Self-adaptive power supply energy-saving configuration method and system based on energy gradient utilization

Abstract

The invention provides a self-adaptive power supply energy-saving configuration method and system based on energy gradient utilization, and relates to the technical field of power supply management, and the method comprises the steps: collecting the operation parameters of electric equipment, building a power utilization model, and carrying out the self-adaptive correction; constructing a multi-stage energy distribution matrix according to the corrected power utilization model, and performing energy efficiency grading on the power utilization equipment; performing spatial-temporal feature extraction by adopting a bidirectional attention mechanism, generating an electric equipment priority strategy in combination with a self-calibration network, and further calculating an energy distribution proportion of each level of electric equipment; and finally, generating a control signal according to the energy distribution proportion, adjusting the power supply power of each stage of electric equipment, distributing the exceeded power to the electric equipment with the secondary high energy conversion efficiency when the power value exceeds a target power threshold value, and outputting a self-adaptive energy balance control instruction according to the prediction control model to adjust the operation state of the electric equipment.

Description

Technical Field

[0001] The invention relates to power management technology, and in particular to an adaptive power energy-saving configuration method and system based on energy cascade utilization. Background Art

[0002] With the continuous increase in energy demand, how to improve power efficiency and reduce energy consumption has become a core issue of global concern. Existing power energy-saving technologies usually rely on static control methods and experience-based load scheduling, which cannot adapt to changes in the operating status of equipment and environmental fluctuations in real time. Traditional methods fail to fully consider the dynamic characteristics of equipment, multi-dimensional load characteristics, and energy efficiency changes, resulting in low energy allocation efficiency and failure to achieve optimal energy-saving effects under changing conditions. Therefore, there is an urgent need for a more intelligent, flexible, and adaptable power energy-saving configuration method.

[0003] At present, most energy-saving solutions rely on preset models or static energy efficiency evaluation indicators, and lack dynamic collection and analysis of real-time equipment operation data. Especially when faced with complex equipment operation and the coexistence of multiple loads, existing technologies are usually unable to effectively optimize power distribution, resulting in energy waste and system inefficiency. In addition, traditional energy-saving systems usually lack adaptive capabilities and cannot make real-time adjustments based on changes in multiple factors such as ambient temperature and equipment operating parameters, which limits their actual application effects.

[0004] Therefore, there is an urgent need for a method that can dynamically monitor and optimize equipment energy efficiency based on real-time data streams and historical databases. Summary of the invention

[0005] The embodiment of the present invention provides an adaptive power supply energy-saving configuration method and system based on energy cascade utilization, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention, Provided is an adaptive power supply energy-saving configuration method based on energy cascade utilization, comprising: Collect the operating parameters of the power-consuming equipment to establish a historical database and a real-time data stream, build a dynamic state observation model based on the historical database, perform state estimation in combination with the real-time data stream to obtain initial energy loss characteristics, build a multi-scale time-frequency analysis model based on the initial energy loss characteristics, use a collaborative iteration method to perform feature decomposition on the multi-scale time-frequency analysis model to obtain energy change characteristics, perform deep feature extraction on the energy change characteristics and equipment load characteristics to establish an initial power consumption model, use a recursive least squares method to adaptively correct the power characteristic parameters, energy conversion parameters and temperature response parameters in the initial power consumption model to obtain a corrected power consumption model, and the corrected power consumption model includes a power characteristic curve, an energy conversion efficiency curve and a temperature response characteristic curve; A multi-level energy allocation matrix is ​​constructed according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, and the energy efficiency evaluation index of the multi-level energy allocation matrix is ​​calculated to perform energy efficiency grading on the power consumption equipment to obtain the equipment grading result. Based on the equipment grading result and the pre-constructed load coupling evaluation index, an adaptive fuzzy neural network is used to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold. According to the target power threshold and the power characteristic curve, an operation timing matrix is ​​constructed, and a bidirectional attention mechanism is used to extract spatiotemporal features of the operation timing matrix and an energy optimization model is constructed in combination with a self-calibration network to generate a priority strategy for power consumption equipment. The energy allocation ratio of each level of power consumption equipment is calculated in combination with the energy conversion efficiency curve. A pulse width modulation control signal is generated according to the energy allocation ratio, the power supply of each level of electrical equipment is adjusted, the power value of the power supply is detected in real time, when the power value exceeds the target power threshold, the excess power is allocated to the electrical equipment with the second highest energy conversion efficiency according to the electrical equipment priority strategy, and the optimal operating power is calculated according to the power characteristic curve, the optimal operating power is set as the updated target power threshold and a predictive control model is generated, an adaptive energy balance control instruction is output according to the predictive control model, and the operating state of the electrical equipment is adjusted according to the adaptive energy balance control instruction.

