A voltage-based intelligent power supply adjustment method, system and storage medium

Abstract

The present invention discloses a voltage-based intelligent power supply adjustment method, system, and storage medium. The method includes: obtaining data related to voltage fluctuations; using the data related to voltage fluctuations to train a voltage fluctuation prediction model to analyze the patterns and trends of voltage fluctuations; using the voltage fluctuation prediction model to input real-time monitoring data to obtain voltage fluctuation prediction results; and dynamically adjusting the operating state of the power supply based on the voltage fluctuation prediction results through an adaptive control strategy. The adaptive control strategy includes automatically adjusting the output voltage and current of the power supply module based on changes in the voltage wave bottom through an energy storage and release system. By using the trained voltage fluctuation prediction model and a multi-feature factor analysis method to analyze the patterns and trends of voltage fluctuations, a more comprehensive and accurate voltage fluctuation prediction can be achieved. The output voltage and current of the power supply module are dynamically adjusted to maintain stable power output.

Description

Technical Field

[0001] The present invention generally relates to the field of artificial intelligence technology. More specifically, the present invention relates to a voltage-based intelligent power supply adjustment method, system, and storage medium. Background Art

[0002] In modern power applications, power modules are undoubtedly the key hub for maintaining the normal operation of the entire power system. From precision machinery on industrial production lines to various electronic devices in homes, from life support systems in the medical field to core base stations in communications networks, power modules are ubiquitous. Their stability is directly related to the reliable operation of these systems. Therefore, the stability of power modules plays a vital role in the reliable operation of the entire power system.

[0003] However, the reality is that frequent voltage fluctuations and abnormal changes can cause serious damage to the power modules themselves. Like a ship sailing in rough seas, the unstable voltage puts additional stress on the electronic components within the power module, significantly accelerating their aging and wear. Power modules that were originally expected to operate stably for long periods of time may prematurely reach the end of their service life due to prolonged exposure to unstable voltage. This not only means more frequent power module replacement, increasing equipment maintenance costs and downtime, but can also pose serious safety risks in some critical application scenarios.

[0004] On the other hand, for electrical devices that rely on power modules, the impact of unstable voltage is even more multifaceted and far-reaching. For example, in a typical computer system, unstable voltage can cause performance degradation in the central processing unit (CPU). When voltage drops below normal, the CPU may not be able to operate at its rated frequency, significantly reducing computing speed. Data tasks that were once fast to process become sluggish, severely impacting work efficiency. Regarding data transmission, unstable voltage can disrupt the signal integrity of the data bus, causing errors during transmission. These errors can manifest as data loss, data tampering, or transmission interruptions. For applications requiring extremely high data accuracy, such as financial trading systems and scientific research data acquisition systems, even the slightest data error can have catastrophic consequences. More seriously, chronic voltage instability can directly damage the hardware of electrical devices. For example, excessively high voltage can break down the insulation of electronic components, causing short circuits and burnout; while excessively low voltage can prevent motors and other equipment from starting properly, causing them to stall for extended periods and subsequently overheat and damage. Similar issues exist in other fields as well. For example, in some continuous industrial processes, equipment failures caused by voltage issues can even cause the entire production line to shut down, resulting in significant economic losses. In the medical field, many medical devices, such as electrocardiographs, monitors, and surgical instruments, require extremely high power supply stability. Even slight voltage fluctuations can affect the measurement accuracy of medical equipment, leading to misdiagnosis of a patient's condition and even endangering their life.

[0005] In view of this, there is an urgent need to provide a voltage-based intelligent power adjustment method to effectively control the stable output of the power supply. Summary of the Invention

[0006] In order to at least solve one or more of the technical problems mentioned above, the present invention proposes a voltage-based intelligent power supply adjustment solution in multiple aspects.

[0007] In a first aspect, the present invention provides a voltage-based intelligent power supply adjustment method, comprising: obtaining relevant data on voltage fluctuations; using the relevant data on voltage fluctuations to train a voltage fluctuation prediction model to analyze the pattern and trend of voltage fluctuations; using the voltage fluctuation prediction model to input real-time monitoring data to obtain prediction results of voltage fluctuations; dynamically adjusting the working state of the power supply through an adaptive control strategy based on the prediction results of the voltage fluctuations; wherein the adaptive control strategy includes automatically adjusting the output voltage and current of the power supply module through an energy storage and release system according to changes in the voltage wave bottom; storing energy when the voltage wave bottom is lower than a set threshold of a reference voltage, and releasing energy when the voltage wave bottom is higher than a set threshold of the reference voltage, so as to smooth the voltage fluctuations.

