Air conditioner starting target frequency prediction method and system based on historical working conditions

Abstract

The application discloses a kind of air conditioner starting target frequency prediction method and system based on historical working condition, method includes collecting air conditioner historical full state working parameter, effective starting condition data is obtained by cleaning screening, based on environmental temperature difference parameter, load grade dynamic quantification model is constructed, associated load grade and condition data generate compressor frequency table, adopt XGBoost gradient boosting algorithm training compressor target frequency prediction model;When detecting starting signal, output initial prediction target frequency, combine real-time load grade to calculate frequency control coefficient and complete correction, issue final target frequency to drive compressor operation, and realize model incremental iteration optimization by running data feedback.Corresponding system is collaboratively constituted by internet of things air conditioner equipment and five big functional platforms, can improve frequency and load matching precision, control real-time and energy-saving effect are considered, equipment characteristics change can be adapted sustainably, effectively reduce vertical air conditioner operating energy consumption.

Description

Technical Field

[0001] This application relates to the field of refrigerator control technology, and more specifically to a method and system for predicting the target frequency of air conditioner startup based on historical operating conditions. Background Technology

[0002] As the core refrigeration component of inverter air conditioners, the compressor's operating frequency directly determines the air conditioner's cooling capacity and overall energy consumption. Precisely controlling the compressor frequency is a key way to reduce energy consumption while ensuring cooling performance. In recent years, with the rapid application of IoT technology in the home appliance sector, floor-standing air conditioners equipped with various communication modules can collect and upload real-time operating data, providing a feasible data foundation and technical conditions for optimizing compressor frequency control strategies based on historical operating data.

[0003] To further improve the energy-saving control effect of air conditioning, existing technologies have proposed a variety of optimization control schemes based on intelligent algorithms. Some schemes collect temperature-related data within the application scenario, predict indoor temperature and total active power through a recurrent neural network model, and then optimize the calculation using a genetic algorithm to obtain the optimal indoor set temperature and supply air temperature parameters, thereby reducing the operating power consumption of the air conditioning unit. Other schemes are designed for computer room air conditioning scenarios, constructing prompt text based on real-time environmental status information, generating a set of air conditioning control strategies through a large language model, and obtaining the target control strategy after verification and screening by a predictive model, which is used to achieve energy-saving control of computer room air conditioning.

[0004] The existing technical solutions mentioned above still have obvious limitations in practical applications. On the one hand, the existing solutions cannot guarantee that the optimized control parameters match the actual operating conditions, and the energy-saving effect is uncertain and difficult to achieve the expected results. On the other hand, the optimization algorithms used in some solutions have high computational complexity, long operation time, and insufficient real-time control response, making it difficult to meet the application requirements of rapid and accurate control of compressor frequency during the air conditioner start-up phase. Summary of the Invention

[0005] To address the issues of inability to guarantee that control parameters match actual operating conditions and the long computation time of optimization algorithms leading to insufficient real-time control response, this application provides a method for predicting the target frequency of air conditioner startup based on historical operating conditions, comprising the following steps: The full-state operating parameters of the vertical air conditioner during its historical operation are obtained through the Internet of Things communication module, and the full-state operating parameters are cleaned and filtered to obtain valid start-up condition data. Based on the ambient temperature difference parameter in the effective start-up condition data, a dynamic quantification model of load level is constructed. The dynamic quantification model of load level is used to classify the cooling load, and the classified load level is associated with the effective start-up condition data to construct a compressor frequency table. A machine learning algorithm is used to train a compressor target frequency prediction model by taking the feature dimension data in the compressor frequency table as input and the corresponding compressor frequency as output. In response to the detection of an air conditioner start-up signal, the real-time operating condition data of the current start-up is obtained, and the real-time operating condition data is input into the compressor target frequency prediction model to obtain the initial predicted target frequency; The load level dynamic quantification model is used to identify the load level of the real-time operating data, and the frequency control coefficient is calculated based on the identified current load level. The initial predicted target frequency is corrected using the frequency control coefficient to obtain the final start-up target frequency, and the final start-up target frequency is sent to the air conditioning control system to drive the compressor to run. The operating data generated during the current startup process is fed back to the compressor frequency table to iteratively optimize the compressor target frequency prediction model.

[0006] In one feasible implementation, the step of cleaning the full-state operating parameters to filter and obtain valid startup condition data includes: Extract the startup data segment under a single cooling mode; Remove abnormal data segments that are too short to run or have missing data; Extract feature fields from the data segment, including device identifier, set temperature, outdoor temperature, indoor temperature, daytime indicator, windshield parameters, compressor maximum frequency, time to reach temperature, and time interval since the last start-up. The combination of these feature fields is defined as the valid start-up condition data.

[0007] In one feasible implementation, the step of constructing a dynamic quantitative model of load level based on the environmental temperature difference parameter in effective start-up operating data includes: The difference between indoor temperature and set temperature, the difference between outdoor temperature and indoor temperature, and room area parameters are selected as characteristic dimensions for load calculation. The effective power-on data is categorized using a clustering algorithm, and the optimal number of load categories is determined based on the Bayesian information criterion. A mapping relationship is established based on the optimal number of load categories, so that any set of input operating data can correspond to a unique output load level.

[0008] In one feasible implementation, the step of training a machine learning algorithm using the feature dimension data in the compressor frequency table as input and the corresponding compressor frequency as output to obtain a compressor target frequency prediction model includes: The linear correlation between the feature dimension data in the compressor frequency table and the compressor frequency was calculated using the Pearson correlation coefficient method. Feature dimensions with relevance higher than a preset threshold are selected as model input variables; The XGBoost gradient boosting algorithm is used to construct the compressor target frequency prediction model, and the model complexity is controlled by a regularization term. The model is trained using the mapping relationship between the model input variables and the compressor frequency until the model converges.

[0009] In one feasible implementation, the formula for calculating the frequency control coefficient based on the identified current load level is as follows: ; In the formula, The frequency control coefficient is an adjustment factor that is greater than 0.

[0010] In one feasible implementation, the step of feeding back the operating data generated during the current startup process to the compressor frequency table to iteratively optimize the compressor target frequency prediction model includes: After the compressor runs at the final target start-up frequency and completes the current cooling cycle, the operating data within the current cooling cycle is extracted, and the operating data is mapped into a standardized data record with the same data structure as the effective start-up condition data. The standardized data records are appended to the compressor frequency table to expand the sample capacity of the compressor frequency table; In response to the expansion of the sample capacity in the compressor frequency table reaching a preset update threshold, the machine learning algorithm is reused to incrementally train the compressor target frequency prediction model based on the expanded compressor frequency table in order to update the internal parameters of the compressor target frequency prediction model.

