A method and system for adaptive adjustment of heating modes

Abstract

The present invention provides a method and system for adaptively adjusting a heating mode, including: installing sensors to collect temperature, humidity, and wind force data; analyzing the collected temperature, humidity, and wind force data to determine the actual perceived temperature locally; determining a first heating mode based on the actual perceived temperature; obtaining the current air temperature and room temperature in the first heating mode and predicting weather changes; adjusting the first heating mode to a second heating mode according to the prediction result; and making a refined adjustment to the second heating mode according to the characteristics of different regions, floors, people, and rooms.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and particularly to a method and system for adaptively adjusting a heating mode. Background Art

[0002] With the progress of technology, the heating industry is gradually developing towards the intelligent direction, and more and more heating energy-saving modes have been proposed. However, at present, most heating systems still adopt a fixed heating temperature, and the heating mode basically remains unchanged regardless of how the external environment changes.

[0003] In the southwestern region of China, the temperature is low and the humidity is high in winter, resulting in the perceived temperature of the human body being several degrees lower than the actual air temperature. Moreover, in different regions, different floors and different rooms, due to different terrains and room layouts, the human perception of cold is different, and the wind force is also different. If a unified heating mode is adopted, some people on some floors and in some rooms will feel hot, while some people on some floors and in some rooms will feel not warm enough. Therefore, the existing heating mode is not flexible enough, which will lead to the inability to ensure the indoor comfort and increase energy consumption. Summary of the Invention

[0004] The present application provides a method and system for adaptively adjusting a heating mode, which is used to solve the problems in the prior art that the existing heating mode is not flexible enough, resulting in low indoor comfort and high energy consumption.

[0005] In view of the above problems, the present application provides a method and system for adaptively adjusting a heating mode.

[0006] The embodiment of the present application provides a method for adaptively adjusting a heating mode, including:

[0007] Install sensors to collect temperature, humidity and wind force data;

[0008] Analyze the collected temperature, humidity and wind force data to determine the actual perceived temperature locally;

[0009] Based on the actual perceived temperature, determine the first heating mode;

[0010] Obtain the current air temperature and room temperature under the first heating mode, and predict the weather change;

[0011] According to the prediction result, adjust the first heating mode to the second heating mode;

[0012] According to the characteristics of different regions, floors, people and rooms, make a refined adjustment to the second heating mode.

[0013] Optionally, analyzing the collected temperature, humidity and wind force data to determine the actual perceived temperature locally includes:

[0014] Clean the outliers from the collected temperature, humidity, and wind data;

[0015] Convert the cleaned data into a standardized format;

[0016] Calculate the perceived temperature, where the perceived temperature WCI is based on the following formula:

[0017] WCI = 13.12 + 0.6215T - 11.37V 0.16 + 0.3965TV 0.16 - 0.15TH - 0.27T 2 + 1.265H 2 + 0.548

[0018] T 2 H 2

[0019] where WCI is the perceived temperature;

[0020] T is the actual temperature;

[0021] V is the wind speed;

[0022] H is the relative humidity.

[0023] Optionally, analyze the collected temperature, humidity, and wind data to determine the local actual perceived temperature, including:

[0024] Clean the collected temperature, humidity, and wind data;

[0025] Convert the cleaned data into a standardized format;

[0026] Construct a regression model and train the regression model using historical data;

[0027] Input the standardized data into the trained regression model and output the predicted perceived temperature;

[0028] where inputting the standardized data into the trained regression model and outputting the predicted perceived temperature includes:

[0029] Extract features from the standardized data;

[0030] Input the extracted features into the trained regression model, and the regression model performs a forward propagation operation based on the weights and the features to generate the predicted perceived temperature;

[0031] Denormalize the predicted perceived temperature and perform semantic interpretation on the denormalized predicted perceived temperature.

[0032] Optionally, predicting weather changes includes:

[0033] Obtaining the indoor temperature, outdoor temperature, humidity, and wind force;

[0034] Preprocessing the indoor temperature, outdoor temperature, humidity, and wind force to obtain first preprocessed data;

[0035] Constructing a recurrent neural network model RNN;

[0036] Training the RNN using historical data;

[0037] Predicting future weather using the trained RNN.

[0038] Optionally, predicting weather changes includes:

[0039] Obtaining the indoor temperature, outdoor temperature, humidity, and wind force;

[0040] Preprocessing the indoor temperature, outdoor temperature, humidity, and wind force to obtain second preprocessed data;

[0041] Analyzing the second preprocessed data and dividing it into linear data and non - linear data;

[0042] For the linear data, performing differencing to obtain a time series;

[0043] Constructing an autoregressive integrated moving average model ARIMA or a seasonal autoregressive integrated moving average model SARIMA;

[0044] Selecting the parameters of the ARIMA or SARIMA model, including the number of autoregressive terms p, the degree of differencing d, and the number of moving average terms q;

[0045] Training the ARIMA or SARIMA model using the selected parameters;

[0046] Predicting the first weather change using the trained ARIMA or SARIMA model;

[0047] Converting the non - linear data into a supervised learning format;

[0048] Constructing an RNN model and defining the number of layers, the number of hidden layer units, and the activation function of the RNN;

[0049] Selecting an optimizer, a loss function, and an evaluation metric, and training the RNN model using the historical data;

[0050] Predicting the second weather change using the trained RNN model;

[0051] Fuse the predicted first weather change and second weather change to obtain the final predicted weather change.

[0052] Optionally, according to the prediction result, adjust the first heating mode to the second heating mode, including:

[0053] Obtain the final predicted weather change;

[0054] Determine the average temperature, maximum temperature and minimum temperature within a certain future time period;

[0055] Optimize the heating strategy according to the average temperature, maximum temperature and minimum temperature within the certain future time period, and adjust it to the second heating mode.

[0056] Optionally, according to the characteristics of different regions, floors, people and rooms, make refined adjustments to the second heating mode, including:

[0057] Collect the real-time temperature, humidity and wind data of each region, floor and room;

[0058] Record the natural attributes of people and the comfort feedback information on temperature;

[0059] Use a camera or sensor to confirm whether there is someone in the room and the activity level information of the person;

[0060] Summarize the real-time temperature, humidity and wind data, the natural attributes of people, the comfort feedback information on temperature, whether there is someone in the room, and the activity level information of the person to form reference data;

[0061] Classify and label the reference data according to time, location, people and room characteristics;

[0062] Generate features related to heating demand based on the above classified and labeled reference data;

[0063] Use the features to train a deep learning model for predicting heating demand;

[0064] Verify the prediction effect of the deep learning model using historical data;

[0065] Use the deep learning model to predict heating demand;

[0066] Automatically adjust the heating temperature of each region, floor and room based on the prediction result.

[0067] Optionally, using the features to train a deep learning model for predicting heating demand, including:

[0068] Collect historical data, including location, time, temperature, humidity, wind speed, pedestrian flow, housing type, housing orientation, and heating equipment type;

[0069] Obtain heating characteristics from the historical data;

[0070] Extract time characteristics from the time;

[0071] Create interaction features based on the heating characteristics;

[0072] Construct a random forest model;

[0073] Input the heating characteristics, time characteristics, and interaction features into the random forest model for training.

