prior art, operation is complex, the response speed is low, authority management is extensive, and the intelligent degree is low are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a method and device for controlling a water conservancy system based on voice interaction. Background Technology

[0002] Water conservancy systems are crucial national infrastructure, and their safe and efficient operation is vital to the national economy and people's livelihoods. Traditional water conservancy system control primarily relies on manual operation by personnel at the control center through computer interfaces and data acquisition and monitoring systems. This model has the following significant drawbacks: 1) Complex operation: The water conservancy system has a large number of devices (including cameras, pumping stations, gates, etc.) and complex logical relationships. Operators need long-term training to master them and are prone to misoperation in emergency situations.

[0003] 2) Slow response speed: From the discovery of the problem to the analysis and judgment, and then to the issuance of operation instructions, the process is cumbersome. Especially in emergency scenarios such as flood control and drought relief, every second counts, and the traditional mode is difficult to meet the needs of rapid response.

[0004] 3) Inadequate access control: Access control is usually based on login accounts, which cannot achieve continuous and seamless verification of the operator's identity, and there is a risk of account theft and abuse of permissions.

[0005] 4) Low level of intelligence: For joint scheduling involving multiple cameras, pumping stations, gates and reservoirs, the system often relies on empirical formulas or preset schemes, making it difficult to make dynamic and globally optimal collaborative decisions based on real-time conditions, resulting in resource waste or poor scheduling effect. Summary of the Invention

[0006] This invention provides a water conservancy system control method and device based on voice interaction, which solves the problems of complex operation, slow response speed, crude permission management and low level of intelligence in the existing technology.

[0007] In a first aspect, embodiments of the present invention provide a water conservancy system control method based on voice interaction, the method comprising: The system monitors ambient audio data and performs wake-up word detection based on a dynamic wake-up threshold generation mechanism. When a preset wake-up word is detected in the ambient audio data, the system uses a voiceprint matching model to perform identity recognition and permission verification on the ambient audio data and generates an identity authentication result. If the identity authentication result meets the requirements, the speech-text recognition model is used to perform speech-text recognition on the environmental audio data, the obtained speech-text is then used to perform intent recognition, the intent recognition result is obtained, and system control commands are generated. Based on the system control commands, a collaborative control model constructed using the QMIX-ITOC algorithm is used to generate and optimize collaborative control decisions, and the execution equipment of the water conservancy system is controlled according to the obtained collaborative control decisions.

[0008] The technical solution provided in this application has at least the following beneficial effects: By replacing complex graphical interfaces with natural voice interaction, operators no longer need to memorize cumbersome device numbers, operating logic, or menu paths. They can simply issue commands in everyday language, and the system will automatically interpret their intentions and generate control strategies. This not only greatly simplifies the operation process, allowing non-professionals or new employees to quickly get started, but also avoids misoperations caused by tension and panic in emergency situations, significantly improving user-friendliness and operational safety. It achieves an end-to-end automated process from "problem detection" to "control execution," monitoring voice commands in real time and immediately initiating a collaborative control model for decision generation and optimization after recognizing voiceprints and intentions. The system completes tasks within seconds or even milliseconds, improving real-time response and interaction efficiency. It introduces voiceprint detection and authentication mechanisms, strongly binding permissions to the user's voiceprint. Before each voice command is issued, seamless, real-time identity verification and permission checks are performed. This completely changes the traditional, crude permission management based on account passwords, effectively eliminating security risks such as account sharing, theft, and unauthorized operations. The collaborative control model possesses intelligent collaborative decision-making capabilities, no longer relying on isolated empirical formulas or preset schemes. Instead, it comprehensively considers the real-time status of all relevant devices and their complex coupling relationships to make global collaborative decisions, reducing resource waste and improving scheduling efficiency.

[0009] In one optional implementation, ambient audio data is monitored, and wake-up word detection is performed based on a dynamic wake-up threshold generation mechanism. When a preset wake-up word is detected in the ambient audio data, a voiceprint matching model is used to perform identity recognition and permission verification on the ambient audio data, generating an identity authentication result, including: Using an audio sensor, ambient audio data is monitored and preprocessed to obtain a continuous, time-ordered sequence of audio feature vectors. The audio feature vector sequence is input into the voiceprint detection model. Based on the user's voiceprint feature library, voiceprint detection is performed to obtain the existence probability of the preset wake-up word and the voiceprint recognition probability vector. Based on the wake-up threshold dynamic generation mechanism, a dynamic wake-up threshold is generated. If the probability of the existence of the preset wake-up word is greater than the dynamic wake-up threshold, the preset wake-up word is detected, the identity recognition and authorization verification process is initiated, and the process proceeds to the next step. Perform identity recognition and authorization verification, and extract the largest voiceprint recognition probability and its corresponding target user ID from the voiceprint recognition probability vector; If the maximum voiceprint recognition probability is greater than the voiceprint confidence threshold, then the target user ID's target permission level is extracted, and an identity authentication result is generated.

[0010] In one alternative implementation, the voiceprint matching model is built based on an improved CRNN algorithm, and the voiceprint matching model includes convolutional layers, recurrent layers, and a dual-task output head, which includes a parallel wake-up detection head and a voiceprint verification head.

