Air conditioner load information management method and system based on B / S architecture

Abstract

The application discloses a kind of based on B / S architecture's air conditioner load information management method and system, including the construction INOA-LSTM model, specifically: step 1: collect original data;Step 2: input variable selection is carried out using Spearman correlation coefficient method, and data set is divided into training set and test set;Step 3: star insect starts in the position of seed storage area found by memory strategy, and finds optimal solution in stored seed.The application adopts common browser / server (B / S) architecture, and divides the overall system into two parts of client and server.The user only needs to understand the basic operation process, and can complete load prediction and air conditioner load period selection through the interactive interface of the device.The prediction result will generate table data, which is convenient for subsequent analysis and processing.

Description

Technical Field

[0001] This invention relates to the field of air conditioning system information management, specifically to an air conditioning load information management method and system based on a B / S architecture and combined with deep learning technology. Background Technology

[0002] The implementation of functions such as load forecasting and air conditioning load design periods relies on corresponding network models in practical applications. However, all network models require a deployment and operating environment for actual load forecasting and have certain performance requirements for computing equipment, which is inconvenient for data management personnel. To solve this problem, this paper designs an air conditioning load information management system. This system integrates the network model proposed in this paper and other high-performance forecasting models. Users only need to understand the basic operation process to complete load forecasting and select air conditioning load periods through the device's interactive interface. The forecast results will generate tabular data for convenient subsequent analysis and processing. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for managing air conditioning load information based on a B / S architecture, a method for predicting and designing air conditioning loads that is easy to deploy and simple to operate, overcomes the technical bottlenecks of existing network models that are highly dependent on the operating environment and have high usage thresholds, and improves the practicality and intelligence level of the system.

[0004] The technical solution adopted in this invention is:

[0005] A method for managing air conditioning load information based on a B / S architecture includes constructing an INOA-LSTM model, specifically: Step 1: Collect raw data; parameters include outdoor meteorological parameters, indoor disturbance parameters and load data. Outdoor meteorological parameters include outdoor daily average dry-bulb temperature, horizontal visibility, outdoor real-time temperature, wind direction, outdoor real-time humidity, total solar radiation on the horizontal surface, average wind speed, weather conditions, surface air pressure, and wind speed. Indoor disturbance parameters include indoor occupancy density. Load data includes real-time load data and historical load data from the previous moment. Step 2: Select input variables using the Spearman correlation coefficient method, and divide the dataset into training and test sets; initialize the sparrow population using the Tent chaotic mapping, mapping the individual positions of the stored seeds to the hyperparameters to be optimized; set parameters such as the maximum number of NOA iterations and population size, and generate the initial values ​​for solution position update optimization; set the upper and lower limits for parameters such as learning rate, batch size, L2 regularization coefficient, number of training rounds, and number of hidden units in the LSTM; initialize the LSTM network, train it using the obtained training dataset, calculate the initial individual fitness, and determine the optimal position of the individual; the sparrows begin searching for other solutions in the collection area, searching for the position of the seed on it, and bringing the found seeds to the storage area for storage; Step 3: The starbird begins to find the location of the stored seeds using a memory strategy and finds the optimal solution among the stored seeds; the latest hyperparameters are input into the LSTM network to calculate the individual fitness of the seeds and determine the optimal location; the optimal seed fitness is compared and the optimal seed location is updated; this continues until the maximum number of iterations is reached; the optimal parameters are output and the LSTM network is trained using the optimal hyperparameter combination for air conditioning load prediction.

[0006] Optional, specifically including; Determine the original dataset The dataset is divided into training and test sets. The INOA-LSTM model is then used to train the training set of the original dataset, resulting in a set of predicted values. Based on this predicted value Compared with actual value residual value Measure the output dimension: , To predict the output dimension; use Bootstrap to process the residual values. , ,conduct A second resampling with replacement is performed to obtain a new set of residuals. Combine residual groups with dataset By adding them together, we obtain a completely new dataset. For a completely new dataset Then, the INOA-LSTM model is trained and predicted again; the resampling is repeated B times to obtain B new datasets, and B INOA-LSTM models are trained at the same time. After using the Bootstrap residual resampling method, a brand new dataset for group B was obtained, as shown below: ; In the formula, Let b represent the b-th dataset. According to the Bootstrap principle, the final prediction result can be obtained by averaging the predicted values ​​of the b-th dataset, as shown in the following expression: ; In the formula, Indicates the first The first dataset One predicted value.