[0074] Optionally, based on the prediction results, automatically adjust the heating temperatures of each region, floor, and room, including:

[0075] Adjust the heating temperature based on the following formula:

[0076] T c = T b + a1(A) + a2(L) + a3(R) + a4(P) + a5(E)

[0077] Where:

[0078] T c is the recommended indoor temperature;

[0079] T b is the baseline temperature;

[0080] A represents the influence of the region;

[0081] L represents the influence of the floor, divided into high, medium, and low;

[0082] R is the influence of the room being occupied or unoccupied;

[0083] P is the influence of people, including the combined influence of age and gender;

[0084] E represents the activity level of people;

[0085] The coefficients a1 to a5 are weights, describing the degree to which each parameter affects the recommended temperature.

[0086] The embodiment of the present application also provides a heating mode adaptive adjustment system. A computer program is stored in the system, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0087] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0088] The technical solution provided by the embodiments of the present application changes from the initial global first heating mode to the new personalized second heating mode, and factors such as different floors, rooms, and personal preferences are taken into consideration under the second heating mode, so as to finely adjust the second heating mode. At the same time, artificial intelligence means and methods are adopted to combine the predicted weather changes with the heating mode. The adaptive adjustment method not only meets the comfort requirements of the vast majority of people, but also saves energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 It is a schematic flow chart of a method for adaptively adjusting a heating mode provided by the present application;

[0090] Figure 2 It is a refined flow chart of S2 provided by the present application;

[0091] Figure 3 It is a refined flow chart of S4 provided by the present application;

[0092] Figure 4 It is a refined flow chart of S5 provided by the present application;

[0093] Figure 5 It is a refined flow chart of S6 provided by the present application;

[0094] Figure 6 It is a schematic diagram of a system structure provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0095] The present application provides a method and system for adaptively adjusting a heating mode, which changes from the initial global first heating mode to the new personalized second heating mode, and factors such as different floors, rooms, and personal preferences are taken into consideration under the second heating mode, so as to finely adjust the second heating mode. At the same time, artificial intelligence means and methods are adopted to combine the predicted weather changes with the heating mode. The adaptive adjustment method not only meets the comfort requirements of the vast majority of people, but also saves energy consumption.

[0096] Embodiment 1

[0097] The embodiment of the present application provides a method for adaptively adjusting a heating mode, as Figure 1 shown, including:

[0098] S1. Install sensors to collect temperature, humidity, and wind data;

[0099] To collect this data, a series of sensors are required. Generally, this includes temperature sensors, humidity sensors, wind speed and direction sensors, etc. These sensors can be connected to the central data collection system wirelessly or by wire, and then specific algorithms are used to combine various data to calculate the perceived temperature.

[0100] Taking a certain community in Guiyang as an example:

[0101] Temperature sensor: A digital temperature sensor such as DS18B20 can be used. This kind of sensor can provide high-precision temperature readings and is suitable for outdoor use.

[0102] Humidity sensor: DHT22 or DHT11 are relatively common humidity sensors, and they can read temperature and humidity simultaneously.

[0103] Wind speed and direction sensor: An Anemometer wind speed and direction meter can be selected, which can measure the wind speed and direction.

[0104] To cover the entire community, a group of sensors can be installed at different locations in the community, such as at the corners in the east, west, south, and north directions of the community, as well as in the central park or square. This can ensure that the collected data is more comprehensive and accurate.

[0105] Data collection method: Each sensor can be connected to a microcontroller (such as Arduino or Raspberry Pi), and the data is sent to the central data processing center through Wi-Fi or LoRaWAN. These microcontrollers read the data of the sensors at regular intervals (such as every 5 minutes) and upload it. In addition, the calculation of the felt temperature can be completed in the central data processing center or on each microcontroller before uploading.

[0106] When installing the equipment, it should be ensured that it is far away from direct sunlight and rain, and waterproof and dustproof measures should be taken to ensure the long-term operation of the equipment.

[0107] S2. Analyze the collected temperature, humidity, and wind data to determine the actual felt temperature locally;

[0108] The felt temperature, also known as the "wind chill index" or "wet bulb temperature", is a quantity that describes the external temperature felt by the human body. This is an index that combines wind speed, humidity, actual atmospheric temperature, and other factors. The felt temperature is not the actual temperature, but represents the relative cold or hot degree felt by the human body due to other environmental factors.

[0109] In one of the embodiments, in S102, analyzing the collected temperature, humidity, and wind data to determine the actual felt temperature locally specifically includes the following steps:

[0110] Clean the outliers in the collected temperature, humidity, and wind data; (for example, the temperature is between -40°C and 50°C, the humidity is between 0% and 100%, etc.)

[0111] Convert the cleaned data into a standardized format; for example, temperature may need to be converted to degrees Celsius, and wind speed may need to be converted to meters per second.

[0112] Calculate the perceived temperature. In cold weather, the wind chill index (WCI) is based on the following formula:

[0113] WCI = 13.12 + 0.6215T - 11.37V 0.16 + 0.3965TV 0.16 - 0.15TH - 0.27T 2 + 1.265H 2 + 0.548

[0114] T 2 H 2

[0115] Where WCI is the wind chill index;

[0116] T is the actual temperature;

[0117] V is the wind speed;

[0118] H is the relative humidity.

[0119] The various coefficients mentioned in the formula, such as 13.12, 0.6215, etc., are empirical values, and other specific values can also be used in the actual formula.

[0120] In another embodiment, in addition to the formula calculation mentioned in the above embodiment, an artificial intelligence method can also be used, such as Figure 2 As shown, analyze the collected temperature, humidity, and wind data to determine the local actual wind chill index, including the following steps:

[0121] A1. Clean the collected temperature, humidity, and wind data;

[0122] Clean the collected data to remove outliers or incomplete data.

[0123] A2. Convert the cleaned data into a standardized form;

[0124] Standardize the data to ensure it is within the same scale range.

[0125] A3. Build a regression model and train the regression model using historical data;

[0126] Taking CNN as an example, the Convolutional Neural Network (CNN) is a deep learning model specially designed to process data with grid structures (such as images and regression data). The following are the detailed steps on how to build and train a CNN model using historical data:

[0127] Building the Convolutional Neural Network model:

[0128] Input layer:

[0129] This layer corresponds to the input data of the model. For images, this is usually a three-dimensional array representing the width, height, and number of channels of the image (for example, an RGB image has 3 channels).

[0130] Convolutional layer:

[0131] Use filters (also known as convolutional kernels) to slide over the input data to extract spatial features.

[0132] These filters can automatically learn important features in the data.

[0133] Activation function layer:

[0134] A non-linear activation function, such as ReLU, is usually applied after each convolutional layer to introduce non-linearity.

[0135] Pooling layer:

[0136] These layers reduce the computational amount and the number of parameters by reducing the spatial dimension of the data, while retaining important feature information.

[0137] Common pooling operations include max pooling and average pooling.

[0138] Fully connected layer:

[0139] After convolutional and pooling operations, the data will be flattened and processed through one or more fully connected layers.

[0140] These layers can capture the relationships between different parts of the model.

[0141] Output layer:

[0142] This layer corresponds to the prediction of the model. For classification tasks, the number of nodes in the output layer is usually the same as the number of classes, and the softmax activation function is used.

[0143] Optimizer, loss function, and evaluation metrics:

[0144] Select an optimizer, such as Adam or SGD, to update the weights of the network.