[0011] In one optional implementation, the audio feature vector sequence is input into the voiceprint detection model. Based on the user's voiceprint feature database, voiceprint detection is performed to obtain the presence probability of a preset wake-up word and the voiceprint recognition probability vector, including: The audio feature vector sequence is input into the speakerprint detection model; The convolutional layer of the voiceprint detection model is used to extract local time-frequency features of the audio feature vector sequence; Temporal features of local time-frequency features are extracted using the recurrent layer of the voiceprint detection model. Based on the temporal features and the audio features corresponding to the preset wake-up words, the wake-up detection head of the dual-task output head of the voiceprint detection model is used to perform wake-up detection and obtain the probability of the existence of the preset wake-up words. Based on the temporal features and the voiceprint features of all users in the user voiceprint feature library, the voiceprint verification head of the dual-task output head of the voiceprint detection model is used to perform voiceprint detection, and a voiceprint recognition probability vector including the voiceprint recognition probability of all users is obtained.

[0012] In one optional implementation, if the authentication result meets the requirements, a speech-to-text recognition model is used to perform speech-to-text recognition on the environmental audio data. Intent recognition is then performed on the obtained speech-to-text to obtain the intent recognition result, and system control commands are generated, including: If the target permission level corresponding to the permission label of the identity authentication result is greater than the minimum permission level threshold, then continue to collect the audio feature vector sequence of the environmental audio data; The audio feature vector sequence is input into the speech-text recognition model to perform speech-text recognition and obtain the speech-recognized text. The speech recognition text and the pre-built knowledge graph of the water conservancy system are input into the pre-built joint extraction model to perform intent recognition and obtain the intent recognition result; The intent recognition results are converted into structured instructions to obtain system control instructions for several key target equipment in the water conservancy system.

[0013] In one alternative implementation, the speech-to-text recognition model is built on an ASR algorithm based on a Conformer-Transducer architecture; The joint extraction model is built on the BERT-BiLSTM-CRF-SVM-GNN algorithm, and includes a word embedding module based on the BERT algorithm, a semantic feature extraction module based on the BiLSTM algorithm, an entity extraction module based on the CRF algorithm, an entity relation extraction module based on the SVM algorithm, and an intent recognition module based on the GNN algorithm.

[0014] In one alternative implementation, the cooperative control model is constructed based on the QMIX-ITOC algorithm, and the cooperative control model includes a cooperative control decision generation module constructed based on the QMIX algorithm and a decision optimization module constructed based on the ITOC algorithm. The cooperative control decision generation module includes a hybrid network, several agents constructed based on the DRQN ​​algorithm, and an experience playback pool. The agents correspond to the key equipment of the water conservancy system.

[0015] In one optional implementation, based on system control commands, a collaborative control decision is generated and optimized using a collaborative control model constructed based on the QMIX-ITOC algorithm. Based on the obtained collaborative control decisions, the execution equipment of the water conservancy system is controlled, including: Based on the system control command, extract the target state data of several key target devices, and input the system control command and several target state data into the collaborative state space of the collaborative control decision generation module of the collaborative control model; Input the action set of several key target equipment into the collaborative action space of the collaborative control decision generation module; Based on the cooperative state space, the target intelligent agent, using a hybrid network and several key target devices, generates cooperative control decisions in the cooperative action space to obtain the initial cooperative control decisions. The decision optimization module of the collaborative control model is used to optimize the initial collaborative control decision to obtain the final collaborative control decision. The final collaborative control decision is published to the execution equipment of the water conservancy system, generating execution instructions for the relevant execution equipment, and controlling the execution equipment of the water conservancy system according to the execution instructions.

[0016] In one optional implementation, the decision optimization module of the cooperative control model is used to optimize the initial cooperative control decision to obtain the final cooperative control decision, including: The collaborative control decision is encoded as an individual vector of the ITOA individuals in the decision optimization module, and the fitness function, ITOA population parameters and maximum number of iterations are set. Based on the ITOA population parameters, the initial ITOA population is obtained by initializing using the Tent chaotic mapping sequence; each ITOA individual in the ITOA population corresponds to an alternative cooperative control decision. The fitness function is used to obtain the fitness value of each initial ITOA individual, and the initial ITOA individual corresponding to the initial collaborative control decision is taken as the optimal solution. Based on the ITOA algorithm for large-scale land movement and concentrated water movement, and by introducing a convergence factor, Levy Flight strategies and dynamic reverse learning mechanisms are used to iteratively update the initial ITOA population to obtain an updated ITOA population. Use the fitness function to obtain the fitness value of each updated ITOA individual, and update the updated ITOA individual with the best fitness value as the optimal solution; When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, the iterative update of the ITOA population is terminated, and the optimal solution of the current iteration is output. The individual vectors of the ITOA individuals corresponding to the optimal solution are decoded to obtain the final collaborative control decision.

[0017] Secondly, embodiments of the present invention provide a voice-interactive water conservancy system control device for implementing an intelligent content recommendation method, the device comprising: The voiceprint detection unit is used to monitor ambient audio data. Based on the wake-up threshold dynamic generation mechanism, it performs wake-up word detection. When a preset wake-up word is detected in the ambient audio data, the voiceprint matching model is used to perform identity recognition and permission verification on the ambient audio data and generate an identity authentication result. The speech-text recognition unit is used to perform speech-text recognition on environmental audio data using a speech-text recognition model if the identity authentication result meets the requirements, perform intent recognition on the obtained speech-text, obtain the intent recognition result, and generate system control commands. The collaborative control unit is used to generate and optimize collaborative control decisions based on the collaborative control model constructed using the QMIX-ITOC algorithm, according to system control commands, and to control the execution equipment of the water conservancy system based on the obtained collaborative control decisions.

