A tourist sentiment analysis method based on a large language model

Abstract

The application discloses a tourist sentiment analysis method based on a large language model, and belongs to the technical field of intelligent scenic spot management, and solves the problems of poor real-time performance, low Chinese context recognition accuracy, insufficient multi-modal fusion, lack of multi-region sentiment linkage prediction and closed-loop management capability in the prior art. The core steps of the method include: S1 data acquisition and preprocessing, S2 basic sentiment calculation, S3 Chinese context enhancement correction, S4 final sentiment value fusion calculation, S5 regional sentiment index construction and early warning, S6 closed-loop feedback regulation, and the S7 multi-region sentiment coupling prediction and pre-intervention step is additionally added. The application supports text and voice multi-modal data acquisition, designs a special sentiment correction mechanism for Chinese context, can realize real-time analysis of tourist sentiment, negative emotion early warning and pre-intervention, and forms an operation optimization closed loop, and provides accurate decision basis for fine management of intelligent scenic spots.

Description

Technical Field

[0001] This invention relates to the field of smart scenic area management technology, specifically a tourist sentiment analysis method based on a large language model. Background Technology

[0002] With the rapid development of the cultural tourism industry, the demand for refined management of large-scale integrated tourist attractions is increasing, and tourists' emotional experience has become a core indicator for measuring the service quality of scenic spots and optimizing operation and management. In the current technology, tourist experience feedback in scenic spots mainly relies on post-event questionnaires and online platform evaluations, which have serious lags and cannot capture tourists' real-time emotional changes in the scenic spot, making it difficult to achieve rapid operational intervention.

[0003] At the level of sentiment analysis technology, traditional sentiment analysis methods mostly use general bag-of-words models or basic pre-trained models, which are not well adapted to the Chinese context: on the one hand, they cannot accurately identify the influence of transitional structures and degree adverbs in Chinese sentences on sentiment tendencies, which easily leads to misjudgment of sentiment scores and low recognition accuracy; on the other hand, most existing solutions only perform independent sentiment analysis on single text data, without considering the sentiment linkage effect between different functional areas in large scenic spots, and cannot predict the cross-regional spread trend of tourists' negative emotions, making it difficult to achieve proactive and forward-looking operation and control.

[0004] Furthermore, existing scenic area sentiment collection solutions are mostly limited to text evaluation input, covering limited scenarios and failing to meet tourists' needs for contactless, voice-based, and real-time feedback. They also lack the ability to integrate and analyze multimodal data, making it difficult to comprehensively and realistically recreate tourists' real-time emotional state, thus failing to make timely operational interventions. Therefore, we propose a tourist sentiment analysis method based on a large language model. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned deficiencies in existing technologies by providing a tourist sentiment analysis method based on a large language model. This method solves the technical problems in existing technologies, such as poor real-time performance of tourist sentiment analysis in scenic areas, low accuracy of Chinese context recognition, lack of multi-regional sentiment linkage prediction capabilities, insufficient multimodal fusion, and inability to form closed-loop management.

[0006] To address the above problems, the technical solution adopted by this invention is: a tourist sentiment analysis method based on a large language model, comprising the following steps: S1. Data Collection and Preprocessing: By deploying tourist information collection terminals in various functional areas of the scenic area, text evaluation data and / or voice feedback data of tourists are collected, and corresponding scene tags, time and anonymous user IDs are matched for each piece of collected data; Chinese word segmentation and semantic vector embedding are performed on the text data, and acoustic feature extraction and speech-to-text processing are performed on the voice data. S2. Basic Sentiment Calculation: Using a pre-trained large language model, a basic sentiment score with values ​​ranging from [-1, 1] is output to the pre-processed text data. For the voice feedback data, a pre-trained acoustic emotion recognition model is simultaneously outputting a base voice emotion value ranging from [-1, 1]. ; S3. Chinese Contextual Enhancement and Correction: Identifies transitional structures and degree adverbs in the text, sets weight coefficients for the sentences before and after the transition, and applies sentiment correction values ​​that match the semantics, thus calculating the contextual correction value. ; S4. Final Sentiment Score Fusion Calculation: Using a pre-set weighted fusion model, the basic sentiment score and context correction value are fused to calculate the final sentiment score for each data point. For voice feedback data, the basic voice emotion value is simultaneously fused in the weighted fusion model. S5. Construction and Early Warning of Regional Sentiment Index: Based on the functional areas of the scenic area, the arithmetic mean of the final sentiment values ​​of all valid data in the corresponding time period is calculated to obtain the regional sentiment index. And construct the trend of sentiment changes over time. ,when <0 or When the user's experience falls below a preset negative warning threshold, a negative experience warning is triggered. S6. Closed-loop feedback control: Based on the triggered negative experience warning, corresponding scenic area operation optimization measures are generated. After the optimization measures are implemented, tourist sentiment data in the corresponding area is collected again to verify the optimization effect, forming a closed-loop operation system of "data collection, sentiment analysis, management optimization, and re-collection".