[0145] Select a loss function, such as cross-entropy loss, to measure the difference between the model's predictions and the true values.

[0146] Select evaluation metrics, such as accuracy, to measure the performance of the model.

[0147] Steps to train the model using historical data:

[0148] A31. Data preprocessing:

[0149] Normalize the historical data so that its values fall between 0 and 1.

[0150] If necessary, augment the data, such as rotation, cropping, and horizontal flipping, to increase the generalization ability of the model.

[0151] A32. Data splitting:

[0152] Divide the dataset into a training set, a validation set, and a test set.

[0153] A33. Model initialization:

[0154] Initialize the weights and biases of the model.

[0155] A34. Batch training:

[0156] Select a batch of data from the training set. Input this batch of data into the model for forward propagation.

[0157] Calculate the loss function; use the backpropagation algorithm to calculate the gradients; use the optimizer to update the weights.

[0158] A35. Validation:

[0159] At the end of each epoch, evaluate the performance of the model using the validation set.

[0160] If the performance of the model on the validation set stops improving or starts to decline, early stopping techniques can be used to end the training.

[0161] A36. Testing:

[0162] After the model training is completed, evaluate its performance using the test set.

[0163] A37. Adjustment:

[0164] According to the validation and test results, adjust the parameters of the model, such as the learning rate, batch size, or model structure.

[0165] A38. Repeat:

[0166] Repeat the above steps until the model achieves satisfactory performance.

[0167] A4. Input the standardized data into the trained regression model to output the predicted perceived temperature.

[0168] Among them, A4 includes:

[0169] A41. Extract features from the standardized data;

[0170] If feature engineering such as feature selection and dimensionality reduction is performed during model training, the same operations need to be performed on the current data.

[0171] A42. Input the extracted features into the trained regression model, and the regression model performs forward propagation operations based on the weights and the features to generate the predicted perceived temperature;

[0172] Input the preprocessed data into the model. The model will perform forward propagation operations based on the weights it has learned and the input data to generate the predicted perceived temperature.

[0173] Obtain the prediction result from the model. This may be a continuous temperature value or several different perceived temperature categories (e.g., "cold", "warm", "hot", etc.). Different from the previous embodiment, empirically, the difference between the perceived temperature and the actual temperature is within a few degrees and is affected by wind speed and humidity. Therefore, the perceived temperature is preferably represented by categories (e.g., "cold", "warm", "hot", etc.).

[0174] A43. Denormalize the predicted perceived temperature and perform semantic interpretation on the denormalized predicted perceived temperature.

[0175] If the data is normalized in the preprocessing step, an inverse operation may be required to convert the prediction result back to the original unit of measurement.

[0176] Provide semantic interpretation or suggestions regarding the perceived temperature based on the prediction result. For example, "Considering the current humidity and wind speed, you may feel a bit cold."

[0177] Compared with traditional technologies, using AI technology to determine the local actual perceived temperature has the following advantages:

[0178] High accuracy: AI can more accurately predict the perceived temperature by learning a large amount of historical data.

[0179] Automation and real-time performance: AI can automatically and real-time perform data analysis and prediction without manual intervention.

[0180] Adaptive ability: As more data is collected, the AI model can continuously learn and optimize to better adapt to different environments and changes.

[0181] Taking multiple factors into account: AI can consider multiple factors affecting the perceived temperature simultaneously, rather than relying solely on one or a few factors.

[0182] Resource saving: Once the model is trained and deployed, it can run for a long time, which may be more economical, saving manpower and material resources compared to traditional methods.

[0183] S3. Determine the first heating mode based on the actual perceived temperature;

[0184] In S3, the first heating mode is a traditional heating mode, that is, through a global heating mode, where every household has the same heating temperature. In this mode, only one global instruction is needed for heating. For example, if the perceived temperature is expressed as "cold", or the specific value of the perceived temperature is only 10 degrees, then a global instruction of "22 degrees Celsius" is issued, and all heated rooms will be 22 degrees.

[0185] Since the actual differences between each floor, the community, and people are not considered in this mode, therefore, fine-tuning is still required to adjust the global first heating mode to a personalized second heating mode.

[0186] S4. Obtain the current air temperature and room temperature in the first heating mode and predict weather changes;

[0187] Under normal circumstances, weather changes can be obtained by retrieving weather forecast information from a third-party API interface. However, the information in the weather forecast, such as the weather conditions released by the meteorological department, has two disadvantages. First, the data released is usually only for a city and not accurate to a specific area. In addition, the data released usually includes the highest temperature, the lowest temperature, and an average temperature, and the highest and lowest temperatures are usually only at the peak and trough periods, with a very short time, which has little reference value for heating. Second, the core focus of the released weather forecast is on the weather (sunny, rainy, snowy, etc.) and temperature, and there is little mention and prediction of data such as humidity and wind speed, let alone a forecast for a specific community.

[0188] Therefore, in the embodiments of the present invention, it is preferably to collect data through sensors and predict weather changes through artificial intelligence, specifically as Figure 3 shown, including the following steps:

[0189] B1. Obtain the indoor temperature, outdoor temperature, humidity, and wind speed;

[0190] Collect the above data through the above sensor group. This data can be real-time data or average data over a period of time.

[0191] B2. Preprocess the indoor temperature, outdoor temperature, humidity, and wind speed to obtain the first preprocessed data;

[0192] Clean the data, delete outliers and missing values, and perform data standardization or normalization. Integrate data from different sources to generate a unified dataset. Define this dataset as the first preprocessed data.

[0193] B3. Construct a Recurrent Neural Network model (RNN);

[0194] Select a suitable model for weather prediction. This can be a traditional statistical model, such as a time series model, or a modern AI model, such as a Recurrent Neural Network (RNN).

[0195] Among them, the construction of the RNN model includes:

[0196] Select the model type: Select an appropriate RNN type according to the complexity of the problem, such as: basic RNN, LSTM, or GRU.

[0197] Define the model structure: Set the input layer, hidden layer, the number of neurons in each layer, output layer, etc.

[0198] Select the loss function and optimizer: For example, for regression problems, MSE can be selected as the loss function, and for classification problems, cross-entropy loss can be selected.

[0199] B4. Use historical data to train the RNN;

[0200] Specifically include:

[0201] B41. Model initialization: Initialize the model weights and biases.

[0202] B42. Batch training: Select a batch size for training according to the data size and computing power.

[0203] B43. Forward propagation: Input the data and obtain the predicted value through the RNN structure.

[0204] B44. Loss calculation: Calculate the loss based on the predicted value and the true value.

[0205] B45. Backward propagation: Calculate the gradient of each weight according to the loss value and update the weights.

[0206] B46. Validation: After each epoch, use the validation set to check the performance of the model to avoid overfitting.

[0207] B47. Early stopping: If the loss of the validation set does not improve significantly in several consecutive epochs, training can be terminated early.

[0208] B5. Use the trained RNN to predict future weather.

[0209] Using the trained RNN model to predict future weather involves the following steps:

[0210] Input data: Input the preprocessed data into the model.

[0211] Model inference: The model will perform a forward propagation operation based on the weights it has learned and the input data to generate a weather prediction.

[0212] Output result: Obtain the prediction result from the model. This could be a continuous temperature value, humidity value, or wind speed, or a more specific weather description such as "sunny", "cloudy", "rainy", etc.