[0018] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the steps of a voice-interactive water system control method provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the functional units of a voice-interactive water conservancy system control device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] The present invention will be further described below with reference to the accompanying drawings.

[0023] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0024] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0025] Those skilled in the art will understand that Figure 1The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0026] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for a voice-interactive water system control device.

[0027] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the electronic program of the voice-interactive water conservancy system control device stored in the memory 1005 through the processor 1001 and executes the voice-interactive water conservancy system control method provided in the embodiment of the present invention.

[0028] Reference Figure 2 The present invention provides a voice-interactive water system control method, the method comprising: S201: Monitor ambient audio data, perform wake-up word detection based on the wake-up threshold dynamic generation mechanism, and when a preset wake-up word is detected in the ambient audio data, use the voiceprint matching model to perform identity recognition and permission verification on the ambient audio data and generate an identity authentication result. S202: If the identity authentication result meets the requirements, use the speech-text recognition model to perform speech-text recognition on the environmental audio data, perform intent recognition on the obtained speech-text, obtain the intent recognition result, and generate system control commands. S203: Based on the system control instructions, use the collaborative control model constructed based on the QMIX-ITOC algorithm to generate and optimize collaborative control decisions, and control the execution equipment of the water conservancy system according to the obtained collaborative control decisions.

[0029] The technical solution provided in this application has at least the following beneficial effects: By replacing complex graphical interfaces with natural voice interaction, operators no longer need to memorize cumbersome device numbers, operating logic, or menu paths. They can simply issue commands in everyday language, and the system will automatically interpret their intentions and generate control strategies. This not only greatly simplifies the operation process, allowing non-professionals or new employees to quickly get started, but also avoids misoperations caused by tension and panic in emergency situations, significantly improving user-friendliness and operational safety. It achieves an end-to-end automated process from "problem detection" to "control execution," monitoring voice commands in real time and immediately initiating a collaborative control model for decision generation and optimization after recognizing voiceprints and intentions. The system completes tasks within seconds or even milliseconds, improving real-time response and interaction efficiency. It introduces voiceprint detection and authentication mechanisms, strongly binding permissions to the user's voiceprint. Before each voice command is issued, seamless, real-time identity verification and permission checks are performed. This completely changes the traditional, crude permission management based on account passwords, effectively eliminating security risks such as account sharing, theft, and unauthorized operations. The collaborative control model possesses intelligent collaborative decision-making capabilities, no longer relying on isolated empirical formulas or preset schemes. Instead, it comprehensively considers the real-time status of all relevant devices and their complex coupling relationships to make global collaborative decisions, reducing resource waste and improving scheduling efficiency.

[0030] In one optional implementation, ambient audio data is monitored, and wake-up word detection is performed based on a dynamic wake-up threshold generation mechanism. When a preset wake-up word is detected in the ambient audio data, a voiceprint matching model is used to perform identity recognition and permission verification on the ambient audio data, generating an identity authentication result, including: S2011: Using an audio sensor, monitor ambient audio data and preprocess the ambient audio data to obtain a continuous, time-ordered sequence of audio feature vectors. In this embodiment, this step serves as the entry point for system voice interaction, integrating wake word detection and voiceprint authentication into a unified step to achieve seamless, efficient, and secure access through wake-up authentication. The processing steps include: data framing: dividing continuous environmental audio data into short frames, such as 25ms per frame with a 10ms frame shift; pre-emphasis: using a high-pass filter to amplify high-frequency components and compensate for the suppression of the speech signal; windowing: applying a Hamming or Hanning window to each frame to reduce spectral leakage; and feature extraction: calculating Mel-frequency cepstral coefficients for each windowed frame, for example, extracting 40-dimensional Mel-frequency cepstral coefficient features, which constitute the audio feature vector at the current moment. S2012: Input the audio feature vector sequence into the voiceprint detection model, perform voiceprint detection based on the user's voiceprint feature library, and obtain the existence probability of the preset wake-up word and the voiceprint recognition probability vector. S2013: Based on the wake-up threshold dynamic generation mechanism, a dynamic wake-up threshold is generated. If the probability of the existence of the preset wake-up word is greater than the dynamic wake-up threshold, the preset wake-up word is detected, the identity recognition and authorization verification process is started, and the next step is entered. The formula is:

[0031] In the formula, Dynamic wake-up threshold; For dynamic wake-up base value; This is the adaptive adjustment coefficient; The energy level of background noise is calculated in real time; The background noise level is used as a reference; the wake-up threshold is dynamically generated based on real-time calculated background noise, giving wake-up word detection environmental adaptability. When the probability of the preset wake word exists When a preset wake word is detected, the identity recognition and authorization verification process is initiated. S2014: Perform identity recognition and authorization verification, and extract the largest voiceprint recognition probability and its corresponding target user ID from the voiceprint recognition probability vector; S2015: If the maximum voiceprint recognition probability is greater than the voiceprint confidence threshold, then extract the target user ID's target permission level (such as operator, administrator, super administrator) and generate the identity authentication result; In this embodiment, the integrated design reduces the response time by more than 30% compared to the traditional serial process of waking up first and then authenticating, and avoids the poor user experience caused by authentication failure after waking up.