[0011] Another aspect of this application provides an air conditioner start-up target frequency prediction system based on historical operating conditions, used to implement the air conditioner start-up target frequency prediction method based on historical operating conditions described in any of the above claims, including: The IoT air conditioning device is equipped with a wireless communication module, which is used to collect and report all-state operating parameters and receive target frequency commands to control the operation of the compressor; A big data storage platform is communicatively connected to the IoT air conditioning device to receive and persistently store the full-state operating parameters; The data cleaning platform is communicatively connected to the big data storage platform. It is used to read the full-state working parameters, filter out invalid data, extract valid power-on operating condition data, and send the valid power-on operating condition data back to the big data storage platform for archiving. The load level quantization platform is communicatively connected to the big data storage platform. It is used to read the archived valid power-on operating condition data to construct a load level dynamic quantization model, and output the load level corresponding to the operating condition data based on the load level dynamic quantization model. The frequency prediction platform is communicatively connected to the big data storage platform and the load level quantification platform, respectively. It is used to retrieve the archived valid start-up condition data and the corresponding load levels to construct the compressor frequency table, and to use the compressor frequency table to train the compressor target frequency prediction model. The platform is connected to the IoT air conditioning device, the frequency prediction platform, and the load level quantification platform respectively. When the power-on signal sent by the IoT air conditioning device is detected, the frequency prediction platform is triggered to output an initial predicted target frequency, and the load level quantification platform is triggered to output the current load level. The initial predicted target frequency is then corrected according to the current load level to generate a final power-on target frequency, and the final power-on target frequency is sent to the IoT air conditioning device. The calling platform is also used to feed back the operating data generated during this startup to the frequency prediction platform via the big data storage platform, so as to incrementally update the compressor target frequency prediction model.

[0012] In one feasible implementation, the data cleaning platform includes a data parsing module and an effective data extraction module; The data parsing module is used to identify single power-on events and operating modes in the full-state operating parameters; The effective data extraction module is used to lock data segments in the cooling mode, remove abnormal data segments with running time below the threshold or missing key fields, and extract feature vectors containing environmental temperature difference parameters and equipment operating parameters from the effective data segments. The data cleaning platform is further configured to output the feature vector as the valid boot condition data to the big data storage platform.

[0013] In one feasible implementation, the load level quantification platform includes a clustering analysis module and a level mapping module; The clustering analysis module is used to read the valid power-on condition data in the big data storage platform, calculate the indoor and outdoor temperature difference and the set temperature difference, and dynamically determine the optimal number of load categories using Bayesian information criteria. The level mapping module has a built-in mapping relationship table generated based on the results of the clustering analysis module. In response to the trigger signal of the calling platform, it receives the current real-time operating data and outputs the corresponding discrete load level value to the calling platform by looking up the table or calculating.

[0014] In one feasible implementation, the frequency prediction platform includes a model training module and an online inference module; The model training module is used to train the compressor target frequency prediction model by using the valid start-up condition data in the big data storage platform as a sample set and combining it with the load level output by the load level quantification platform. The online inference module responds to the trigger signal of the calling platform, receives real-time operating condition data, loads the compressor target frequency prediction model for calculation, and outputs the initial predicted target frequency to the calling platform; The online inference module is also used to feed back the real-time operating condition data and actual operating frequency of this startup to the model training module, so as to update the compressor target frequency prediction model through incremental learning.

[0015] As described above, this application uses historical air conditioner operating data and big data filtering to obtain effective operating conditions and corresponding compressor frequency data for each start-up. Based on this, a compressor frequency table including cooling load levels is constructed, providing filtered and effective training data for the frequency prediction model. This solves the problem of unfiltered training data leading to prediction results deviating from actual energy-saving requirements, ensuring the adaptability of prediction results to actual operating conditions. The solution uses the XGBoost algorithm to train the compressor start-up target frequency prediction model. Leveraging the algorithm's parallel computing capabilities and regularization optimization mechanism, it improves computational efficiency while maintaining prediction accuracy, adapting to the real-time control requirements of air conditioner start-up scenarios and avoiding control delay issues caused by complex swarm intelligence optimization algorithms. Matching the compressor operating frequency with the current cooling load allows for a rapid and stable decrease in indoor temperature, reducing frequent start-ups and shutdowns caused by excessively high frequencies and slow temperature reaching caused by excessively low frequencies. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the implementation of the invention and, together with the description, serve to explain the principles of the embodiments of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0017] Figure 1 This is a flowchart illustrating the air conditioner start-up target frequency prediction method based on historical operating conditions, as shown in the embodiments of this application. Figure 2 This is an architecture diagram of an air conditioner start-up target frequency prediction system based on historical operating conditions, as shown in an embodiment of this application. Detailed Implementation

[0018] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the embodiments of the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of how embodiments of the invention are carried out.

[0019] Traditional floor-standing air conditioners commonly use PID control based on factory test experience values ​​to adjust the compressor frequency. This fails to fully consider the differences in operating conditions caused by different users' equipment installation environments and usage habits, resulting in insufficient matching between the compressor operating frequency and the actual cooling load. This easily leads to problems such as excessively high frequency causing temperature overshoot and frequent start-stop, or excessively low frequency causing slow temperature reaching and poor cooling experience, ultimately resulting in increased air conditioning energy consumption and a decreased user experience. Meanwhile, existing air conditioning energy-saving solutions based on intelligent optimization algorithms often suffer from shortcomings such as ineffective selection of training data, high algorithm computational complexity, large real-time control response delays, and weak adaptability to single devices, making it difficult to achieve stable energy-saving effects while ensuring real-time performance. To address these issues, this application provides a method for predicting the target start-up frequency of air conditioners based on historical operating conditions. Through a complete process of historical data screening, load quantification and grading, machine learning prediction, load coefficient correction, and data feedback iteration, it achieves accurate matching between the compressor start-up target frequency and the cooling load, balancing control real-time performance and energy-saving effects.

[0020] The air conditioner start-up target frequency prediction method based on historical operating conditions provided in this application embodiment refers to... Figure 1 As shown, it includes the following steps: S100: Obtain the full-state operating parameters of the vertical air conditioner during its historical operation through the Internet of Things communication module, and clean the full-state operating parameters to filter out valid start-up condition data.

[0021] The IoT communication module can be integrated into the main control circuit of a vertical air conditioner, establishing electrical connections with various sensors and actuators inside the air conditioner. It can synchronously collect all status parameters throughout the entire operating cycle of the air conditioner and upload them to the cloud data processing node via a wireless network. The full-state operating parameters cover all collectable operating dimensions, including device identity information, on / off status, operating mode, ambient temperature parameters, fan parameters, and compressor operating parameters, completely recording the actual operation process of the air conditioner. Cleaning the full-state operating parameters involves filtering, denoising, and extracting the massive amounts of raw data according to preset data filtering rules. Invalid data that cannot reflect normal cooling operation patterns is removed, and data segments that can completely characterize the characteristics of a single start-up operation are retained, forming structured and effective start-up condition data.

