Air conditioner recommendation method and device based on lstm model and air conditioner

Abstract

The application relates to the technical field of information recommendation, and discloses an air conditioner recommendation method and device based on an LSTM model and an air conditioner. The method comprises the following steps: collecting user operation data, environment data and third-party weather data, and pre-processing the collected data; training a double-layer LSTM model by using the pre-processed data; performing air conditioner setting prediction according to current time window data and the trained double-layer LSTM model; and displaying recommendation information according to the comprehensive confidence of the prediction result. In this way, more accurate and personalized air conditioner setting suggestions can be provided for users, the intelligent level of the recommendation system and the user experience are significantly improved, and the demand of users for accurate and personalized recommendation is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information recommendation, for example to an air conditioner recommendation method and device based on an LSTM model and an air conditioner. BACKGROUND

[0002] With the continuous progress of air conditioning technology and the popularity of the smart home concept, air conditioning equipment has developed from traditional single-function models to intelligent air conditioners with intelligent control and personalized services. At the same time, the rapid development of artificial intelligence large model technology has promoted the intelligent process in the field of smart home, making the air conditioner recommendation system an important part of improving user experience. However, the existing air conditioner recommendation system still has deficiencies in meeting the dynamic changes of user needs, and it is difficult to fully adapt to complex use scenarios.

[0003] At the present stage, in order to fully adapt to complex application scenarios, the related technology discloses an air conditioner recommendation method based on rules, which collects user operation data, extracts recent operation features, and combines environmental temperature to make threshold judgment to generate recommendation parameters. This method improves the intelligent level of the recommendation system to some extent, so that the air conditioner can make basic recommendations according to the user's recent operation and environmental temperature.

[0004] In the process of implementing the embodiments of the present disclosure, it is found that at least the following problems exist in the related art:

[0005] Although the related technology can improve the intelligent level of the recommendation system to some extent, the user's needs and habits will change over time, season, etc. This method is difficult to capture these dynamic changes, resulting in low recommendation accuracy and failing to meet the user's demand for accurate personalized recommendation.

[0006] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present application, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0007] In order to have a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not a general review, nor does it determine the key / important elements or describe the scope of protection of these embodiments, but as a prelude to the detailed description below.

[0008] The embodiments of the present disclosure provide an air conditioner recommendation method and device based on an LSTM model and an air conditioner to meet the user's demand for accurate personalized recommendation.

[0009] In some embodiments, the air conditioner recommendation method based on the LSTM model comprises: collecting user operation data, environment data and third-party weather data, and preprocessing the collected data; training a double-layer LSTM model using the preprocessed data; performing air conditioner setting prediction according to current time window data and the trained double-layer LSTM model; and displaying recommendation information according to the comprehensive confidence of the prediction result.

[0010] In some embodiments, the air conditioner recommendation device based on the LSTM model comprises: a preprocessing module configured to collect user operation data, environment data and third-party weather data, and preprocess the collected data; a training module configured to train a double-layer LSTM model using the preprocessed data; a prediction module configured to perform air conditioner setting prediction according to current time window data and the trained double-layer LSTM model; and a display module configured to display recommendation information according to the comprehensive confidence of the prediction result; wherein the comprehensive confidence is obtained by weighting the mode prediction probability, the temperature prediction standard deviation and the wind speed prediction probability.

[0011] In some embodiments, the air conditioner recommendation device based on the LSTM model comprises: a processor and a memory storing program instructions, the processor being configured to execute the air conditioner recommendation method based on the LSTM model when running the program instructions.

[0012] In some embodiments, the air conditioner comprises: an air conditioner body; and the air conditioner recommendation device based on the LSTM model as described above, which is installed on the air conditioner body.

[0013] The embodiments of the present disclosure provide an air conditioner recommendation method, device and air conditioner based on an LSTM model, which can achieve the following technical effects:

[0014] The present scheme collects user operation data, environment data and third-party weather data, and preprocesses the data, thereby providing more comprehensive and multi-dimensional input information for the model. The powerful time series modeling capability of the double-layer LSTM model is utilized to deeply mine the dynamic change rules of user behavior patterns and accurately capture the dynamic evolution of user demand with factors such as time and season. The air conditioner setting prediction is then performed according to the current time window data, and the recommendation information is displayed according to the comprehensive confidence of the prediction result. In this way, more accurate and personalized air conditioner setting suggestions can be provided for the user, the intelligent level of the recommendation system and the user experience are significantly improved, and the demand of the user for accurate and personalized recommendation is met.

[0015] The general description above and the following description below are exemplary and explanatory only and are not intended to be limiting. BRIEF DESCRIPTION OF DRAWINGS

[0016] One or more embodiments are illustrated by way of example in the figures that are not intended to be limiting of the present embodiments. Like numbers refer to like elements throughout the drawings, which are not necessarily to scale, and in which:

[0017] Figure 1 is a schematic diagram of an air conditioner recommendation method based on an LSTM model provided by an embodiment of the present disclosure;

[0018] Figure 2 is a schematic diagram of a method for data preprocessing provided by an embodiment of the present disclosure;

[0019] Figure 3 is a schematic diagram of a method for model training provided by an embodiment of the present disclosure;

[0020] Figure 4 is a schematic diagram of another air conditioner recommendation method based on an LSTM model provided by an embodiment of the present disclosure;

[0021] Figure 5 is a schematic diagram of an air conditioner recommendation device based on an LSTM model provided by an embodiment of the present disclosure;

[0022] Figure 6 is a schematic diagram of a structure of an air conditioner provided by an embodiment of the present disclosure.

[0023] Reference signs:

[0024] 20: processor; 21: memory; 22: communication interface; 23: bus; 100: air conditioner; 200: air conditioner recommendation device based on an LSTM model. DETAILED DESCRIPTION

[0025] In order to enable a more detailed understanding of the features and technical content of the embodiments of the present disclosure, the implementation of the embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings, which are only used for reference and do not limit the embodiments of the present disclosure. In the following technical description, in order to facilitate explanation, a plurality of details are provided to provide a full understanding of the disclosed embodiments. However, one or more embodiments can still be implemented without these details. In other cases, well-known structures and devices can be simplified to facilitate the drawings.

[0026] The terms "first", "second", and the like in the specification and claims of the embodiments of the present disclosure and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion.