[0213] Denormalization / Inverse standardization: If the data was normalized or standardized in the preprocessing step, an inverse operation may be required to convert the prediction result back to the original unit of measurement.

[0214] Result interpretation: Provide specific suggestions or explanations to the user based on the prediction result. For example: "It is predicted that the temperature tomorrow will be 25°C, and there is a 60% probability of rain. It is recommended to carry an umbrella when going out."

[0215] In addition, in one of the embodiments, when predicting weather changes, by combining traditional statistical methods and modern AI models, more accurate and stable prediction results can often be obtained compared to the aforementioned RNN model.

[0216] Traditional statistical methods, such as the autoregressive integrated moving average model (ARIMA) or the seasonal autoregressive integrated moving average model (SARIMA), are very good at capturing the linear trends and periodicities in the data.

[0217] Modern AI models, such as the recurrent neural network (RNN) or the long short-term memory network (LSTM) in deep learning, can capture the non-linear patterns in the data, especially in the case of a large amount of data.

[0218] By combining the two, the respective advantages can be fully utilized to improve the accuracy and robustness of the prediction. The specific approach can be: First, use statistical methods for preliminary prediction, and then use the prediction result as an input or feature and input it together with other relevant data into the AI model to obtain the final prediction result.

[0219] Specifically, it includes the following steps:

[0220] The steps of using the autoregressive integrated moving average model (ARIMA) or the seasonal autoregressive integrated moving average model (SARIMA) to capture the linear trends and periodicities in the data include C1 - C8:

[0221] C1. Obtain the indoor temperature, outdoor temperature, humidity, and wind force;

[0222] Collect relevant data such as temperature, humidity, and wind speed for the past few weeks, months, or years.

[0223] C2. Preprocess the indoor temperature, outdoor temperature, humidity, and wind force to obtain second preprocessed data;

[0224] Among them, the preprocessing includes data cleaning, format standardization, etc.

[0225] C3. Analyze the second preprocessed data and divide it into linear data and non-linear data;

[0226] Among the received data, a part of the data has obvious time linearity or seasonality, and this part of the data can be defined as linear data, while the data without obvious temporality or seasonality is defined as non-linear data. Specifically, matlab can be used to plot different types of data according to the time axis, so that linear data and non-linear data can be easily obtained.

[0227] C4. For linear data, perform differencing processing on it to obtain a time series;

[0228] If the data has obvious seasonality, differencing processing can be performed on it first to obtain a stable time series.

[0229] C5. Construct an autoregressive moving average model ARIMA or a seasonal autoregressive moving average model SARIMA model;

[0230] ARIMA (Autoregressive Integrated Moving Average Model):

[0231] ARIMA is a model for predicting a time series. It combines three methods:

[0232] AR (AutoRegressive): Use its own lag values (or past observations) to predict the time series. For example, if (p) is the number of lag values (i.e., time steps), it is denoted as (AR(p)).

[0233] I (Integrated): This means that the time series is differenced, or compared with the values of the previous period, to make it stationary. (d) is the number of differencing times.

[0234] MA (Moving Average): Use the lag error terms of the time series to predict. For example, if (q) is the number of lag error terms, then it is denoted as (MA(q)).

[0235] Therefore, the ARIMA model is usually expressed as (ARIMA(p,d,q)).

[0236] SARIMA (Seasonal AutoRegressive Integrated Moving Average Model):

[0237] SARIMA adds a seasonal component to the basis of ARIMA. In addition to the above three parameters of ARIMA, SARIMA also has three seasonal parameters:

[0238] SAR (Seasonal AutoRegressive): Used to predict time series based on seasonal lags.

[0239] SI (Seasonal Integrated): This indicates that the time series has been seasonally differenced.

[0240] SMA (Seasonal Moving Average): Uses error terms based on seasonal lags to predict time series.

[0241] The SARIMA model is usually expressed as (SARIMA(p,d,q)(P,D,Q)s), where (s) is the seasonal period. For example, for annual data, s = 12 represents 12 months.

[0242] ARIMA and SARIMA are models for predicting time series data. ARIMA focuses on non-seasonal trends, while SARIMA adds a seasonal component on this basis, making it more suitable for data with obvious seasonal fluctuations.

[0243] C6. Select the parameters of the ARIMA or SARIMA model, including the number of autoregressive terms p, the number of differencing times d, and the number of moving average terms q;

[0244] Select p (the number of autoregressive terms), d (the number of differencing times), and q (the number of moving average terms) of the ARIMA or SARIMA model.

[0245] Use ACF and PACF plots to assist in determining the values of p and q.

[0246] The following are the step-by-step instructions on how to use these plots to determine the values of p and q:

[0247] C61. Determine the value of d

[0248] First, it is necessary to determine whether the time series is stationary. A non-stationary time series has a mean and / or variance that changes over time.

[0249] If a time series appears to have an obvious trend or seasonal pattern, it may be necessary to difference it until it becomes stationary. The number of times of differencing is the value of d.

[0250] C62. Using ACF and PACF Plots

[0251] Once the time series becomes stationary, ACF and PACF plots can be drawn to determine the values of p and q.

[0252] For an AR(p) model:

[0253] The PACF plot will show a cut-off pattern after lag p.

[0254] The ACF plot will decay gradually.

[0255] For an MA(q) model:

[0256] The PACF plot will decay gradually.

[0257] The ACF plot will show a cut-off pattern after lag q.

[0258] C63. Selecting the Values of p and q

[0259] Observe the PACF plot and find the last lag before which all other lags are significant. This can be used as the value of p.

[0260] Observe the ACF plot and find the last lag before which all other lags are significant. This can be used as the value of q.

[0261] C64. Parameters of the SARIMA Model

[0262] For the SARIMA model, the values of P (the number of seasonal autoregressive terms), D (the number of seasonal differencing times), and Q (the number of seasonal moving average terms) also need to be selected. The same principles of ACF and PACF apply to the seasonal components, but seasonal lags should be considered (e.g., if the seasonal period is 12 months, consider lags 12, 24, 36, etc.).

[0263] C65. Modeling and Validation

[0264] After selecting the parameters, use these parameters to train an ARIMA or SARIMA model. Then, methods such as cross-validation, time series splitting, or rolling forecasting can be used to evaluate the performance of the model. If the results are not satisfactory, it may be necessary to adjust the values of p, d, q and try again.

[0265] It is important to remember that choosing the parameters for an ARIMA or SARIMA model may require multiple attempts and validations. The ACF and PACF plots provide a good starting point, but the actual modeling and validation are key parts of determining the parameters.

[0266] Due to its unique climate characteristics (cold, damp, and gloomy), winter in Guiyang may have different time series patterns. To determine the parameters of an ARIMA or SARIMA model, this plan can follow the steps below, based on the actual climate data of Guiyang:

[0267] Data collection

[0268] First, obtain the time series data of daily temperature, humidity, and other possible meteorological indicators in winter in Guiyang. At least several years of data are needed to obtain an accurate model.

[0269] Data visualization

[0270] Plot the data graph and observe whether there are obvious trends or seasonality. This is very helpful for deciding whether differencing is needed and the number of times of differencing (i.e., the value of d).

[0271] Determine the value of d

[0272] If the data has an obvious upward or downward trend, differencing may be needed at least once.

[0273] If the data has a seasonal pattern (e.g., there is a trend of temperature drop or humidity increase at a certain specific time every year), then seasonal differencing may be needed. This is especially important for the SARIMA model.