[0022] S200: Based on the ambient temperature difference parameter in the effective start-up condition data, construct a dynamic quantification model of load level, use the dynamic quantification model of load level to classify the cooling load, and associate the classified load level with the effective start-up condition data to construct a compressor frequency table.

[0023] The ambient temperature difference parameter, a temperature-related parameter directly related to the cooling load in the effective operating condition data, can intuitively reflect the cooling pressure that the air conditioning system needs to bear under the current operating conditions. The dynamic load level quantification model can classify the cooling load under different operating conditions through unsupervised classification. The total number of levels is not a fixed value but is dynamically determined based on the actual distribution characteristics of the effective operating condition data to adapt to the differences in operating conditions of different equipment. Associating the graded load levels with the effective operating condition data means matching a corresponding load level label to each set of effective operating condition data, combined with the compressor operating frequency corresponding to that set of data, to form a complete sample record. All sample records are integrated to form a compressor frequency table. The compressor frequency table, as the training dataset for subsequent machine learning algorithms, integrates three core types of information: operating condition characteristics, load level, and compressor frequency, establishing a correlation mapping from operating condition to load and then to frequency.

[0024] S300: Using a machine learning algorithm, the model is trained with the feature dimension data in the compressor frequency table as input and the corresponding compressor frequency as output to obtain a compressor target frequency prediction model.

[0025] The machine learning algorithm employs supervised learning suitable for continuous value prediction. By learning the inherent mapping relationship between operating condition features and compressor frequencies in the compressor frequency table, a prediction model is constructed that can output the corresponding predicted frequency based on input operating condition parameters. Feature dimension data consists of input variables selected from various parameters included in the compressor frequency table to characterize the operating condition characteristics; the corresponding compressor frequency is the model's prediction target, i.e., the label data. The training process iteratively optimizes the model parameters, reducing the deviation between the model's predicted values ​​and the true frequency values ​​until the model converges and reaches the preset accuracy requirements. The trained compressor target frequency prediction model can be directly used for frequency prediction under new operating conditions; the online inference process only requires forward computation, without complex iterative optimization.

[0026] S400: In response to detecting an air conditioner start-up signal, acquire the real-time operating condition data of the current start-up, and input the real-time operating condition data into the compressor target frequency prediction model to obtain the initial predicted target frequency.

[0027] The air conditioner start-up signal is generated by the main control unit of the floor-standing air conditioner upon receiving the user's start-up command and is synchronously sent to the cloud control terminal via the IoT communication module. Real-time operating condition data consists of operating parameters collected at the moment of start-up, consistent with the model input dimensions, ensuring compatibility of the model input format. After receiving the start-up signal and real-time operating condition data, the cloud control terminal inputs the real-time operating condition data into the pre-trained compressor target frequency prediction model. The model then quickly outputs the initial predicted target frequency for the corresponding operating condition through forward computation.

[0028] S500: The load level dynamic quantification model is used to identify the load level of the real-time operating data, and the frequency control coefficient is calculated based on the identified current load level.

[0029] The real-time operating data of the current unit is input into a pre-built dynamic quantification model of the load level. The model can output the cooling load level corresponding to this start-up based on the input operating parameters. The frequency control coefficient is a proportional adjustment parameter that corresponds one-to-one with the load level. Different load levels correspond to different coefficient values, which are used to proportionally correct the initial predicted target frequency. The higher the load level, the greater the current cooling demand, and the corresponding frequency control coefficient can be adjusted accordingly to make the corrected frequency better match the high load demand.

[0030] S600: The initial predicted target frequency is corrected using the frequency control coefficient to obtain the final start-up target frequency, and the final start-up target frequency is sent to the air conditioning control system to drive the compressor to run.

[0031] The correction process is achieved through numerical calculations using a frequency control coefficient and the initial predicted target frequency. The result is the final start-up target frequency. This final start-up target frequency is transmitted to the local control system of the vertical air conditioner via an IoT communication link. The local control system controls the compressor's operating speed based on this target frequency, driving the compressor to start and run at the target frequency. By correcting the initial predicted frequency using a load control coefficient, precise matching between the compressor's target frequency and the current cooling load can be achieved. This avoids problems such as excessively high compressor frequency leading to rapid cooling, temperature overshoot, and frequent start-stop cycles, as well as problems such as excessively low frequency causing the temperature to fail to reach the target level. This ensures cooling performance while reducing unnecessary energy consumption.

[0032] S700: Feedback the operating data generated during the current startup process to the compressor frequency table to iteratively optimize the compressor target frequency prediction model.

[0033] The operational data generated during the current startup process includes the operating parameters of this startup and the actual compressor frequency data. After standardization, this data is appended to the compressor frequency table as new training samples. Iterative optimization is triggered after a certain number of new samples have accumulated, updating the internal parameters of the prediction model through incremental training, allowing the model to learn new operating characteristics. It is understandable that as equipment usage time increases, the equipment's operating characteristics and installation environment conditions may slowly change, leading to a gradual decrease in the adaptability of a fixed prediction model. By continuously feeding back operational data and iteratively optimizing the model, the prediction model can continuously adapt to the actual operating characteristics of the equipment and environmental changes, maintaining high prediction accuracy over the long term and achieving the effect of increasing energy efficiency with continued use of the air conditioner.

[0034] The air conditioner start-up target frequency prediction method provided in this embodiment constructs a technical process from historical data processing to model training, from real-time prediction correction to closed-loop iterative optimization. On the one hand, it achieves rapid matching of the start-up target frequency through effective data screening and machine learning prediction; on the other hand, it achieves accurate adaptation of frequency and load through load level quantification and coefficient correction. Combined with the mechanism of continuous model iterative optimization, it ensures the long-term energy-saving effect of the solution. This method combines real-time control with energy-saving effectiveness, can adapt to the personalized operating conditions of different equipment, and effectively reduces the operating energy consumption of vertical air conditioners while ensuring the user's cooling experience.

[0035] In some embodiments of this application, step S100 of cleaning the full-state operating parameters to filter and obtain valid startup operating condition data includes: S110: Extract the startup data segment in a single cooling mode.

[0036] Specifically, by identifying the switch flag field in the full-state operating parameters, the time points of state transitions can be marked. When the switch flag changes from the off state to the on state, it is recorded as the start point of a power-on event; when the switch flag changes from the on state to the off state, it is recorded as the end point of this power-on event. All data between the start and end points constitute a complete single power-on data segment. After dividing all single power-on segments, the operating mode field in each data segment is further identified, and segments with the operating mode of cooling mode are selected as the basis data for subsequent processing.