[0027] The term "plurality" means two or more, unless otherwise specified.

[0028] In the embodiments of the present disclosure, the character " / " represents a "or" relationship between the preceding and following objects. For example, A / B represents: A or B.

[0029] The term "and / or" is a description of the association relationship between objects, which means that there can be three relationships. For example, A and / or B means: A or B, or, A and B, three relationships.

[0030] The term "corresponding" can refer to an association relationship or a binding relationship. A and B correspond to each other means that A and B have an association relationship or a binding relationship.

[0031] Figure 1 FIG. 1 is a schematic diagram of an air conditioner recommendation method based on an LSTM model provided by an embodiment of the present disclosure. In combination with FIG. 1, Figure 1 The embodiment of the present disclosure provides an air conditioner recommendation method based on an LSTM model, which comprises the following steps:

[0032] S11, the air conditioner collects user operation data, environment data and third-party weather data, and pre-processes the collected data.

[0033] S12, the air conditioner trains a double-layer LSTM model using the pre-processed data.

[0034] S13, the air conditioner performs air conditioner setting prediction according to the current time window data and the trained double-layer LSTM model.

[0035] S14, the air conditioner displays recommendation information according to the comprehensive confidence of the prediction result.

[0036] In the present scheme, the air conditioner can collect user operation data, environmental data and third-party weather data, and preprocess the collected data. Specifically, the user operation data includes various operation behaviors of the user on the air conditioner through the air conditioner remote controller or mobile phone application, such as air conditioner mode switching (cooling, heating, dehumidifying, air supply, etc.), temperature adjustment value or adjustment range, wind speed adjustment range, deflector angle change and on-off time record, etc. These operation data are collected directly through the built-in controller of the air conditioner or extracted from the user's mobile phone APP (Application) operation log, and the accurate time stamp and parameter change of each operation are recorded, which can reflect the user's specific use preference and habit of the air conditioner under different situations. The environmental data are collected by the built-in sensor of the air conditioner in real time, including indoor temperature, humidity, light intensity, thermal comfort index and other parameters. These data are collected by the built-in sensor of the air conditioner in real time, with a sampling frequency of 1 time per minute, which can provide the specific conditions of the current indoor environment for the recommendation system, so as to better understand the user's demand under a specific environment. The third-party weather data are obtained through the weather API (Application Programming Interface) interface and aligned with the user operation time. Mainly including real-time temperature, humidity, wind speed and other information, these data can help the system understand the potential influence of the external environment on the indoor temperature and user comfort, and further accurately predict the user's demand for air conditioner settings. This multi-source data collection method is more comprehensive than the traditional method, which can provide a richer information basis for subsequent model training.

[0037] Further, after collecting these data, the air conditioner will preprocess the collected data. First, the air conditioner uses the sliding window method to detect outliers to filter out continuous repeated operations with operation intervals less than the set time length. This method can effectively remove abnormal data caused by misoperation or system failure, ensuring the quality and accuracy of the data. Next, the air conditioner fills in the missing data to obtain the complete data after filling. The filling of missing data is an important link in data preprocessing. As an example, for time series data missing, a prediction method based on a bidirectional LSTM (Long Short-Term Memory) model can be used for filling, which can accurately predict the value of missing data according to the existing data trend and pattern; for environmental data missing, an XGBoost (eXtreme Gradient Boosting) model is used for prediction and filling, which has high accuracy and stability in processing environmental data, and can effectively improve the integrity of the data. Further, the air conditioner can extract time features, environmental features, and user operation features from the complete data. Among them, time features include but are not limited to hour features, week features, and season features extracted from user operation timestamps and environmental data sampling times. Environmental features include but are not limited to temperature, humidity, light intensity, and THI (Temperature Humidity Index), as well as the difference features between outdoor weather data and indoor environment obtained through weather API, such as indoor and outdoor temperature difference and rainfall warning state. User operation features include but are not limited to temperature adjustment amplitude, mode switching information, wind speed adjustment parameters, and deflector adjustment parameters parsed from remote control and mobile phone APP operation logs. Further, the air conditioner can standardize each feature extracted respectively. In this way, by standardizing these features, the differences in dimensions and orders of magnitude between different features can be effectively eliminated, making the data more suitable for model training and prediction. Finally, the air conditioner divides the standardized features into dynamic windows to obtain preprocessed data. Specifically, dynamic window division is to automatically adjust the data window length according to the season. For example, a 7-day data window is used in summer, a 14-day data window is used in winter, and a 10-day data window is used in transition seasons. This dynamic window division method can better adapt to the influence of seasonal changes on user air conditioner usage habits and improve the prediction accuracy of the model.

[0038] With this scheme, by comprehensively collecting user operation data, environmental data, and third-party weather data, and using advanced preprocessing methods, the user's usage habits and demand changes in different situations can be more accurately captured, providing a higher quality data foundation for subsequent model training and prediction.

[0039] Further, the air conditioner can divide the pre-processed data into a training set and a test set. The training set is used for the training process of the model, while the test set is used to verify the performance of the model after training to ensure that the model not only performs well on the training data, but also accurately predicts on new, unseen data.

[0040] Specifically, the air conditioner can input the training set data into a double-layer LSTM model. Here, the double-layer LSTM model is designed to better capture long-term dependencies in the data. Compared with a single-layer LSTM model, the double-layer structure can more deeply understand and process complex patterns in time series data. Specifically, during the training process, an optimizer with learning rate decay and early stopping mechanism can be used for iterative training. Learning rate decay is a strategy to adjust the learning rate, which gradually decreases as the training process progresses. This helps the model to fine-tune the parameters in the later training period, thereby improving the accuracy of the model. The early stopping mechanism is a technique to prevent overfitting. When the performance of the model on the validation set no longer improves after a certain number of iterations, the training process is stopped early, thereby avoiding overfitting of the model to the training data and improving the generalization ability of the model.