[0274] Use the ACF and PACF plots

[0275] Use the stationary data (possibly the differenced data) to plot the ACF and PACF plots.

[0276] Observe the PACF plot and find the last significant lag, which can give the value of p.

[0277] Observe the ACF plot and find the last significant lag, which can give the value of q.

[0278] Select the seasonal parameters

[0279] If the data shows obvious seasonality (e.g., there are similar patterns every year), then it is necessary to consider using the SARIMA model. Observe the ACF and PACF plots again, but this time focus on the seasonal lags. For example, if the seasonality is once a year and 12 - month data are used, pay attention to the places with lags of 12, 24, 36, etc.

[0280] Model fitting and validation

[0281] Fit an ARIMA or SARIMA model using the selected parameter values and validate it. Based on the validation results, the parameters may need to be adjusted and refitted.

[0282] C7. Train the ARIMA or SARIMA model using the selected parameters;

[0283] The following is a detailed supplement regarding C7:

[0284] C71 Parameter Estimation

[0285] At the beginning of the model, a preliminary estimation of the parameters is carried out. ARIMA and SARIMA models include autoregressive terms, moving average terms, and differencing, so these parameters need to be estimated. Maximum Likelihood Estimation (MLE) or Bayesian Estimation is usually used to determine the optimal values of these parameters.

[0286] C72 Optimization Algorithm

[0287] To find the parameters that minimize the model error, optimization algorithms such as BFGS, LBFGS, Newton-Conjugate-Gradient, etc. are usually used. These algorithms will try to update the model parameters in each iteration to minimize the prediction error.

[0288] C73 Model Diagnosis

[0289] After the model training is completed, the diagnostic information can be checked to evaluate the fitness of the model. This includes checking the residuals, QQ plots, ACF and PACF plots, etc. If the residuals show obvious non-randomness, then the model may need to be modified.

[0290] C74 Model Selection

[0291] The finally selected model should be the one with the minimum prediction error, but the complexity of the model should also be considered. An overly complex model may lead to overfitting and poor performance of the model on new data.

[0292] C75 Model Validation

[0293] After the model is selected, the "rolling prediction" method can be used to validate the performance of the model. In rolling prediction, the model uses known data for prediction and then observes the error between the predicted value and the actual value. This can help confirm whether the model is really suitable for future predictions.

[0294] C8. Use the trained ARIMA or SARIMA model to predict the first weather change;

[0295] The steps of using a Recurrent Neural Network (RNN) to capture non - linear patterns in data include C9 - C13:

[0296] C9. Convert the non - linear data into a supervised learning format;

[0297] That is, create an input sequence and corresponding output from the non - linear data.

[0298] C10. Build an RNN model and define the number of layers, number of hidden layer units, and activation function of the RNN;

[0299] Specifically, as shown in B3, it will not be elaborated here.

[0300] C11. Select an optimizer, loss function, and evaluation metric, and use the historical data to train the RNN model;

[0301] C12. Use the trained RNN model to predict the second - day weather change;

[0302] C13. Fuse the predicted first - day weather change and the second - day weather change to obtain the final predicted weather change.

[0303] Perform weighted averaging or other ensemble strategies on the prediction results of the two models, such as stacking, model fusion, etc.

[0304] Among them, result integration is a commonly used strategy for combining the predictions of multiple models to improve the prediction accuracy. The following are some commonly used result integration methods:

[0305] Weighted average

[0306] If there are two models in this solution, weights can be assigned to their prediction results and the weighted average can be calculated.

[0307] Suppose the prediction of model A is P A , and the prediction of model B is P B :

[0308] P final = w A ×P A + w B ×P B

[0309] Where w A and w B are the weights assigned to the two models, and satisfy w A + w B = 1.

[0310] The weights can be determined according to the historical performance of the models or the results of cross - validation.

[0311] Stacking

[0312] Stacking is to use a new model to combine the prediction results of multiple models. The prediction results of the base models are used as the input features of the new model.

[0313] 1. Divide the data into two parts: the training set and the validation set.

[0314] 2. Use the training set to train all the base models.

[0315] 3. Use the base models to make predictions on the validation set.

[0316] 4. Use the prediction results of the base models as features to train the true results of the validation set to obtain a new model.

[0317] Model Fusion

[0318] Model fusion is to combine the structures and parameters of different models into a new model.

[0319] For example, for neural networks, the outputs of the middle layers of two models can be merged and a new output layer can be trained; or, the parameters of the two models can be weighted averaged.

[0320] Voting

[0321] If the prediction problem is a classification problem, the voting strategy can be used. The prediction of each model for the class can be regarded as a vote, and the class with the most votes is finally selected.

[0322] Simple voting: The predictions of each model have the same weight.

[0323] Weighted voting: Different weights are assigned to the predictions of each model based on the performance or trustworthiness of the model.

[0324] To implement the above strategy, tools such as `VotingClassifier` or `VotingRegressor` in `sklearn.ensemble` of Python can be used.

[0325] In addition, in one of the embodiments, dual-model training and cascading can also be considered, for example:

[0326] Train using ARIMA / SARIMA and RNN / LSTM models respectively.

[0327] Use ARIMA / SARIMA to make a preliminary prediction on the time series.

[0328] Use the prediction results of ARIMA / SARIMA and other relevant data as inputs and feed them into the RNN / LSTM model for further prediction.

[0329] S5. Adjust the first heating mode to the second heating mode according to the prediction result;

[0330] In one embodiment, as Figure 4 shown, S5 adjusts the first heating mode to the second heating mode according to the prediction result, including the following steps D1 - D3:

[0331] D1. Obtain the final predicted weather change;

[0332] Obtain the weather prediction for the next few days from the ARIMA / SARIMA and RNN models in the previous step, paying particular attention to temperature, humidity, and wind speed.

[0333] D2. Determine the average temperature, maximum temperature, and minimum temperature within a certain future time period;

[0334] D3. Optimize the heating strategy according to the average temperature, maximum temperature, and minimum temperature within the certain future time period, and adjust it to the second heating mode.

[0335] For example, if the prediction data shows that it will be warm during the day but very cold at night, the heating can be slightly reduced during the day, but heated in advance at night to ensure a comfortable indoor temperature at night.

[0336] Consider the requirements of different indoor areas. For example, if a certain room is usually colder at night, a higher heating temperature can be set for that room.

[0337] In another embodiment, the heating mode will be adjusted in advance based on the predicted weather information, that is, adjusted to the second heating mode. Such an adjustment can ensure the comfort of the indoor temperature and energy conservation. The following are the example steps of how to adjust the heating mode according to the prediction:

[0338] E1. Obtain weather prediction information:

[0339] Obtain the weather prediction for the next few days from the ARIMA / SARIMA and RNN models in the previous step, paying particular attention to temperature, humidity, and wind speed.

[0340] E2. Analyze the prediction data:

[0341] Check if there are any abnormal temperature fluctuations or other weather events (such as cold snaps, heavy rains, etc.).

[0342] Analyze the prediction data to determine the average temperature, maximum temperature, and minimum temperature for the next few days.

[0343] E3. Determine the heating demand:

[0344] If the prediction shows that the temperature will drop significantly, it may be necessary to increase the heating temperature.