[0037] S120: Remove abnormal data segments that have too short a running time or are missing data.

[0038] A minimum effective runtime threshold can be preset. The total runtime of each cooling mode startup segment is compared with this threshold. If the runtime is shorter than the preset threshold, it is determined to be an invalid startup segment and is removed. At the same time, the integrity of the core parameter fields in each data segment is checked, and the missing rate of key fields is calculated. If the missing rate exceeds the preset fault tolerance threshold, it is determined to be an abnormal segment with missing data and is removed.

[0039] Short startup times are typically caused by user errors, testing, or sudden power outages, and fail to reflect the compressor's normal operating conditions. Missing data segments lack complete operational characteristics and are therefore unworthy of being used as training samples. Removing these abnormal data segments further purifies the training sample set, preventing abnormal data from misleading the model into learning incorrect mapping relationships. This improves the model's convergence speed and final prediction accuracy to some extent, while reducing the waste of computational resources caused by invalid data.

[0040] S130: Extract feature fields from the data segment, including device identifier, set temperature, outdoor temperature, indoor temperature, daytime indicator, windshield parameters, compressor maximum frequency, temperature reach time, and time interval since last start-up, and define the combination of the feature fields as the valid start-up condition data.

[0041] The equipment identifier uniquely distinguishes different vertical air conditioning units. The set temperature is the user-defined target indoor temperature, while the indoor temperature is the ambient indoor temperature at the initial startup. The difference between these two temperatures directly reflects the initial temperature difference that needs to be eliminated, a core factor affecting the cooling load. The outdoor temperature is the ambient outdoor temperature at startup; the difference between this and the indoor temperature determines the rate of heat transfer from the outside to the inside, indirectly affecting the cooling load. The daytime indicator distinguishes between daytime and nighttime operating conditions, adapting to different environmental thermal characteristics and user habits at different times. The fan speed parameters include the set fan speed and the actual fan speed, reflecting the air conditioner's air delivery capacity, which directly affects the indoor air circulation speed and cooling efficiency. The compressor maximum frequency is the maximum stable operating frequency reached by the compressor during this startup, a core indicator representing the optimal frequency under this condition, and serves as label data for subsequent model training. The time to reach the set temperature is the time from startup to the indoor temperature dropping to the set temperature, reflecting the cooling efficiency under this condition. The time interval since the last startup reflects the residual heat accumulation in the room; the shorter the interval, the less residual heat in the room, and the lower the initial cooling load.

[0042] By extracting and combining the aforementioned multi-dimensional feature fields to form effective startup condition data, the operating condition characteristics of a single startup can be comprehensively characterized from multiple dimensions such as environment, equipment, and user habits, providing rich feature support for subsequent load quantification and model training. At the same time, by eliminating irrelevant and redundant fields, the data dimensionality of subsequent calculations can be reduced, thereby improving the efficiency of data processing and model training.

[0043] In some embodiments of this application, step S200 of constructing a dynamic quantitative model of load level based on the environmental temperature difference parameter in effective power-on operating condition data includes: S210: Select the difference between indoor temperature and set temperature, the difference between outdoor temperature and indoor temperature, and room area parameters as characteristic dimensions for load calculation.

[0044] The difference between the indoor temperature and the set temperature is the difference between the actual indoor temperature and the user-set temperature at the initial moment of startup. The larger the difference, the higher the initial cooling demand, making it the most direct parameter reflecting the cooling load. The difference between the outdoor temperature and the indoor temperature is the difference between the outdoor ambient temperature and the initial indoor temperature. The larger the difference, the more heat is transferred into the room per unit time, and the higher the continuous heat load that the cooling system needs to bear. The room area parameter refers to the indoor area of ​​the room where the air conditioner is installed. The larger the area, the greater the heat capacity of the indoor space, the higher the total cooling capacity required to reach the set temperature, and the greater the overall cooling load.

[0045] The three types of parameters mentioned above cover the main factors affecting cooling load from three dimensions: initial temperature difference, environmental heat transfer, and space scale. Selecting them as feature dimensions for load calculation can comprehensively and accurately reflect the cooling load level of the operating condition. At the same time, the number of features is reduced, avoiding excessive computational complexity. It is understood that, in addition to the above core features, other load-related parameters can be included as supplements according to the actual application scenario. The three types of parameters selected in this embodiment are basic features, which can control the calculation dimensions while ensuring quantitative accuracy, balancing precision and efficiency.

[0046] S220: The effective power-on condition data is classified into categories using a clustering algorithm, and the optimal number of load categories is determined based on the Bayesian information criterion.

[0047] Clustering algorithms can employ K-means clustering or Gaussian mixture model clustering. Each set of valid operating condition data, corresponding to its load feature vector, is used as input. Clustering is performed according to different preset numbers of clusters, ensuring that operating condition data within the same cluster have similar load characteristics, while load characteristics differ significantly between different clusters. The Bayesian information criterion can be used to comprehensively evaluate the fitting effect and model complexity of the clustering model; a lower criterion value indicates better overall model performance.

[0048] By calculating the Bayesian information criterion values ​​corresponding to different cluster numbers in the clustering model, the cluster number corresponding to the optimal criterion value is selected as the optimal number of load categories. This dynamic approach to determining the optimal number of categories automatically adjusts the number of load levels based on the distribution characteristics of different equipment's operating data. For equipment with a large span of operating conditions, the number of levels can be automatically increased to improve the granularity of the classification; for equipment with a concentrated operating condition distribution, the number of levels can be automatically reduced to avoid overly fine classification. This dynamic classification method avoids the problems of overly coarse or overly fine classification caused by a fixed number of levels, adapting to the differences in operating conditions of different equipment and improving the rationality and accuracy of load level classification.

[0049] S230: Establish a mapping relationship based on the optimal number of load categories, so that any set of input operating data can correspond to a unique output load level.

[0050] The mapping relationship can be established based on the cluster center vectors obtained from clustering. For any input operating condition data, after extracting its load feature vector, the distance between this vector and the cluster center vectors is calculated, and the data is assigned to the nearest cluster. The number corresponding to this cluster is the load level of the operating condition data. Alternatively, the mapping relationship can be implemented by training a lightweight classification model. The level results obtained from clustering are used as labels to train a simple classifier, which outputs the load level corresponding to the input operating condition. After establishing a clear mapping relationship, for newly input real-time operating condition data, the corresponding load level can be quickly output through simple calculation without repeating the clustering process, ensuring the response speed of real-time load identification and meeting the latency requirements of real-time control.