[0041] Further, when the training set prediction index reaches the preset accuracy threshold, the test set is used to verify the performance of the model. Understandably, the preset accuracy threshold is set according to the specific application scenario and requirements. When the model's prediction index on the training set reaches this threshold, it means that the model has sufficient learning ability. Then, the test set is used to verify the model to evaluate the model's performance on unseen data. If the model's performance on the test set also meets expectations, it can be considered that the model has been successfully trained. And in the case of passing the verification, the trained double-layer LSTM model is output. It should be noted that the output model contains all the parameters and structural information learned during the training process, which can be used to predict new user operation data, environmental data and third-party weather data, thereby providing personalized air conditioner setting recommendations. With this scheme, the double-layer LSTM model adopted combines advanced optimization strategies and verification mechanisms, which can more accurately capture long-term patterns of user behavior and environmental changes, thereby providing more accurate recommendations. In addition, through strict training and verification process, the model's generalization ability has been significantly improved, which can better adapt to different users and environmental conditions, providing more reliable and personalized air conditioner recommendation services.

[0042] In this solution, the current time window data refers to the user operation data, environmental data, and third-party weather data collected in the current period. These data reflect the user's air conditioner usage habits and preferences, as well as the current indoor environmental conditions and external weather conditions. Specifically, the time window data includes user operations such as switching air conditioner modes, adjusting temperatures, adjusting wind speeds, and changing deflector angles within the last period of time (e.g., within 24 hours); it also includes current indoor temperature, humidity, light intensity, and other environmental data, as well as real-time external temperature, humidity, wind speed, and other third-party weather data. These data are collected in real time through the air conditioner's sensors and network interface and are preprocessed as inputs to the model. In this way, the air conditioner inputs these current time window data into the trained double-layer LSTM model, which predicts the air conditioner settings based on the user behavior patterns and environmental characteristics learned during the training phase. The prediction results include air conditioner mode, target temperature, wind speed, and deflector angle parameters. With this solution, the double-layer LSTM model can capture changes in user usage habits over different time periods and provide more accurate and personalized air conditioner setting predictions by combining current environmental and weather conditions. This prediction not only improves user comfort but also reduces energy consumption by optimizing air conditioner usage parameters. In addition, the model of this embodiment can dynamically adjust the prediction results based on real-time data, with stronger adaptability and flexibility, better meeting user needs in different situations.

[0043] In this solution, the comprehensive confidence is an index calculated by multiple factors, which reflects the model's confidence in the current prediction results. Specifically, the multiple factors include mode prediction confidence, temperature prediction confidence, wind speed prediction confidence, and deflector angle prediction confidence. The confidence of each prediction result is calculated by a specific algorithm, for example, the mode prediction confidence is derived from the Softmax probability output, while the temperature prediction confidence depends on the standard deviation of the predicted value.

[0044] It should be noted that when calculating the comprehensive confidence, different weights are assigned according to different prediction types, and these weights are set based on a comprehensive consideration of the importance of user usage habits and air conditioner operations. In one example, the weight of mode prediction can be set to 0.4, the weight of temperature prediction to 0.3, the weight of wind speed prediction to 0.2, and the weight of deflector angle prediction to 0.1. In this way, the confidence of each prediction result is combined by weighted averaging to obtain a comprehensive confidence value. This value is usually between 0 and 1, and the closer to 1, the more confident the model is about the prediction results.

[0045] Further, the air conditioner can adopt different recommendation strategies according to the value of the comprehensive confidence. As an example, if the comprehensive confidence is higher than a first preset threshold, it means that the model is very confident about the prediction result, at this time the air conditioner will directly display the model prediction value, and the user can directly accept the recommendation without additional adjustment. If the comprehensive confidence is between the second preset threshold and the first preset threshold, it means that the model has a certain confidence about the prediction result, but there is a certain uncertainty. In this case, the air conditioner will display the model prediction value and suggest parameter adjustment. If the comprehensive confidence is lower than the second preset threshold, it means that the model is not confident enough about the prediction result, at this time the air conditioner will display the historical average value and suggest that the user manually set the air conditioner parameters to ensure the user's comfort and satisfaction. With this scheme, through the evaluation of the comprehensive confidence, not only more accurate recommendations can be provided, but also different operation suggestions can be given to the user according to the confidence, thereby improving the user's use experience and the trust of the recommendation system.

[0046] By adopting the air conditioner recommendation method based on the LSTM model provided in the embodiments of the present disclosure, user operation data, environment data and third-party weather data are collected, and the data is preprocessed, thereby providing more comprehensive and multi-dimensional input information for the model. The powerful time series modeling capability of the double-layer LSTM model is utilized to deeply mine the dynamic change law of the user behavior pattern, and accurately capture the dynamic evolution of the user demand with time, season and other factors. Thus, the air conditioner setting prediction is performed according to the current time window data, and the recommended information is displayed according to the comprehensive confidence of the prediction result. In this way, more accurate and personalized air conditioner setting suggestions can be provided for the user, the intelligent level of the recommendation system and the user experience are significantly improved, and the demand of the user for accurate and personalized recommendation is met.

[0047] Figure 2 is a method schematic diagram for data preprocessing provided in the embodiments of the present disclosure. As shown in Figure 2 , optionally, in the S11 process, the air conditioner pre-processes the collected data, including:

[0048] S21, the air conditioner detects abnormal values by using a sliding window method to filter continuous repeated operations with an operation interval less than a set time length.

[0049] S22, the air conditioner fills in the missing data to obtain complete data after filling.

[0050] S23, the air conditioner extracts time features, environment features and user operation features from the complete data, and respectively performs standardization processing on the extracted features.

[0051] S24, the air conditioner divides the standardized processed features into dynamic windows to obtain preprocessed data.

[0052] In this solution, the air conditioner can use a sliding window method to detect outliers, with a window size set to 5. This method identifies unreasonable fluctuations or outliers by analyzing consecutive data points. As an example, set the time length to 30 seconds. In this way, during the detection process, the air conditioner pays special attention to consecutive repeated operations with an operation interval less than 30 seconds, which may be caused by misoperation or system failure and is generally considered invalid. In this way, by filtering these invalid operations, the quality and reliability of the data can be improved, thereby providing more accurate data basis for subsequent model training.

[0053] In addition, in addition to using the sliding window method to detect outliers, statistical-based methods can also be considered to identify and filter outliers. For example, the Z-score method is used to calculate the standardized distance of each data point from the mean to identify outliers. The IQR (Interquartile Range) method is used to determine the range of outliers using the quartiles of the data. These alternatives can be used as alternatives to the sliding window method to adapt to different data characteristics and application scenarios, further improving the flexibility and accuracy of data preprocessing.