[0008] In some embodiments, a voltage fluctuation prediction model is trained using data related to voltage fluctuations to analyze patterns and trends of voltage fluctuations, including the following data preparation stages: processing missing or abnormal data in the data related to voltage fluctuations to obtain original training data; transforming or combining feature data in the original training data to generate valid training data containing new features after the transformation or combination, wherein the new features include frequency features extracted using Fourier transform; normalizing the valid training data to eliminate dimensional differences between different features; analyzing the correlation between different features in the valid training data and voltage fluctuations, and adding features whose influence coefficients on prediction results exceed a threshold to the valid training data; wherein analyzing the correlation between different features in the valid training data and voltage fluctuations includes extracting static features and time series features from the valid training data, and fusing the two features using a context feature fusion network;

[0009] In some embodiments, the training of the voltage fluctuation prediction model includes: dividing the valid training data into a training set, a validation set and a test set; using the training set data to train the voltage fluctuation prediction model according to the time series prediction model to analyze the pattern and trend of voltage fluctuation; using the validation set data to verify the trained voltage fluctuation prediction model to obtain a verification result; adjusting the voltage fluctuation prediction model according to the verification result to optimize the voltage fluctuation prediction model; the adjustment of the voltage fluctuation prediction model includes adjusting model parameters, feature combinations or model structures; and using the test set data to evaluate the prediction result indicators of the optimized voltage fluctuation prediction model to obtain an evaluation result.

[0010] In some embodiments, the time series prediction model includes: normalizing all input training set data to the range of [-1,1]; constructing a multi-layer long short-term memory network, the long short-term memory network including at least one layer for capturing short-term dependencies and one layer for capturing long-term dependencies; wherein, an added attention mechanism is also included before or after each layer of the long short-term memory network to weight the input features, and dynamically adjust the attention weights of different time steps according to the current input and context vector; using the long short-term memory network, inputting a multidimensional feature vector at time point t, the multidimensional feature vector including at least one or more of the following features: voltage value, temperature, humidity, load change; according to the multidimensional feature vector, capturing the long-term and short-term dependencies in the time series data to obtain a voltage fluctuation prediction result.

[0011] In some embodiments, the extraction of static features and temporal features from the effective training data and the fusion of the two features using a context feature fusion network include: extracting original static features from the effective training data; the original static features include the rated power, voltage range, and environmental indicators of the smart power supply; using a statistical analysis algorithm to calculate the context features of the original static features in the effective training data to obtain a static feature context vector; using an Embedding layer in a deep learning model to process the original static features to obtain an original static feature vector; and combining the original static feature vector and the static feature context vector. Weighted splicing is performed to obtain a fused static feature vector; time series features are extracted from the effective training data through a time series analysis algorithm; the time series features include historical data of voltage and current; the time series features are processed using the LSTM layer in the deep learning model to obtain a time series feature vector; the context features of the time series features in the effective training data are calculated using a statistical analysis algorithm to obtain a time series feature context vector; the time series feature vector and the time series feature context vector are weighted spliced ​​to obtain a fused time series feature vector; the fused static feature vector and the fused time series feature vector are weighted spliced ​​and fused to obtain a fused feature vector.

[0012] In some embodiments, the operating state of the power supply is dynamically adjusted through an adaptive control strategy based on the predicted results of voltage fluctuations, including: judging the changes in the voltage bottom based on the predicted results of voltage fluctuations to automatically adjust the output voltage and current of the power module; the automatic adjustment includes adjusting the output current and voltage of the inverter in the power module, and controlling the charging and discharging process of the energy storage system of the power module.

[0013] In a second aspect, the present invention provides a voltage-based intelligent power supply adjustment system, comprising: a data acquisition module for acquiring relevant data on voltage fluctuations; a prediction model training module for training a voltage fluctuation prediction model using the relevant data on voltage fluctuations to analyze the pattern and trend of voltage fluctuations; a real-time prediction module for using the voltage fluctuation prediction model to input real-time monitoring data to obtain a prediction result of voltage fluctuations; a power supply adjustment module for dynamically adjusting the working state of the power supply through an adaptive control strategy based on the prediction result of voltage fluctuations; wherein the adaptive control strategy includes automatically adjusting the output voltage and current of the power supply module through an energy storage and release system according to changes in the voltage wave bottom; storing energy when the voltage wave bottom is lower than a set threshold of the reference voltage, and releasing energy when the voltage wave bottom is higher than a set threshold of the reference voltage to smooth the voltage fluctuations.

[0014] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the voltage-based intelligent power supply adjustment method as described in any one of the embodiments described in the first aspect is implemented.

[0015] Through the voltage-based intelligent voltage adjustment method provided above, the embodiment of the present invention can ensure the accuracy of intelligent power supply adjustment by preprocessing and analyzing the acquired voltage fluctuation related data. By adding an attention mechanism before or after each layer of the long short-term memory network to weight the input features, and then using these weights to construct a context vector, the feature weights of different time steps are dynamically adjusted. This mechanism enables the model to focus on key information more flexibly and improve the accuracy of prediction. According to the input multi-dimensional feature vector, the long-term and short-term dependencies in the time series data are captured to obtain the voltage fluctuation prediction results. Through training, the model can learn the pattern and trend of voltage fluctuations, improve the accuracy and response speed of voltage bottom prediction, and thus dynamically adjust the working state of the power supply to adapt to voltage fluctuations.