[0007] Optionally, it also includes an assessment of the uncertainty of the air conditioning load forecast results, specifically: Replace the predicted values ​​with the noise variance in the dataset to obtain a new dataset. ; : This is noise error; By analyzing the dataset To train a new model, the objective function is as follows: After estimating the noise variance and model variance, the corresponding upper and lower limits of the interval can be calculated to quantify the uncertainty of the prediction results. Optionally, the selection of the air conditioning load design period can also be included, specifically: By calculating the deviation between the daily cumulative distribution function (CDF) values ​​of various meteorological parameters for each time period and the CDF values ​​of long-term meteorological data at the 50%-90th percentile, appropriate weights are selected based on the RF method. Then, the weighted sum (WS) of the FS values ​​of each parameter is calculated, and the candidate with the smallest WS is selected as the air conditioning load design period. The weighted sum of the FS data is calculated as follows: ; ; ; In the formula, Meteorological parameters exist of value; For Heaven; For a period of time; Meteorological parameters exist of value; The number of possible values ​​for a certain meteorological parameter The number of meteorological parameters selected; For each Value at sky Time-weighted sum mean; Meteorological parameters The weight value.

[0008] Optionally, in step 2, three strategies, including OBL, are used to initialize the starbird population, specifically including: OBL Opposite Learning Strategy: For the initially generated starbird population, construct corresponding opposite individuals, evaluate the fitness of the original individuals and the opposite individuals, and select individuals with better fitness to form the optimized population. Tent Chaotic Mapping Strategy: Use Tent chaotic mapping to generate chaotic sequences and map the chaotic sequences to the search space of the algorithm to redistribute the positions of individuals in the population, thereby improving the uniformity and diversity of the population in the search space. Solution position update strategy: Based on the current best individual in the population and randomly selected individuals, update and adjust the positions of individuals in the population to move the population towards the potential optimal solution region, thereby obtaining the final initialized star sparrow population.

[0009] An air conditioning load information management system based on a B / S architecture, characterized in that it is used to run the air conditioning load information management method based on a B / S architecture according to any one of claims 1-5 of the present invention, specifically configured as follows: The data acquisition layer collects information by having users upload load and meteorological data files. The data storage layer uses a MySQL database to store user information, model files, uploaded data, and prediction results. The service layer is responsible for implementing the core functions of the system, covering model construction, training, evaluation and optimization, while also providing interface support for external data sources; The transport layer is responsible for the secure transmission of data. Efficiency and confidentiality can be improved through encryption, compression, and transmission protocol optimization. The presentation layer is dedicated to providing a user-friendly interface experience, using HTML, CSS, and JavaScript technologies to build the front-end interface, which includes modules such as login, prediction result display, and data management. The access layer is the entry point for users to log in to the system via PC. It is mainly responsible for authentication and access control to ensure system security and the privacy of user information.

[0010] The beneficial effects of this invention are: The system adopts a common browser / server (B / S) architecture, dividing the overall system into two main parts: client and server. In the air conditioning load forecasting system, this architecture is further subdivided into six layers. The data acquisition layer primarily collects information through user-uploaded load and meteorological data files; the data storage layer utilizes a MySQL database to store user information, model files, uploaded data, and forecast results. The service layer is responsible for implementing the core functions of the system, covering model construction, training, evaluation, and optimization, while also providing interface support for external data sources. The transmission layer is responsible for secure data transmission, improving efficiency and confidentiality through encryption, compression, and transmission protocol optimization. The presentation layer focuses on providing a user-friendly interface experience, using HTML, CSS, and JavaScript technologies to build a front-end interface including modules for login, forecast result display, and data management. The access layer serves as the entry point for users to log in to the system via PC, primarily responsible for authentication and access control, ensuring system security and the privacy of user information. Users only need to understand the basic operating procedures to complete load forecasting and select air conditioning load periods through the device's interactive interface. The forecast results will generate tabular data for convenient subsequent analysis and processing. Attached Figure Description

