Dynamic power management system based on central air conditioning load characteristics

Abstract

The application relates to the technical field of central air conditioning system management, and discloses a dynamic power management system based on central air conditioning load characteristics, which mainly comprises a load characteristic analysis module, a power demand prediction module, an energy distribution optimization module and a control execution module. The load characteristic analysis module classifies load characteristics by analyzing historical operation data of the central air conditioning system; the power demand prediction module predicts future power demand in combination with load characteristic classification results and weather forecast data; the energy distribution optimization module generates an optimal energy distribution scheme according to predicted power demand and current power supply conditions by using a genetic algorithm; and the control execution module adjusts the operation of the central air conditioning system according to the optimization results. The system accurately analyzes load characteristics, predicts power demand and optimizes energy distribution, effectively reduces power cost, meets the power demand of the central air conditioning system, improves energy utilization efficiency, and has remarkable economic and environmental protection benefits.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of central air conditioning system management, in particular to a dynamic power management system based on central air conditioning load characteristics. BACKGROUND

[0002] With the continuous expansion of modern building scale and the increasing complexity of functions, central air conditioning system as an important part of building energy consumption, its energy consumption problem is increasingly prominent. Central air conditioning system in the process of operation, its load characteristics are influenced by many factors, such as indoor and outdoor temperature, humidity, temperature difference change, passenger flow and equipment running state, etc. These factors lead to the dynamic change of the load demand of central air conditioning system, and the traditional static power management method has been difficult to meet the demand of high efficiency and energy saving.

[0003] In the traditional central air conditioning power management, fixed operation strategy and energy distribution scheme are often used, which cannot be dynamically adjusted according to the actual load change. This management method may lead to energy waste when the load is low, and may not meet the power demand of the system when the load is high, thereby affecting the normal operation of the central air conditioning system and the comfort in the building.

[0004] In order to solve this problem, people began to study dynamic power management system based on load characteristics. The existing system mostly only adjusts according to the current load state, lacks prediction of future load change and optimization of energy distribution. This leads to the system unable to respond in time when facing sudden load change, and still has the problem of low energy utilization efficiency. Therefore, it is urgent to need a power management system which can be dynamically managed based on the load characteristics of central air conditioning. The system needs to be able to analyze the load characteristics of central air conditioning system in real time, predict the future power demand, and optimize the energy distribution scheme according to the current power supply condition, so as to realize the minimization of power cost and meet the power demand of central air conditioning system. SUMMARY

[0005] The purpose of the present application is to provide a dynamic power management system based on the load characteristics of central air conditioning to solve the problems raised in the background.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: a dynamic power management system based on the load characteristics of central air conditioning, the system comprises:

[0007] A load characteristic analysis module is configured to build a load characteristic analysis model, wherein the input data of the load characteristic analysis model is historical operation data of the central air conditioning system, including temperature, humidity, indoor-outdoor temperature difference, passenger flow, and equipment operation state, and the output data is a load characteristic classification result of the central air conditioning system, including high load, medium load, and low load; and the model training uses a central air conditioning system operation data set containing different load states, and the corresponding label is a load state category.

[0008] A power demand prediction module is configured to build a power demand prediction model, wherein the input data of the power demand prediction model is the output result of the load characteristic analysis module and weather forecast data, and the output data is a predicted power demand of the central air conditioning system in a future period of time; and the model training uses a time series data set containing the load characteristic classification result and weather forecast data, and the corresponding label is the actual power consumption of the central air conditioning system at each time point.

[0009] An energy distribution optimization module is configured to build an optimization model using a genetic algorithm, wherein the input data of the optimization model is the prediction result of the power demand prediction model and the current power supply condition, and the output data is an optimal energy distribution scheme to minimize the power cost and meet the power demand of the central air conditioning system; and the model training uses a data set containing predicted power demand and actual power supply condition, and the corresponding label is a cost and efficiency index of energy distribution.

[0010] A control execution module is connected to the energy distribution optimization module and adjusts the operation parameters and energy distribution of the central air conditioning system according to the optimization result.

[0011] Preferably, the load characteristic analysis model is built using a support vector machine (SVM) algorithm, and the network structure thereof includes:

[0012] An input layer is configured to receive historical operation data of the central air conditioning system.