[0007] Preferably, the specific steps of the Chinese context enhancement and correction in step S3 are as follows: S3.1, Transition Structure Recognition: Detect transition keywords in the text. If a transition structure is identified, set a weight coefficient for the statement preceding the transition. The weight coefficient of the statement after the transition. ,in + =1; S3.2 Degree Adverb Correction: Identify degree adverbs in the text, distinguish between negative and positive semantic degrees, and set corresponding negative correction values. With positive correction value ; S3.3, Calculation of Contextual Correction Value: via formula Calculate the context correction value .

[0008] Preferably, in step S4, for the pure text input scenario, the formula for the weighted fusion model is: ; in, Based on the weighting of the basic sentiment score, As a context-corrected value weight, Weighted by the user's historical sentiment average, and + + =1; This represents the average sentiment of the current anonymous user ID across the last N experiences in the corresponding scenario.

[0009] Preferably, in step S4, for the multimodal speech input scenario, the formula for the weighted fusion model is: ; in, Based on the weighting of the basic sentiment score, As a context-corrected value weight, The weights are the base values ​​for voice emotion, and + + =1.

[0010] Preferably, it also includes a multi-regional emotion coupling prediction and pre-intervention step S7, specifically: S7.1, Area Division and State Definition: The scenic area is divided into four functional areas: entrance ticket check area, main scenic area, dining area, and performance area. The emotional index of each area at time t is defined as follows: , And construct the region state vector ; S7.2 Constructing the regional influence matrix M: matrix elements This represents the weight of the influence of region j on region i. ∈[0,1], the diagonal elements of the matrix are 1; S7.3, Emotional State Prediction: Through formula Predicting the regional sentiment state in the next moment, among which To manually adjust the input vector; S7.4, Pre-intervention trigger: Preset propagation threshold When the predicted arbitrary region In such cases, the system can trigger cross-regional pre-intervention in advance and generate corresponding optimization and adjustment measures.

[0011] Compared with existing technologies, this invention provides a tourist sentiment analysis method based on a large language model, which has the following beneficial effects: 1. This invention enables real-time, full-scene collection and analysis of tourists' emotions in scenic areas. It supports two modes: touchscreen text input and microphone voice input. It covers all functional areas of the scenic area, including the ticket check area, main scenic area, catering area, and performance area. It can capture tourists' instantaneous emotional changes and solves the problem of delayed feedback in traditional solutions. 2. This invention has designed a unique enhancement and correction mechanism for Chinese contexts. By allocating weights to transitional structures and correcting the sentiment of degree adverbs, it significantly improves the accuracy of sentiment recognition in Chinese scenic spot evaluation scenarios, reduces semantic misjudgments, and can more accurately restore tourists' true emotional tendencies. 3. This invention constructs a multi-regional emotional coupling prediction model, which can quantify the emotional linkage effect between different functional areas of a scenic spot, predict the cross-regional spread trend of tourists' negative emotions in advance, realize the upgrade from passive response to proactive pre-intervention, effectively avoid the spread of collective negative experiences, and enhance the initiative of scenic spot management. 4. This invention forms a complete closed-loop management system, which automatically generates operational optimization suggestions based on sentiment analysis results. After the measures are implemented, the optimization effect is verified by data re-collection, realizing two-way optimization of tourist emotional experience and scenic area operation management, and providing accurate and implementable decision-making basis for the refined and intelligent management of scenic areas. Attached Figure Description

[0012] Figure 1 This is a flowchart of the method framework of the present invention. Detailed Implementation

[0013] This invention discloses a tourist sentiment analysis method based on a large language model. The following detailed description, in conjunction with specific embodiments, further illustrates the invention. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.