[0054] In this solution, for the processing of missing data, two different methods can be used, depending on the type of data. For the missing part of the time series data, a bidirectional LSTM-based prediction method can be used for filling. The bidirectional LSTM can consider both past and future information of the data, thereby more accurately predicting the missing values. For the missing part of the environmental data, an XGBoost model is used for prediction and filling. XGBoost is a high-efficiency machine learning algorithm that can handle complex nonlinear relationships and has good robustness to missing values. Through the combination of these two methods, the gaps in the data can be effectively filled, obtaining a complete and accurate data set.

[0055] In an optimized solution, for the missing of time series data, linear interpolation or time series decomposition method can be used for filling; for the missing of environmental data, K-nearest neighbor (K-NN) interpolation or physical model-based prediction method can be used. These alternatives can be selected according to the specific data characteristics and application scenarios to improve the integrity and accuracy of the data, thereby providing a more reliable data basis for model training.

[0056] In this solution, time features, environment features, and user operation features can also be extracted from the complete data. Specifically, time features include but are not limited to hour features (0~23), day features (0~6), and season features (0~3). These features can reflect the user's usage habits and preferences at different time scales. Specifically, the hour feature (0~23) represents the number of hours in a day, ranging from 0 to 23, which is used to capture the user's air conditioner usage habits at different time periods in a day. For example, the user may be more inclined to set the air conditioner temperature lower in the evening (such as 20~23) and higher in the daytime (such as 10~15). By dividing a day into 24 hours, the model can learn the user's preferences at each specific hour, thus making more accurate predictions. The day feature (0~6) represents the number of days in a week, ranging from 0 to 6, where 0 usually represents Sunday, 1 represents Monday, and so on, until 6 represents Saturday. This feature is used to capture the user's usage habits on different days of the week. For example, the user may have different air conditioner usage patterns on weekdays (Monday to Friday) and weekends (Saturday and Sunday). By dividing a week into 7 days, the model can learn the user's preferences at each specific day, thus making more accurate predictions. The season feature (0~3) represents the season of the year, ranging from 0 to 3, where 0 represents winter, 1 represents spring, 2 represents summer, and 3 represents autumn. This feature is used to capture the user's usage habits in different seasons. For example, the user may use the cooling mode more frequently in summer and the heating mode more frequently in winter. By dividing a year into 4 seasons, the model can learn the user's preferences in each season, thus making more accurate predictions. Environment features include but are not limited to temperature, humidity, light intensity, and THI (Temperature Humidity Index), as well as the difference between outdoor weather data and indoor environment obtained through weather API, which can describe the specific conditions of the current environment and provide important context information for the model. User operation feature data includes but is not limited to temperature adjustment amplitude, mode switching frequency, and wind speed adjustment frequency, which can reflect the user's operation habits and specific needs for air conditioner settings. By extracting these multi-dimensional features, the model can more comprehensively understand the user's usage patterns and environmental changes. With this solution, by setting time feature data, combined with environment feature data and user operation feature data, the model can comprehensively understand the user's usage patterns and environmental changes from multiple dimensions. This multi-dimensional feature extraction method provides rich information for the model, enabling it to more accurately predict the user's air conditioner setting needs in different situations, thus providing more personalized and intelligent recommendation services.

[0057] Further, to improve the training effect and prediction accuracy of the model, the air conditioner can perform standardization processing on the extracted features. Specifically, the standardization processing formula for temperature is T norm =(T-16) / 14, the standardization processing formula for wind speed is V norm =V / 3, and the standardization processing formula for the angle of the air deflector is A norm =A / 180. Here, T is the collected temperature value, V is the collected wind speed value, and A is the collected air deflector angle value. In this way, the standardization processing method can eliminate the differences in dimensions and orders of magnitude between different features, making the data more suitable for model training and prediction. By normalizing the data to the same scale, the model can more effectively learn and process patterns and relationships in the data.

[0058] Further, the air conditioner can perform dynamic window division on the data according to the season. This dynamic window division method can better adapt to the influence of seasonal changes on user air conditioner usage habits and improve the prediction accuracy of the model. By adjusting the length of the data window according to the season, the model can more accurately capture user usage patterns and preferences in different seasons, thereby providing more accurate personalized recommendation services.

[0059] Optionally, in the S22 process, the air conditioner fills in the missing data, including:

[0060] The air conditioner uses a recurrent neural network to predict and fill in the missing parts of the time series data, which is the timestamp log of user operation data.

[0061] The air conditioner uses a gradient boosting decision tree model to predict and fill in the missing parts of the environmental data, which is the observed value of indoor temperature, humidity, and light intensity.

[0062] The user operation data includes mode switching, temperature setting adjustment, wind speed and air deflector angle change.

[0063] In this scheme, for the missing parts of the time series data, the air conditioner uses a recurrent neural network to predict and fill in. The time series data here mainly refers to the timestamp log of user operation data, which records the specific time points of each user operation on the air conditioner, including mode switching, temperature setting adjustment, wind speed, and air deflector angle change. These operation records have accurate time markers and can reflect the user's specific settings and adjustments on the air conditioner at different time points.

[0064] To more accurately predict the missing values in the time series data, the air conditioner adopts a bidirectional LSTM model whose input features include not only the historical operation sequence but also the timestamp and environmental parameters. The historical operation sequence can provide the user's past behavior patterns, the timestamp accurately records the time points at which these operations occur, and the environmental parameters provide the historical operation environment for the model. By integrating these information, the bidirectional LSTM model can more accurately predict the missing time series data, thereby filling in the blanks in the data.

[0065] For the missing part of the environmental data, the air conditioner adopts a gradient boosting decision tree model for prediction and filling. Specifically, the environmental data mainly includes the observed values of indoor temperature, humidity, and light intensity, which reflect the specific conditions of the current indoor environment. The input features of the XGBoost model include weather data, adjacent time period environmental data mean, and season identifier. The weather data provides external environmental information, the adjacent time period environmental data mean can reflect the trend of environmental parameter changes, and the season identifier considers the influence of seasonal changes on the indoor environment. Through these input features, the XGBoost model can effectively predict the missing environmental data, ensuring the integrity and accuracy of the data.