[0016] Furthermore, in some embodiments, real-time monitoring and prediction of voltage bottoms can enable adaptive adjustment of power modules, improving the stability and efficiency of the power system. The integration of energy storage and release systems enables energy storage and release, effectively smoothing voltage fluctuations and improving the reliability of the power system.

[0017] Therefore, the voltage-based intelligent voltage adjustment method provided by this invention can accurately monitor real-time voltage changes, quickly analyze the causes and trends of voltage instability, and promptly implement effective adjustment measures to ensure that power output remains stable within a safe and reliable range. This ensures the stable operation of various power systems in modern power applications, avoids the many risks and losses caused by power supply problems, and provides a solid power supply guarantee foundation for scientific and technological progress and social development. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an illustrative and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0019] Figure 1 An exemplary flow chart of a voltage-based intelligent voltage adjustment method according to an embodiment of the present invention is shown;

[0020] Figure 2 An exemplary structural block diagram of a voltage-based intelligent voltage adjustment system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0022] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present invention indicate the presence of the described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their collections. It should be further understood that the term "and / or" used in the specification and claims of the present invention refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0023] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Figure 1 FIG. 1 shows an exemplary flow chart of a voltage-based intelligent voltage adjustment method 100 according to an embodiment of the present invention.

[0025] like Figure 1 As shown, in method 100, step S101: obtaining data related to voltage fluctuations. The voltage fluctuation data collected mainly includes: 1) historical data, which is achieved by collecting past voltage fluctuation data, including input / output voltage, current, temperature, etc. 2) real-time monitoring data: sensors monitor current voltage, current, temperature, etc. in real time.

[0026] In one embodiment, the voltage fluctuation related data includes: historical data and real-time monitoring data including at least one of the following characteristics: input voltage, output voltage, current, temperature, and load.

[0027] The foundation for predicting voltage fluctuations is formed by collecting historical and real-time monitoring data, including input / output voltage, current, temperature, and load changes. This data can be used to analyze the current state and historical performance of the power module, providing training data for subsequent prediction models.

[0028] Furthermore, in step S102: using the voltage fluctuation related data, a voltage fluctuation prediction model is trained to analyze the pattern and trend of the voltage fluctuation.

[0029] In one embodiment, voltage fluctuation prediction models are trained using voltage fluctuation data to analyze patterns and trends of voltage fluctuations. The training includes the following data preparation stages:

[0030] First, data cleaning is performed. Missing or abnormal data in the voltage fluctuation data is processed, and outliers and missing values ​​in historical data and real-time monitoring data are removed to obtain raw training data. Prediction-enhancing features are extracted from the raw data, such as statistical features such as the mean, standard deviation, maximum, and minimum values ​​of the voltage.

[0031] Next, feature selection is performed to select features that influence the prediction results. Based on the selected features and their correlations, the feature data in the original training data is transformed or combined to generate valid training data containing new features after the transformation or combination, where the new features include frequency features extracted using Fourier transform. If the number of features is too large, dimensionality reduction can be performed using methods such as principal component analysis (PCA) to reduce the complexity of the model.

[0032] Then, the effective training data is normalized to eliminate the dimensional differences between different features. By normalizing the features so that they have the same dimension, it is easier to process the model.

[0033] Finally, according to the correlation between different features in the effective training data and voltage fluctuations, feature fusion is performed to obtain fused features, and the fused features whose influence coefficient on the prediction results exceeds a threshold are added to the effective training data; wherein, according to the correlation between different features in the effective training data and voltage fluctuations, feature fusion includes: extracting static features and temporal features in the effective training data, and fusing the two features using a context feature fusion network.

[0034] Specifically, in one embodiment, extracting static features and temporal features from the valid training data and fusing the two features using a context feature fusion network includes:

[0035] Extract original static features from the valid training data. These original static features are features that do not change over time, including, for example, the rated power, voltage range, and environmental indicators of the power module. These features can be converted into continuous vector representations through the Embedding layer.

[0036] Utilizing a statistical analysis algorithm, the contextual features of the original static features in the valid training data are calculated to obtain a static feature context vector. In non-image data, global contextual features typically refer to features that provide overall environmental or background information. For static features, global contextual features may refer to global statistical information related to the static features, or their relationship to other static features.

[0037] The original static features are processed using the Embedding layer in the deep learning model to obtain a raw static feature vector. The original static feature vector and the static feature context vector are then weighted concatenated to obtain a fused static feature vector. Concatenation involves concatenating two feature vectors along the same dimension. This is accomplished by analyzing the relationships between static features or calculating the statistical characteristics of static features across the entire dataset and concatenating this information with the static feature vector output by the Embedding layer.