[0032] In one alternative implementation, the voiceprint matching model is built on an improved Convolutional Recurrent Neural Network (CRNN) algorithm, and the voiceprint matching model includes convolutional layers, recurrent layers, and a dual-task output head, which includes a parallel wake-up detection head and a voiceprint verification head. In this embodiment, the convolutional layer uses several small convolutional kernels (such as 3x3) and pooling layers to extract local time-frequency features from the audio feature vector sequence, and is not sensitive to short-term speech changes; The recurrent layer uses a bidirectional gated recurrent unit (GRU) to capture long-term dependencies and analyze the temporal features of wake words; Dual-task output header: The output of the loop layer is connected to two parallel fully connected layers. Wake-up Detector: A neuron with a sigmoid activation function that outputs a probability value between 0 and 1, representing the probability of the presence of the currently input preset wake word.

[0033] Voiceprint verification head: A fully connected layer with a Softmax activation function, the number of neurons of which is equal to the number of authorized users (N) in the user's voiceprint feature database, and outputs an N-dimensional voiceprint recognition probability vector, representing the probability that the current speaker is an authorized user; Streaming inference: The model processes feature sequences in a sliding window manner. For example, the model maintains a 1.5-second context window, and every 10ms of new data comes in, it recalculates the presence probability of the preset wake word and the voiceprint recognition probability vector within the window.

[0034] In one optional implementation, the audio feature vector sequence is input into the voiceprint detection model. Based on the user's voiceprint feature database, voiceprint detection is performed to obtain the presence probability of a preset wake-up word and the voiceprint recognition probability vector, including: S20121: Input the audio feature vector sequence into the speakerprint detection model; S20122: Use the convolutional layer of the voiceprint detection model to extract local time-frequency features of the audio feature vector sequence; S20123: Using the recurrent layer of the voiceprint detection model, extract the temporal features of local time-frequency features; S20124: Based on the temporal features and the audio features corresponding to the preset wake-up word, the wake-up detection head of the dual-task output head of the voiceprint detection model is used to perform wake-up detection and obtain the existence probability of the preset wake-up word. S20125: Based on the temporal features and the voiceprint features of all users in the user voiceprint feature library, use the voiceprint verification head of the dual-task output head of the voiceprint detection model to perform voiceprint detection and obtain a voiceprint recognition probability vector that includes the voiceprint recognition probability of all users.

[0035] In one optional implementation, if the authentication result meets the requirements, a speech-to-text recognition model is used to perform speech-to-text recognition on the environmental audio data. Intent recognition is then performed on the obtained speech-to-text to obtain the intent recognition result, and system control commands are generated, including: S2021: If the target permission level corresponding to the permission label of the identity authentication result is greater than the minimum permission level threshold, then continue to collect the audio feature vector sequence of the environmental audio data. S2022: Input the audio feature vector sequence into the speech-text recognition model to perform speech-text recognition and obtain the speech-recognized text; S2023: Input the speech recognition text and the pre-built knowledge graph of the water conservancy system into the pre-built joint extraction model, perform intent recognition, and obtain the intent recognition result; S2024: Perform structured instruction conversion on the intent recognition results to obtain system control instructions for several key target equipment in the water conservancy system.

[0036] In this embodiment, the speech recognition text is "Help me check the current situation of the upper reaches of the Dongjiang River." The key entity is extracted: {Location: Upper reaches of the Dongjiang River, Time: Now}. The water conservancy system knowledge graph is accessed to understand that "upper reaches of the Dongjiang River" is a regional concept, the intent type is "video retrieval", and it is associated with all camera nodes in this region. Finally, a structured system control command is generated: {Intent: Video retrieval, Entity: {Region: Upper reaches of the Dongjiang River}, Scope: All}.

[0037] In one alternative implementation, the speech-to-text recognition model is built on an Automatic Speech Recognition (ASR) algorithm based on the Conformer-Transducer architecture; In this embodiment, the Conformer algorithm of the speech-text recognition model combines the advantages of CNN and Transformer, and can capture local and global dependencies at the same time. The Transducer algorithm supports streaming recognition, ensuring the real-time performance and accuracy of recognition, and obtaining speech-recognized text. The joint extraction model is built on the following algorithms: Bidirectional Encoder Representations from Transformers (BERT), Bidirectional Long Short-Term Memory (BiLSTM), Conditional Random Field (CRF), Support Vector Machine (SVM), and Graph Neural Network (GNN). The joint extraction model includes a word embedding module based on the BERT algorithm, a semantic feature extraction module based on the BiLSTM algorithm, an entity extraction module based on the CRF algorithm, an entity relation extraction module based on the SVM algorithm, and an intent recognition module based on the GNN algorithm. In this embodiment, the word embedding module uses BERT to perform deep context encoding on the text to generate word embedding vectors rich in semantic information. The pre-training advantage of BERT enables it to understand professional vocabulary in the field of water conservancy very well. Semantic feature extraction module: Input the word embedding vectors output by BERT into BiLSTM to further capture the forward and backward semantic dependencies in the text sequence and obtain more refined semantic features; Entity extraction module: The feature sequence output by BiLSTM is fed into CRF. CRF can learn the constraint relationship between entity labels, thereby extracting key entities in the instruction more accurately, such as {equipment: 'Dongjiang gate', location: 'Dongjiang reservoir', value: '2.5 meters'}. Entity Relationship Extraction Module: Simultaneously, the [CLS] vector (representing the semantics of the entire sentence) output by BiLSTM is fed into an SVM classifier. SVM is robust in small sample and multi-class tasks and is used to determine the relationships between entities, such as <'Dongjiang Sluice Gate', 'belongs to', 'Dongjiang Reservoir'>, <'Dongjiang Sluice Gate', 'target opening', '2.5 meters'>.