[0051] In some embodiments of this application, step S300, which involves training a machine learning algorithm using feature dimension data from the compressor frequency table as input and the corresponding compressor frequency as output to obtain a compressor target frequency prediction model, includes: S310: The linear correlation between the data of each feature dimension in the compressor frequency table and the compressor frequency is calculated using the Pearson correlation coefficient method.

[0052] The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two continuous variables, and its value ranges from -1 to 1. The closer the absolute value of the correlation coefficient is to 1, the stronger the linear correlation between the two variables. If it is -1, it means that they are perfectly negatively correlated; if it is 1, it means that they are perfectly positively correlated. The closer the absolute value of the correlation coefficient is to 0, the weaker the linear correlation between the two variables.

[0053] For each feature dimension in the compressor frequency table, its Pearson correlation coefficient with the compressor frequency can be calculated using the formula above; where, the feature dimension is defined. and compressor frequency The Pearson correlation coefficient between feature dimension A and compressor frequency B can be calculated: .

[0054] in, It is the dimension of variable features. The i-th data point, It is the compressor frequency. The i-th data point, It is the feature dimension The mean, It is the compressor frequency. The mean.

[0055] By calculating the Pearson correlation coefficient between each feature dimension and the compressor frequency, we can quantify the influence of each feature on the compressor frequency, clarify the contribution of different features in frequency prediction, provide an objective quantitative basis for subsequent feature selection, and avoid the blindness of subjective feature selection.

[0056] S320: Select feature dimensions with a relevance higher than a preset threshold as model input variables.

[0057] A correlation threshold can be pre-set. Feature dimensions with an absolute Pearson correlation coefficient higher than this threshold are retained as input variables for the compressor target frequency prediction model; feature dimensions with a correlation lower than the threshold are discarded and not included in the model input. Filtering highly correlated feature dimensions simplifies the model's input structure, reduces unnecessary computational overhead, and improves the speed of model training and online inference. At the same time, removing features that are weakly correlated with or irrelevant to frequency can reduce the interference of noisy data on model training and improve the model's prediction accuracy to a certain extent.

[0058] S330: The XGBoost gradient boosting algorithm is used to construct the compressor target frequency prediction model, and the model complexity is controlled by a regularization term. The model is trained using the mapping relationship between the model input variables and the compressor frequency until the model converges.

[0059] XGBoost is an ensemble learning algorithm based on the gradient boosting framework. It employs an additive model, building a final strong learner by continuously accumulating weak learners in a decision tree. The model's prediction is a weighted sum of the predictions from multiple weak learners. This algorithm supports multi-threaded parallel processing, improving training speed and adapting to the training needs of large datasets. A regularization term is added to the algorithm's objective function, penalizing the number of leaf nodes and their weights. This controls the growth complexity of the decision tree, preventing overfitting of the training data and improving the model's generalization ability under unknown conditions. During training, the selected model input variables are used as input features, and the corresponding compressor frequencies are used as prediction labels. Each iteration adds a new weak learner to the decision tree, optimizing the model along the gradient descent direction of the loss function, gradually reducing the error between the predicted and true values ​​until the loss function converges or the preset maximum number of iterations is reached.

[0060] Essentially, the XGBoost algorithm determines the structure and parameters of the new weak learner by calculating the first-order gradient and second-order Hessian matrix information of the loss function. Compared with traditional gradient boosting trees, utilizing second-order information can accelerate convergence and improve training accuracy. The prediction model built using the XGBoost algorithm combines the advantages of fast training speed, high prediction accuracy, and strong generalization ability. It can efficiently learn the nonlinear mapping relationship between operating condition features and compressor frequency, while also exhibiting fast online inference speed, fully meeting the accuracy and real-time requirements of air conditioner start-up frequency prediction scenarios.

[0061] This embodiment uses Pearson correlation coefficient to screen highly correlated features and combines it with regularized XGBoost gradient boosting algorithm to train the prediction model. This effectively simplifies the model input dimension, improves training and inference efficiency, and ensures that the model has high prediction accuracy and good generalization ability, providing reliable model support for accurate and fast prediction of startup target frequency.

[0062] In some embodiments of this application, the formula for calculating the frequency control coefficient based on the identified current load level is as follows: ; In the formula, γ is the frequency control coefficient, which is an adjustment factor greater than 0.

[0063] The current load level is the load level value corresponding to the current start-up condition identified by the load level dynamic quantification model. Its value ranges from 1 to a positive integer equal to the total number of load categories; a larger value indicates a higher cooling load. The total number of load categories is the optimal number of load categories determined by the load level dynamic quantification model, i.e., the maximum value of the load level. According to the formula derivation, the frequency control coefficient changes linearly between 0.9 and 0.98 with the load level. When the load level is the lowest level (value 1), the frequency control coefficient takes its minimum value, adjusting the initial predicted target frequency downwards to accommodate the smaller cooling demand under low load conditions. When the load level is the highest level (value equal to the total number of load categories), the frequency control coefficient takes its maximum value of 0.98, adjusting the initial predicted target frequency upwards to accommodate the larger cooling demand under high load conditions.

[0064] In some embodiments of this application, the step S700 of feeding back the operating data generated during the current startup process to the compressor frequency table to iteratively optimize the compressor target frequency prediction model includes: S710: After the compressor runs at the final start-up target frequency and completes the current cooling cycle, extract the operating data within the current cooling cycle and map the operating data into a standardized data record with the same data structure as the effective start-up condition data.

[0065] The end of each cooling cycle is marked by the air conditioner shutting down or switching out of cooling mode. After the cycle ends, operational data from the entire startup process is collected, including environmental parameters, equipment operating parameters, time parameters, and other comprehensive data. The collected operational data is extracted, organized, and calculated according to the field structure of valid startup data, retaining feature fields completely consistent with valid startup data and supplementing necessary derived fields. This ensures that the generated data records have the exact same field structure, data format, and order as the original data in the compressor frequency table. Mapping the operational data to standardized data records ensures the consistency of format between new data and existing training samples, allowing new data to be directly added to the training set without additional format conversion, improving data reuse efficiency and the convenience of model updates.

[0066] S720: The standardized data records are appended to the compressor frequency table to expand the sample capacity of the compressor frequency table.

[0067] Each completed cooling cycle generates a corresponding standardized data record, which is appended to the end of the compressor frequency table. The sample size of the compressor frequency table continuously expands as the air conditioner is turned on more times. This expansion enriches the operating condition coverage of the training set, incorporating operational data from different environmental conditions and usage scenarios, providing more comprehensive sample support for model optimization. It is understandable that as equipment is used over time, its heat exchange efficiency and compressor performance will slowly degrade due to aging, and the thermal characteristics of the installation environment may also change with the seasons and surrounding environment. Continuously added operational data accurately reflects these changes, providing a data foundation for the model to adapt to new equipment states and environmental conditions.

