Intelligent landscape space building system based on emotion data

Abstract

The invention provides a landscape space intelligent building system based on emotion data, and the system comprises a data collection module which obtains the physiological data and environment data of a user in real time through a wearable sensor and environment monitoring equipment, and guarantees the privacy through a user authorization mechanism; the data processing module carries out anonymization encryption processing on the user data, environment data are fused to generate an emotion label vector, and meanwhile, intensity values of landscape elements are extracted based on academic literature emerging word analysis; the decision-making module calculates a landscape element confidence coefficient weight in combination with the strength value, and constructs a reinforcement learning framework to generate a landscape element adjustment scheme; the execution module implements dynamic regulation and control through an adjustable landscape component; and the closed-loop feedback module monitors the change of the emotional state of the user, optimizes a confidence coefficient weight and reinforces a learning strategy. According to the system, real-time self-adaptive adjustment of the landscape space is driven through the emotion data, and the matching degree of the environment and the user emotion state is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent environment interaction technology, and in particular to a landscape space intelligent creation system based on emotion data. Background Art

[0002] As a crucial vehicle for human activity, the quality of landscape environments directly impacts users' emotional experiences and psychological states. Traditional landscape design relies primarily on the designer's subjective experience and static planning schemes, making it difficult to dynamically adjust to users' real-time emotional needs after the space is operational. While modern intelligent buildings have introduced automated control technologies for environmental parameters, these systems generally suffer from two major limitations: First, the control logic is based on a preset physical threshold response mechanism, failing to fully account for the dynamic changes in users' physiological and psychological states; second, the single-dimensional adjustment of environmental parameters lacks the closed-loop learning capability to respond to user emotional feedback, leading to a disconnect between the spatial atmosphere created and the actual user experience.

[0003] In recent years, the development of wearable devices and computer vision technology has made it possible to capture user physiological data and facial expressions in real time. Some studies have attempted to use this data to build environmental regulation systems, but existing solutions still have significant shortcomings. First, most systems focus solely on user data itself, failing to integrate it with environmental parameters for multimodal analysis, thus ignoring the interactive impact of environmental factors on emotional state. Second, regulation strategies often rely on static rule bases or simple machine learning models, lacking the ability to continuously track academic research in the landscape field, resulting in decision-making that lags behind cutting-edge knowledge. Furthermore, imperfect privacy protection mechanisms, insufficient universality of emotion quantification models, and the lack of a closed-loop optimization chain between landscape element regulation and user feedback further restrict the practicality and sustainable evolution of such systems. Therefore, there is an urgent need for an intelligent landscape space creation system that integrates multi-source emotion data, a dynamic academic knowledge base, and a closed-loop reinforcement learning mechanism. Summary of the Invention

[0004] In order to solve the technical problems mentioned in the current background technology, the purpose of the present invention is to provide an intelligent landscape space creation system based on emotional data.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The landscape space intelligent construction system based on emotional data includes:

[0007] The data collection module collects emotional data including user data and environmental data in real time, and implements the user authorization mechanism when collecting user data;

[0008] The data processing module anonymizes the user data to generate an encrypted feature vector; fuses the encrypted feature vector with the environmental data to generate an emotional label vector; obtains literature in the field of landscape emotion, extracts the emergent words and life cycles in the literature based on the emergent detection algorithm, and calculates the strength value of the emergent words;

[0009] A decision module calculates the confidence weight of the landscape element according to the intensity value and the life cycle; constructs a reinforcement learning framework, and generates an adjustment plan for the landscape element by combining the confidence weight and the sentiment label vector;

[0010] The execution module arranges the adjustable landscape components in the landscape space and adjusts the landscape components according to the adjustment scheme;

[0011] The closed-loop feedback module monitors the changes in the user's emotional state in real time after execution, and optimizes the confidence weight and reinforcement learning framework based on the changes in the emotional state.