[0013] A feature mapping layer is configured to map the input data to a high-dimensional space through an RBF kernel function.

[0014] A support vector layer is configured to find an optimal hyperplane in the high-dimensional space for classifying load characteristics.

[0015] An output layer is configured to output the load characteristic classification result of the central air conditioning system.

[0016] Preferably, the training steps of the load characteristic analysis model include:

[0017] S101: initializing the parameters of the SVM model, including kernel function parameters, penalty parameters, and learning rate.

[0018] S102: dividing the data set for training into a training set and a validation set.

[0019] S103: iteratively train the load characteristic analysis model using the training set, in each iteration, calculate the model output and calculate the loss according to the loss function; the loss function adopts the Hinge loss function, and the calculation formula is:

[0020]

[0021] wherein y i is the real class label, f(x i ) is the output of the model decision function, and N is the number of samples;

[0022] S104: update the model parameters by gradient descent method according to the loss value;

[0023] S105: repeat steps S103 to S104 until the loss converges or the preset number of iterations is reached;

[0024] S106: verify the trained model using the verification set to evaluate the classification accuracy of the model;

[0025] S107: when the model performance reaches the preset standard, save the model parameters to obtain the trained load characteristic analysis model.

[0026] Preferably, the power demand prediction model is constructed using a feedforward neural network FNN algorithm, and the network structure includes:

[0027] Input layer: receiving load characteristic classification results and weather forecast data;

[0028] Hidden layer: containing multiple layers of neurons, each layer of neurons is followed by a ReLU activation function, for feature extraction;

[0029] Output layer: output the power demand prediction value of the central air conditioning system in the future period.

[0030] Preferably, the training steps of the power demand prediction model include:

[0031] S201: standardize the input data;

[0032] S202: initialize the parameters of the FNN model, including weights, bias terms and learning rate;

[0033] S203: divide the data set into training set and verification set;

[0034] S204: iteratively train the model using the training set, in each iteration, calculate the model output and calculate the loss according to the loss function;

[0035] S205: update the model parameters by backpropagation algorithm and gradient descent method according to the loss value;

[0036] S206: repeating the steps S204 to S205 until the loss converges or the preset number of iterations is reached.

[0037] S207: verifying the trained model using the verification set to evaluate the prediction accuracy of the model.

[0038] S208: saving the model parameters when the model performance reaches the preset standard to obtain the trained power demand prediction model.

[0039] Preferably, in the step S204, the mean square error (MSE) function is used as the loss function, and the calculation formula is:

[0040]

[0041] where y i is the true power demand of the i-th sample, is the predicted power demand of the i-th sample, and M is the number of samples.

[0042] Preferably, the genetic algorithm of the energy distribution optimization model includes:

[0043] Initialization step: generating an initial population, each individual representing an energy distribution scheme;

[0044] Fitness evaluation step: calculating the fitness of each individual according to the power demand prediction results and the current power supply situation, representing the cost and efficiency indicators of energy distribution;

[0045] Selection step: using the roulette wheel selection method to select individuals for reproduction according to the fitness;

[0046] Crossing step: using single-point crossing method to generate new individuals;

[0047] Mutation step: randomly mutating part of the individuals to increase population diversity;

[0048] Iteration step: repeating the fitness evaluation, selection, crossing, and mutation steps until the preset number of iterations is reached or the optimal solution is found;

[0049] Output step: outputting the optimal energy distribution scheme.

[0050] Preferably, the calculation formula of the fitness is:

[0051] F = αC + βE

[0052] where C is the total cost of energy distribution, E is the efficiency indicator of energy distribution, and α and β are weight coefficients used to balance the importance of cost and efficiency.

[0053] Preferably, the time resolution of the historical operation data is not less than once per hour, and the data quality is pre-processed, including missing value filling and outlier detection.

[0054] Preferably, the data monitoring module is further included for monitoring the operation data and power supply status of the central air conditioning system in real time, and inputting the data to the load characteristic analysis module and the power demand prediction module to realize closed-loop control of dynamic power management.