[0011] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of the air conditioning load information management system; Figure 2 This is a system framework structure diagram; Figure 3 This is a screenshot of the login page; Figure 4 It is a data visualization chart before processing; Figure 5 It is a visualization of the processed data; Figure 6 This is a visualization of the point prediction results; Figure 7 This is a visualization of the interval prediction results; Figure 8 This is a diagram showing the design results for air conditioning load during different time periods. Figure 9 The prediction results are shown in the image. Figure 10 A graph showing the prediction error; Figure 11 Here is a flowchart of the INOA-LSTM load forecasting process; Figure 12 Here is a flowchart of the Bootstrap interval prediction process; Figure 13This is a flowchart showing the selection process for air conditioning systems that meet design time periods; Figure 14 This is a diagram showing the results of random initialization; Figure 15 This is a diagram showing the initialization results of the Tent chaotic mapping; Figure 16 It is an adaptive factor Change curve graph. Detailed Implementation

[0012] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0013] Part One: 1. System framework design: combining Figure 1-3 This system adopts a common browser / server (B / S) architecture, dividing the overall system into two main parts: client and server. In the air conditioning load forecasting system, this architecture is further subdivided into six layers. The data acquisition layer primarily collects information through user-uploaded load and meteorological data files; the data storage layer utilizes a MySQL database to store user information, model files, uploaded data, and forecast results. The service layer is responsible for implementing the core functions of the system, covering model construction, training, evaluation, and optimization, while also providing interface support for external data sources. The transmission layer is responsible for the secure transmission of data, improving efficiency and confidentiality through encryption, compression, and transmission protocol optimization. The presentation layer focuses on providing a user-friendly interface experience, using HTML, CSS, and JavaScript technologies to build the front-end interface, including modules for login, forecast result display, and data management. The access layer is the entry point for users to log in to the system via PC, primarily responsible for authentication and access control, ensuring system security and the privacy of user information.

[0014] 2. Database Design: To support the efficient operation of the load forecasting system, the database design aims to build an optimized data storage and management architecture to ensure the effective storage, management, and retrieval of data related to air conditioning load, forecasting models, user information, and system configuration. Based on different data types and their functional roles in the system, the database is divided into five functional module data tables. The specific data table structure design is shown below: (1) User Information Table In the database design of the load forecasting system, the user information table is used to store basic data of system users, covering core fields such as login account, user identity and permissions, and contact information, to support user identification and permission management functions. Typical fields are shown in Table 1.

[0015] Table 1 User Information Table

[0016] (2) Load data acquisition table The main fields of this data table include the prediction time, the predicted value, and the type of prediction model applied. Its detailed structure is shown in Table 2. Table 2 Data Collection Table

[0017] (3) Prediction Result Table To avoid the time overhead of repeatedly performing model training when querying the same prediction results, thus reducing system efficiency, the system design stores relevant information for each prediction action in a database for easy access and retrieval by users. This data table mainly includes fields such as prediction time, prediction result, and the name of the prediction model used.

[0018] The specific design is shown in Table 3.

[0019] Table 3 Prediction Results

[0020] Part Two: Data Description: This study uses a university laboratory in Xi'an, Shaanxi Province as the research object. Parameters include outdoor meteorological parameters, indoor disturbance parameters, and load data. To facilitate observation of data distribution, the raw data is visualized, such as... Figure 4 As shown in the figure. Outdoor meteorological parameters include daily average dry-bulb temperature, horizontal visibility, real-time outdoor temperature, wind direction, real-time outdoor humidity, total solar radiation at horizontal level, average wind speed, weather conditions, surface air pressure, and wind speed. Indoor disturbance parameters include indoor occupancy density. Load data includes real-time load data and historical load data from the previous moment. The air conditioning operation mode is intermittent, meaning it only operates during working hours, with one sampling point every hour. A total of 736 sets of hourly data were recorded daily from 9:00 AM to 12:00 PM and from 2:00 PM to 6:00 PM during the summer air conditioning season of 2023, from June 1st to August 31st.