[0012] Furthermore, the user data includes the user's heart rate data, skin electrical data, and emotion vector data;

[0013] Using wearable physiological sensors, the user's heart rate data and skin electrical data are collected in real time;

[0014] By configuring the camera network, the user's facial image is captured at a specific frame rate;

[0015] By adopting a distributed strategy to deploy environmental monitoring sensors in the landscape space, including illuminance meters, temperature and humidity sensors, and sound level meters, environmental data including light data, temperature and humidity data, and noise data can be obtained in real time.

[0016] Furthermore, the facial image is input into a preset convolutional neural network model to obtain emotion vector data, where the convolutional neural network model includes an input layer, a backbone layer, a mapping layer, and an output layer;

[0017] The input layer inputs the facial image;

[0018] The backbone layer includes shallow convolution and deep convolution:

[0019] The shallow convolution extracts the edge texture of the facial image to obtain semantic features, and the deep convolution aggregates the semantic features to generate the basic feature tensor ;

[0020] The mapping layer transforms the base feature tensor Projected to the latent space, through the nonlinear transformation function Generate emotion prototype vector :

[0021]

[0022] in, is the preset weight matrix, To vectorize the basic feature tensor, The bias term preset for the model;

[0023] The output layer transforms the emotion prototype vector into Perform illumination domain adaptation calibration to obtain the emotion vector P and output it. The formula is as follows:

[0024]

[0025] in, and are the mean and standard deviation of the emotion prototype vector, and Scaling parameters related to preset lighting conditions;

[0026] After obtaining the emotion vector P, destroying the facial image;

[0027] The user authorization mechanism is implemented by displaying the data to be collected to the user before collecting the user data, and the user can choose whether to authorize the data to be collected.

[0028] Furthermore, differential privacy technology is combined with Gaussian noise to anonymize the user data and generate an irreversible encrypted feature vector. The formula is as follows:

[0029]

[0030] in, is the encrypted feature vector, and and is Gaussian noise, For heart rate data, The data is the electrical skin data.

[0031] Furthermore, the steps for obtaining the emotion label vector are as follows:

[0032] 1) Set the environment vector ;

[0033] 2) The encrypted feature vector With the environment vector Input into a preset emotion quantification model, wherein the emotion quantification model includes an input layer, a fusion layer and an output layer;

[0034] The encrypted feature vector With the environment vector Input to the input layer;

[0035] The fusion layer fuses dense feature vectors and environmental vector :

[0036]

[0037] in, is the sentiment label vector, is the preset weight matrix, is the bias vector;

[0038] The output layer outputs the sentiment label vector .

[0039] Furthermore, the calculation formula of the strength value R of the emergent word is as follows:

[0040] R

[0041] in, is the frequency of the emergent word in its life cycle, For emergent words in non-life cycle

[0042] The frequency of occurrence.

[0043] Furthermore, the calculation formula of the confidence weight is as follows:

[0044]

[0045] in, is the confidence weight of the a-th landscape element, is the preset time decay constant, For the life cycle, is an exponential function.

[0046] Furthermore, the reinforcement learning framework is constructed based on the Q-learning algorithm, including a state space, an action space, a Q function, a policy function, and an online learning mechanism. The confidence weight and the sentiment label vector are combined to generate a landscape element adjustment scheme. The landscape elements include lighting elements, water features, and plant elements. The specific steps are as follows:

[0047] 1) The state space integrates the confidence weight into the sentiment label vector to obtain :

[0048]

[0049] in, and and are pleasure, stress, and concentration, respectively, and T is the transposition operator;

[0050] 2) The action space constructs an action set based on landscape elements:

[0051]

[0052] in, Indicates lighting adjustment, Indicates waterscape adjustment, Indicates plant regulation;

[0053] 3) The Q function is used to evaluate the action value, evaluating the specific emotion label vector The expected cumulative reward of performing action a is:

[0054]

[0055] in, is the parameter vector corresponding to the preset action a, The value is represented in the state The expected cumulative reward of performing action a under

[0056] 4) The policy function is used to select actions:

[0057]

[0058] in, is the preset exploration rate, To adjust the plan, To select the action that can make The optimal action with the largest function value;

[0059] 5) The online learning mechanism is triggered in the closed-loop feedback module.