[0038] For the static eigenvector, let S be the static eigenvector, G S is the global context feature vector of the static feature, and the weighted splicing operation can be expressed as:

[0039] Fusion static feature vector F S =αS+βG S +λ SG S⊙G s , where α, β are learnable weight coefficients that can be optimized through the training process. SG is the weight of the interaction term between the static feature vector and the context feature vector, which can also be optimized through the training process. ⊙ represents the element-wise product (Hadamard product) of the vectors.

[0040] Compared to extracting static features, time series features can be extracted from the valid training data using a time series analysis algorithm. These time series features refer to features that change over time, including historical data such as voltage and current. These features can be processed through an LSTM layer to extract long-term dependencies in the time series data.

[0041] The time series features are processed using the LSTM layer in the deep learning model to obtain a time series feature vector.

[0042] Utilizing a statistical analysis algorithm, the context features of the time series features in the effective training data are calculated to obtain a time series feature context vector.

[0043] The time series feature vector and the time series feature context vector are weightedly concatenated to obtain a fused time series feature vector. For time series data, the global context feature can be a statistical feature of the entire time series, such as the average value, standard deviation, maximum value, and minimum value.

[0044] For the time series feature vector, T is the time series feature vector obtained by processing the LSTM layer, G T is the global context feature vector of the time series feature, including the overall statistical features of the time series. Weighted fusion can be performed using the following formula:

[0045] Fusion time series feature vector F T =γT+δG T +λ TG T⊙G T , where F T is the fused time series feature vector, which includes the time series features and their global context features. γ,δ are learnable weight coefficients that can be optimized through the training process. TG is the weight of the interaction term between the time series feature vector and the context feature vector.

[0046] Then, the fused static feature vector and the fused temporal feature vector are weightedly spliced ​​and fused to obtain a fused feature vector. The fused feature vector contains information about the fused static feature vector and the fused temporal feature vector, and can be specifically weighted fused using the following formula:

[0047] F=òF S +ζF T , where F is the final fusion feature vector, which is F S and F T ∈ and ζ are the weights of the fused static feature vector and the fused temporal feature vector, which are used to adjust the importance of the two in the final fused feature vector.

[0048] The final fused feature vector will include at least the original static feature vector, the temporal feature vector, and the corresponding global contextual features. This allows the model to not only capture the information of static and temporal features, but also understand the contextual relationship between these features in the entire dataset.

[0049] The algorithm provided by the present invention not only incorporates static and temporal features and their corresponding global context features, but also adds interaction terms between features, enabling the model to capture the interactions between different features. Those skilled in the art will appreciate that global context features provide additional information to the model, enabling it to more comprehensively utilize information in the input data and improve model performance and predictive accuracy.

[0050] In one embodiment, the step of training the voltage fluctuation prediction model may include:

[0051] First, the valid training data is divided into a training set, a validation set, and a test set. A new sequence feature dataset is constructed using historical data. Given a dataset of total length L, the dataset is divided into training, validation, and test sets according to a certain ratio. When sampling the training set, the first sliding window is first selected at the head of the training set, with an input window time step of T and an output prediction time step of τ. The first window is then slid backward by τ steps, and the second window is selected. This process is repeated until the boundary between the divided training and validation sets is reached.

[0052] Secondly, the training set data is used to train a voltage fluctuation prediction model based on a time series prediction model to analyze the pattern and trend of voltage fluctuation.

[0053] In one embodiment, the time series prediction model includes normalizing all input training set data to the range of [-1, 1]. The following formula can be used:

[0054]

[0055] Among them, x normalized is the normalized standard data, x min and x max are the minimum and maximum values ​​of the original data x, respectively.

[0056] Based on the teachings of the embodiments provided by the present invention, those skilled in the art should understand that, in some cases, linearly mapping data to the range of [-1, 1] is not the only option, and the normalization range needs to be adjusted based on the distribution of the data and the performance feedback of the model. For example, if the distribution of the data is mainly concentrated in the interval [0, 1], then using [0, 1] as the normalization range may be more appropriate. Similarly, if the model is more sensitive to a specific numerical range during training, the normalization strategy can be adjusted to accommodate this sensitivity.

[0057] Furthermore, a multi-layered long-short-term memory (LSTM) network is constructed. The LSTM network includes at least one layer for capturing short-term dependencies and one layer for capturing long-term dependencies. An attention mechanism is added before or after each LSTM layer to weight input features and dynamically adjust attention weights at different time steps based on the current input and context vector.

[0058] Specifically, in one embodiment scenario of constructing a multi-layer long short-term memory network, in order to increase the multiple factors and weights that reflect the dependencies, in addition to considering characteristic factors such as voltage value, temperature, humidity and load changes, in order to more comprehensively reflect all factors affecting voltage fluctuations, more relevant features can be added, such as time factors (such as time period of a day), holiday information, or other environmental factors.