[0038] Intent recognition and graph linking: Extracted entities and relationships are matched and associated with nodes in the water conservancy system knowledge graph using entity linking technology. The knowledge graph stores the hierarchical relationships, physical attributes, and operational logic of all key equipment, reservoirs, and rivers. After linking, the originally isolated text information is placed in a structured knowledge network. Finally, GNN (Graph Neural Network) is used for message passing and aggregation on this linked subgraph. GNN can infer the user's deep intent based on the graph structure and node attributes. For example, if a user says, "Open the gates of the Dongjiang Reservoir and retrieve the video," the GNN can combine the knowledge graph showing that the Dongjiang Reservoir has multiple gates, as well as real-time data such as the current water level and flow rate, to infer that the user's true intention is likely to open the main floodgate of the Dongjiang Reservoir, rather than a small gate, and retrieve the monitoring video of the main floodgate. The structured system control command is: {Intent Type: Open gate, retrieve video, Entity: {Location: Dongjiang Reservoir, Scope: gate monitoring area}, Linkage: None}.

[0039] In one optional implementation, the speech-recognized text and a pre-built knowledge graph of the water conservancy system are input into a pre-built joint extraction model for intent recognition, yielding intent recognition results, including: S20231: Input the speech recognition text and the pre-built knowledge graph of the water conservancy system into the pre-built joint extraction model; S20232: Use the word embedding module of the joint extraction model to perform word embedding on the speech recognition text and obtain word embedding vectors; S20233: The semantic feature extraction module of the joint extraction model is used to extract the semantic features of word embedding vectors; S20234: Based on semantic features, use the entity extraction module of the joint extraction model to generate several entities; S20235: Based on semantic features, use the entity relation extraction module of the joint extraction model to generate entity relations for several entities; S20236: Based on entity relationships, link several entities to corresponding nodes in the water conservancy system knowledge graph to obtain the linked water conservancy system knowledge graph; S20237: Based on the linked knowledge graph of the water conservancy system, use the intent recognition module of the joint extraction model to perform intent recognition and obtain the intent recognition result.

[0040] In one alternative implementation, the cooperative control model is constructed based on Q-value Mixing (QMIX) - Improved Tornado Optimization Algorithm (ITOC), and the cooperative control model includes a cooperative control decision generation module based on the QMIX algorithm and a decision optimization module based on the ITOC algorithm. The cooperative control decision generation module includes a hybrid network, several agents constructed based on the Deep Recurrent Q-Network (DRQN) algorithm, and an experience playback pool. The agents correspond to the key equipment of the water conservancy system. Intelligent agents include gate intelligent agents, pump station intelligent agents, video intelligent agents, emergency response plan intelligent agents, and resource graph intelligent agents; The agent-based DRQN ​​introduces a recurrent neural network on top of the Deep Q-Network (DQN), enabling it to handle partially observable problems and possess memory capabilities. This is crucial for handling water control tasks with time delays and continuity. Agents receive the relevant portion of the global state (or the global state weighted by an attention mechanism) as their observations. Each agent inputs its observations and actions from the previous time step into its DRQN ​​network, outputting a Q-value vector for its own action set. This vector represents the expected reward for performing each action under the current observation. The hybrid network is the core innovation of QMIX. It receives the Q-value vectors of all agents as input and mixes them into a global joint action value function through a non-linear neural network that satisfies monotonicity constraints. The hybrid network also receives the global state as additional input, allowing it to adjust the weights of different agents' Q-values ​​in global decision-making based on global conditions (such as total rainfall). During the inference phase (i.e., in practical applications), the system employs a greedy strategy, selecting the joint action that maximizes the mixed Q-value from the cooperative action space based on the mixed Q-value output by the hybrid network, as the initial cooperative control decision. The decision optimization module takes the initial collaborative control decision as the objective, constructs an independent, small-scale infinite-time optimal control problem for each device that needs continuous control, and iteratively optimizes within the search space to obtain the optimal collaborative control decision. In this embodiment, the user's voice may be more complex, such as "Xiaota, Xiaota, the water level in the upper reaches of the Dongjiang River is critical, activate the emergency plan." At this time, the collaborative control model will simultaneously activate the video agent, the emergency plan agent, and the resource graph agent. The task of the video agent is to immediately retrieve real-time video from all key locations along the upper reaches of the Dongjiang River. The task of the emergency response intelligent agent is to match and retrieve the "Dongjiang River Basin Flood Control Emergency Plan" from the emergency response plan database based on the "water level alarm" event type and the "upstream of Dongjiang River" location, and display its content on one side of the large screen; The task of the resource map intelligent agent is to automatically mark the locations of emergency rescue teams, material reserve points, and refuge sites within a 5-kilometer radius on a GIS map, centered on the "upper reaches of the Dongjiang River," forming a dynamic "single map of emergency resources."

[0041] QMIX-ITOC Decision Making: The QMIX algorithm is responsible for coordinating the actions of these agents from a global perspective, ensuring the logic and effectiveness of information display. For example, the decision prioritizes the camera in the most dangerous position, with the contingency plan and resource map displayed in a split-screen format. The ITOC algorithm then optimizes the decision, such as adjusting the layout of the video feed to avoid overlap or optimizing the label density of the resource map to ensure clarity. Ultimately, a comprehensive emergency command interface integrating "on-site footage, decision-making basis, and resource support" is presented to the commanders.