[0066] With this scheme, by considering the missing conditions of time series data and environmental data and using the corresponding model for prediction and filling, the missing problem in the data can be more comprehensively handled. This method not only improves the quality and integrity of the data, but also provides a more reliable data basis for subsequent model training and prediction. Compared with the prior art, this scheme has higher flexibility and accuracy in handling missing data, and can better adapt to different data characteristics and application scenarios, thereby providing more accurate and personalized air conditioner recommendation services for users.

[0067] Optionally, in the S24 process, the air conditioner divides the standardized processed features into dynamic windows, including:

[0068] In the case where the current season is summer, the dynamic window duration is determined as the first duration.

[0069] In the case where the current season is the transition season, the dynamic window duration is determined as the second duration.

[0070] In the case where the current season is winter, the dynamic window duration is determined as the third duration.

[0071] Among them, the first duration is less than the second duration, and the second duration is less than the third duration.

[0072] In this scheme, when the current season is summer (June to August), the length of the dynamic window is determined as the first length, i.e., 7 days. This shorter data window can quickly respond to the frequent changes in air conditioner usage patterns of users in summer, capturing the immediate preferences and needs of users in the high-temperature season. When the season is the transition season (March to May and September to November), the length of the dynamic window is determined as the second length, i.e., 10 days. This medium-length data window can adapt to the gradual changes in user usage habits during the seasonal transition, while maintaining a certain data stability to provide enough information for the model to make accurate predictions. When the season is winter (December to February), the length of the dynamic window is determined as the third length, i.e., 14 days. The longer data window in winter can better reflect the relatively stable and long-term usage habits of users in the cold season, ensuring that the model can fully learn the preferences of users in this season, thereby providing more accurate recommendations.

[0073] With this scheme, it can more flexibly adapt to the changes in user air conditioner usage habits in different seasons, improve the prediction accuracy and adaptability of the model. Compared with the fixed data window length method in the prior art, the dynamic window division scheme can better capture the usage patterns and preferences of users in different seasons, thereby providing more personalized and accurate air conditioner recommendation services. Not only improves the performance of the recommendation system, but also enhances the user experience, reduces the frequency of manual adjustment of air conditioner settings by users, and also improves the energy utilization efficiency.

[0074] In an optimized scheme, after dynamic window division is performed on the standardized features, time decay weighting processing is further performed on the data in the dynamic time window. In this scheme, the summer data decay coefficient is set to 0.6-0.8, and the winter data decay coefficient is set to 0.85-0.95. Through the exponential decay function w(t)=α (T-t)The data weighting is implemented, where a is an attenuation coefficient, T is the current time, and t is the data time. In this way, in the specific implementation, a fixed attenuation coefficient of 0.7 is used in summer, and a fixed attenuation coefficient of 0.9 is used in winter, while the attenuation coefficient range is dynamically adjusted according to the user operation frequency and the environmental parameter change amplitude, a smaller attenuation coefficient (such as adjusting to 0.6 in summer) is used when the user operation is frequent, and a larger attenuation coefficient (such as adjusting to 0.95 in winter) is used when the environmental parameter changes drastically. With this scheme, by differentiating the seasonal attenuation coefficient, the model pays more attention to the recent data while retaining the long-term trend characteristics. The faster attenuation in summer can better capture the user's frequent adjustment behavior pattern, and the slower attenuation in winter is conducive to maintaining stable use habits, so as to input the weighted data into the double-layer LSTM model for training and prediction. In this way, the test shows that this scheme improves the temperature prediction accuracy by 12%, the mode prediction accuracy by 8%, and reduces the energy consumption by 15%, significantly improving the timeliness and personalization level of the recommendation system.

[0075] Figure 3 is a method for model training provided by an embodiment of the present disclosure. In combination with Figure 3 , optionally, in the S12 process, the air conditioner trains the double-layer LSTM model using the preprocessed data, including:

[0076] S31, the air conditioner divides the preprocessed data into a training set and a test set.

[0077] S32, the air conditioner inputs the training set data into the double-layer LSTM model, and iteratively trains using an optimizer with learning rate decay and early stopping mechanism.

[0078] S33, the air conditioner uses the test set to verify the model performance when the training set prediction index reaches the preset accuracy threshold.

[0079] S34, the air conditioner outputs the trained double-layer LSTM model when the verification is passed.

[0080] In this scheme, the preprocessed data can be divided into a training set and a test set, where the training set accounts for 80%, and the test set accounts for 20%. This division ratio can ensure that the model can fully learn the patterns and features in the data during the training phase, and at the same time objectively evaluate the performance of the model through the test set to verify the generalization ability of the model.

[0081] Further, when initializing the double-layer LSTM model, the input dimension is set to an 8-dimensional feature vector, which integrates time features, environmental features, and user operation features, providing the model with rich information. Specifically, in the double-layer LSTM model with a specific structure, the number of units in the first hidden layer is 128, and the number of units in the second hidden layer is 64. This structure design can capture the main features in the data through the first layer, and the second layer further refines and refines these features, thereby improving the model's ability to recognize complex patterns.

[0082] During training, data input with a batch size of 32 can be used, and the training period is 50 times. This setting can ensure training efficiency while allowing the model to fully learn the features in the data. In one example, the Adam optimizer is used for parameter updating, with an initial learning rate of 0.001, and an early stopping mechanism is configured to terminate training when the validation set loss does not decrease for 10 consecutive times. This early stopping mechanism can effectively prevent overfitting and ensure that the model always maintains good generalization ability during training. In addition, the learning rate is decayed to 0.8 times of the original value every 5 training cycles, which helps the model adjust parameters more finely in the later training period, further improving the model's accuracy.

[0083] During training, the mean squared error (MSE) loss between the predicted value and the true value can be calculated, and the model parameters are updated through backpropagation. This loss function selection can effectively measure the difference between the model's predicted value and the true value, guiding the model's training direction. Specifically, the training set prediction indicators include the temperature prediction mean absolute error (MAE) of the training set validation, the pattern prediction accuracy, and the wind speed prediction score. In this way, the training set prediction indicators reaching the preset accuracy threshold include: when the temperature prediction MAE of the training set validation is less than 0.5℃, the pattern prediction accuracy is higher than 92%, and the wind speed prediction F1 score is higher than 0.85, it is determined that the training set prediction indicators reach the preset accuracy threshold, then the test set is used to verify the model performance. If the performance on the test set also meets these preset accuracy thresholds, i.e., the temperature prediction MAE is less than 0.5℃, the pattern prediction accuracy is higher than 92%, and the wind speed prediction F1 score is higher than 0.85, the model training is successful, and the trained double-layer LSTM model is output. The setting of these evaluation indicators ensures that the model meets high standards in multiple key performance indicators, thereby ensuring the high quality and reliability of the model.