[0060] Furthermore, the landscape components include lighting devices with adjustable color temperature, waterscape equipment with controllable water flow intensity, and modular plant carriers with movable positions;

[0061] The landscape components are adjusted based on the adjustment scheme.

[0062] Furthermore, after the landscape component is adjusted, the data collection and data processing operations are re-executed to generate a new emotion label vector ;

[0063] Based on the sentiment label vector Calculate the updated reward value r, the formula is as follows:

[0064]

[0065] in, and and They are the adjusted pleasure, pressure and concentration respectively. and and is the preset weight coefficient;

[0066] Based on the change of emotion Optimize the confidence weight, the formula is as follows:

[0067]

[0068] in, is the optimized confidence weight, is the confidence weight of the original landscape element, is the preset attenuation factor;

[0069] Reintegrate through reinforcement learning framework and get ;

[0070] Based on the updated reward value r and Perform online learning mechanism updates in reinforcement learning frameworks :

[0071]

[0072] in, is the preset learning rate, is the preset discount factor, is the reward value, For the next state The maximum Q value among all possible actions a′ is, For the new state s′, the executed action Q value.

[0073] Compared with the prior art, the advantages of the present invention are:

[0074] 1. This invention breaks through the limitations of traditional environmental systems that rely solely on physical parameters or one-way user data. By integrating user physiological indicators, real-time emotion vectors, and multi-dimensional environmental data, it constructs a quantitative model of emotion label vectors. It also introduces differential privacy technology to ensure data anonymization. Combined with a user authorization mechanism, it achieves multi-dimensional and accurate perception of the user's emotional state while ensuring privacy. This multimodal data collaboration mechanism significantly improves the comprehensiveness of the analysis of the relationship between the environment and user emotions.

[0075] 2. This invention innovatively incorporates academic research dynamics into the decision-making system. Using a emergent word strength detection algorithm, it extracts cutting-edge knowledge from the field and generates confidence weights for landscape elements. Based on this, it constructs a reinforcement learning framework, integrating the academic weights and real-time sentiment tag vectors into a Q-learning model to dynamically generate landscape element adjustment plans. This architecture avoids the cognitive lag of purely data-driven models while overcoming the mechanical nature of static rule bases, enabling decision-making to combine the foresight of domain knowledge with the adaptability of specific scenarios.

[0076] 3. The present invention uses a closed-loop feedback module to track changes in users' emotional states after landscape adjustments in real time. Based on the emotional changes, the confidence weights and reinforcement learning parameters are dynamically optimized. This mechanism transforms the execution effect into a reward function, driving the online update strategy of the Q-learning model to achieve a complete closed loop. This self-iterative capability enables the system to continuously adapt to individual differences and environmental changes, breaking through the bottleneck of traditional system adjustment strategies that are rigid and lack evolutionary capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0078] Figure 1 Schematic diagram of the system workflow of the present invention;

[0079] Figure 2 Schematic diagram of the flow of the regulation scheme of the present invention;

[0080] Figure 3 Schematic diagram of the closed-loop feedback module of the present invention. DETAILED DESCRIPTION

[0081] To achieve the above objectives, the present invention is implemented through the following technical solutions. The present invention provides a landscape space intelligent construction system based on emotional data. Figures 1 to 3 , the system comprises:

[0082] The data collection module collects emotional data including user data and environmental data in real time, and implements the user authorization mechanism when collecting user data;

[0083] User data includes the user's heart rate data, skin electrical data, and emotion vector data;

[0084] Environmental data refers to the environmental data of the landscape space;

[0085] Use wearable physiological sensors (such as wrist heart rate monitors and fingertip skin conductance sensors) to collect user heart rate (HR) data and skin conductance (GSR) data in real time;