[0055] Preferably, the training step of the strategy adjustment model comprises:

[0056] Step 301: Collect training data, including the difference between the planning result output by the path planning model and the actual behavior of the logistics robot in the task execution process, including path deviation, time delay and energy consumption difference; at the same time, collect the optimal task execution strategy labeled by experts under the given planning result and actual behavior difference as the label of the training data;

[0057] Step 302: Construct a random forest RF algorithm model, set the parameters of RF, including the number of trees, the maximum features and the maximum depth;

[0058] Step 303: Train the RF algorithm model using the training data, improve the model performance through bagging method, and make the model accurately predict the optimal task execution strategy under the given input difference;

[0059] Step 304: After training, the strategy adjustment model is obtained.

[0060] Preferably, the specific execution strategy output by the task execution strategy generation model is generated according to the optimal path output by the path planning model, and the logistics knowledge content related to the logistics robot is created and displayed, including strategy cases, operation specifications and safety guidelines for different path planning.

[0061] Preferably, the present application further comprises a multi-robot cooperation module, which utilizes the path planning model to plan the optimal path avoiding collision and efficient cooperation according to the real-time state and position information of multiple logistics robots, and further utilizes the task execution strategy generation model to generate the cooperative execution strategy for multiple robots according to the planned path.

[0062] Compared with the prior art, the present application has the following beneficial effects:

[0063] The present application accurately classifies the load state (high load, medium load, low load) of the central air conditioning system through the load characteristic analysis module, providing accurate basic data for subsequent energy management. The power demand prediction module combines load characteristics and weather forecast data to accurately predict power demand in the future, thereby avoiding energy waste and excess supply. The energy distribution optimization module uses a genetic algorithm to dynamically adjust the energy distribution scheme based on the prediction results and current power supply conditions, ensuring that the power demand of the central air conditioning system is met while minimizing power costs, significantly improving energy utilization efficiency.

[0064] The system can analyze the load characteristics of the central air conditioning system in real time and dynamically adjust the power demand and energy distribution scheme according to load changes and fluctuations in external environment (such as weather). This dynamic adjustment capability enables the system to adapt to energy management needs in different scenarios, whether it is a commercial building, office space or public facility, and can achieve efficient energy utilization. Through accurate load characteristic analysis and power demand prediction, the system can avoid unnecessary energy waste and reduce electricity costs during peak power periods. The energy distribution optimization module aims to minimize power costs to reduce overall operating costs and improve economic efficiency while meeting demand.

[0065] The system can ensure that the central air conditioning system operates stably under various load conditions and provides a comfortable indoor environment. By dynamically adjusting operating parameters and energy distribution, the system can reduce the risk of failure and energy shortages, improving user satisfaction and comfort. The present application helps to reduce energy consumption and carbon emissions by improving energy utilization efficiency and reducing operating costs, in line with the concept of sustainable development. In the context of global energy shortages and increasing environmental protection awareness, the promotion and application of the system will play a positive role in promoting green buildings and sustainable development. BRIEF DESCRIPTION OF DRAWINGS

[0066] Fig. 1 The working principle diagram of the dynamic power management system based on the load characteristics of the central air conditioning system described in the present application;

[0067] Fig. 2 The training flowchart of the load characteristic analysis model;

[0068] Fig. 3 The training process diagram of the power demand prediction model. DETAILED DESCRIPTION

[0069] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0070] Please refer to Figs. 1-3 The present application provides a technical solution: a dynamic power management system based on the load characteristics of central air conditioning, the system includes:

[0071] Implementation of the load characteristic analysis module: a load characteristic analysis model is constructed, which takes the historical operation data of the central air conditioning system as input, including temperature, humidity, indoor and outdoor temperature difference, passenger flow and equipment running state and other parameters. Through data preprocessing and feature extraction, the historical operation data is converted into a format recognizable by the model. Use the central air conditioning system operation data set containing different load states to train the model, and the corresponding label is the load state category (high load, medium load, low load). After the model training is completed, according to the real-time input data, the load characteristic classification result of the central air conditioning system is output.

[0072] Implementation of the power demand prediction module: a power demand prediction model is constructed, and the input data of the model is the output result of the load characteristic analysis module and the weather forecast data. The load characteristic classification result and the weather forecast data are fused to form time series data. Use the time series data set containing the load characteristic classification result and the weather forecast data to train the model, and the corresponding label is the actual power consumption of the central air conditioning system at each time point. After the model training is completed, according to the real-time input data, the power demand of the central air conditioning system in the future period of time is predicted.