[0068] S730: In response to the expansion of the sample capacity in the compressor frequency table reaching a preset update threshold, the machine learning algorithm is reused to incrementally train the compressor target frequency prediction model based on the expanded compressor frequency table in order to update the internal parameters of the compressor target frequency prediction model.

[0069] Specifically, the update threshold can be set to a fixed number of new samples, triggering a model update after every fixed number of new samples; or it can be set to a fixed time period, triggering a model update after a fixed period of time. Incremental training refers to iteratively training the model using newly added sample data based on the existing model parameters, without having to retrain from scratch. This significantly reduces the computational load and time required for training while maintaining the effectiveness of model updates, thus improving the efficiency of model updates. By updating the model's internal parameters through incremental training, the model can learn the working condition characteristics in the new samples, adapt to the slow changes in equipment and environment, and maintain consistently high prediction accuracy.

[0070] Another aspect of this application provides an air conditioner start-up target frequency prediction system based on historical operating conditions, used to implement the air conditioner start-up target frequency prediction method based on historical operating conditions in the above embodiments, referring to... Figure 2 As shown, the system includes: IoT air conditioning equipment, big data storage platform, data cleaning platform, load level quantification platform, frequency prediction platform, and call platform.

[0071] The IoT-enabled air conditioning device is equipped with a wireless communication module for collecting and reporting full-state operating parameters and receiving target frequency commands to control the compressor operation. In one embodiment, the IoT air conditioning device is a vertical air conditioning terminal with IoT communication capabilities. It internally includes various data acquisition components such as temperature sensors, fan sensors, and compressor detection units, as well as control components such as a main control unit and a drive unit, enabling real-time collection of full-state operating parameters throughout the entire operating cycle. The wireless communication module may include any one or more of Wi-Fi, narrowband IoT, 2G, 3G, 4G, and 5G modules, used to achieve bidirectional data interaction between the air conditioning device and various cloud platforms.

[0072] As the terminal node of the system, the IoT air conditioning device undertakes the functions of raw data collection and reporting, serving as the data source entry point of the system. On the other hand, it undertakes the function of executing control commands, driving the compressor to run according to the target frequency commands issued by the cloud, serving as the final execution end of energy-saving control and realizing the practical application of energy-saving solutions.

[0073] The big data storage platform communicates with the IoT air conditioning device to receive and persistently store the full-state operating parameters. Deployed on a cloud server cluster, the big data storage platform features high capacity, high availability, and high concurrency. It can simultaneously receive a large number of real-time reported full-state operating parameters from IoT air conditioning devices and classify, store, and index them according to device identifier and time dimension.

[0074] The data cleaning platform communicates with the big data storage platform to read the full-state operating parameters, filter out invalid data, extract valid startup condition data, and then transmit the valid startup condition data back to the big data storage platform for archiving. Specifically, the data cleaning platform retrieves the original full-state operating parameters from the big data storage platform, performs event segmentation, pattern filtering, anomaly removal, and feature extraction on the data according to preset cleaning rules, obtains structured, high-quality valid startup condition data, and transmits the processed data back to the big data storage platform for archiving and storage, for subsequent platform access.

[0075] The load level quantization platform is communicatively connected to the big data storage platform. It is used to read the archived valid power-on condition data to construct a load level dynamic quantization model, and output the load level corresponding to the condition data based on the load level dynamic quantization model.

[0076] The load level quantification platform retrieves archived valid operating condition data from the big data storage platform, constructs a dynamic load level quantification model through cluster analysis and optimal number of clusters, and establishes mapping rules from operating conditions to load levels. During the real-time control phase, the load level quantification platform can receive real-time operating condition data sent by the calling platform, quickly calculate and output the corresponding load level using the mapping rules, and feed it back to the calling platform.

[0077] The frequency prediction platform is communicatively connected to the big data storage platform and the load level quantification platform, respectively, and is used to retrieve the archived valid start-up condition data and the corresponding load levels to construct a compressor frequency table, and use the compressor frequency table to train a compressor target frequency prediction model.

[0078] The frequency prediction platform retrieves valid start-up operating condition data from the big data storage platform and obtains corresponding load level labels from the load level quantification platform. It integrates these two types of data to construct a compressor frequency table containing operating conditions, load, and frequency, which serves as the dataset for model training. Using the compressor frequency table as training samples, the platform trains a compressor target frequency prediction model through machine learning algorithms and stores the model file for real-time access. During the real-time control phase, the frequency prediction platform receives real-time operating condition data from the calling platform, quickly calculates and outputs the initial predicted target frequency using the prediction model, and feeds it back to the calling platform.

[0079] The platform is connected to the IoT air conditioning device, the frequency prediction platform, and the load level quantification platform. When the power-on signal sent by the IoT air conditioning device is detected, the platform triggers the frequency prediction platform to output an initial predicted target frequency and the load level quantification platform to output the current load level. The platform then corrects the initial predicted target frequency based on the current load level to generate the final power-on target frequency and sends the final power-on target frequency to the IoT air conditioning device.

[0080] The platform is responsible for receiving power-on requests from terminal devices, coordinating the frequency prediction platform and load level quantification platform to perform processing tasks in parallel, obtaining the predicted frequency and load level, completing frequency correction calculations, and finally sending control commands to the terminal devices. The platform uniformly schedules the collaborative work of each platform, optimizes the timing of the control process, minimizes power-on response delays, ensures real-time control, and improves the user experience.

[0081] Furthermore, the calling platform is also used to feed back the operating data generated during this startup to the frequency prediction platform via the big data storage platform, so as to incrementally update the compressor target frequency prediction model. Specifically, after the startup operation is completed, the calling platform collects the complete operating data of this startup and uploads it to the big data storage platform for archiving. The frequency prediction platform can periodically retrieve newly added operating data from the big data storage platform to incrementally update the prediction model. Through the closed-loop mechanism of data feedback and model update, the model can continuously self-optimize, enabling the system to continuously adapt to changes in the operating characteristics of the equipment and maintain stable energy-saving effects in the long term.

[0082] The energy-saving system provided in this embodiment constructs a closed-loop technology system through six platforms, encompassing data acquisition, data processing, model building, real-time control, and iterative optimization. This enables personalized energy-saving control of vertical air conditioners based on historical operating conditions. Furthermore, the clear division of labor among the platforms and smooth data flow ensure fast real-time control response and system reliability, effectively reducing the energy consumption of vertical air conditioners while possessing continuous optimization capabilities.

[0083] In some embodiments of this application, the data cleaning platform includes a data parsing module and an effective data extraction module.