[0021] Introduction to main research methods: (1) Spearman correlation coefficient method: In order to optimize the input features and improve the interpretability of the model, the above parameters need to be correlated before predicting the air conditioning load. Commonly used correlation analysis methods include Pearson correlation analysis and Spearman correlation coefficient method, among which Spearman correlation coefficient method is a non-parametric statistical method. This method does not have strict requirements on the data distribution and is particularly suitable for analyzing the monotonic relationship between variables. Compared with Pearson correlation analysis, Spearman correlation coefficient method does not rely on the linear assumption of the data and can provide more accurate results when the data distribution is more complex or there are outliers. Therefore, this paper selects Spearman correlation coefficient method for the selection of input variables and visualizes the processed data, such as Figure 5 As shown.

[0022] (2) Long Short-Term Memory (LSTM) is a special type of deep learning neural network, which is developed by improving RNN and has powerful time series prediction memory capabilities. Compared with RNN, LSTM has a deeper architecture. This architecture introduces the concept of storage units to determine the information that needs to be retained and discarded. LSTM selectively transmits information through the storage unit structure consisting of forget gate, input gate and output gate.

[0023] (3) The Nutcracker Optimization Algorithm (NOA) is a swarm intelligence optimization algorithm inspired by the foraging behavior of larks in nature. Larks explore food resources in a vast search space, store the discovered resources in appropriate locations, and retrieve food from the storage locations according to their needs. This behavior demonstrates the larks' balancing strategy between global search and local exploitation. NOA abstracts this process into an optimization mechanism, where foraging behavior corresponds to global search, and resource storage and retrieval behavior embodies local exploitation. By simulating this intelligent behavior, NOA efficiently explores the solution space of the problem globally, while simultaneously performing deep optimization on high-quality solutions. In LSTM hyperparameter optimization problems, NOA is used to find the optimal combination of model parameters (such as learning rate, number of hidden layer units, batch size, etc.) to improve the model's performance and generalization ability.

[0024] The NOA algorithm consists of four phases: foraging phase (search), storage phase, storage-search phase, and optimal development phase, in order to find the optimal solution to the globally optimal problem. Phase 1: Foraging Phase In the initial stage of the NOA algorithm, a target solution is generated. The initial value is calculated using the following formula: ; In the formula, and Latitude The spatial boundary, (wherein) (D represents the total number of latitudes). This is a random value between 0 and 1. The starbird will then bring suitable seeds back to its seed storage location if it finds them. If the starbird cannot find a good seed, it will update its position to search for one. This behavior can be mathematically modeled using a position update strategy, calculated as follows: ; In the formula, It is the current The Middle The new position of the star sparrow It is the first The first star sparrow One location, and It is the first of the optimization objective parameters The upper and lower bounds of a dimension; For random values, middle This is the th best solution obtained so far. Dimensions A, C, and B are three different indicators randomly selected from the population to explore high-quality food sources. , , and It is a random number in the range [0, 1]. It is iteration The first of all solutions in the current population The average value of the dimension. Based on normal distribution Levy Flight random values ​​between 0 and 1 The formula for calculating the randomly generated number is as follows: ; In the formula, and It is a random number in the range [0,1].

[0025] Phase Two: Storage Phase The starling begins by transporting food obtained in the previous foraging phase to a temporary storage location. This behavior can be represented as: ; In the formula, The current iteration Location of food storage for the finches It is based on the data generated by Levy's flight. It is a random number between 0 and 1. It is a number that decreases linearly from 1 to 0.

[0026] Phase 3: Storage Search When food is scarce, starfinches switch from search mode to exploration and scouting mode. This phase is generally referred to as the second exploration. Starfinches use spatial memory strategies to locate their food. These objects are defined as reference points ( In NOA, each starbird in the population has two... The following matrix can be used for definition: ; In the formula, and Indicates the current The generation The cache location of each star sparrow of The starbird can thus accurately locate food. The first one... It is generated by updating the current position within the adjacent region to find the hidden cache around the starbird. The first one is generated. The mathematical formula is: ; The second It is generated by updating the current solution within the problem's search space; the second one... Calculate using the following formula: ; ; In the formula, for index; , For the current iteration The Middle The cache location of each star sparrow and They are Upper and lower bounds of the dimension problem; Decrease linearly from 1 to 0; It's the second one. A random number in the range [0,1]. It is a random value in the range [0, π]. It is used to determine the percentage of other regions to be explored globally within the search space.