[0086] By configuring a camera network, the user's facial image is captured at a frame rate of 30fps, and the facial image is input into a preset convolutional neural network model to obtain emotion vector data;

[0087] The convolutional neural network model includes input layer, backbone layer, mapping layer and output layer;

[0088] The input layer inputs the facial image;

[0089] The backbone layer includes shallow convolution and deep convolution:

[0090] Shallow convolution extracts edge textures (such as eye corner lines and mouth corner contours) to obtain semantic features;

[0091] Deep convolution aggregates semantic features (such as eyelid opening and closing, nasolabial groove displacement) to generate basic feature tensors ;

[0092] The mapping layer transforms the basic feature tensor Projected to the latent space, through the nonlinear transformation function Generate emotion prototype vector :

[0093]

[0094] in, is the preset weight matrix, To vectorize the basic feature tensor, The bias term preset for the model;

[0095] The output layer transforms the emotion prototype vector into Perform illumination domain adaptation calibration:

[0096]

[0097] in, and are the mean and standard deviation of the emotion prototype vector, and Scaling parameters related to preset lighting conditions;

[0098] For example, a user is in an indoor landscape space with weak ambient light. The system captures the user's facial image through the camera and processes it through the backbone layer and mapping layer of the convolutional neural network model to obtain the emotion prototype vector. , preset scaling parameters under this lighting condition .

[0099] Calculate the mean and standard deviation :μ= (0.8 + (-0.2) + 0.5) / 3 = 0.3667, σ=√[ ( (0.8-0.3667)² + (-0.2-0.3667)² + (0.5-0.3667)² ) / 3 ] = 0.400

[0100] Perform illumination domain adaptation calibration, the formula is :

[0101] Happiness:

[0102] Pressure value:

[0103] Focus:

[0104] The final output emotion vector .

[0105] Output standardized 128-dimensional emotion feature vector , each dimension of which corresponds to a specific emotion (such as pleasure, surprise, disgust, etc.);

[0106] Image data is destroyed immediately after processing is completed.

[0107] Environmental monitoring sensors, including illuminance meters (range 0-100klx), temperature and humidity sensors (accuracy ±0.5°C / ±3%RH), and sound level meters (A-weighted mode), are deployed in a distributed manner to obtain real-time environmental data including light, temperature, humidity, and noise.

[0108] Before data collection, a list of data categories that need to be collected is displayed to the user, and the user can freely choose which data to authorize.

[0109] The data processing module anonymizes the user data to generate an encrypted feature vector; fuses the encrypted feature vector with the environmental data to generate an emotional label vector; obtains literature in the field of landscape emotion, extracts the emergent words and life cycles in the literature based on the emergent detection algorithm, and calculates the strength value of the emergent words;

[0110] Use differential privacy technology to add Gaussian noise to anonymize user data and generate irreversible encrypted feature vectors:

[0111]

[0112] in, is the encrypted feature vector, and and is Gaussian noise with mean 0 and variance The normal distribution of is the preset data sensitivity, is the preset privacy budget, For heart rate data, For skin electrical data;

[0113] Anonymized encrypted feature vector , cannot be associated with a specific natural person.

[0114] Based on the preset sentiment quantification model, dense feature vector is integrated Combined with the environmental data, the emotion label vector is obtained. The specific steps are as follows:

[0115] 1) Set the environment vector :

[0116]

[0117] Where L is the light intensity, is temperature, I is humidity, is the noise decibel value;

[0118] 2) The dense feature vector and environmental vector Input to the emotion quantification model, which includes input layer, fusion layer and output layer;

[0119] The fused dense feature vector and environmental vector Input to the input layer;

[0120] The fusion layer fuses dense feature vectors and environmental vector :

[0121]

[0122] in, is the emotion label vector with a dimension of 3, representing pleasure, pressure and concentration. is the preset weight matrix, The bias vector is trained and calibrated by the sentiment quantization model to adjust the baseline value of the output label;

[0123] The output layer outputs the sentiment label vector ;

[0124] In this embodiment, the official API interface of Web of Science is used to obtain literature on landscape emotionalization.