[0059] Assuming that the time feature t (such as hours) and the holiday flag Holiday are added as new features, the new feature vector can be expressed as: X t =[V t ,T t ,H t ,L t ,t,Holiday], where V t is the voltage, T t is the temperature, H t is the humidity, L t is the load change.

[0060] By the above new eigenvector X t , the long short-term memory network can capture the impact of different time periods and holidays on voltage fluctuations.

[0061] The input layer of the long short-term memory network receives the enhanced multidimensional feature vector X at time t t After that, specifically, the state formula at the first layer is:

[0062] f t =σ(W f ·[h t-1 ,x t ]+b f ), i t =σ(W i ·[h t-1 ,x t ]+b i ), o t =σ(W o ·[h t-1 ,x t ]+b o ), h t =o t *tanh(C t ),

[0063] Among them, f t is the activation vector of the forget gate at time t to determine the information to be retained or forgotten in the cell state; i tis the activation vector of the input gate at time t, which determines the information stored in the cell state; is the candidate value of the cell state at time t, used to update the cell state; C t is the cell state at time t, which is used to store information in the sequence and pass it to the next time step; t is the activation vector of the output gate at time t, which is used to determine the amount of information output from the unit state to the hidden state; h t is the hidden state at time t, based on the output gate o t and cell state C t The weighted sum of is used as the input of the next time step or the output of the output layer.

[0064] W f ,W i ,W C ,W o : They are the weight matrices of the forget gate, input gate, unit state, and output gate, respectively, which are used to respond to the input and the previous hidden state. Assume that the input multidimensional feature vector X t If it is 6-dimensional, the weight matrix is ​​a 6*128 matrix, where 128 is the number of LSTM units.

[0065] b f ,b i ,b C ,b o : They are the bias vectors of the forget gate, input gate, unit state, and output gate, respectively, providing a baseline value for the calculation of the corresponding gate.

[0066] σ is the activation function used for the forget gate, input gate, and output gate. Its output range is between 0 and 1, indicating the degree of gate opening. Tanh is the hyperbolic tangent activation function used for candidate memory units. Its output range is between -1 and 1, indicating the strength of new information. t-1 ,x t ] is the hidden state h of the previous time step t-1 and the input x at the current time step t * is an element-by-element multiplication operation used to calculate the update of the unit state. t-1 is the hidden state at time t-1, which contains the information of the previous time step. t-1 is the cell state at time t-1, which contains information from the previous time steps in the sequence.

[0067] Each of these gates has its own weight matrix and bias vector. These parameters can be learned through training to adapt to the specific data and task. The size of the weight matrix depends on the dimensionality of the input and the number of units in the hidden layer.

[0068] To allow the model to automatically learn which features are more important at the current time step during training, we can add an attention mechanism to weight the input features. This allows the model to not only rely on a fixed hidden state at each step, but also adaptively adjust the attention weights for different time steps based on the current input and previous context, allowing the model to automatically learn which features are more important at the current time step during training.

[0069] Specifically, the attention weight α t It is calculated through an additional fully connected layer and its formula can be expressed as:

[0070] α t =softmax(W a ·[h t ;h t-1 ;...;h1]+b a ),

[0071] Among them, W a is the weight matrix of the attention layer, b a is the bias vector. The vector α t Each element of corresponds to the weight of the hidden state at the previous time step.

[0072] Through the explanation of the above embodiments, those skilled in the art can understand that the output h of the LSTM layer t Afterwards, an attention vector α can be introduced t , weighting the hidden state so that the model can focus on important information at different time steps. Further, through the attention weight α t and hidden state h t , build a context vector c t , which contains the historical information that the model considers important at the current time step.

[0073] Specifically, in one embodiment, assuming that an attention layer is added after the second LSTM layer, the specific steps are as follows:

[0074] Calculate attention weight: α t =softmax(W a ·[h t (2) ,x t ]+b a ),in, is the hidden state of the second layer of LSTM.

[0075] Then, the formula for constructing the context vector includes:

[0076]

[0077] Among them, the context vector c t is the weighted sum of all previous hidden states, with weights given by α t Decide.

[0078] Further, the context vector is combined with the current hidden state: c t Directly as input for subsequent processing, or use it with After splicing, it is used as new input:

[0079]

[0080] Finally, the weighted eigenvector can be It can be passed to the third LSTM layer or used directly in the output layer for prediction.

[0081] By introducing an additional fully connected layer to calculate attention weights and then using these weights to construct a context vector, the feature weights at different time steps can be dynamically adjusted. This mechanism enables the model to focus on key information more flexibly and improves prediction accuracy.