[0014] refer to Figure 1 All embodiments of this invention are applied to large-scale integrated tourist attractions, which include four core functional areas: an entrance ticket check area, a main scenic area, a dining area, and a performance area. This method is used to analyze tourists' emotional experiences in real time and provide decision-making basis for optimizing scenic area management. Each functional area within the scenic area is equipped with a tourist information collection terminal, which includes a touch screen evaluation terminal, a microphone voice acquisition module, and a voice-to-text module.

[0015] The endpoints and any values ​​of the ranges disclosed herein are not limited to the precise ranges or values, and these ranges or values ​​should be understood to include values ​​close to these ranges or values. For numerical ranges, the endpoint values ​​of the various ranges, the endpoint values ​​of the various ranges and individual point values, and individual point values ​​can be combined with each other to obtain one or more new numerical ranges, which should be considered as specifically disclosed herein.

[0016] As a specific embodiment 1 of the present invention: In this embodiment, for the scenario where tourists submit text evaluations through the touch-screen evaluation terminal, tourist sentiment analysis is performed. The specific steps are as follows: 1. Data collection A tourist submits an evaluation text at the dining area through the terminal: "The queue is a bit long, but the waiter has a very good attitude and the taste is also good." The system synchronously collects and records the associated information of this data: text content T, scenario label S = dining area, time t, anonymous user ID.

[0017] 2. Semantic preprocessing 2.1 Word segmentation: Perform Chinese word segmentation on the above text content to obtain a word sequence: queue / a bit / long / but / waiter / attitude / very good / taste / good; 2.2 Semantic vector embedding: Encode the segmented text through a pre-trained large language model to obtain a semantic vector , where T is the input text content and the vector dimension is 1024 dimensions.

[0018] 3. Basic sentiment calculation Use a pre-trained large language model to output a basic sentiment score , and the value range is [-1, 1]. In this embodiment, the model outputs = 0.35.

[0019] 4. Chinese context enhancement calculation 4.1 Recognition of turning structure: Detect the turning keyword "but" in the text, determine that there is a turning structure, and set the weight coefficient of the previous sentence = 0.4, and the weight coefficient of the subsequent sentence = 0.6, ensuring that the sentiment weight of the sentence after turning is greater than that of the previous sentence; 4.2 Modification of degree adverbs: Identify the degree modifiers in the text. Among them, "a bit long" is mildly negative, and the negative correction value = -0.2; "very good" and "good" are positive enhancements, and the positive correction value = +0.4; 4.3 Calculation of context correction value: Calculate the context correction value through the formula Substitute the values to obtain: = 0.4×(-0.2) + 0.6×0.4 = -0.08 + 0.24 = 0.16.

[0020] 5. Calculation of final sentiment value The system calculates the average emotional experience of this anonymous user ID in the dining area over the past three times. =0.2, the final sentiment score is calculated using a text-scene weighted fusion model. The model formula is: ; In this embodiment, a weighting coefficient is set: =0.5, =0.3, =0.2, substitute the values ​​to calculate: =0.5×0.35+0.3×0.16+0.2×0.2=0.175+0.048+0.04=0.263; Final Emotional Value =0.263, indicating that the tourist had a mildly positive experience.

[0021] 6. Construction and Early Warning of Scenic Area Experience Index In this embodiment, 50 valid sentiment data points collected from the dining area during a certain period are analyzed, and the regional sentiment index of the dining area during that period is calculated. The arithmetic mean of the final sentiment values ​​of all data is used. For example, if the average value is 0.12, it indicates that the overall experience of tourists in the dining area during that period is slightly neutral to positive.

[0022] The system constructs a time series of sentiment changes and calculates trend values. ,like A value less than 0 indicates a declining trend in visitor experience in the area, triggering an alert.