[0125] Based on the BDA algorithm in CiteSpace, we extract the emergent words and life cycle in the document and calculate the strength value R of the emergent words. The formula is as follows:

[0126] R

[0127] in, is the frequency of the emergent word in its life cycle, For emergent words in non-life cycle

[0128] The frequency of occurrence.

[0129] The emergent word represents the current research hotspot, the intensity value reflects the importance of the word in the literature set, and the life cycle represents the time span from the jump to the return of the frequency of the emergent word to the normal state.

[0130] A decision module calculates the confidence weight of the landscape element according to the intensity value and the life cycle; constructs a reinforcement learning framework, and generates an adjustment plan for the landscape element by combining the confidence weight and the sentiment label vector;

[0131] Based on the intensity value and life cycle, the confidence weight of the landscape element is calculated using the following formula:

[0132]

[0133] in, is the confidence weight of the a-th landscape element, is the preset time decay constant, For the life cycle, is an exponential function;

[0134] In this embodiment, the landscape elements include lighting elements, water features elements, and plant elements;

[0135] A reinforcement learning framework is constructed based on the Q-learning algorithm. The reinforcement learning framework includes a state space, an action space, a Q function, a policy function, and an online learning mechanism. The confidence weight and the sentiment label vector are combined to generate a landscape element adjustment plan. The specific steps are as follows:

[0136] 1) State space integrates confidence weights into sentiment label vectors :

[0137]

[0138] in, and and are happiness, stress, and concentration, respectively. T represents the transpose operator, which converts a row vector into a column vector (or vice versa).

[0139] 2) Action space: Construct an action set a based on landscape elements, including:

[0140] Lighting adjustment : {-1 (lower color temperature), 0 (keep), +1 (increase color temperature)}

[0141] Waterscape Adjustment : {-1 (decrease water flow strength), 0 (maintain), +1 (increase water flow strength)}

[0142] Plant Regulation :{(move left), ,

[0143] Therefore, the complete action set is a three-dimensional vector;

[0144] 3) Q function is used to evaluate the action value and evaluate the specific emotion label vector The expected cumulative reward of performing action a is:

[0145]

[0146] in, is the parameter vector corresponding to the preset action a, The value is represented in the state The expected cumulative reward of performing action a under

[0147] 4) Based on The greedy policy function selects actions to balance exploration and exploitation:

[0148]

[0149] in, is the preset exploration rate (typical value ), For adjustment schemes (e.g. It means increasing the lighting color temperature, reducing the water flow intensity, and maintaining the plant position. To select the action that can make The optimal action with the largest function value;

[0150] 5) Online learning mechanism, triggered in the closed-loop feedback module.

[0151] The execution module arranges the adjustable landscape components in the landscape space and adjusts the landscape components according to the adjustment scheme;

[0152] Landscape components include lighting devices with adjustable color temperature, waterscape equipment with controllable water flow intensity, and modular plant carriers with movable positions;

[0153] The landscape element adjustment scheme controls the adjustable landscape components. The specific steps are as follows:

[0154] 1) Set control benchmark parameters, including:

[0155] is the current color temperature of the lighting device (unit: Kelvin / K), The current water flow intensity of the waterscape equipment (unit: percentage / %), is the current position coordinate of the plant carrier (unit: meter / m);

[0156] 2) Adjustment of lighting devices: Color temperature adjustment follows a dynamic step size mechanism, and the formula is as follows:

[0157]

[0158] in, is the target color temperature of the lighting, is the preset color temperature adjustment step, and The maximum and minimum color temperatures supported by the device (for example: ), is the limiting function, ensuring Not exceeding and ;

[0159] 3) Waterscape equipment control, water flow intensity is adjusted proportionally, the formula is as follows:

[0160]

[0161] in, is the target water flow intensity, Adjust the water flow intensity step for the preset value (for example: );

[0162] 4) Plant carrier control: Position movement is based on direction and step length. The formula is as follows:

[0163]

[0164] in, Plant carrier target location, is the preset carrier movement step, and For the carrier movement range (for example, track length constraint ).