[0082] Furthermore, in one embodiment, a Dropout layer may be added after the third LSTM layer, and the dropout rate may be preset to 0.2. The formula may include:

[0083]

[0084] Furthermore, in one embodiment, the output layer can output the prediction result through a fully connected layer. Formulas can include:

[0085]

[0086] Finally, the model can be trained using mean squared error as the loss function and the Adam optimizer. During training, early stopping can be used to prevent overfitting.

[0087] After training is completed, you can use the trained model to make predictions:

[0088] In one embodiment, the long short-term memory network can be used to input a multidimensional feature vector X at time point t. tThe multidimensional feature vector contains at least one or more of the following features: voltage value, temperature, humidity, load change, time point (e.g., time of day, daytime / nighttime), and other external factors (e.g., holidays). Based on the input multidimensional feature vector, the long-term and short-term dependencies in the time series data are captured to obtain voltage fluctuation prediction results. Through training, the model can learn the patterns and trends of voltage fluctuations and thus predict future voltage changes.

[0089] The trained voltage fluctuation prediction model is then further validated using the validation set data to obtain a validation result. Based on the validation result, the voltage fluctuation prediction model can be adjusted to optimize the voltage fluctuation prediction model; the adjustment of the voltage fluctuation prediction model includes adjusting model parameters, feature combinations, or model structure.

[0090] Finally, the test set data is used to evaluate the prediction result indicators of the optimized voltage fluctuation prediction model to obtain an evaluation result.

[0091] Further, in step S103: using the voltage fluctuation prediction model, inputting real-time monitoring data to obtain a prediction result of voltage fluctuation;

[0092] By inputting real-time monitoring data into a trained voltage fluctuation prediction model, voltage fluctuation prediction results can be obtained, which can provide decision support for adaptive control strategies and is the prerequisite for achieving dynamic adjustment.

[0093] Further, in step S104: according to the prediction result of the voltage fluctuation, the working state of the power supply is dynamically adjusted through an adaptive control strategy; wherein, the adaptive control strategy includes automatically adjusting the output voltage and current of the power module through an energy storage and release system according to the change of the voltage wave bottom; storing energy when the voltage wave bottom is lower than the set threshold of the reference voltage, and releasing energy when the voltage wave bottom is higher than the set threshold of the reference voltage, so as to smooth the voltage fluctuation.

[0094] Dynamically adjust the power supply's operating state: Based on the voltage fluctuation prediction results obtained in step S103, the power supply module's operating state is dynamically adjusted through an adaptive control strategy. This step is the core of intelligent power management, enabling the power supply module to adapt to voltage fluctuations and maintain stable output.

[0095] In one embodiment, dynamically adjusting the working state of the power supply through an adaptive control strategy based on the prediction result of the voltage fluctuation includes:

[0096] Based on the predicted results of voltage fluctuations, the changes in the voltage wave bottom are judged to automatically adjust the output voltage and current of the power module; the automatic adjustment includes adjusting the output current and voltage of the inverter in the power module, and controlling the charging and discharging process of the power module energy storage system.

[0097] Adaptive control strategies involve automatically adjusting the output voltage and current of power modules based on changes in the voltage base through an energy storage and release system. This strategy effectively smooths voltage fluctuations and improves the stability and efficiency of the power system.

[0098] In one embodiment, one or more reference voltage levels are first set based on the grid's standard operating voltage and historical voltage data. The reference voltage level can be the grid's average voltage or a voltage range defined by the power system operator. Real-time monitoring of the grid's voltage level can then be used to capture voltage fluctuations. This data can be used to determine whether the current voltage is at its bottom. The real-time monitored voltage level is compared with the reference voltage. If the real-time voltage is below a certain threshold of the reference voltage, the voltage bottom is considered low; if it is above another threshold, the voltage bottom is considered high.

[0099] The reference voltage threshold is dynamically set based on the power grid's conditions and historical data. For example, during peak hours, the reference voltage may vary, depending on specific grid conditions and safety standards. This allows for efficient management of the power supply's energy storage and release to respond to voltage fluctuations, ensuring stable and reliable power output.

[0100] Furthermore, some embodiments also include a fault analysis model. By using the voltage prediction model to input real-time monitoring data, in addition to obtaining voltage fluctuation prediction results, the fault analysis model can also be used to determine the cause and location of power module failures, thereby improving the reliability and maintenance efficiency of the smart power supply.

[0101] In summary, real-time monitoring of grid voltage and load changes provides the necessary input data for the power module. Using the established voltage prediction model, trends and patterns of voltage fluctuations are predicted. Based on these predictions, the power module's output voltage and current can be dynamically adjusted to maintain stable power output. Maintaining stable voltage output reduces adverse effects on connected devices and extends their service life.

[0102] Figure 2 FIG. 2 shows an exemplary structural block diagram of a voltage-based intelligent voltage adjustment system 200 according to an embodiment of the present invention.

[0103] like Figure 2As shown, a voltage-based intelligent power adjustment system is provided in system 200, including:

[0104] Data acquisition module 201: used to obtain data related to voltage fluctuations.