[0027] Phase 4: Optimal Development When a lark searches for its cache, it may encounter two situations: the food is present or it is not. This behavior can be mathematically modeled using the following equation: ; ; In the formula, For the first Each star sparrow is iterating The new position of the time For the current iteration The Middle The current dimension of each star sparrow. For iteration The best position in the middle For the current iteration The Middle The first star sparrow at its current position ; , , The result is a random number between 0 and 1. Finally, output the optimal value. .

[0028] (4) Bootstrap residual resampling is an improved strategy based on traditional Bootstrap resampling. This method calculates the residual between the model estimate and the actual load value, and randomly adds these residuals to the fitted output to generate new response variables to construct a resampled dataset.

[0029] (5) The random forest algorithm assesses the relative importance of each meteorological factor to the air conditioning load by analyzing the contribution of a single meteorological factor to the improvement of the purity of all decision tree nodes. After the model training is completed, the variable importance of each meteorological parameter is evaluated, and the results are normalized so that the sum of the importance of all parameters is 1. The analysis of the characteristic importance of meteorological parameters helps to quantify the contribution of each parameter to energy performance prediction, and the mean squared error or out-of-bag error is usually used as the evaluation basis.

[0030] Part Three: To further investigate the predictive performance of the INOA-LSTM model, experiments were conducted using NOA, Gray Wolf Optimizer (GWO), Sparrow Search Algorithm (SSA), Whale Optimization (WOA), and Particle Swarm Optimization (PSO). Using the same dataset, their prediction results were compared, and the performance of each model was analyzed. The prediction results and errors of each model are shown below. Figure 9 and Figure 10 As shown.

[0031] Figure 11 The flowchart for INOA-LSTM load forecasting shows the specific implementation steps of the load forecasting layer, including steps 1 to 9. Figure 12 This is the Bootstrap interval prediction flowchart, which includes the following steps 10 to 12; Figure 13 This is a flowchart for selecting air conditioning systems that meet design time periods, including step 13.

[0032] Step 1: Collect raw data. Taking a university laboratory in Xi'an, Shaanxi Province as the research object, the parameters include outdoor meteorological parameters, indoor disturbance parameters, and load data. Outdoor meteorological parameters include daily average dry-bulb temperature, horizontal visibility, real-time outdoor temperature, wind direction, real-time outdoor humidity, total solar radiation at horizontal level, average wind speed, weather conditions, surface air pressure, and wind speed. Indoor disturbance parameters include indoor occupancy density. Load data includes real-time load data and historical load data from the previous moment. The air conditioning system operates intermittently, meaning it only operates during working hours, with one sampling point every hour. A total of 736 hourly data sets were recorded from 9:00 AM to 12:00 PM and from 2:00 PM to 6:00 PM daily during the summer air conditioning season of 2023, from June 1st to August 31st. Step 2: Select input variables using the Spearman correlation coefficient method. Choose variables with higher importance as input parameters for the model. Then, divide the dataset into training and test sets based on the selected influential parameters; Step 3: Initialize the starbird population using three strategies, including OBL (Opposite Learning). The specific process is as follows: OBL Opposite Learning Strategy: For each initially generated starbird individual, construct corresponding opposite individuals and evaluate the fitness of both the original and opposite individuals, selecting the individuals with better fitness to form the optimized population; Tent Chaotic Mapping Strategy: Use Tent chaotic mapping to generate a chaotic sequence and map this sequence to the algorithm's search space, redistributing the positions of population individuals to improve the uniformity and diversity of the population in the search space; Solution Position Update Strategy: Update and adjust the positions of population individuals based on the current best individual and randomly selected individuals, moving the population towards potential optimal solution regions, thereby obtaining the final initialized starbird population. These three improved strategies effectively enhance the performance of the starbird optimization algorithm.

[0033] Specifically, the OBL (Optimal Learning Blockchain) opposition learning strategy involves constructing corresponding opposition individuals for each initially generated starbird population, evaluating the fitness of both the original and opposition individuals, and selecting individuals with better fitness to form the optimized population. For the current population, the opposing species cluster obtained based on OBL The calculation formula is as follows: ; In the formula, Indicates the first individual The corresponding individual in the opposing space, that is, the opposing individual. and These represent the upper and lower limits of the variable's value range, respectively.