[0165] The closed-loop feedback module monitors the changes in the user's emotional state in real time after execution, and optimizes the confidence weight and reinforcement learning framework based on the changes in the emotional state.

[0166] After adjusting the landscape components, re-execute the data collection and data processing operations to output a new emotion label vector ;

[0167] Based on the new emotional label vector Calculate the updated reward value r, the formula is as follows:

[0168]

[0169] in, and and They are the adjusted pleasure, pressure and concentration respectively. and and is the preset weight coefficient (default );

[0170] Reward design logic: Increased pleasure, increased concentration, and reduced stress are positive rewards.

[0171] Dynamic optimization of confidence weights;

[0172] Based on the change of emotion (Euclidean distance) Optimize confidence weights:

[0173]

[0174] in, is the optimized confidence weight, is the confidence weight of the original landscape element; is the emergent word strength value, is the preset attenuation factor (preset ), control the historical weight retention ratio, To optimize the mechanism, the more significant the emotional improvement and the higher the research interest, the greater the weight increase;

[0175] Reintegrate through reinforcement learning framework and :

[0176]

[0177] Use the updated reward value r and Perform online learning mechanism updates in reinforcement learning frameworks :

[0178]

[0179] in, is the preset learning rate (typical value ), is the preset discount factor (typical value ), is the reward value (provided by the closed-loop feedback module and calculated based on the change in emotional state), For the next state The maximum Q value among all possible actions a′ is, For the new state s′, the executed action The Q value of .

[0180] In summary, the present invention breaks through the limitations of a single data dimension, innovatively coordinates user physiological indicators, real-time emotion recognition and environmental parameters to construct a dynamic emotion label vector, and simultaneously realizes data collection anonymization through user authorization mechanism and differential privacy technology, thereby improving the accuracy of emotion perception while ensuring privacy security.

[0181] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The landscape space intelligent construction system based on emotional data is characterized by: include; The data collection module collects emotional data including user data and environmental data in real time, and implements the user authorization mechanism when collecting user data; The data processing module anonymizes the user data to generate an encrypted feature vector; fuses the encrypted feature vector with the environmental data to generate an emotional label vector; obtains literature in the field of landscape emotion, extracts the emergent words and life cycles in the literature based on the emergent detection algorithm, and calculates the strength value of the emergent words; A decision module calculates the confidence weight of the landscape element according to the intensity value and the life cycle; Constructing a reinforcement learning framework and generating a regulation scheme for landscape elements by combining the confidence weights and the sentiment label vectors; The execution module arranges the adjustable landscape components in the landscape space and adjusts the landscape components according to the adjustment scheme; The closed-loop feedback module monitors the changes in the user's emotional state in real time after execution, and optimizes the confidence weight and reinforcement learning framework based on the changes in the emotional state.

2. The system according to claim 1, wherein: The user data includes the user's heart rate data, skin electrical data, and emotion vector data; Using wearable physiological sensors, the user's heart rate data and skin electrical data are collected in real time; By configuring the camera network, the user's facial image is captured at a specific frame rate; By adopting a distributed strategy to deploy environmental monitoring sensors in the landscape space, including illuminance meters, temperature and humidity sensors, and sound level meters, environmental data including light data, temperature and humidity data, and noise data can be obtained in real time.