[0007] In an optional embodiment, A dynamic state observation model is constructed according to the historical database, and the state estimation is performed in combination with the real-time data stream to obtain initial energy loss characteristics. A multi-scale time-frequency analysis model is constructed based on the initial energy loss characteristics, and the energy change characteristics are obtained by feature decomposing the multi-scale time-frequency analysis model in a collaborative iteration manner, including: An autoregressive moving average model is constructed according to the historical database to obtain a dynamic state observation model, wherein the dynamic state observation model includes a state transfer equation and an observation equation, and the order of the dynamic state observation model is determined by the Akaike information criterion based on the white noise sequence of the state transfer equation and the observation equation; Input the real-time data stream into the dynamic state observation model, use the particle filter algorithm based on importance sampling to perform state estimation, calculate the particle weight value according to the likelihood function of the state observation value, correct the particle weight value by the system resampling method to eliminate particle degradation, and use the weighted mean estimation method to obtain the initial energy loss characteristic sequence, the initial energy loss characteristic sequence includes power loss amount, energy conversion efficiency and temperature change amount; A multi-scale time-frequency analysis model of wavelet packet transform is constructed based on the initial energy loss feature sequence, an optimal wavelet basis function is selected to perform multi-layer decomposition on the initial energy loss feature sequence to obtain a wavelet packet coefficient matrix, an adaptive threshold function is constructed according to the noise standard deviation of each frequency band of the wavelet packet coefficient matrix, and the wavelet packet coefficient matrix is ​​optimized to obtain an optimized time-frequency feature matrix; Constructing a characteristic correlation matrix of the optimized time-frequency characteristic matrix, performing eigenvalue decomposition on the characteristic correlation matrix using a singular value decomposition method to obtain an eigenvalue sequence and an eigenvector group, and determining the number of principal components based on the energy contribution rate of the eigenvalue sequence; A eigenvector group corresponding to the number of principal components is selected to construct an initial multidimensional energy change feature, the initial multidimensional energy change feature is multiplied by the feature correlation matrix to obtain a feature projection amount, the sum of the squares of the differences between the feature projection amount and the corresponding eigenvalue is calculated to obtain an iterative objective function value, the iteration is terminated when the iterative objective function value is less than a preset threshold, the current eigenvector group is determined as the optimal eigenvector group, and a final energy change feature is constructed based on the optimal eigenvector group, the final energy change feature including a power characteristic change amount, an energy conversion efficiency change amount, and a temperature response change amount.