[0105] The prediction model training module 202 is configured to train a voltage fluctuation prediction model using the voltage fluctuation related data to analyze the pattern and trend of the voltage fluctuation.

[0106] The real-time prediction module 203 is used to utilize the voltage fluctuation prediction model and input the real-time monitoring data to obtain the prediction result of the voltage fluctuation.

[0107] Power supply adjustment module 204: used to dynamically adjust the working state of the power supply through an adaptive control strategy based on the predicted results of voltage fluctuations; wherein, the adaptive control strategy includes automatically adjusting the output voltage and current of the power supply module through an energy storage and release system according to the changes in the voltage wave bottom; storing energy when the voltage wave bottom is lower than the set threshold of the reference voltage, and releasing energy when the voltage wave bottom is higher than the set threshold of the reference voltage, so as to smooth the voltage fluctuations.

[0108] It can be understood that the system 200 is a specific implementation of the method 100, so the above Figure 1 The features described can be applied analogously here.

[0109] Based on the above, the present invention further discloses a computer-readable storage medium, which contains computer program instructions storing a voltage-based intelligent power supply adjustment method. When the computer program instructions are executed by a processor, the method described in the above multiple embodiments or implementation methods is implemented.

[0110] In some implementation scenarios, the computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium. Any such computer storage medium may be part of, or accessible to, or connectable to, a device or apparatus. Any application or module described herein may be implemented using computer-readable / executable instructions stored or otherwise retained by such a computer-readable medium.

[0111] In summary, the embodiment provided by the present invention combines historical data with real-time monitoring data (input / output voltage, current, temperature, etc.) to train a voltage fluctuation prediction model to analyze the patterns and trends of voltage fluctuations. This multi-feature analysis method can provide a more comprehensive and accurate voltage fluctuation prediction and can better adapt to the complex and changing power grid environment compared to traditional single-feature prediction models. In addition, based on the predicted results of voltage fluctuations, the solution of the present invention not only considers the changes in the voltage bottom through the designed adaptive control strategy, but also automatically adjusts the output voltage and current of the power module through the energy storage and release system. This integrated solution can more effectively smooth voltage fluctuations and improve the stability and efficiency of the power supply system. Therefore, through the real-time monitoring and adjustment of the power supply operating status of the present invention, more stable and reliable digital intelligent power management can be achieved. Moreover, through the integration of the adaptive control strategy and the energy storage and release system, it is also possible to achieve high power density and miniaturization of power modules based on high frequency, magnetic integration, packaging, modularization and other technologies, thereby improving the packaging technology of power modules.

[0112] Although a number of embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art may devise numerous modifications, variations, and alternatives without departing from the concept and spirit of the present invention. It should be understood that in practicing the present invention, various alternatives to the embodiments of the present invention described herein may be employed. The appended claims are intended to define the scope of the present invention and therefore cover equivalents or alternatives within the scope of these claims.

Claims

1. A voltage-based intelligent power supply adjustment method, characterized in that: include: Obtain data related to voltage fluctuations; Using the voltage fluctuation related data, a voltage fluctuation prediction model is trained to analyze the pattern and trend of voltage fluctuation; Using the voltage fluctuation prediction model, inputting real-time monitoring data to obtain a voltage fluctuation prediction result; Dynamically adjusting the operating state of the power supply through an adaptive control strategy based on the predicted results of the voltage fluctuation; wherein the adaptive control strategy includes automatically adjusting the output voltage and current of the power supply module through an energy storage and release system based on changes in the voltage wave bottom; storing energy when the voltage wave bottom is lower than a set threshold of the reference voltage, and releasing energy when the voltage wave bottom is higher than the set threshold of the reference voltage, so as to smooth the voltage fluctuation; The voltage fluctuation prediction model is trained using voltage fluctuation data to analyze the patterns and trends of voltage fluctuations. The data preparation phase includes the following: Processing missing or abnormal data in the voltage fluctuation related data to obtain original training data; Transforming or combining feature data in the original training data to generate effective training data containing new features after the transformation or combination, wherein the new features include frequency features extracted using Fourier transform; Normalizing the effective training data to eliminate dimensional differences between different features; According to the correlation between different features in the effective training data and voltage fluctuations, feature fusion is performed to obtain fused features, and fused features whose influence coefficient on the prediction results exceeds a threshold are added to the effective training data; wherein, according to the correlation between different features in the effective training data and voltage fluctuations, feature fusion includes: extracting static features and temporal features in the effective training data, and fusing the two features using a context feature fusion network.

2. The intelligent power supply adjustment method according to claim 1, wherein: The voltage fluctuation related data includes: historical data and real-time monitoring data including at least one of the following characteristics: input voltage, output voltage, current, temperature, and load.