[0034] Tent chaotic mapping strategy: A chaotic sequence is generated using Tent chaotic mapping, and this chaotic sequence is mapped to the algorithm's search space to redistribute the positions of individuals in the population, thereby improving the uniformity and diversity of the population within the search space; Tent chaotic mapping generates chaotic sequences. The expression is as follows: ; In the formula, Let represent the chaotic value of the i-th dimension of the i-th individual generated after the Tent chaotic mapping transformation. This represents the seed value, which is randomly generated within the range of 0-1. This represents the Tent mapping control parameter, used to divide the segmentation points of the mapping function. This paper sets this parameter to 0.6 to control the random generation range of each element in the initial solution vector, limiting it to the interval centered on the first randomly generated value, thereby increasing the likelihood that the initial population is close to the solution space of the actual problem.

[0035] Individuals in the population after chaotic mapping can be represented as: ; In the formula, Indicates the first The individual The actual value of each dimension and They are respectively The upper and lower limits.

[0036] Solution position update strategy: Based on the current best individual in the population and randomly selected individuals, the positions of individuals in the population are updated and adjusted, causing the population to move towards the potential optimal solution region, thereby obtaining the final initialized starbird population. In the standard NOA algorithm, due to poor control over the step size, the search position update process during initial value generation can easily deviate from the search direction, miss the optimal foraging region, and cause the algorithm to get stuck in local optima. Therefore, this paper proposes an adaptive inertia weight factor ω to improve the position update of generated solutions. Its calculation formula is as follows: ; In the formula, It is the maximum number of iterations. This is the current iteration number.

[0037] During the iteration process, inertia weight The continuous adaptive changes help to improve the balance between the algorithm's search space and its development capability. The improved formula for generating the initial solution is as follows: ; Figure 14 and Figure 15 The comparison between the population distribution generated by the Tent chaotic mapping and random initialization is shown, and it can be seen that the initial population distribution generated by the Tent mapping is more uniform. Figure 16 The curve of the adaptive inertia weight factor ω changing with the number of iterations is shown; Table 4 shows the performance comparison of INOA and NOA on the standard test function, and the results show that INOA is better than the original NOA in terms of convergence speed and optimization accuracy.

[0038] Table 4. Test Results of INOA and NOA

[0039] Step 4: Set parameters such as the maximum number of NOA iterations and population size, and update the initial values ​​of the solution location optimization. Set the upper and lower limits of the following parameters in LSTM: Learning Rate, Batch Size, L2 Regularization Coefficient, Number of Epochs, and Number of Hidden Units. Step 5: Initialize the LSTM network and train it using the training dataset obtained in Step 2, calculating the initial individual fitness and determining the optimal position of each individual. Update the solution positions, and the star sparrow begins searching for other solutions in the collection area, searching for the location of the seed there, and bringing the found seed to the storage area for storage; Step 6: The starbird begins searching for the location of the stored seeds using a memory strategy, and finds the optimal solution among the stored seeds. The latest hyperparameters are input into the LSTM network. The individual fitness of the seeds is calculated to determine the optimal location; Step 7: Compare the fitness of the best seed and update the optimal seed position; Step 8: Repeat steps (6)-(7) until the maximum number of iterations is reached; Step 9: Output the optimal parameters and train an LSTM network using the optimal hyperparameter combination for air conditioning load prediction. The prediction results are as follows: Figure 6 As shown; Step 10: First, determine the original dataset. And divide the training set and test set accordingly; Step 11: Train the INOA-LSTM model on the training set of the original dataset to obtain a set of predicted values. Based on this predicted value Compared with actual value residual value Measure output dimension: , To predict the output dimension; use Bootstrap to process the residual values. , ,conduct A second resampling with replacement is performed to obtain a new set of residuals. , and combine the residual groups with the dataset By adding them together, we obtain a completely new dataset. For a completely new dataset Then, the INOA-LSTM model is trained and predicted again; the resampling is repeated B times to obtain B new datasets, and B INOA-LSTM models are trained at the same time. Step 12: After using the Bootstrap residual resampling method, a brand new dataset for group B was obtained, as shown below: ; In the formula, Let represent the b-th dataset. According to the Bootstrap principle, averaging the predicted values ​​of dataset B yields the final prediction result, expressed as follows: ; In the formula, Indicates the first The first dataset One predicted value.