[0023] 7. Closed-loop feedback mechanism When the regional sentiment index <0, or When a warning is triggered by a value less than 0, the system will automatically perform the following actions: (1) Send early warning information to the scenic area management terminal; (2) Optimization suggestions: increase the number of service personnel to provide emotional guidance and comfort, dynamically adjust the queue diversion, and add temporary service windows; After the optimization measures are implemented, the system re-collects visitor sentiment data in the dining area, repeats the above sentiment analysis steps, and judges whether the visitor experience has improved, forming a closed-loop operation system of "data collection, sentiment analysis, management optimization, and re-collection".

[0024] 8. Model training process in this embodiment 8.1 Training Data: 5000 manually annotated Chinese scenic spot evaluation texts were used, with sentiment labels ranging from [-1, 1]; 8.2 Loss Function: The mean squared error loss function is used. ,in The model predicts sentiment scores. These are the actual sentiment values ​​labeled by humans. 8.3 Optimization method: The Adam optimizer was used, the learning rate was set to 0.001, and the number of training rounds was 30.

[0025] Through the above training, the model can effectively improve the accuracy of Chinese contextual emotion recognition, reduce the false judgment rate, reflect changes in the scenic area experience in real time, and support the refined management decision-making of the scenic area.

[0026] As a specific embodiment 2 of the present invention: This embodiment targets the scenario where tourists submit voice feedback through the microphone voice acquisition module, and performs multimodal fusion tourist sentiment analysis. The specific steps are as follows: 1. Voice data acquisition and preprocessing A tourist submitted a voice message through the terminal: "There were a lot of people today, and the queue was long, but overall I still had a lot of fun." The system processes the raw speech signal. Perform preprocessing: 1.1 Acoustic Feature Extraction: Extract 13-dimensional MFCC features, pitch F0, energy intensity, and speech rate from the speech signal to form a 20-dimensional acoustic feature vector A; 1.2 Automatic Speech-to-Text (ASR): Converts the original speech signal into text content using a speech-to-text module. "There were a lot of people today, and the queues were long, but overall I still had a lot of fun." The system synchronously records the scene label S for this data entry as: main scenic area, time t, and anonymous user ID.

[0027] 2. Calculation of baseline voice emotion values Calculate the baseline value of speech emotion using a pre-trained acoustic emotion recognition model. , The value range is [-1, 1]. In this embodiment, the model output is... =0.15 indicates that the tourist's tone was slightly positive.

[0028] 3. Text Semantic Foundation Sentiment Computing A pre-trained large language model outputs a basic sentiment score for the converted text. In this embodiment =0.3.

[0029] 4. Enhanced Chinese context correction 4.1 Semantic Feature Recognition: Recognize keywords and structures in the text, where "a lot" is a mild negative, "long queue time" is a negative, and the transition keywords "however" and "quite happy" are positive reinforcements; 4.2 Weighting and Correction Value Settings: Set negative correction values =-0.3, positive correction value =+0.5; Weight of the sentence preceding the transitional structure =0.4, weight of the second sentence =0.6; 4.3 Calculation of contextual correction value: using the formula Calculations and substitution of values ​​yield: =0.4×(-0.3)+0.6×0.5=-0.12+0.30=0.18.

[0030] 5. Final Emotional Value Fusion Calculation This embodiment uses a speech multimodal weighted fusion model, the formula of which is: Set the weighting coefficients: =0.5, =0.3, =0.2, substitute the values ​​to calculate: =0.5×0.3+0.3×0.18+0.2×0.15=0.15+0.054+0.03=0.234; Final Emotional Value =0.234, this value is included in the regional sentiment index statistics of the main scenic area. If the original average sentiment index of the main scenic area during this period was 0.05, after adding this data, =0.08, indicating that the overall mood has improved slightly.

[0031] 6. Abnormal early warning judgment Preset negative experience warning threshold =-0.2, when < A negative experience warning is triggered at any time; in this embodiment, the current... =0.08 > -0.2, no warning is triggered.

[0032] If negative expressions related to queuing account for more than 40% of the voice samples during a certain period, the system will automatically generate optimization suggestions: add temporary windows, increase the number of guides, and adjust the tourist flow. After the measures are implemented, voice data will be collected again and the regional sentiment index will be recalculated.