[0073] Implementation of the energy distribution optimization module: an optimization model is constructed using a genetic algorithm, and the input data of the model is the prediction result of the power demand prediction model and the current power supply condition. According to the predicted power demand and the current power supply condition, the optimization goal is set to minimize the power cost and meet the power demand of the central air conditioning system. Use the data set containing the predicted power demand and the actual power supply condition to train the optimization model, and the corresponding label is the cost and efficiency index of energy distribution. After the model training is completed, according to the real-time input data, the optimal energy distribution scheme is output.

[0074] Implementation of the control execution module: the control execution module is connected to the energy distribution optimization module and receives the optimization result as input. According to the optimization result, the running parameters and energy distribution of the central air conditioning system are adjusted to realize the optimal energy utilization and power cost control.

[0075] The application will be further described below in combination with Examples 1 to 3:

[0076] Example 1:

[0077] The load characteristic analysis model is constructed using the support vector machine (SVM) algorithm, and the detailed implementation is as follows:

[0078] 1.1 Network structure:

[0079] Input layer: receives historical operation data of the central air conditioning system, including temperature, humidity, indoor and outdoor temperature difference, passenger flow and equipment operation state, etc.

[0080] Feature mapping layer: the input data is mapped to a high-dimensional space through a radial basis function (RBF) kernel function, so as to find linearly separable features in the high-dimensional space.

[0081] Support vector layer: find the optimal hyperplane in the high-dimensional space, which can maximize the interval between different load characteristic categories, and is used for classifying load characteristics.

[0082] Output layer: outputs the load characteristic classification results of the central air conditioning system, including high load, medium load and low load.

[0083] 1.2 Training steps:

[0084] S101: initialize the parameters of the SVM model. This includes the parameters of the RBF kernel function (such as the value of γ), the penalty parameter C and the learning rate (if online learning or incremental learning is used). The selection of these parameters will affect the performance and convergence speed of the model.

[0085] S102: divide the data set used for training into training set and validation set. The training set is used to train the model, and the validation set is used to evaluate the performance of the model to ensure that the model can also perform well on unseen data.

[0086] S103: use the training set to iteratively train the load characteristic analysis model. In each iteration, the output of the model is calculated, and the loss is calculated according to the Hinge loss function. The calculation formula of the Hinge loss function is:

[0087]

[0088] Where y i is the true class label, f(x i ) is the output of the model decision function, and N is the number of samples. This loss function encourages the model to give higher confidence to correctly classified samples, while penalizing incorrectly classified samples.

[0089] S104: Update the model parameters by gradient descent or other optimization algorithms according to the loss value. This step aims to minimize the loss function, making the model more accurate in classifying load characteristics.

[0090] S105: Repeat steps S103 to S104 until the loss converges or reaches a preset number of iterations. During this process, the performance of the model will gradually improve, and the loss value will gradually decrease.

[0091] S106: Validate the trained model using the validation set. Evaluate the classification accuracy of the model, i.e., the proportion of correctly classified samples to the total number of samples. This step is used to verify the generalization ability of the model.

[0092] S107: When the model performance reaches the preset standard (such as classification accuracy exceeding a certain threshold), save the model parameters to obtain the trained load characteristic analysis model. This model can be used in actual load characteristic classification tasks.

[0093] Example 2:

[0094] The power demand prediction model constructed by the feedforward neural network (FNN) algorithm of the present application has the following detailed implementation method:

[0095] 2.1 Network structure:

[0096] Input layer: receives load characteristic classification results (such as high load, medium load, low load) and weather forecast data (such as temperature, humidity, wind speed, etc.). These data are used as input features of the model to predict future power demand.

[0097] Hidden layer: contains multiple layers of neurons, each followed by a ReLU (Rectified Linear Unit) activation function. The ReLU function can introduce nonlinearity, allowing the model to learn complex feature representations. The design of multiple hidden layers can extract high-level features from input data, improving the prediction ability of the model.

[0098] Output layer: outputs the predicted value of the power demand of the central air conditioning system in the future period. This prediction value is a continuous variable, representing the power consumption of the central air conditioning system in the future time period.