3. The system according to claim 1 and 2, characterized in that Inputting the facial image into a preset convolutional neural network model to obtain emotion vector data, the convolutional neural network model including an input layer, a backbone layer, a mapping layer and an output layer; The input layer inputs the facial image; The backbone layer includes shallow convolution and deep convolution: The shallow convolution extracts the edge texture of the facial image to obtain semantic features, and the deep convolution aggregates the semantic features to generate the basic feature tensor ; The mapping layer transforms the base feature tensor Projected to the latent space, through the nonlinear transformation function Generate emotion prototype vector : in, is the preset weight matrix, To vectorize the basic feature tensor, The bias term preset for the model; The output layer transforms the emotion prototype vector into Perform illumination domain adaptation calibration to obtain the emotion vector P and output it. The formula is as follows: in, and are the mean and standard deviation of the emotion prototype vector, and Scaling parameters related to preset lighting conditions; After obtaining the emotion vector P, destroying the facial image; The user authorization mechanism is implemented by displaying the data to be collected to the user before collecting the user data, and the user can choose whether to authorize the data to be collected.

4. System according to claims 2 and 3, characterized in that Differential privacy technology combined with Gaussian noise is used to anonymize the user data and generate an irreversible encrypted feature vector. The formula is as follows: in, is the encrypted feature vector, and and is Gaussian noise, For heart rate data, The data is the electrical skin data.

5. The system according to claim 4, characterized in that The steps for obtaining the emotion label vector are as follows: 1) Set the environment vector ; 2) The encrypted feature vector With the environment vector Input into a preset emotion quantification model, wherein the emotion quantification model includes an input layer, a fusion layer and an output layer; The encrypted feature vector With the environment vector Input to the input layer; The fusion layer fuses the encrypted feature vector With the environment vector : in, is the sentiment label vector, is the preset weight matrix, is the bias vector; The output layer outputs the emotion label vector .

6. The system according to claim 1, wherein: The calculation formula of the strength value R of the emergent word is as follows: R in, is the frequency of the emergent word in its life cycle, For emergent words in non-life cycle The frequency of occurrence.

7. The system according to claims 1 and 6, characterized in that The calculation formula of the confidence weight is as follows: in, is the confidence weight of the a-th landscape element, is the preset time decay constant, For the life cycle, is an exponential function.

8. The system according to claims 1 and 7, characterized in that The reinforcement learning framework is built based on the Q-learning algorithm, including state space, action space, Q function, policy function and online learning mechanism. It combines confidence weights and sentiment label vectors to generate landscape element adjustment schemes. The landscape elements include lighting elements, water features elements and plant elements. The specific steps are as follows: 1) The state space integrates the confidence weight into the sentiment label vector to obtain : in, and and are pleasure, stress, and concentration, respectively, and T is the transposition operator; 2) The action space constructs an action set based on the landscape elements: in, Indicates lighting adjustment, Indicates waterscape adjustment, Indicates plant regulation; 3) The Q function is used to evaluate the action value, evaluating the specific emotion label vector The expected cumulative reward of performing action a is: in, is the parameter vector corresponding to the preset action a, The value is represented in the state The expected cumulative reward of performing action a under 4) The policy function is used to select actions: in, is the preset exploration rate, To adjust the plan, To select the action that can make The optimal action with the largest function value; 5) The online learning mechanism is triggered in the closed-loop feedback module.

9. The system according to claim 1, wherein: The landscape components include lighting devices with adjustable color temperature, waterscape equipment with controllable water flow intensity, and modular plant carriers with movable positions; The landscape components are adjusted based on the adjustment scheme.

10. The system according to claim 1, wherein: After the landscape component is adjusted, the data collection and data processing operations are re-executed to generate a new emotion label vector ; Based on the sentiment label vector Calculate the updated reward value r, the formula is as follows: in, and and They are the adjusted pleasure, pressure and concentration respectively. and and is the preset weight coefficient; Based on the change of emotion Optimize the confidence weight, the formula is as follows: in, is the optimized confidence weight, is the confidence weight of the original landscape element, is the preset attenuation factor; Reintegrate through reinforcement learning framework and get ; Based on the updated reward value r and Perform online learning mechanism updates in reinforcement learning frameworks : in, is the preset learning rate, is the preset discount factor, is the reward value, For the next state The maximum Q value among all possible actions a′ is, For the new state s′, the executed action Q value.

Images