[0033] 7. Supplementary Model Training in this Example 7.1 Training Data: 3000 manually labeled voice samples of tourists in scenic areas were used, with emotion labels ranging from [-1, 1]; 7.2 Multimodal Loss Function: ; 7.3 Training with fused parameters: The Adam optimizer was used, with a learning rate of 0.0005, a batch size of 64, and 40 training epochs.

[0034] Compared to Example 1, the model in this example enhances the ability to recognize tourists' true tone, reduces misjudgment of text semantics, improves the accuracy of negative experience warnings, and supports feedback from tourists without touchscreen input devices, covering more comprehensive scenarios.

[0035] The core algorithm framework of Example 1 and Example 2 is the same, only the input modality is different. Both are based on large language models for semantic analysis and have formed a complete closed-loop management and optimization mechanism.

[0036] As a specific embodiment 3 of the present invention: This embodiment addresses the emotional linkage effect between different functional areas within a large scenic area by constructing a multi-regional emotional coupling prediction model. This model is used to predict the cross-regional spread trend of tourists' emotions in advance, enabling proactive intervention. The specific implementation method is as follows: 1. Scenic Area Division and Status Definition The scenic area is divided into four core functional areas, represented as a set: ; in: Entrance ticket check area; Main scenic area; Dining area; Performance area; Define the sentiment index of each region at time t as: The range of values ​​is .

[0037] 2. Construct the regional influence matrix M The regional influence matrix M is a 4×4 matrix, and the matrix elements are... This represents the weight of the influence of region j on region i. The matrix has 1 as the diagonal element, and its matrix form is as follows: ; In this embodiment, weights are set according to the operational patterns of the scenic area: the catering area has a greater impact on the main scenic area, so weights are set accordingly. =0.4; The performance area has a significant impact on the entrance area, so a setting is needed. =0.3; the remaining weights are set according to the actual operation data of the scenic area.

[0038] 3. Prediction of emotional state in the next moment Define the region state vector: ; The following formula can be used to predict the regional sentiment state in the next moment: ; in, For the region coupling matrix, The input vector is adjusted manually.

[0039] In this embodiment, let the current region state vector be... The catering area has an S3(t) value of -0.3, indicating a significantly negative state. Substituting this into the regional influence matrix, we obtain the predicted state vector for the next time step: ; The forecast results show that the main scenic area S2(t+1)=-0.02, indicating a negative transmission trend.

[0040] 4. Judgment of abnormal propagation threshold Preset threshold for the spread of negative emotions =-0.15, if any region is predicted If so, the system will trigger regional intervention in advance.

[0041] Meanwhile, this embodiment integrates semantic trend analysis from a large language model during the prediction phase: constructing a semantic trend vector from the N most recent tourist reviews. The prediction results can be optimized using the following formula: ; in + =1, set in this embodiment =0.7, =0.3.

[0042] 5. Closed-loop regulation model and intervention measures If a negative propagation trend is predicted, the system calculates the optimal adjustment vector: ; Where K is the diagonal adjustment matrix. In this embodiment, the target sentiment index vector is preset. .

[0043] In this embodiment, based on the predicted spread of negative emotions from the dining area to the main landscape area, the system automatically generates preliminary intervention measures: (1) Open temporary service windows in the catering area; (2) Push mobile food carts to the surrounding areas of the catering area; (3) Divert tourists to low-load areas such as performance areas through scenic area broadcasts and mini-programs; After the intervention measures were implemented, the system re-collected tourist sentiment data from each region, repeated the sentiment analysis and prediction steps, and verified the intervention effect.

[0044] 6. Technical effects of this embodiment This embodiment achieves accurate modeling of the spread of emotions between different areas of the scenic area, enabling early prediction of declining tourist experiences, effectively reducing the spread of negative group experiences, and significantly improving the initiative and foresight of scenic area management.

[0045] The three embodiments of this invention present a clear technological progression: Embodiment 1 achieves accurate sentiment analysis of a single text; Embodiment 2 achieves fusion sentiment recognition of multimodal speech; and Embodiment 3 achieves multi-regional sentiment linkage prediction and proactive regulation. Together, these three embodiments construct a complete closed-loop system of "collection, analysis, index construction, prediction, regulation, and re-collection," comprehensively covering the entire process of visitor sentiment management in smart scenic areas. This improves the visitor experience, benefits scenic area management, and provides ideas and suggestions for improvement.