[0040] However, to obtain the confidence interval of the model, it is also necessary to estimate the uncertainty variance of the model, the expression of which is shown below: ; In the formula, This represents the model variance, which is the variance caused by differences in the model's parameters during training, given a confidence level of [value missing]. In this case, the confidence interval of the model can be obtained as follows: ; However, in practical engineering applications, the confidence intervals constructed often fail to meet the requirements. To construct intervals that meet the requirements, estimation of the noise variance is essential. Therefore, the noise variance is estimated using a resampled dataset, and its expression is as follows: ; In the formula, This represents noise error, but since noise cannot be negative, the noise variance estimate is... Since the noise variance set contains a large number of zero values, in order to reduce the impact of the randomness of the noise variance, the predicted values ​​in the dataset are replaced with the noise variance, thus obtaining a new dataset. .

[0041] By analyzing the dataset To obtain the noise variance, a new model is trained. Therefore, in order to maximize the probability of noise variance occurring in the load observation samples, the new model can be trained using the maximum likelihood method. The objective function is shown in the following equation: After estimating the noise variance and model variance, the corresponding upper and lower limits of the interval can be calculated to quantify the uncertainty of the prediction results. The interval prediction results are as follows: Figure 7 As shown.

[0042] Step 13: By calculating the deviation between the daily cumulative distribution function (CDF) values ​​of various meteorological parameters for each time period (i.e., 9:00-12:00 AM and 3:00-6:00 PM) and the CDF values ​​of long-term meteorological data at the 50%-90th percentile, the appropriate weights are selected based on the RF method. Then, the weighted sum (WS) of the FS values ​​of each parameter is calculated, and the candidate with the smallest WS is selected as the design period for air conditioning load. The selection results for the design period are as follows: Figure 8 As shown. The weighted total of FS data is calculated as follows: ; ; ; In the formula, Meteorological parameters exist of value; For Heaven; For a period of time; Meteorological parameters exist of value; The number of possible values ​​for a certain meteorological parameter The number of meteorological parameters selected; For each Value at sky Time-weighted sum mean; Meteorological parameters The weight value.

[0043] Through the above modules and specific implementation steps, a deep learning-based air conditioning load information management system with a B / S architecture can be obtained. Furthermore, the prediction algorithm of the load prediction layer can be modified to achieve other high-precision, parallel prediction algorithms.

[0044] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for managing air conditioning load information based on a B / S architecture, characterized in that, This includes constructing the INOA-LSTM model, specifically: Step 1: Collect raw data; parameters include outdoor meteorological parameters, indoor disturbance parameters and load data. Outdoor meteorological parameters include outdoor daily average dry-bulb temperature, horizontal visibility, outdoor real-time temperature, wind direction, outdoor real-time humidity, total solar radiation on the horizontal surface, average wind speed, weather conditions, surface air pressure, and wind speed. Indoor disturbance parameters include indoor occupancy density. Load data includes real-time load data and historical load data from the previous moment. Step 2: Select input variables using the Spearman correlation coefficient method, and divide the dataset into training and test sets; initialize the sparrow population using the Tent chaotic mapping, mapping the individual positions of the stored seeds to the hyperparameters to be optimized; set parameters such as the maximum number of NOA iterations and population size, and generate the initial values ​​for solution position update optimization; set the upper and lower limits for parameters such as learning rate, batch size, L2 regularization coefficient, number of training rounds, and number of hidden units in the LSTM; initialize the LSTM network, train it using the obtained training dataset, calculate the initial individual fitness, and determine the optimal position of the individual; the sparrows begin searching for other solutions in the collection area, searching for the position of the seed on it, and bringing the found seeds to the storage area for storage; Step 3: The starbird begins to find the location of the stored seeds using a memory strategy and finds the optimal solution among the stored seeds; the latest hyperparameters are input into the LSTM network to calculate the individual fitness of the seeds and determine the optimal location; the optimal seed fitness is compared and the optimal seed location is updated; this continues until the maximum number of iterations is reached; the optimal parameters are output and the LSTM network is trained using the optimal hyperparameter combination for air conditioning load prediction.