[0099] 2.2 Training steps:

[0100] S201: Standardize the input data. Since the dimensions and ranges of input data may be different, standardization can convert data to the same scale, improving the training efficiency and prediction accuracy of the model.

[0101] S202: Initialize the parameters of the FNN model. This includes weights, bias terms, and learning rate. Weights and bias terms are learnable parameters in the model, and learning rate controls the step size of model parameter updates.

[0102] S203: Divide the dataset into training set and validation set. The training set is used to train the model, and the validation set is used to evaluate the performance of the model, to ensure that the model can also perform well on unseen data.

[0103] S204: Use the training set to iteratively train the model. In each iteration, calculate the output of the model, and calculate the loss according to the mean square error (MSE) function. The formula of the MSE function is:

[0104]

[0105] where y i is the true power demand of the i-th sample, is the predicted power demand of the i-th sample, and M is the number of samples. The MSE function measures the average of the squared difference between the predicted value and the true value, and is a commonly used loss function in regression problems.

[0106] S205: According to the loss value, update the model parameters by backpropagation algorithm and gradient descent method. The backpropagation algorithm calculates the gradient of the loss function with respect to the model parameters, and the gradient descent method updates the parameters in the opposite direction of the gradient to minimize the loss function.

[0107] S206: Repeat steps S204 to S205 until the loss converges or reaches the preset number of iterations. During this process, the performance of the model will gradually improve, and the loss value will gradually decrease.

[0108] S207: Use the validation set to validate the trained model. Evaluate the prediction accuracy of the model, such as using mean square error, mean absolute error, etc. This step is used to verify the generalization ability of the model, to ensure that the model can accurately predict the power demand in actual application.

[0109] S208: When the performance of the model reaches the preset standard (such as the prediction accuracy is lower than a certain threshold), save the model parameters, and get the trained power demand prediction model. This model can be used in actual power demand prediction tasks, and provides accurate data support for energy distribution optimization.

[0110] Example 3:

[0111] The present application adopts genetic algorithm to construct energy distribution optimization model, and the specific steps include:

[0112] Initialization Step: Generate an initial population, each individual in the population represents an energy allocation scheme. These schemes may include the allocation of electricity to central air conditioning systems, the activation and deactivation strategies of backup power sources, etc. The size of the initial population can be determined according to the complexity of the problem and the computing resources to ensure sufficient search space and population diversity.

[0113] Fitness Evaluation Step: Calculate the fitness of each individual according to the power demand prediction results and the current power supply situation. The fitness function considers the total cost (C) and efficiency index (E) of energy allocation, and its calculation formula is:

[0114] F = αC + βE

[0115] Where C is the total cost of energy allocation, which may include electricity purchase cost, equipment operation and maintenance cost, etc.; E is the efficiency index of energy allocation, which may represent energy utilization rate, system stability, etc.; α and β are weight coefficients, used to balance the importance of cost and efficiency in fitness evaluation. By calculating the fitness, the pros and cons of each energy allocation scheme can be quantified, providing a basis for subsequent selection, crossover and mutation steps.

[0116] Selection Step: Use the roulette wheel selection method to select individuals for reproduction according to fitness. The roulette wheel selection method allocates selection probability according to the proportion of individual fitness, the higher the fitness of the individual, the greater the probability of being selected, so that its excellent genes are more likely to be passed on to the next generation.

[0117] Crossover Step: Use single-point crossover method to generate new individuals. Single-point crossover method randomly selects a crossover point, exchanges the gene fragments of two parent individuals after the point to generate two new offspring individuals. Through crossover operation, the excellent genes of parent individuals can be combined to generate offspring individuals with potentially better performance.

[0118] Mutation Step: Randomly mutate part of the individuals to increase population diversity. Mutation operation can randomly change some gene values in the individual, or use other mutation strategies to introduce new gene combinations. Through mutation operation, new areas in the search space can be explored to avoid the algorithm falling into local optimal solution.

[0119] Iteration Step: Repeat the fitness evaluation, selection, crossover and mutation steps until the preset number of iterations is reached or the optimal solution is found. In each iteration, a new population is generated, and through fitness evaluation, excellent individuals are selected for reproduction and mutation. Through continuous iteration, the energy allocation scheme can be gradually optimized, and the cost and efficiency performance of the system can be improved.