[0042] In one optional implementation, based on system control commands, a collaborative control decision is generated and optimized using a collaborative control model constructed based on the QMIX-ITOC algorithm. Based on the obtained collaborative control decisions, the execution equipment of the water conservancy system is controlled, including: S2031: Based on the system control command, extract the target state data of several key target devices, and input the system control command and several target state data into the collaborative state space of the collaborative control decision generation module of the collaborative control model; For example, the gate status data includes: [upstream water level (m), downstream water level (m), current opening degree (%), motor temperature (°C), whether it is faulty (0 / 1)]; The pump station's status data includes: [water level in the collection tank (m), outlet pressure (Pa), operating frequency (Hz), unit status (running / stopped / faulty)]; The camera's status data includes: [Online status (0 / 1), current preset position, encoding format, network latency (ms)]; S2032: Input the action set of several target key equipment into the collaborative action space of the collaborative control decision generation module; S2033: Based on the cooperative state space, the target intelligent agent, using a hybrid network and several target key devices, generates cooperative control decisions in the cooperative action space to obtain the initial cooperative control decisions; S2034: The decision optimization module of the cooperative control model is used to optimize the initial cooperative control decision and obtain the final cooperative control decision. S2035: The final collaborative control decision is published to the execution equipment of the water conservancy system, execution instructions are generated for the relevant execution equipment, and the execution equipment of the water conservancy system is controlled according to the execution instructions; In this embodiment, the execution device corresponds to the key device, such as the data query device for the camera, the gate opening control motor device, etc.

[0043] In one optional implementation, the decision optimization module of the cooperative control model is used to optimize the initial cooperative control decision to obtain the final cooperative control decision, including: S20341: Encode the collaborative control decision into an individual vector of the ITOA individuals in the decision optimization module, and set the fitness function, ITOA population parameters and maximum number of iterations; The formula is:

[0044] In the formula, For ITOA individuals X fitness value; For ITOA individuals X The corresponding safety score is evaluated through rapid simulation using a hydrodynamic model (or simplified model) of the hydraulic system. X After implementation, it is crucial to check whether the water level at key points (such as low-lying urban areas and important dams) exceeds the warning line, and whether the water flow velocity causes erosion of the riverbed. If any strict safety constraints are violated, It can be set to a very large negative value (penalty term) to directly eliminate the scheme; For ITOA individuals X Corresponding economic score, calculation scheme X Total energy consumption, such as gate opening and closing energy consumption, pump station operation energy consumption, video data query energy consumption, etc., in order to minimize energy consumption, It can be set to the negative value of the total energy consumption, so that The larger the value, the lower the energy consumption. For ITOA individuals X Corresponding timeliness score, evaluation scheme X The time required to achieve the control objective (such as lowering the water level in the target area to a safe level within a specified time, or finding and retrieving target video footage and displaying it on a large screen) is also... It can be set to a negative value of the required time; the shorter the time, the higher the score. The deviation is calculated by Euclidean distance between the current solution and the initial decision. This term is introduced to ensure that the final decision does not deviate too far from the macro-intention of QMIX, thus ensuring the rationality and interpretability of the decision. It has a small weight and plays a fine-tuning role. The fitness weighting coefficient can be adjusted according to actual needs; X For ITOA individual reference parameters; S20342: Based on the ITOA population parameters, the initial ITOA population is obtained by initializing using the Tent chaotic mapping sequence; each ITOA individual in the ITOA population corresponds to an alternative cooperative control decision. The formula is:

[0045] In the formula, For the initial ITOA population, the first i An initial ITOA individual; For the first i One chaotic variable; These are the upper and lower bounds of the search space; i For ITOA individual indicators;

[0046] In the formula, For the first i- One chaotic variable; compared with random initialization, chaotic initialization can ensure that the population is evenly distributed in the solution space, thus enhancing diversity. S20343: Use the fitness function to obtain the fitness value of each initial ITOA individual, and take the initial ITOA individual corresponding to the initial collaborative control decision as the optimal solution; S20344: Based on the ITOA algorithm for large-scale land movement and concentrated water movement, and introducing a convergence factor, Levy Flight strategies and dynamic reverse learning mechanisms are used to iteratively update the initial ITOA population to obtain an updated ITOA population. Based on the current iteration number, a control parameter is introduced to determine whether the individual is in the "large-scale land movement" or "concentrated water movement" stage. P (Simulating a tornado) The formula is:

[0047] In the formula, This represents the maximum number of iterations. t This represents the current iteration number; P When the value is less than 0.5, the formula for simulating the large-scale movement of a tornado over land is:

[0048] In the formula, Number of iterations t+ The first large-scale land movement obtained by 1 i A newer ITOA individual; The convergence factor; Number of iterations t The optimal solution; For randomly selected ITOA individuals; Number of iterations t The i A new ITOA individual, initially the initial ITOA individual corresponding to the initial collaborative control decision; These are random numbers distributed according to a standard normal distribution. for Levy Distribute random numbers; b for Levy Step length, and b ∈[1,2];

[0049] In the formula, These are the maximum and minimum values ​​of the convergence factor; This represents the maximum number of iterations. t This represents the current iteration number; , To adjust the parameters; It is the hyperbolic tangent function; when P When ≥0.5, the formula for simulating the concentrated movement of a tornado over water is:

[0050] In the formula, Number of iterations t+ The first water-concentrated movement obtained i A newer ITOA individual; A uniformly random number in the range [0,1]. Number of iterations t The updated mean of individual ITOA; To develop an intensity coefficient and control the accuracy of local searches; The mean guiding coefficient is used to enhance convergence; Randomly select several updated ITOA individuals for dynamic reverse analysis, and use a greedy strategy to retain the better ITOA individuals;

[0051] In the formula, Number of iterations t+ 1 of i An updated ITOA individual after dynamic reversal; For the updated ITOA population, the first i A newer ITOA individual; For the updated ITOA population, the updated first i The reverse solution for each ITOA individual; The fitness function;

[0052] In the formula, Number of iterations t The reverse center point determined by the dynamic boundary of the current search space; It is a dynamic inverse factor that increases with iteration (intensified inverse exploration in later stages). The formula for boundary handling of all ITOA individuals that exceed the boundary is:

[0053] In the formula, Number of iterations t+ 1 of i The updated ITOA individual after boundary processing; Number of iterations t+ 1 of i A newer ITOA individual; The updated ITOA population is obtained by integrating the updated ITOA individuals obtained from large-scale land movement, the updated ITOA individuals obtained from concentrated water movement, the updated ITOA individuals after dynamic reversal, and the updated ITOA individuals after boundary processing. S20345: Use the fitness function to obtain the fitness value of each updated ITOA individual, and update the updated ITOA individual with the best fitness value as the optimal solution; S20346: When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, terminate the iterative update of the ITOA population and output the optimal solution of the current iteration; S20347: Decode the individual vector of the ITOA individual corresponding to the optimal solution to obtain the final collaborative control decision.

[0054] This invention also provides a voice-interactive water system control device, referring to... Figure 3 The diagram shows a functional unit diagram of a voice-interactive water conservancy system control device 300 according to the present invention. The device may include the following units: The voiceprint detection unit 301 is used to monitor ambient audio data. Based on the wake-up threshold dynamic generation mechanism, it performs wake-up word detection. When a preset wake-up word is detected in the ambient audio data, the voiceprint matching model is used to perform identity recognition and permission verification on the ambient audio data to generate an identity authentication result. The speech-text recognition unit 302 is used to perform speech-text recognition on environmental audio data using a speech-text recognition model if the identity authentication result meets the requirements, perform intent recognition on the obtained speech-text, obtain intent recognition results, and generate system control commands. The collaborative control unit 303 is used to generate and optimize collaborative control decisions using a collaborative control model based on the QMIX-ITOC algorithm according to system control commands, and to control the execution equipment of the water conservancy system according to the obtained collaborative control decisions.

[0055] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; The processor, when executing the program stored in the memory, implements the voice-interactive water system control method of the present invention.

[0056] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0057] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0058] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the voice-interactive water system control method of the embodiments of the present invention.

[0059] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0064] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A voice-interactive control method for a water conservancy system, characterized in that, The method includes: The system monitors ambient audio data and performs wake-up word detection based on a dynamic wake-up threshold generation mechanism. When a preset wake-up word is detected in the ambient audio data, the system uses a voiceprint matching model to perform identity recognition and permission verification on the ambient audio data and generates an identity authentication result. If the identity authentication result meets the requirements, the speech-text recognition model is used to perform speech-text recognition on the environmental audio data, the obtained speech-text is then used to perform intent recognition, the intent recognition result is obtained, and system control commands are generated. Based on the system control commands, a collaborative control model constructed using the QMIX-ITOC algorithm is used to generate and optimize collaborative control decisions, and the execution equipment of the water conservancy system is controlled according to the obtained collaborative control decisions.

2. The water conservancy system control method based on voice interaction according to claim 1, characterized in that, The system monitors ambient audio data and performs wake-up word detection based on a dynamic wake-up threshold generation mechanism. When a preset wake-up word is detected in the ambient audio data, a voiceprint matching model is used to perform identity recognition and permission verification on the ambient audio data, generating an identity authentication result, including: Using an audio sensor, ambient audio data is monitored and preprocessed to obtain a continuous, time-ordered sequence of audio feature vectors. The audio feature vector sequence is input into the voiceprint detection model. Based on the user's voiceprint feature library, voiceprint detection is performed to obtain the existence probability of the preset wake-up word and the voiceprint recognition probability vector. Based on the wake-up threshold dynamic generation mechanism, a dynamic wake-up threshold is generated. If the probability of the existence of the preset wake-up word is greater than the dynamic wake-up threshold, the preset wake-up word is detected, the identity recognition and authorization verification process is initiated, and the process proceeds to the next step. Perform identity recognition and authorization verification, and extract the largest voiceprint recognition probability and its corresponding target user ID from the voiceprint recognition probability vector; If the maximum voiceprint recognition probability is greater than the voiceprint confidence threshold, then the target user ID's target permission level is extracted, and an identity authentication result is generated.

3. The voice-interactive water system control method according to claim 2, characterized in that, The voiceprint matching model is constructed based on an improved CRNN algorithm and includes convolutional layers, recurrent layers, and a dual-task output head, which includes a parallel wake-up detection head and a voiceprint verification head.