[0345] If the prediction shows that the temperature will be stable or rise slightly, it may be necessary to maintain or slightly lower the heating temperature.

[0346] Consider the humidity situation, as high humidity usually makes people feel colder.

[0347] E4. Adjust the heating mode:

[0348] According to the analysis results, set the temperature of the thermostat in advance.

[0349] Consider slightly lowering the temperature when people are not at home or sleeping to save energy.

[0350] If the prediction shows that it will be very cold in the next few days, consider activating the preheating function of the heating system.

[0351] E5. Optimize the heating strategy:

[0352] If the prediction data shows that it will be warm during the day but very cold at night, the heating can be slightly reduced during the day, but heated in advance at night to ensure a comfortable indoor temperature at night.

[0353] Consider the needs of different areas indoors. For example, if a certain room is usually colder at night, a higher heating temperature can be set for that room.

[0354] E6. Monitoring and feedback:

[0355] Once the mode is adjusted, continuously monitor the actual indoor temperature and compare it with the predicted value.

[0356] If there is a large deviation between the actual temperature and the predicted value, consider making further mode adjustments or checking whether the heating system is working properly.

[0357] In this way, adjusting the heating mode in advance according to the weather forecast can not only ensure that the indoor temperature always remains within a comfortable range, but also effectively save energy, thus achieving the dual goals of economy and comfort.

[0358] S6. Refined adjustment of the second heating mode according to the characteristics of different regions, floors, people and rooms.

[0359] Taking the humid and cold characteristics in winter in Guiyang as an example, the refined adjustment can start from the following aspects:

[0360] Classification of people and demand analysis:

[0361] Set different comfortable temperature ranges according to different populations (the elderly, children, adults) and activity types (such as rest, work, exercise, etc.).

[0362] Considering the humid and cold weather in Guiyang, pay attention to avoiding excessive indoor humidity to reduce the feeling of cold and the growth of mold.

[0363] Adjustment of heating mode:

[0364] Temperature setting: Based on human comfort and outdoor temperature, intelligently adjust the indoor temperature. For example, if the external temperature is 5°C, the indoor temperature can be set between 20 - 22°C.

[0365] Humidity control: Since the humidity is relatively high in winter in Guiyang, a dehumidifier can be used to reduce the indoor humidity and keep it within the range of 50% - 60% relative humidity to avoid excessive dampness indoors.

[0366] Wind speed adjustment: Based on the room size and the number of people, adjust the wind speed of the heating equipment to ensure uniform indoor temperature and reduce the feeling of cold wind.

[0367] Humanized adjustment:

[0368] Time period adjustment: According to daily living habits, raise the room temperature in the morning and evening, and make fine adjustments during the day according to personnel activities and outdoor temperature.

[0369] Zone heating: According to the room usage, such as the living room, bedroom, bathroom, etc., set different heating modes respectively.

[0370] Vacant mode: When it is detected that a room has been unoccupied for a period of time, automatically reduce the heating, which is energy-saving and environmentally friendly.

[0371] Prediction and automatic adjustment:

[0372] Use AI technology to predict weather changes in the next few hours or days, such as temperature, humidity, and wind speed, etc., and automatically adjust the heating mode accordingly.

[0373] Through deep learning, continuously optimize the heating mode according to historical user habits and feedback to better meet user comfort.

[0374] Specifically, based on different regions, floors, people, and room characteristics, the heating requirements are very different. To achieve higher comfort and economic benefits, the embodiments of the present invention need to finely regulate the heating mode. In S6, as Figure 5 shown, according to the characteristics of different regions, floors, people, and rooms, finely adjust the second heating mode, including the following steps:

[0375] F1. Collect real-time temperature, humidity, and wind data of each region, floor, and room;

[0376] F2. Record the natural attributes of the person and the feedback information on temperature comfort;

[0377] F3. Use a camera or sensor to confirm whether there is someone in the room and the activity level information of the person;

[0378] F4. Summarize the real-time temperature, humidity, wind force data, the natural attributes of the person, the feedback information on temperature comfort, whether there is someone in the room, and the activity level information of the person to form reference data;

[0379] F5. Classify and label the reference data according to time, location, person, and room characteristics;

[0380] F6. Generate features related to heating demand based on the above classified and labeled reference data;

[0381] Generate features related to heating demand based on the above classified data, such as average temperature, activity level of people, room orientation, etc.

[0382] The steps of F6 can be abbreviated as feature engineering. Feature engineering is a very crucial step in deep learning, which involves extracting, creating, and selecting those features that are helpful for improving the model performance from the original data. The following are the specific steps:

[0383] F61. Data exploration

[0384] First, it is necessary to understand the relevant data, including data distribution, missing values, outliers, etc.

[0385] F61.1: View the basic statistical information of the data, such as maximum value, minimum value, average value, median, etc.

[0386] F61.2: Use visualization tools (such as histogram, boxplot, etc.) to view the data distribution.

[0387] F61.3: Check whether there are missing values or outliers in the data.

[0388] F62. Feature generation

[0389] Based on the understanding of the data, new features can be created or existing features can be transformed.

[0390] F62.1: Time features: If there is a timestamp in the data, time-related features such as "time of day", "day of the week", etc. can be extracted from it.

[0391] F62.2: Interaction features: Based on business understanding, consider combining two or more features. For example, if there are two features, temperature and humidity, a new feature such as "temperature-humidity index" can be considered.

[0392] F62.3: Statistical features: For continuous features, statistical features can be generated, such as the average temperature, maximum temperature, and minimum temperature in the past hour or day.

[0393] F62.4: Domain knowledge features: Based on the professional knowledge of heating, consider other factors that may affect heating demand. For example, according to the orientation of the room (south-facing or north-facing), the sunshine duration and intensity can be estimated, and thus the natural heating amount of the room can be estimated.

[0394] F63. Feature Selection

[0395] Sometimes, not all features are useful. Some features may have little relation to the target variable or be highly correlated with other features.

[0396] F63.1: Use statistical tests (such as Pearson correlation coefficient) to check the relationship between features and the target variable.

[0397] F63.2: Use feature selection algorithms, such as recursive feature elimination (RFE), feature importance, etc. to select the most important features.

[0398] F64. Feature Transformation

[0399] To enable the model to better understand the data, sometimes it is necessary to transform the features.

[0400] F64.1: Standardize or normalize continuous features.

[0401] F64.2: Encode categorical features, for example, using one-hot encoding.

[0402] Through the above feature engineering process, a set of features related to heating demand can be generated, thus helping the model to better learn and predict.

[0403] F7. Use the described features to train a deep learning model for predicting heating demand;

[0404] In one of the embodiments, using the described features to train a deep learning model for predicting heating demand includes:

[0405] Collect historical data, including location, time, air temperature, humidity, wind speed, pedestrian flow, housing type, housing orientation, heating equipment type;

[0406] Obtain heating features from the historical data;

[0407] Extract time features from the said time;

[0408] Create interaction features based on heating characteristics;

[0409] Construct a random forest model;

[0410] Input the said heating characteristics, time features and interaction features into the random forest model for training.

[0411] Specifically, when predicting heating demand, various factors involved need to be considered first. It is crucial to abstract these factors into features so that the AI model can learn patterns from the data. The following are the steps on how to use domain knowledge features, time features and interaction features to train an AI model (such as a random forest model) for heating demand:

[0412] F71. Data preprocessing

[0413] Collect historical data, including location, time, temperature, humidity, wind speed, pedestrian flow, house type, house orientation, heating equipment type, etc.