[0084] The data parsing module is used to identify individual power-on events and operating modes in the full-state operating parameters. Specifically, the data parsing module receives full-state time-series data retrieved from the big data storage platform, performs structured parsing on the raw data stream, identifies the state switching nodes of the switch flags, divides each independent power-on event, determines the start and end timestamps of each power-on, and splits the continuous time-series data into multiple independent power-on event units. Simultaneously, the data parsing module extracts the operating mode field corresponding to each power-on event to identify the operating mode type of that power-on.

[0085] The effective data extraction module is used to lock data segments in cooling mode, remove abnormal data segments with runtime below a threshold or missing key fields, and extract feature vectors containing environmental temperature difference parameters and equipment operating parameters from the effective data segments. The effective data extraction module first filters all startup events divided by the data parsing module, identifies events with cooling mode, locks the corresponding data segments as objects to be processed, and excludes irrelevant data from other operating modes. Subsequently, the effective data extraction module performs validity checks on each cooling mode data segment. The checks include runtime checks and data integrity checks: segments with runtime shorter than a preset duration threshold are judged as short-duration abnormal segments, and segments with a key field missing ratio exceeding a preset fault tolerance ratio are judged as data missing abnormal segments; both types of abnormal segments are removed. For effective data segments that pass the validity checks, the effective data extraction module extracts preset core feature fields, including environmental temperature difference parameters and equipment operating parameters, and combines them to form a feature vector with unified dimensions.

[0086] Furthermore, the data cleaning platform is also configured to output the feature vector as the valid startup condition data to the big data storage platform. The feature vector generated by the valid data extraction module has a unified field structure, data format, and sorting. The data cleaning platform uses this feature vector as the final valid startup condition data and sends it back to the big data storage platform for archiving and storage, for subsequent use by the load quantization platform and frequency prediction platform.

[0087] In some embodiments of this application, the load level quantification platform includes a clustering analysis module and a level mapping module. The clustering analysis module is used to read valid operating condition data from the big data storage platform, calculate the indoor-outdoor temperature difference and the set temperature difference, and dynamically determine the optimal number of load categories using Bayesian information criteria. The level mapping module has a built-in mapping relationship table generated based on the results of the clustering analysis module. In response to the trigger signal of the calling platform, it receives the current real-time operating condition data and outputs the corresponding discrete load level value to the calling platform by looking up the table or calculating.

[0088] In one implementation, the clustering analysis module retrieves valid power-on condition data from the big data storage platform, extracts core temperature parameters such as indoor temperature, outdoor temperature, and set temperature, calculates the difference between indoor temperature and set temperature (i.e., the set temperature difference), and the difference between outdoor temperature and indoor temperature (i.e., the indoor-outdoor temperature difference), and combines these with room area parameters to form a load feature vector.

[0089] Furthermore, after the clustering analysis module completes the load grading model construction, it transforms the classification rules into a mapping table and stores it in the memory of the grading mapping module, thus achieving pre-loading. During the real-time control phase, when the calling platform sends a load query trigger signal along with real-time operating data, the grading mapping module extracts the load characteristic parameters from the real-time operating conditions. Through table lookup or simple distance calculation, it quickly obtains the corresponding discrete load grading values ​​and returns the results to the calling platform.

[0090] Understandably, the clustering analysis module can periodically reread newly added valid operating condition data, update the load grading model and mapping table, and enable the load grading rules to be continuously optimized as data accumulates, further improving the accuracy of grading. The separation design of offline modeling and online inference allows complex model calculations to be completed offline, while only simple query calculations are performed online, balancing model accuracy and real-time performance.

[0091] In some embodiments of this application, the frequency prediction platform includes a model training module and an online inference module. The model training module is used to train the compressor target frequency prediction model using valid start-up condition data in the big data storage platform as a sample set and combined with the load level output by the load level quantification platform. The online inference module responds to the trigger signal of the calling platform, receives real-time operating condition data, loads the compressor target frequency prediction model for calculation, and outputs the initial predicted target frequency to the calling platform.

[0092] Specifically, the model training module retrieves valid start-up operating condition data from the big data storage platform and obtains the corresponding load level labels from the load level quantification platform. These two types of data are integrated to construct a compressor frequency table as the training sample set. The model training module uses machine learning algorithms to train the sample set, iteratively optimizing and continuously learning the mapping relationship between operating condition characteristics and compressor frequency. This generates a compressor target frequency prediction model file, which is then persistently stored.

[0093] The online inference module pre-loads the trained compressor target frequency prediction model into memory and runs resident, awaiting invocation requests. Upon receiving a trigger signal and real-time operating data from the invocation platform, the online inference module inputs the real-time operating data into the pre-loaded model, quickly outputs the initial predicted target frequency through forward computation, and returns the result to the invocation platform. The online inference module undertakes the core function of real-time frequency prediction, employing model pre-loading to achieve rapid inference response, meeting the high real-time requirements of power-on control, and avoiding startup delays caused by model loading.

[0094] Furthermore, the online inference module is also used to feed back the real-time operating condition data and actual operating frequency of this startup to the model training module, so as to update the compressor target frequency prediction model through incremental learning. After each startup operation, the online inference module collects the real-time operating condition data and the actual operating compressor frequency data, and synchronously feeds them back to the model training module. After the received new samples accumulate to a preset threshold, the model training module continues to train on the basis of the original model parameters using incremental learning, updates the internal parameters of the model, generates an optimized model file, and synchronizes it to the online inference module to realize the iterative update of the model. Updating the model through incremental learning eliminates the need for full retraining, which can significantly reduce computational overhead, improve the efficiency of model updates, and ensure that the model can continuously adapt to changes in equipment and environment.

[0095] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0096] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for predicting the target start-up frequency of air conditioners based on historical operating conditions, characterized in that, Includes the following steps: The full-state operating parameters of the vertical air conditioner during its historical operation are obtained through the Internet of Things communication module, and the full-state operating parameters are cleaned and filtered to obtain valid start-up condition data. Based on the ambient temperature difference parameter in the effective start-up condition data, a dynamic quantification model of load level is constructed. The dynamic quantification model of load level is used to classify the cooling load, and the classified load level is associated with the effective start-up condition data to construct a compressor frequency table. A machine learning algorithm is used to train a compressor target frequency prediction model by taking the feature dimension data in the compressor frequency table as input and the corresponding compressor frequency as output. In response to the detection of an air conditioner start-up signal, the real-time operating condition data of the current start-up is obtained, and the real-time operating condition data is input into the compressor target frequency prediction model to obtain the initial predicted target frequency; The load level dynamic quantification model is used to identify the load level of the real-time operating data, and the frequency control coefficient is calculated based on the identified current load level. The initial predicted target frequency is corrected using the frequency control coefficient to obtain the final start-up target frequency, and the final start-up target frequency is sent to the air conditioning control system to drive the compressor to run. The operating data generated during the current startup process is fed back to the compressor frequency table to iteratively optimize the compressor target frequency prediction model.