4. The voice-interactive water system control method according to claim 3, characterized in that, The audio feature vector sequence is input into the voiceprint detection model. Based on the user's voiceprint feature database, voiceprint detection is performed to obtain the presence probability of a preset wake-up word and the voiceprint recognition probability vector, including: The audio feature vector sequence is input into the speakerprint detection model; The convolutional layer of the voiceprint detection model is used to extract local time-frequency features of the audio feature vector sequence; Temporal features of local time-frequency features are extracted using the recurrent layer of the voiceprint detection model. Based on the temporal features and the audio features corresponding to the preset wake-up words, the wake-up detection head of the dual-task output head of the voiceprint detection model is used to perform wake-up detection and obtain the probability of the existence of the preset wake-up words. Based on the temporal features and the voiceprint features of all users in the user voiceprint feature library, the voiceprint verification head of the dual-task output head of the voiceprint detection model is used to perform voiceprint detection, and a voiceprint recognition probability vector including the voiceprint recognition probability of all users is obtained.

5. The voice-interactive water system control method according to claim 4, characterized in that, If the identity authentication result meets the requirements, the speech-to-text recognition model is used to perform speech-to-text recognition on the environmental audio data. Intent recognition is then performed on the obtained speech-to-text to obtain the intent recognition result, and system control commands are generated, including: If the target permission level corresponding to the permission label of the identity authentication result is greater than the minimum permission level threshold, then continue to collect the audio feature vector sequence of the environmental audio data; The audio feature vector sequence is input into the speech-text recognition model to perform speech-text recognition and obtain the speech-recognized text. The speech recognition text and the pre-built knowledge graph of the water conservancy system are input into the pre-built joint extraction model to perform intent recognition and obtain the intent recognition result; The intent recognition results are converted into structured instructions to obtain system control instructions for several key target equipment in the water conservancy system.

6. The voice-interactive water system control method according to claim 5, characterized in that, The speech-to-text recognition model is built based on the ASR algorithm of the Conformer-Transducer architecture; The joint extraction model is constructed based on the BERT-BiLSTM-CRF-SVM-GNN algorithm, and includes a word embedding module based on the BERT algorithm, a semantic feature extraction module based on the BiLSTM algorithm, an entity extraction module based on the CRF algorithm, an entity relation extraction module based on the SVM algorithm, and an intent recognition module based on the GNN algorithm.

7. The voice-interactive water system control method according to claim 6, characterized in that, The cooperative control model is constructed based on the QMIX-ITOC algorithm, and includes a cooperative control decision generation module based on the QMIX algorithm and a decision optimization module based on the ITOC algorithm. The cooperative control decision generation module includes a hybrid network, several agents constructed based on the DRQN ​​algorithm, and an experience playback pool. The agents correspond to the key equipment of the water conservancy system.

8. The voice-interactive water system control method according to claim 7, characterized in that, Based on system control commands, a collaborative control model constructed using the QMIX-ITOC algorithm is used to generate and optimize collaborative control decisions. Based on the obtained collaborative control decisions, the execution equipment of the water conservancy system is controlled, including: Based on the system control command, extract the target state data of several key target devices, and input the system control command and several target state data into the collaborative state space of the collaborative control decision generation module of the collaborative control model; Input the action set of several key target equipment into the collaborative action space of the collaborative control decision generation module; Based on the cooperative state space, the target intelligent agent, using a hybrid network and several key target devices, generates cooperative control decisions in the cooperative action space to obtain the initial cooperative control decisions. The decision optimization module of the collaborative control model is used to optimize the initial collaborative control decision to obtain the final collaborative control decision. The final collaborative control decision is published to the execution equipment of the water conservancy system, generating execution instructions for the relevant execution equipment, and controlling the execution equipment of the water conservancy system according to the execution instructions.

9. The voice-interactive water system control method according to claim 8, characterized in that, The decision optimization module of the cooperative control model optimizes the initial cooperative control decision to obtain the final cooperative control decision, including: The collaborative control decision is encoded as an individual vector of the ITOA individuals in the decision optimization module, and the fitness function, ITOA population parameters and maximum number of iterations are set. Based on the ITOA population parameters, the initial ITOA population is obtained by initializing using the Tent chaotic mapping sequence; each ITOA individual in the ITOA population corresponds to an alternative cooperative control decision. The fitness function is used to obtain the fitness value of each initial ITOA individual, and the initial ITOA individual corresponding to the initial collaborative control decision is taken as the optimal solution. Based on the ITOA algorithm for large-scale land movement and concentrated water movement, and by introducing a convergence factor, Levy Flight strategies and dynamic reverse learning mechanisms are used to iteratively update the initial ITOA population to obtain an updated ITOA population. Use the fitness function to obtain the fitness value of each updated ITOA individual, and update the updated ITOA individual with the best fitness value as the optimal solution; When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, the iterative update of the ITOA population is terminated, and the optimal solution of the current iteration is output. The individual vectors of the ITOA individuals corresponding to the optimal solution are decoded to obtain the final collaborative control decision.

10. A voice-interactive water system control device, used to implement the water system control method as described in any one of claims 1-9, characterized in that, The device includes: The voiceprint detection unit is used to monitor ambient audio data. Based on the wake-up threshold dynamic generation mechanism, it performs wake-up word detection. When a preset wake-up word is detected in the ambient audio data, the voiceprint matching model is used to perform identity recognition and permission verification on the ambient audio data and generate an identity authentication result. The speech-text recognition unit is used to perform speech-text recognition on environmental audio data using a speech-text recognition model if the identity authentication result meets the requirements, perform intent recognition on the obtained speech-text, obtain the intent recognition result, and generate system control commands. The collaborative control unit is used to generate and optimize collaborative control decisions based on the collaborative control model constructed using the QMIX-ITOC algorithm, according to system control commands, and to control the execution equipment of the water conservancy system based on the obtained collaborative control decisions.

Images