2. The air conditioning load information management method based on B / S architecture according to claim 1, characterized in that, Specifically, it includes; Determine the original dataset The dataset is divided into training and test sets. The INOA-LSTM model is then used to train the training set of the original dataset, resulting in a set of predicted values. Based on this predicted value Compared with actual value residual value Measure the output dimension: , To predict the output dimension; use Bootstrap to process the residual values. , ,conduct A second resampling with replacement is performed to obtain a new set of residuals. Combine residual groups with dataset By adding them together, we obtain a completely new dataset. For a completely new dataset Then, the INOA-LSTM model is trained and predicted again; the resampling is repeated B times to obtain B new datasets, and B INOA-LSTM models are trained at the same time. After using the Bootstrap residual resampling method, a brand new dataset for group B was obtained, as shown below: ； In the formula, Let b represent the b-th dataset. According to the Bootstrap principle, the final prediction result can be obtained by averaging the predicted values ​​of the b-th dataset, as shown in the following expression: ； In the formula, Indicates the first The first dataset One predicted value.

3. The air conditioning load information management method based on B / S architecture according to claim 2, characterized in that, It also includes an assessment of the uncertainty of air conditioning load forecast results, specifically: Replace the predicted values ​​with the noise variance in the dataset to obtain a new dataset. ; : This is noise error; By analyzing the dataset To train a new model, the objective function is as follows: After estimating the noise variance and model variance, the corresponding upper and lower limits of the interval can be calculated to quantify the uncertainty of the prediction results.

4. The air conditioning load information management method based on B / S architecture according to claim 2, characterized in that, This also includes the selection of air conditioning load design periods, specifically: By calculating the deviation between the daily cumulative distribution function (CDF) values ​​of various meteorological parameters for each time period and the CDF values ​​of long-term meteorological data at the 50%-90th percentile, appropriate weights are selected based on the RF method. Then, the weighted sum (WS) of the FS values ​​of each parameter is calculated, and the candidate with the smallest WS is selected as the air conditioning load design period. The weighted sum of the FS data is calculated as follows: ； ； ； In the formula, Meteorological parameters exist of value; For Heaven; For a period of time; Meteorological parameters exist of value; The number of possible values ​​for a certain meteorological parameter The number of meteorological parameters selected; For each Value at sky Time-weighted sum mean; Meteorological parameters The weight value.

5. The air conditioning load information management method based on B / S architecture according to any one of claims 1-4, characterized in that, In step 2, three strategies, including OBL, are used to initialize the starbird population, specifically including: OBL Opposite Learning Strategy: For the initially generated starbird population, construct corresponding opposite individuals, evaluate the fitness of the original individuals and the opposite individuals, and select individuals with better fitness to form the optimized population. Tent Chaotic Mapping Strategy: Use Tent chaotic mapping to generate chaotic sequences and map the chaotic sequences to the search space of the algorithm to redistribute the positions of individuals in the population, thereby improving the uniformity and diversity of the population in the search space. Solution position update strategy: Based on the current best individual in the population and randomly selected individuals, update and adjust the positions of individuals in the population to move the population towards the potential optimal solution region, thereby obtaining the final initialized star sparrow population.

6. An air conditioning load information management system based on a B / S architecture, characterized in that, For running the air conditioning load information management method based on B / S architecture as described in any one of claims 1-5, the specific settings are as follows: The data acquisition layer collects information by having users upload load and meteorological data files. The data storage layer uses a MySQL database to store user information, model files, uploaded data, and prediction results. The service layer is responsible for implementing the core functions of the system, covering model construction, training, evaluation and optimization, while also providing interface support for external data sources; The transport layer is responsible for the secure transmission of data. Efficiency and confidentiality can be improved through encryption, compression, and transmission protocol optimization. The presentation layer is dedicated to providing a user-friendly interface experience, using HTML, CSS, and JavaScript technologies to build the front-end interface, which includes modules such as login, prediction result display, and data management. The access layer is the entry point for users to log in to the system via PC. It is mainly responsible for authentication and access control to ensure system security and the privacy of user information.

Images