[0084] With this scheme, the double-layer LSTM model structure can more effectively capture the long-term dependencies in time series data, and compared with the traditional single-layer LSTM model, it can better handle complex user behavior patterns and environmental changes. Secondly, through fine-tuning of hyperparameters and optimization strategies, such as the Adam optimizer, early stopping mechanism and learning rate decay, the model can be trained more efficiently, improving the training speed and accuracy of the model. In addition, through a strict evaluation index system, the model has reached a high standard in multiple key performance indicators, providing more accurate and reliable air conditioner recommendation services for users. In this way, not only the performance of the recommendation system is improved, but also the user experience is enhanced, reducing the frequency of manual adjustment of air conditioner settings, while also improving energy efficiency.

[0085] In an optimized scheme, the feature vector output by the LSTM can also be input into the causal graph reasoning module to generate a causal weight matrix by calculating the conditional independence between variables, and then the original features are gate fused to finally generate a feature representation with causal explanation. In one example, a causal graph reasoning module is introduced after the double-layer LSTM model to learn the causal relationship between user operations and environmental parameters from historical data through the PC algorithm. In specific implementation, first, the time series features output by the LSTM layer are used as input to construct a directed acyclic graph containing causal edges such as outdoor temperature → set temperature, and light intensity → wind speed. The weight of each edge is calculated by conditional independence test (e.g., the causal weight of outdoor temperature on set temperature is 0.82). Then, the causal weight matrix is gate fused with the hidden state of the LSTM, and the proportion of causal information injection is controlled by the sigmoid function, and finally a feature representation is output that contains both time series patterns and causal explanations. In this way, the introduction of the causal reasoning module enables the air conditioner to distinguish between user-initiated adjustments and environmental passive influences, providing interpretable recommendations such as "suggest lowering the set temperature due to the increase in outdoor temperature" while maintaining prediction accuracy, significantly improving user trust.

[0086] Figure 4 is another LSTM model-based air conditioner recommendation method provided by the embodiments of the present disclosure. As shown in Figure 4 The embodiments of the present disclosure also provide another LSTM model-based air conditioner recommendation method, which includes:

[0087] S41, the air conditioner collects user operation data, environmental data and third-party weather data, and pre-processes the collected data.

[0088] S42, the air conditioner trains a double-layer LSTM model using the pre-processed data.

[0089] S43, the air conditioner performs air conditioner setting prediction based on the current time window data and the trained double-layer LSTM model.

[0090] S44, the air conditioner calculates a comprehensive confidence of the prediction result.

[0091] S451, in the case where the comprehensive confidence is greater than a first preset threshold, the air conditioner displays the model prediction value.

[0092] S452, in the case where the comprehensive confidence is greater than or equal to a second preset threshold and less than or equal to the first preset threshold, the air conditioner displays the model prediction value and suggests parameter adjustment.

[0093] S453, in the case where the comprehensive confidence is less than the second preset threshold, the air conditioner displays the historical use average value and suggests manual setting.

[0094] In this scheme, after calculating the comprehensive confidence, the air conditioner can take different recommendation strategies according to the calculated comprehensive confidence. If the comprehensive confidence is greater than the first preset threshold, it means that the model is very confident about the prediction result, at this time the air conditioner directly displays the model prediction value, and the user can directly accept these suggestions without additional adjustment. If the comprehensive confidence is between the second preset threshold and the first preset threshold, it means that the model has a certain confidence in the prediction result, but there is a certain uncertainty. In this case, the air conditioner displays the model prediction value while suggesting that the user adjust the parameters according to the actual situation. If the comprehensive confidence is less than the second preset threshold, it means that the model is not confident enough about the prediction result, at this time the air conditioner displays the historical use average value and suggests that the user manually set the air conditioner parameters to ensure user comfort and satisfaction. In this scheme, the setting standards of the first preset threshold and the second preset threshold are based on the comprehensive evaluation of the confidence of the model prediction result to ensure the reliability of the recommendation system and user experience. Specifically, the first preset threshold is set to 0.7, indicating that the model is very confident about the prediction result and can directly recommend to the user; the second preset threshold is set to 0.5, indicating that the model has a certain confidence in the prediction result, but there is a certain uncertainty, which needs to prompt the user to adjust.

[0095] With this scheme, through the evaluation of the comprehensive confidence, not only more accurate recommendations can be provided, but also different operation suggestions can be given to the user according to the level of confidence, thereby improving the user's use experience and the trust of the recommendation system. In addition, this embodiment introduces a confidence-driven recommendation algorithm, further enhancing the adaptability and flexibility of the recommendation system, better meeting the needs of users in different situations, while improving energy utilization efficiency and reducing unnecessary energy waste.

[0096] In one optimized scheme, real-time environmental change rates (such as temperature and humidity fluctuation gradients) and user behavior consistency indicators (such as recent operation deviations) are introduced as dynamic weight adjustment factors, enabling the calculation of comprehensive confidence to have environmental adaptability. For example, when the environment changes drastically, the temperature weight is automatically increased to 0.4, and when user operations are unstable, the pattern weight is increased to 0.5. Simultaneously, the static threshold is replaced with a dynamic threshold based on a sliding window (0.7±0.1), automatically adjusting the threshold range by analyzing historical user acceptance rates. Tests show that this optimized scheme further improves the adoption rate of high-confidence recommendation scenarios to 95%, increases prediction accuracy under extreme weather conditions by 18%, and reduces the false trigger rate of low-confidence alarms by 22%, achieving a more accurate "prediction-feedback" closed-loop optimization.

[0097] Optionally, in the S44 process, the overall confidence level of the air conditioning calculation and prediction results includes:

[0098] The maximum class probability of the working mode output by the air conditioner determination model is used as the first confidence component.