[0046] The technical solutions of the various embodiments can be combined with each other, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0047] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A tourist sentiment analysis method based on a large language model, characterized in that, Includes the following steps: S1. Data Collection and Preprocessing: By deploying tourist information collection terminals in various functional areas of the scenic area, text evaluation data and / or voice feedback data of tourists are collected, and corresponding scene tags, time and anonymous user IDs are matched for each piece of collected data. The system performs Chinese word segmentation and semantic vector embedding on text data, and acoustic feature extraction and speech-to-text processing on speech data. S2. Basic Sentiment Calculation: Using a pre-trained large language model, a basic sentiment score with values ​​ranging from [-1, 1] is output to the pre-processed text data. ; For the voice feedback data, a pre-trained acoustic emotion recognition model is simultaneously outputting a base voice emotion value ranging from [-1, 1]. ; S3. Chinese Contextual Enhancement and Correction: Identifies transitional structures and degree adverbs in the text, sets weight coefficients for the sentences before and after the transition, and applies sentiment correction values ​​that match the semantics, thus calculating the contextual correction value. ; S4. Final Sentiment Score Fusion Calculation: Using a pre-set weighted fusion model, the basic sentiment score and context correction value are fused to calculate the final sentiment score for each data point. ; For voice feedback data, the basic voice emotion value is simultaneously fused in the weighted fusion model; S5. Construction and Early Warning of Regional Sentiment Index: Based on the functional areas of the scenic area, the arithmetic mean of the final sentiment values ​​of all valid data in the corresponding time period is calculated to obtain the regional sentiment index. And construct the trend of sentiment changes over time. ,when <0 or When the user's experience falls below a preset negative warning threshold, a negative experience warning is triggered. S6. Closed-loop feedback control: Based on the triggered negative experience warning, corresponding scenic area operation optimization measures are generated. After the optimization measures are implemented, tourist sentiment data in the corresponding area is collected again to verify the optimization effect, forming a closed-loop operation system of "data collection, sentiment analysis, management optimization, and re-collection".

2. The tourist sentiment analysis method based on a large language model according to claim 1, characterized in that, The specific steps of the Chinese context enhancement and correction in step S3 are as follows: S3.1, Transition Structure Recognition: Detect transition keywords in the text. If a transition structure is identified, set a weight coefficient for the statement preceding the transition. The weight coefficient of the statement after the transition. ,in + =1; S3.2, Degree Adverb Correction: Identify degree adverbs in the text, distinguish between negative and positive semantic degrees, and set corresponding negative correction values. With positive correction value ; S3.3, Calculation of Contextual Correction Value: via formula Calculate the context correction value .

3. The tourist sentiment analysis method based on a large language model according to claim 1, characterized in that, In step S4, for the plain text input scenario, the formula for the weighted fusion model is: ； in, Based on the weighting of the basic sentiment score, As a context-corrected value weight, Weighted by the user's historical sentiment average, and + + =1; This represents the average sentiment of the current anonymous user ID across the last N experiences in the corresponding scenario.

4. The tourist sentiment analysis method based on a large language model according to claim 1, characterized in that, In step S4, for the multimodal speech input scenario, the formula for the weighted fusion model is: ； in, Based on the weighting of the basic sentiment score, As a context-corrected value weight, The weights are the base values ​​for voice emotion, and + + =1.

5. The tourist sentiment analysis method based on a large language model according to claim 1, characterized in that, It also includes step S7, which involves multi-regional emotion coupling prediction and pre-intervention, specifically: S7.1, Area Division and State Definition: The scenic area is divided into four functional areas: entrance ticket check area, main scenic area, dining area, and performance area. The emotional index of each area at time t is defined as follows: , And construct the region state vector ; S7.2 Constructing the regional influence matrix M: matrix elements This represents the weight of the influence of region j on region i. ∈[0,1], the diagonal elements of the matrix are 1; S7.3, Emotional State Prediction: Through formula Predicting the regional sentiment state in the next moment, among which To manually adjust the input vector; S7.4, Pre-intervention trigger: Preset propagation threshold When the predicted arbitrary region In such cases, the system can trigger cross-regional pre-intervention in advance and generate corresponding optimization and adjustment measures.

Images