3. The intelligent power supply adjustment method according to claim 1, wherein: The step of extracting static features and temporal features from the effective training data and fusing the two features using a context feature fusion network includes: Extracting original static features from the valid training data; the original static features include the rated power, voltage range, and environmental indicators of the smart power supply; Calculate the context features of the original static features in the valid training data using a statistical analysis algorithm to obtain a static feature context vector; Processing the original static features using an Embedding layer in a deep learning model to obtain an original static feature vector; Performing weighted concatenation on the original static feature vector and the static feature context vector to obtain a fused static feature vector; Extracting time series features from the effective training data using a time series analysis algorithm; the time series features include historical data of voltage and current; Processing the time series features using an LSTM layer in a deep learning model to obtain a time series feature vector; Calculate the context features of the time series features in the valid training data using a statistical analysis algorithm to obtain a time series feature context vector; Performing weighted concatenation on the time series feature vector and the time series feature context vector to obtain a fused time series feature vector; The fused static feature vector and the fused temporal feature vector are weightedly concatenated and fused to obtain a fused feature vector.

4. The intelligent power supply adjustment method according to claim 1, wherein: The training voltage fluctuation prediction model includes: Dividing the valid training data into a training set, a validation set, and a test set; Using the training set data, a voltage fluctuation prediction model is trained according to a time series prediction model to analyze patterns and trends of voltage fluctuations; Validating the trained voltage fluctuation prediction model using the validation set data to obtain a validation result; According to the verification result, the voltage fluctuation prediction model is adjusted to optimize the voltage fluctuation prediction model; the adjusting of the voltage fluctuation prediction model includes adjusting model parameters, feature combinations or model structures; The test set data is used to evaluate the prediction result indicators of the optimized voltage fluctuation prediction model to obtain an evaluation result.

5. The intelligent power supply adjustment method according to claim 4, wherein: The time series prediction model includes: Normalize all input training set data to the range of [-1, 1]; Constructing a multi-layer long short-term memory network, the long short-term memory network including at least one layer for capturing short-term dependencies and one layer for capturing long-term dependencies; wherein an attention mechanism is added before or after each layer of the long short-term memory network to weight input features and dynamically adjust attention weights at different time steps based on the current input and context vector; Using the long short-term memory network, a multidimensional feature vector is input at time point t, wherein the multidimensional feature vector includes at least one or more of the following features: voltage value, temperature, humidity, and load change; According to the multi-dimensional feature vector, the long-term and short-term dependencies in the time series data are captured to obtain the voltage fluctuation prediction result.

6. The intelligent power supply adjustment method according to claim 1, wherein: The method of dynamically adjusting the working state of the power supply through an adaptive control strategy based on the prediction result of the voltage fluctuation includes: Based on the predicted results of voltage fluctuations, the changes in the voltage wave bottom are judged to automatically adjust the output voltage and current of the power module; the automatic adjustment includes adjusting the output current and voltage of the inverter in the power module, and controlling the charging and discharging process of the power module energy storage system.

7. The intelligent power supply adjustment method according to claim 1, wherein: The threshold value of the reference voltage is set dynamically according to the power grid conditions and historical data of the power supply.

8. A voltage-based intelligent power supply adjustment system, characterized in that: include: Data acquisition module: used to obtain relevant data of voltage fluctuation; Prediction model training module: used to train a voltage fluctuation prediction model using the voltage fluctuation related data to analyze the pattern and trend of voltage fluctuation; Real-time prediction module: used to use the voltage fluctuation prediction model and input real-time monitoring data to obtain the prediction result of voltage fluctuation; Power supply adjustment module: used to dynamically adjust the operating state of the power supply through an adaptive control strategy based on the predicted results of voltage fluctuations. The adaptive control strategy includes automatically adjusting the output voltage and current of the power supply module based on changes in the voltage wave bottom through an energy storage and release system. When the voltage wave bottom is lower than the set threshold of the reference voltage, energy is stored, and when the voltage wave bottom is higher than the set threshold of the reference voltage, energy is released to smooth out voltage fluctuations. The voltage fluctuation prediction model is trained using voltage fluctuation data to analyze the patterns and trends of voltage fluctuations. The data preparation phase includes the following: Processing missing or abnormal data in the voltage fluctuation related data to obtain original training data; Transforming or combining feature data in the original training data to generate effective training data containing new features after the transformation or combination, wherein the new features include frequency features extracted using Fourier transform; Normalizing the effective training data to eliminate dimensional differences between different features; According to the correlation between different features in the effective training data and voltage fluctuations, feature fusion is performed to obtain fused features, and fused features whose influence coefficient on the prediction results exceeds a threshold are added to the effective training data; wherein, according to the correlation between different features in the effective training data and voltage fluctuations, feature fusion includes: extracting static features and temporal features in the effective training data, and fusing the two features using a context feature fusion network.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the voltage-based intelligent power supply adjustment method according to any one of claims 1 to 7 is implemented.

Images