[0099] The air conditioner calculates the second confidence component based on the temperature prediction value output by the model.

[0100] The air conditioner uses the predicted probability value of wind speed as the third confidence component and the predicted probability value of the air guide angle as the fourth confidence component.

[0101] The air conditioner calculates the overall confidence level of the prediction result by weighting and summing the first confidence level components to the fourth confidence level components according to the weight ratio of 0.4:0.3:0.2:0.1.

[0102] In this approach, the air conditioner first determines the highest class probability of the operating mode output by the model as the first confidence component. Understandably, the last layer of the model uses the Softmax activation function, which transforms the output into a probability distribution. For example, if the model output is [cooling: 0.7, heating: 0.2, ventilation: 0.1], then the confidence of the cooling mode, which has the highest probability, can be determined as 0.7, which is the first confidence component.

[0103] Next, the air conditioner calculates the second confidence component based on the temperature prediction value output by the model. Understandably, the aforementioned model outputs not only the temperature value but also the standard deviation σ of that prediction. The formula for calculating the confidence level of the temperature prediction is 1 / (σ+0.1). The 0.1 is added here for smoothing and to avoid the denominator being zero. For example, if the model predicts a temperature of 26℃ and the standard deviation σ is 0.3, then the confidence level of the temperature prediction is 1 / (0.3+0.1)=2.5. After normalization, it might be 0.8, which is the second confidence component.

[0104] In addition, the air conditioner takes the wind speed prediction probability value as the third confidence component and the deflector angle prediction probability value as the fourth confidence component. These probability values are also output by the model and reflect the confidence level of the model in predicting the wind speed and deflector angle.

[0105] Finally, the air conditioner weights and sums the four confidence components in a weight ratio of 0.4:0.3:0.2:0.1 to obtain the comprehensive confidence. This weighting method is based on a comprehensive consideration of the importance of different parameters, ensuring that the comprehensive confidence can fully reflect the overall confidence level of the model in predicting the results.

[0106] In this way, the weighting method of the present solution fully considers the importance of different parameters in the user's use process, ensuring that the comprehensive confidence can fully and objectively reflect the overall confidence level of the model in predicting the results. Not only does this improve the accuracy and reliability of the recommendation system, but it also enhances the user experience, reduces the frequency of manual adjustment of air conditioner settings by the user, and improves energy utilization efficiency and reduces unnecessary energy waste.

[0107] The embodiment of the present disclosure provides an air conditioner recommendation device based on an LSTM model, which includes a preprocessing module, a training module, a prediction module, and a display module. The preprocessing module is configured to collect user operation data, environmental data, and third-party weather data, and preprocess the collected data; the training module is configured to train a double-layer LSTM model using the preprocessed data; the prediction module is configured to predict air conditioner settings according to current time window data and the trained double-layer LSTM model; and the display module is configured to display recommendation information according to the comprehensive confidence of the prediction results.

[0108] The air conditioner recommendation device based on an LSTM model provided by the embodiment of the present disclosure collects user operation data, environmental data, and third-party weather data, and preprocesses the data, thereby providing more comprehensive and multi-dimensional input information for the model. The powerful time series modeling capability of the double-layer LSTM model is used to deeply mine the dynamic change rules of user behavior patterns and accurately capture the dynamic evolution of user demand with factors such as time and season. The air conditioner settings are then predicted according to the current time window data, and the recommendation information is displayed according to the comprehensive confidence of the prediction results. In this way, more accurate and personalized air conditioner setting suggestions can be provided for the user, significantly improving the intelligent level of the recommendation system and the user experience, and meeting the user's demand for accurate and personalized recommendations.

[0109] Figure 5 is a schematic diagram of an air conditioner recommendation device based on an LSTM model provided by the embodiment of the present disclosure. Combined with the above description of the air conditioner recommendation device based on an LSTM model, the working principle of the air conditioner recommendation device based on an LSTM model is described in detail. Figure 5As shown, the embodiment of the present disclosure provides an air conditioner recommendation device 200 based on an LSTM model, comprising a processor 20 and a memory 21. Optionally, the device can also comprise a communication interface 22 and a bus 23. Wherein the processor 20, the communication interface 22, the memory 21 can complete the communication among each other through the bus 23. The communication interface 22 can be used for information transmission. The processor 20 can call the logic instructions in the memory 21 to execute the air conditioner recommendation method based on the LSTM model of the above-mentioned embodiment.

[0110] In addition, the logic instructions in the memory 21 described above can be realized in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium.

[0111] The memory 21 as a kind of computer readable storage medium can be used to store software programs, computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor 20 executes the program instructions / modules stored in the memory 21, thereby executing function application and data processing, i.e. realizing the air conditioner recommendation method based on the LSTM model in the above-mentioned embodiment.

[0112] The memory 21 can include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function; The data storage area can store data created according to the use of the terminal device, etc. In addition, the memory 21 can include a high-speed random access memory, and can also include a non-volatile memory.

[0113] Figure 6 It is a structure schematic diagram of an air conditioner provided by the embodiment of the present disclosure. In combination with the above-mentioned Figure 6 As shown, the embodiment of the present disclosure provides an air conditioner 100, comprising an air conditioner body and the above-mentioned air conditioner recommendation device 200 based on an LSTM model. The air conditioner recommendation device 200 based on an LSTM model is installed on the air conditioner body. The installation relationship described herein is not limited to placing in the interior of the air conditioner body, but also includes installation connection with other components of the air conditioner 100, including but not limited to physical connection, electrical connection or signal transmission connection, etc. Those skilled in the art can understand that the air conditioner recommendation device 200 based on an LSTM model can be adapted to a feasible air conditioner body, and thus realize other feasible embodiments.

[0114] The embodiment of the present disclosure provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are set to execute the above-mentioned air conditioner recommendation method based on an LSTM model.