[0414] For missing data, perform interpolation or other methods for processing.

[0415] Normalize or standardize the data to ensure that different features are on the same dimension.

[0416] F72. Feature engineering

[0417] Extract factors related to heating from domain knowledge. For example, the insulation condition of the house, whether it is close to the window, house orientation, etc.

[0418] Extract features from time data. For example, decompose the date into year, month, day, season, whether it is a weekday / holiday, etc.

[0419] Create interaction features based on existing features. For example, consider the interaction between temperature and humidity, or the interaction between house orientation and temperature.

[0420] F73. Model selection

[0421] Select a suitable model according to the quantity and complexity of the data. For example, you can start with a simple linear regression and then consider using a complex model such as a random forest model.

[0422] F74. Model training

[0423] Use 80% of the data as the training set and 20% as the validation set.

[0424] Input the feature data into the selected model and start training.

[0425] Monitor the training process to ensure that the model does not overfit. If overfitting occurs, consider increasing regularization, adjusting model parameters, or stopping training early.

[0426] F75. Model Validation

[0427] Evaluate the performance of the model using the validation set. Mean squared error, root mean squared error, or other evaluation metrics can be used.

[0428] If the model performs poorly on the validation set, return to step F73 for model selection or return to step F74 to further optimize the model.

[0429] F76. Model Deployment

[0430] Once the model is verified and achieves satisfactory performance, it can be deployed into the actual environment for predicting heating demand.

[0431] Regularly retrain and optimize the model using new data to ensure it always remains efficient and accurate.

[0432] Through the above steps, an AI model for predicting heating demand can be trained using various feature data including domain knowledge features, time features, and interaction features.

[0433] F8. Validate the prediction effect of the deep learning model using historical data;

[0434] Model validation is a key step in ensuring the performance of the deep learning model. Taking Guiyang as an example, the following are the specific steps:

[0435] F81. Data Preprocessing

[0436] Clean the historical data of central heating in Guiyang, deleting or filling in missing values.

[0437] Separate the data into features (such as temperature, humidity, wind speed, etc.) and target variables (e.g., actual heating demand or perceived temperature).

[0438] F82. Dataset Division

[0439] Divide the historical data into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to adjust model parameters, and the test set is used to finally evaluate the model performance.

[0440] F83. Model Training

[0441] Train the AI model using the training set data.

[0442] F84. Model Validation

[0443] Predict the target variable using the validation set data.

[0444] Calculate the error between the predicted results of the computational model and the true values in the validation set, such as the mean squared error (MSE) or other appropriate evaluation metrics.

[0445] F85. Model Optimization

[0446] Based on the performance feedback from the validation set, adjust the hyperparameters of the model. For example, if using a neural network, the learning rate, batch size, number of neurons, etc. may be adjusted.

[0447] Feature selection may be required, that is, deleting or adding certain features to see if it can improve the performance of the model.

[0448] If a complex model is used and overfitting occurs, try adding a regularization term or reducing the complexity of the model.

[0449] F86. Model Evaluation

[0450] When the performance of the model on the validation set reaches a satisfactory level, use the test set to evaluate the final performance of the model.

[0451] If the performance of the test set is similar to that of the validation set and both meet the expectations, the model is successfully validated. Otherwise, the model structure or features may need to be reconsidered.

[0452] F87. Model Robustness Testing

[0453] To test the robustness of the model, noise can be added to the input data to see if the prediction performance of the model remains stable.

[0454] Different data sources or data within different time periods can be used to test the performance of the model to ensure that the model can maintain good performance in various scenarios.

[0455] Through the above verification and optimization steps, it can be ensured that the AI model has good prediction performance, accuracy, and robustness on the central heating data in Guiyang.

[0456] F9. Use the described deep learning model for heating demand prediction;

[0457] After training, the model can be used to predict heating demand, such as predicting the specific values of temperature, humidity, and wind force for the next few days, etc.

[0458] F10. Automatically adjust the heating temperature for each region, floor, and room based on the prediction results.

[0459] The logic of adjustment is as follows:

[0460] Based on the perceived temperature and heating demand predicted by the model, formulate specific heating temperature strategies for each region, floor, and room. For example, further adjust the heating temperature according to characteristics such as the age, gender, and activity level of the people to ensure the comfort of each individual. For unoccupied rooms, the heating temperature can be appropriately reduced to save energy.

[0461] For example, if it is predicted that the perceived temperature in rooms on high floors will increase in the future, the heating temperature on that floor may be reduced. The elderly and children may require a higher indoor temperature, while the heating in unoccupied rooms can be appropriately reduced. In the humid and cold southern regions, for a child's room on the middle floor. The system will combine humidity to provide a moderate heating temperature and keep the indoor environment dry.

[0462] Through the above-mentioned automated and intelligent heating control steps, it can be ensured that the residents in each region, floor, and room can enjoy a comfortable and economical heating environment, while also greatly improving the energy utilization efficiency.

[0463] In addition, in order to integrate these complex parameters into a formula to describe the heating strategy, it is first necessary to define the coefficients of each parameter. These coefficients are usually determined based on experience, research, and / or the results of the model.

[0464] Adjust the heating temperature based on the following formula:

[0465] T c =T b +a1(A)+a2(L)+a3(R)+a4(P)+a5(E)

[0466] Where:

[0467] T c is the recommended indoor temperature;

[0468] T b is the reference temperature, which is a basic temperature, such as 18 °C;

[0469] A represents the influence of the region, and its value may be based on the average winter temperature and humidity of the region;

[0470] L represents the influence of the floor, divided into high, medium, and low, and each level corresponds to a different value;

[0471] R is the influence of whether the room is occupied or not. For example, it is positive when occupied and negative when unoccupied;

[0472] P is the influence of the people, including the combined influence of age and gender. For example, the elderly and children are negative, women and men are positive, and the value of men is higher than that of women, etc.;

[0473] E represents the activity level of a person, which can be divided into three levels: working, sleeping, and active, and each level corresponds to a different value.

[0474] The coefficients a1 to a5 are weights that describe the degree to which each parameter affects the recommended temperature.

[0475] For example, for a region, if a region is usually colder, the value of (A) may be higher; for a floor, the higher floor may be warmer than the lower floor, so the value of (L) may be negative for the higher floor; for an occupied room and an unoccupied room, (R) may increase or decrease the recommended temperature.

[0476] Determining the exact values of these coefficients requires data, experiments, and / or the results of models. In addition, the above formula is a simplified linear model; the actual model may involve non-linear effects, interaction effects, etc.

[0477] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0478] The technical solution provided in the embodiment of this application changes from the initial global first heating mode to the new personalized second heating mode, and factors such as different floors, rooms, and people's preferences are taken into account in the second heating mode, so as to finely adjust the second heating mode. At the same time, the means and methods of artificial intelligence are adopted to combine the predicted weather changes with the heating mode. The adaptive adjustment method not only meets the comfort requirements of the vast majority of people but also saves energy consumption.

[0479] Embodiment 2

[0480] Based on the same inventive concept as a heating mode adaptive adjustment method in the foregoing embodiment, this application also provides a system. A computer program is stored on the system, and when the computer program is executed by a processor, the method in Embodiment 1 is implemented.