[0120] Output step: output the optimal energy allocation scheme with the highest fitness value, which maximizes efficiency while minimizing cost. The optimal scheme can serve as a decision basis for actual energy allocation, guiding the power dispatch and energy management of the central air conditioning system.

[0121] To ensure the accuracy of power demand prediction and energy allocation optimization, the time resolution of historical operation data is not less than once per hour, meaning that the system can obtain the hourly operation state data of the central air conditioning system, such as temperature, humidity, energy consumption, etc., providing detailed data basis for subsequent load characteristic analysis and power demand prediction. In historical operation data, due to various reasons (such as equipment failure, data transmission error, etc.), some data may be missing. The system uses appropriate methods (such as interpolation method, mean filling method or machine learning-based prediction filling method) to fill in the missing values to ensure the integrity and continuity of the data. The invention integrates an outlier detection mechanism to identify and process outliers in the data through statistical methods (such as 3σ rule) or machine learning algorithms (such as Isolation Forest, Support Vector Machine, etc.). Outliers may be caused by equipment failure, operation error or external interference, etc., and their processing can improve the quality and reliability of the data.

[0122] To realize the closed-loop control of dynamic power management, the invention also includes a data monitoring module, which has the following specific functions:

[0123] Real-time monitoring: The data monitoring module can collect real-time operation data and power supply conditions of the central air conditioning system. These data include but are not limited to system energy consumption, temperature, humidity, device status, and voltage, current, etc. parameters of the power network.

[0124] Data transmission and processing: The collected real-time data is transmitted to the system's data processing center through the communication network. After cleaning, checking and formatting, etc. in the data processing center, the data is input to the load characteristic analysis module and the power demand prediction module.

[0125] Closed-loop control implementation:

[0126] Load characteristic analysis: The load characteristic analysis module uses real-time data and historical data to conduct in-depth analysis of the load characteristics of the central air conditioning system. Through analysis, the system's load pattern, peak period and the correlation between load and external environmental factors (such as weather, time, etc.) can be identified.

[0127] Power demand prediction: The power demand prediction module uses advanced algorithms (such as feedforward neural network FNN, etc.) to predict future power demand based on the results of load characteristic analysis and real-time data. The prediction results can provide decision support for energy allocation optimization.

[0128] Energy distribution optimization: Based on the power demand prediction results and the current power supply situation, the energy distribution optimization module uses genetic algorithm and other optimization methods to optimize and adjust the energy distribution scheme of the central air conditioning system. The optimized scheme is executed through the control system to realize closed-loop control of dynamic power management.

[0129] Through the above embodiments, the application can ensure the quality and integrity of historical operation data, providing a reliable data foundation for load characteristic analysis and power demand prediction. At the same time, through real-time monitoring and closed-loop control mechanism of the data monitoring module, the system can dynamically adjust the energy distribution scheme of the central air conditioning system, improving the energy efficiency and economy of the system.

[0130] It should be noted that, in this paper, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article or apparatus.

[0131] Although embodiments of the application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic power management system based on central air conditioning load characteristics, characterized in that, The system comprises: a load characteristic analysis module, which constructs a load characteristic analysis model, the input data of the load characteristic analysis model being historical operation data of the central air conditioning system, including temperature, humidity, indoor-outdoor temperature difference, passenger flow and equipment operation state, and the output data being a load characteristic classification result of the central air conditioning system, including high load, medium load and low load; model training uses a central air conditioning system operation data set containing different load states, and the corresponding label is the load state category; a power demand prediction module, which constructs a power demand prediction model, the input data of the power demand prediction model being the output result of the load characteristic analysis module and weather forecast data, and the output data being a predicted power demand of the central air conditioning system in a future period of time; model training uses a time series data set containing load characteristic classification results and weather forecast data, and the corresponding label is the actual power consumption of the central air conditioning system at each time point; an energy distribution optimization module, which adopts a genetic algorithm to construct an optimization model, the input data of the optimization model being the prediction result of the power demand prediction model and the current power supply condition, and the output data being an optimal energy distribution scheme to minimize power cost and meet the power demand of the central air conditioning system; model training uses a data set containing predicted power demand and actual power supply condition, and the corresponding label is the cost and efficiency index of energy distribution; a control execution module connected to the energy distribution optimization module, which adjusts the operation parameters and energy distribution of the central air conditioning system according to the optimization result; The load characteristic analysis model is constructed using a support vector machine (SVM) algorithm, and the network structure comprises: an input layer that receives historical operation data of the central air conditioning system; a feature mapping layer that maps the input data to a high-dimensional space through an RBF kernel function; a support vector layer that finds an optimal hyperplane in the high-dimensional space for classifying load characteristics; an output layer that outputs the load characteristic classification result of the central air conditioning system; The training steps of the load characteristic analysis model comprise: S101: initializing the parameters of the SVM model, including kernel function parameters, penalty parameters and learning rate; S102: dividing the data set for training into a training set and a validation set; S103: iteratively training the load characteristic analysis model using the training set, in each iteration, calculating the model output and calculating the loss according to the loss function; the loss function uses a Hinge loss function, and the calculation formula is: ， wherein, is the true class label, is the output of the model decision function, N is the number of samples; S104: updating the model parameters by gradient descent method according to the loss value; S105: repeating steps S103 to S104 until the loss converges or the preset number of iterations is reached; S106: verifying the trained model using the validation set to evaluate the classification accuracy of the model; S107: when the model performance reaches the preset standard, saving the model parameters to obtain the trained load characteristic analysis model.