[0115] The technical solutions of the embodiments of the present disclosure can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes one or more instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method disclosed in the embodiments of the present disclosure. The aforementioned storage medium can be a non-transitory storage medium, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0116] The above description and drawings sufficiently illustrate the embodiments of the present disclosure to enable one skilled in the art to practice them. Other embodiments can include structural, logical, electrical, process, and other changes. The embodiments represent only a few of the possible variations. Individual components and functions are optional unless explicitly required, and the order of operations can be changed. Parts and features of some embodiments can be included or replaced by parts and features of other embodiments. Also, the words used in this application are only used to describe the embodiments and not to limit the claims. As used in the description of the embodiments and the claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of one or more associated listed items. In addition, when used in this application, the term "comprise" and its variants "comprises" and / or comprises" and the like mean the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, or device that includes the stated element. In this document, each embodiment focuses on the differences from other embodiments, and the same or similar parts between embodiments can be referred to each other. For the method, product, etc. disclosed in the embodiments, if it corresponds to the method part disclosed in the embodiments, the relevant part can be referred to the description of the method part.

[0117] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods for each specific application to realize the described functions, but such implementation should not be considered beyond the scope of the embodiments of the present disclosure. The skilled person can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0118] In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units can only be a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms. The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to implement the embodiments. In addition, each functional unit in the embodiments of the present disclosure can be integrated in one processing unit, or each unit can be a physically independent unit, or two or more units can be integrated in one unit.

[0119] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

Claims

1. An air conditioner recommendation method based on an LSTM model, characterized in that, include: Collect user operation data, environmental data, and third-party weather data, and preprocess the collected data. The user operation data includes air conditioning mode switching, temperature adjustment value or adjustment range, wind speed adjustment range, change of air guide plate angle, or start-up and shutdown time records. Train a two-layer LSTM model using the preprocessed data; Predict air conditioning settings based on current time window data and a trained two-layer LSTM model; Recommended information is displayed based on the overall confidence level of the prediction results; The prediction results include wind speed, deflector angle, pattern, and temperature; the overall confidence level of the prediction results is calculated, including: The maximum class probability of the working mode output by the model is determined as the first confidence component; Calculate the second confidence component based on the temperature prediction value output by the model; The wind speed prediction probability value is used as the third confidence component, and the wind guide plate angle prediction probability value is used as the fourth confidence component. The first confidence component to the fourth confidence component are weighted and summed in a weight ratio of 0.4:0.3:0.2:0.1 to calculate the overall confidence of the prediction result; The confidence level of temperature prediction is calculated using the formula 1 / (σ+0.1), where σ is the standard deviation of the predicted temperature value. The step of displaying recommendation information based on the comprehensive confidence level of the prediction results includes: If the overall confidence level is greater than a first preset threshold, the model prediction value is displayed; If the overall confidence level is greater than or equal to the second preset threshold and less than or equal to the first preset threshold, the model prediction value is displayed and parameter adjustment is suggested. If the overall confidence level is less than the second preset threshold, the historical usage average is displayed and manual setting is suggested.

2. The method according to claim 1, characterized in that, The preprocessing of the collected data includes: Outliers are detected using a sliding window method to filter out consecutive repetitive operations with an interval shorter than a set time. Fill in the missing data to obtain the complete data after filling; Time features, environmental features, and user operation features are extracted from the complete data, and each extracted feature is standardized. Dynamic windowing is performed on each feature after standardization to obtain preprocessed data.

3. The method according to claim 2, characterized in that, The filling of missing data includes: For missing portions of the time series data, a recurrent neural network is used for prediction and filling. The time series data is the timestamp log of the user operation data. For the missing environmental data, a gradient boosting decision tree model is used to predict and fill in the missing data. The environmental data includes the observed values ​​of indoor temperature, humidity and light intensity. The user operation data includes mode switching, temperature setting adjustment, and changes in wind speed and air guide plate angle.

4. The method according to claim 2, characterized in that, The dynamic window partitioning of each feature after standardization includes: Given that the current season is summer, the dynamic window duration is set to the first duration. If the current season is a transitional season, the dynamic window duration is determined to be a second duration; If the current season is winter, the duration of the dynamic window is determined to be the third duration. Wherein, the first duration is less than the second duration, and the second duration is less than the third duration.

5. The method according to claim 1, characterized in that, The method of training a two-layer LSTM model using preprocessed data includes: The preprocessed data is divided into a training set and a test set; The training set data is input into a two-layer LSTM model, and iterative training is performed using an optimizer with learning rate decay and early stopping mechanism. When the prediction metric of the training set reaches the preset accuracy threshold, the test set is used to verify the model performance. If the validation passes, output the trained two-layer LSTM model.

6. An air conditioner recommendation device based on an LSTM model, characterized in that, include: The preprocessing module is configured to collect user operation data, environmental data, and third-party weather data, and to preprocess the collected data. The user operation data includes air conditioning mode switching, temperature adjustment value or adjustment range, wind speed adjustment range, change of air guide plate angle, or start-up and shutdown time records. The training module is configured to train a two-layer LSTM model using preprocessed data; The prediction module is configured to predict air conditioning settings based on the current time window data and the trained two-layer LSTM model. The display module is configured to show recommendation information based on the overall confidence level of the prediction results; The prediction results include wind speed, deflector angle, pattern, and temperature; the overall confidence level of the prediction results is calculated, including: The maximum class probability of the working mode output by the model is determined as the first confidence component; Calculate the second confidence component based on the temperature prediction value output by the model; The wind speed prediction probability value is used as the third confidence component, and the wind guide plate angle prediction probability value is used as the fourth confidence component. The first confidence component to the fourth confidence component are weighted and summed in a weight ratio of 0.4:0.3:0.2:0.1 to calculate the overall confidence of the prediction result; The confidence level of temperature prediction is calculated using the formula 1 / (σ+0.1), where σ is the standard deviation of the predicted temperature value. The step of displaying recommendation information based on the comprehensive confidence level of the prediction results includes: If the overall confidence level is greater than a first preset threshold, the model prediction value is displayed; If the overall confidence level is greater than or equal to the second preset threshold and less than or equal to the first preset threshold, the model prediction value is displayed and parameter adjustment is suggested. If the overall confidence level is less than the second preset threshold, the historical usage average is displayed and manual setting is suggested.

7. An air conditioning recommendation device based on an LSTM model, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute the air conditioning recommendation method based on the LSTM model as described in any one of claims 1 to 5 when running the program instructions.

8. An air conditioner, characterized in that, include: Air conditioner unit; The air conditioner recommendation device based on the LSTM model as described in claim 6 or 7 is installed on the air conditioner body.

Images