[0481] Embodiment 3

[0482] The embodiment of this application also provides a system 6000, as Figure 6 shown, including a memory 64 and a processor 61. Computer-executable instructions are stored on the memory, and when the processor runs the computer-executable instructions on the memory, the above method is implemented. In practical applications, the system may also separately include other necessary components, including but not limited to any number of input systems 62, output systems 63, processors 61, controllers, memories 64, etc., and all systems that can implement the heating mode adaptive adjustment method in the embodiment of this application are within the protection scope of this application.

[0483] The memory includes but is not limited to a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), or a compact disc read-only memory (CD-ROM), and this memory is used for relevant instructions and data.

[0484] The input system 62 is used for inputting data and / or signals, and the output system 63 is used for outputting data and / or signals. The output system 63 and the input system 62 can be independent devices or an integrated device.

[0485] The processor can include one or more processors. For example, it includes one or more central processing units (CPUs). When the processor is a CPU, the CPU can be a single-core CPU or a multi-core CPU. The processor can also include one or more dedicated processors, and the dedicated processors can include GPUs, FPGAs, etc., for acceleration processing.

[0486] The memory is used for storing the program code and data of the network device.

[0487] The processor is used to call the program code and data in this memory and execute the steps in the above method embodiments. For details, please refer to the description in the method embodiments and will not be elaborated here.

[0488] In several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The displayed or discussed couplings, direct couplings, or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the systems or units can be in electrical, mechanical, or other forms.

[0489] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0490] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in the system or transmitted through the system. The computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The system can be any available medium accessible by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a read-only memory (ROM), a random access memory (RAM), a magnetic medium, such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium, such as a digital versatile disc (DVD), or a semiconductor medium, such as a solid state disk (SSD), etc.

[0491] This specification and the drawings are merely exemplary descriptions of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A method for adaptively adjusting a heating mode, characterized in that Including: Install sensors to collect temperature, humidity, and wind data; Analyze the collected temperature, humidity, and wind data to determine the local actual perceived temperature; Based on the actual perceived temperature, determine the first heating mode; Obtain the current air temperature and room temperature in the first heating mode and predict weather changes; According to the prediction result, adjust the first heating mode to the second heating mode; Fine-tune the second heating mode according to the characteristics of different regions, floors, people, and rooms; Among them, predicting weather changes includes: Obtain indoor temperature, outdoor temperature, humidity, and wind; Preprocess the indoor temperature, outdoor temperature, humidity, and wind to obtain second preprocessed data; Analyze the second preprocessed data and divide it into linear data and non-linear data; For linear data, perform differencing on it to obtain a time series; Construct an autoregressive integrated moving average model ARIMA or a seasonal autoregressive integrated moving average model SARIMA; Select the parameters of the ARIMA or SARIMA model, including the number of autoregressive terms p, the number of differencing times d, and the number of moving average terms q; Use the selected parameters to train the ARIMA or SARIMA model; Use the trained ARIMA or SARIMA model to predict the first weather change; Convert the non-linear data into a supervised learning format; Construct an RNN model and define the number of layers, the number of hidden layer units, and the activation function of the RNN; Select an optimizer, a loss function, and an evaluation metric, and use historical data to train the RNN model; Use the trained RNN model to predict the second weather change; Fuse the predicted first weather change and second weather change to obtain the final predicted weather change.

2. The method according to claim 1, wherein Analyze the collected temperature, humidity, and wind data to determine the local actual perceived temperature, including: Clean the outliers in the collected temperature, humidity, and wind data; Convert the cleaned data into a standardized format; Calculate the perceived temperature, and the perceived temperature WCI is based on the following formula: WCI = 13.12 + 0.6215T - 11.37V 0.16 + 0.3965TV 0.16 - 0.15TH - 0.27T 2 + 1.265H 2 + 0.548 T 2 H 2 Where WCI is the perceived temperature; T is the actual temperature; V is the wind speed; H is the relative humidity.

3. The method according to claim 1, characterized in that Analyze the collected temperature, humidity, and wind data to determine the local actual perceived temperature, including: Clean the collected temperature, humidity, and wind data; Convert the cleaned data into standardization; Construct a regression model and use historical data to train the regression model; Input the standardized data into the trained regression model and output the predicted perceived temperature; Among them, inputting the standardized data into the trained regression model and outputting the predicted perceived temperature includes: Extract features from the standardized data; Input the extracted features into the trained regression model, and the regression model performs a forward propagation operation based on the weights and the features to generate the predicted perceived temperature; Denormalize the predicted perceived temperature and perform semantic interpretation on the denormalized predicted perceived temperature.

4. The method according to claim 1, characterized in that, Predicting weather changes includes: Obtain the indoor temperature, outdoor temperature, humidity, and wind force; Preprocess the indoor temperature, outdoor temperature, humidity, and wind force to obtain the first preprocessed data; Construct a recurrent neural network model RNN; Train the RNN using historical data; Use the trained RNN to predict future weather.

5. The method according to claim 1, characterized in that According to the prediction results, adjust the first heating mode to the second heating mode, including: Obtain the final predicted weather changes; Determine the average temperature, maximum temperature, and minimum temperature within a certain future time period; Optimize the heating strategy based on the average temperature, maximum temperature, and minimum temperature within the certain future time period, and adjust it to the second heating mode.

6. The method according to claim 1, characterized in that, According to the characteristics of different regions, floors, people, and rooms, make a refined adjustment to the second heating mode, including: Collect real-time temperature, humidity, and wind force data for each region, floor, and room; Record the natural attributes of people and their comfort feedback information regarding temperature; Use a camera or sensor to confirm whether there is someone in the room and the activity level information of people; Summarize the real-time temperature, humidity, and wind force data, the natural attributes of people, their comfort feedback information regarding temperature, whether there is someone in the room, and the activity level information of people to form reference data; Classify and label the reference data according to time, location, people, and room characteristics; Generate features related to heating demand based on the above classified and labeled reference data; Use the features to train a deep learning model for predicting heating demand; Verify the prediction effect of the deep learning model using historical data; Use the deep learning model to predict heating demand; Automatically adjust the heating temperature for each region, floor, and room based on the prediction results.

7. The method according to claim 6, wherein Use the features to train a deep learning model for predicting heating demand, including: Collect historical data, including location, time, temperature, humidity, wind speed, pedestrian flow, house type, house orientation, heating equipment type; Obtain heating features from the historical data; Extract time features from the time; Create interaction features based on the heating features; Construct a random forest model; Input the heating features, time features, and interaction features into the random forest model for training.

8. The method according to claim 6, characterized in that, Automatically adjust the heating temperature for each region, floor, and room based on the prediction results, including: Adjust the heating temperature based on the following formula: T c = T b + a1(A) + a2(L) + a3(R) + a4(P) + a5(E) Where: T c is the recommended indoor temperature; T b is the reference temperature; A represents the influence of the region; L represents the influence of the floor, divided into high, medium, and low; R is the influence of whether there is someone in the room; P is the influence of people, including the combined influence of age and gender; E represents the activity level of people; The coefficients a1 to a5 are weights, describing the degree to which each parameter affects the recommended temperature.

9. A system for adaptively adjusting a heating mode, characterized in that A computer program is stored in the system, and when the computer program is executed by a processor, it implements the steps of any one of the methods in claims 1-8.

Images