2. The dynamic power management system based on central air conditioning load characteristics according to claim 1, wherein, The power demand prediction model is constructed using a feedforward neural network (FNN) algorithm, and the network structure comprises: an input layer that receives load characteristic classification results and weather forecast data; a hidden layer that contains multiple layers of neurons, each layer of neurons being followed by a ReLU activation function for feature extraction; Output layer: output the predicted value of the central air conditioning system's power demand in the future period.

3. The dynamic power management system based on central air conditioning load characteristics of claim 2, wherein, The training step of the power demand prediction model includes: S201: standardize the input data; S202: initialize the parameters of the FNN model, including weights, bias terms, and learning rates; S203: divide the data set into training set and validation set; S204: use the training set to iteratively train the model, and in each iteration, calculate the model output and loss according to the loss function; S205: update the model parameters according to the loss value through the back propagation algorithm and gradient descent method; S206: repeat steps S204 to S205 until the loss converges or reaches the preset number of iterations; S207: use the validation set to verify the trained model and evaluate the prediction accuracy of the model; S208: when the model performance reaches the preset standard, save the model parameters to obtain the trained power demand prediction model.

4. The dynamic power management system based on central air conditioning load characteristics of claim 3, wherein, In step S204, the mean square error (MSE) function is used as the loss function, and its calculation formula is: ， wherein, Ptrue(i) is the true power demand for the i-th sample, Ppred(i) is the predicted power demand for the i-th sample, and M is the number of samples.

5. The dynamic power management system based on central air conditioning load characteristics of claim 1, wherein, The genetic algorithm of the energy distribution optimization model includes: Initialization step: generate an initial population, each individual represents an energy distribution scheme; Fitness evaluation step: according to the power demand prediction results and the current power supply situation, calculate the fitness of each individual, which represents the cost and efficiency indicators of energy distribution; Selection step: adopt roulette selection method, select individuals for breeding according to fitness; Cross step: adopt single-point crossover method to generate new individuals; Mutation step: randomly mutate part of the individuals to increase population diversity; Iteration step: repeat the fitness evaluation, selection, crossover and mutation steps until the preset number of iterations is reached or the optimal solution is found; Output step: output the optimal energy distribution scheme.

6. The dynamic power management system based on central air conditioning load characteristics according to claim 5, wherein, The calculation formula of the fitness is: ， Where C is the total cost of energy distribution, E is the efficiency indicator of energy distribution, and α and β are weight coefficients to balance the importance of cost and efficiency.

7. The dynamic power management system based on central air conditioning load characteristics of claim 1, wherein, The time resolution of the historical operation data is not less than once per hour, and the data quality has been preprocessed, including missing value filling and outlier detection.

8. The dynamic power management system based on central air conditioning load characteristics of claim 1, wherein, It also includes a data monitoring module for real-time monitoring of the operation data and power supply situation of the central air conditioning system, and inputting the data to the load characteristic analysis module and the power demand prediction module to realize closed-loop control of dynamic power management.

Images