2. The method for predicting the target frequency of air conditioner operation based on historical operating conditions according to claim 1, characterized in that, The step of cleaning and filtering the full-state operating parameters to obtain valid start-up operating condition data includes: Extract the startup data segment under a single cooling mode; Remove abnormal data segments that are too short to run or have missing data; Extract feature fields from the data segment, including device identifier, set temperature, outdoor temperature, indoor temperature, daytime indicator, windshield parameters, compressor maximum frequency, time to reach temperature, and time interval since the last start-up. The combination of these feature fields is defined as the valid start-up condition data.

3. The method for predicting the target frequency of air conditioner operation based on historical operating conditions according to claim 1, characterized in that, The steps for constructing a dynamic quantitative model of load level based on the ambient temperature difference parameter in valid start-up operating data include: The difference between indoor temperature and set temperature, the difference between outdoor temperature and indoor temperature, and room area parameters are selected as characteristic dimensions for load calculation. The effective power-on data is categorized using a clustering algorithm, and the optimal number of load categories is determined based on the Bayesian information criterion. A mapping relationship is established based on the optimal number of load categories, so that any set of input operating data can correspond to a unique output load level.

4. The method for predicting the target frequency of air conditioner operation based on historical operating conditions according to claim 1, characterized in that, The step of training a compressor target frequency prediction model using a machine learning algorithm with feature dimension data from the compressor frequency table as input and the corresponding compressor frequency as output includes: The linear correlation between the feature dimension data in the compressor frequency table and the compressor frequency was calculated using the Pearson correlation coefficient method. Feature dimensions with relevance higher than a preset threshold are selected as model input variables; The XGBoost gradient boosting algorithm is used to construct the compressor target frequency prediction model, and the model complexity is controlled by a regularization term. The model is trained using the mapping relationship between the model input variables and the compressor frequency until the model converges.

5. The method for predicting the target frequency of air conditioner operation based on historical operating conditions according to claim 1, characterized in that, The formula for calculating the frequency control coefficient based on the identified current load level is as follows: ； In the formula, The frequency control coefficient is an adjustment factor that is greater than 0.

6. The method for predicting the target frequency of air conditioner operation based on historical operating conditions according to claim 1, characterized in that, The step of feeding back the operating data generated during the current startup process to the compressor frequency table to iteratively optimize the compressor target frequency prediction model includes: After the compressor runs at the final target start-up frequency and completes the current cooling cycle, the operating data within the current cooling cycle is extracted, and the operating data is mapped into a standardized data record with the same data structure as the effective start-up condition data. The standardized data records are appended to the compressor frequency table to expand the sample capacity of the compressor frequency table; In response to the expansion of the sample capacity in the compressor frequency table reaching a preset update threshold, the machine learning algorithm is reused to incrementally train the compressor target frequency prediction model based on the expanded compressor frequency table in order to update the internal parameters of the compressor target frequency prediction model.

7. A target frequency prediction system for air conditioner startup based on historical operating conditions, used to implement the target frequency prediction method for air conditioner startup based on historical operating conditions as described in any one of claims 1-6, characterized in that, include: The IoT air conditioning device is equipped with a wireless communication module, which is used to collect and report all-state operating parameters and receive target frequency commands to control the operation of the compressor; A big data storage platform is communicatively connected to the IoT air conditioning device to receive and persistently store the full-state operating parameters; The data cleaning platform is communicatively connected to the big data storage platform. It is used to read the full-state working parameters, filter out invalid data, extract valid power-on operating condition data, and send the valid power-on operating condition data back to the big data storage platform for archiving. The load level quantization platform is communicatively connected to the big data storage platform. It is used to read the archived valid power-on operating condition data to construct a load level dynamic quantization model, and output the load level corresponding to the operating condition data based on the load level dynamic quantization model. The frequency prediction platform is communicatively connected to the big data storage platform and the load level quantification platform, respectively. It is used to retrieve the archived valid start-up condition data and the corresponding load levels to construct the compressor frequency table, and to use the compressor frequency table to train the compressor target frequency prediction model. The platform is connected to the IoT air conditioning device, the frequency prediction platform, and the load level quantification platform respectively. When the power-on signal sent by the IoT air conditioning device is detected, the frequency prediction platform is triggered to output an initial predicted target frequency, and the load level quantification platform is triggered to output the current load level. The initial predicted target frequency is then corrected according to the current load level to generate a final power-on target frequency, and the final power-on target frequency is sent to the IoT air conditioning device. The calling platform is also used to feed back the operating data generated during this startup to the frequency prediction platform via the big data storage platform, so as to incrementally update the compressor target frequency prediction model.

8. The air conditioner start-up target frequency prediction system based on historical operating conditions according to claim 7, characterized in that, The data cleaning platform includes a data parsing module and an effective data extraction module; The data parsing module is used to identify single power-on events and operating modes in the full-state operating parameters; The effective data extraction module is used to lock data segments in the cooling mode, remove abnormal data segments with running time below the threshold or missing key fields, and extract feature vectors containing environmental temperature difference parameters and equipment operating parameters from the effective data segments. The data cleaning platform is further configured to output the feature vector as the valid boot condition data to the big data storage platform.

9. The air conditioner start-up target frequency prediction system based on historical operating conditions according to claim 7, characterized in that, The load level quantification platform includes a cluster analysis module and a level mapping module; The clustering analysis module is used to read the valid power-on condition data in the big data storage platform, calculate the indoor and outdoor temperature difference and the set temperature difference, and dynamically determine the optimal number of load categories using Bayesian information criteria. The level mapping module has a built-in mapping relationship table generated based on the results of the clustering analysis module. In response to the trigger signal of the calling platform, it receives the current real-time operating data and outputs the corresponding discrete load level value to the calling platform by looking up the table or calculating.

10. The air conditioner start-up target frequency prediction system based on historical operating conditions according to claim 7, characterized in that, The frequency prediction platform includes a model training module and an online inference module; The model training module is used to train the compressor target frequency prediction model by using the valid start-up condition data in the big data storage platform as a sample set and combining it with the load level output by the load level quantification platform. The online inference module responds to the trigger signal of the calling platform, receives real-time operating condition data, loads the compressor target frequency prediction model for calculation, and outputs the initial predicted target frequency to the calling platform; The online inference module is also used to feed back the real-time operating condition data and actual operating frequency of this startup to the model training module, so as to update the compressor target frequency prediction model through incremental learning.

Images