[0008] In an optional embodiment, Constructing a multi-level energy allocation matrix according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, calculating the energy efficiency evaluation index of the multi-level energy allocation matrix, and performing energy efficiency grading on power consumption equipment to obtain equipment grading results include: Collecting the operating parameters of the electrical equipment to obtain the power characteristic curve and energy conversion efficiency curve of the electrical equipment, calculating the product of the physical layout parameters of the electrical equipment and the energy conversion efficiency curve to obtain the energy transfer function, performing time-series mapping on the energy transfer function and the power characteristic curve to obtain the energy transfer coefficient between the electrical equipment; mapping the energy transfer coefficient to the corresponding matrix position based on the physical connection topology relationship of the electrical equipment to construct a multi-level energy allocation matrix; Performing a ratio operation on the measured efficiency value of the electrical equipment and the industry benchmark efficiency value to obtain a static energy efficiency index, performing a time domain integration operation on the operating efficiency curve of the electrical equipment and the rated efficiency value to obtain a dynamic energy efficiency index, and normalizing the energy transfer coefficient in the multi-level energy allocation matrix to obtain a coupling energy efficiency index; performing a weighted combination operation on the static energy efficiency index, the dynamic energy efficiency index and the coupling energy efficiency index to obtain a comprehensive energy efficiency evaluation index of the electrical equipment; An initial fuzzy membership matrix of electrical equipment for different energy efficiency levels is constructed, the cluster center of each energy efficiency level is calculated based on the comprehensive energy efficiency evaluation index, the fuzzy membership matrix is ​​updated according to the Euclidean distance between the cluster center and the comprehensive energy efficiency evaluation index, and iteration is repeated until the membership difference between two adjacent iterations is less than a preset membership threshold, so as to obtain the energy efficiency grading result of the electrical equipment.

[0009] In an optional embodiment, Based on the equipment classification result and the pre-constructed load coupling evaluation index, an adaptive fuzzy neural network is used to fuse the ambient temperature parameters, the equipment operation parameters and the load characteristic parameters to calculate the target power threshold, and an operation timing matrix is ​​constructed according to the target power threshold and the power characteristic curve. A bidirectional attention mechanism is used to extract the spatiotemporal features of the operation timing matrix and an energy optimization model is constructed in combination with a self-calibration network to generate a priority strategy for electrical equipment. The energy allocation ratio of each level of electrical equipment is calculated in combination with the energy conversion efficiency curve, including: Collect physical layout data, process flow data and operation sequence data of electrical equipment, calculate the distance matrix between electrical equipment according to the physical layout data to obtain the physical space coupling coefficient, construct the upstream and downstream dependency diagram of electrical equipment according to the process flow data to obtain the functional coupling coefficient, and calculate the opening time overlap of electrical equipment according to the operation sequence data to obtain the timing coupling coefficient; perform product operation on the physical space coupling coefficient, the functional coupling coefficient and the timing coupling coefficient to obtain the load coupling evaluation index; The equipment classification result, the load coupling evaluation index, the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter are input into an adaptive fuzzy neural network, the adaptive fuzzy neural network performs fuzzification processing on each parameter to obtain a membership value, performs a neural network reasoning operation based on the membership value, and generates a target power threshold; Constructing a running time matrix according to the target power threshold and the power characteristic curve; applying a bidirectional attention mechanism to the running time matrix, calculating the weight coefficients of different time steps through the time dimension attention network to obtain time series features, calculating the correlation strength between different devices through the space dimension attention network to obtain space features, and performing tensor fusion of the time series features and the space features to obtain a multi-dimensional feature matrix; The multidimensional feature matrix is ​​input into a self-calibration network, the self-calibration network performs feature enhancement on the multidimensional feature matrix through a residual connection module, selectively filters the enhanced features through a gating unit, and outputs an optimized feature vector after batch normalization processing; The optimized feature vector is input into the feature dimension reduction network to obtain a priority scoring vector, a priority constraint matrix is ​​generated based on the functional coupling coefficient and the timing coupling coefficient, and the priority scoring vector and the priority constraint matrix are weightedly fused to construct an energy optimization model; the energy conversion efficiency curve is piecewise linearly fitted to obtain an equipment energy consumption characteristic function, and the energy optimization model is substituted into the equipment energy consumption characteristic function to calculate the energy allocation ratio of each level of electrical equipment in different load ranges.

[0010] In an optional embodiment, Performing piecewise linear fitting on the energy conversion efficiency curve to obtain a device energy consumption characteristic function, substituting the energy optimization model into the device energy consumption characteristic function to calculate the energy distribution ratio of each level of electrical equipment in different load intervals includes: According to the historical operation data of the equipment, the operating power and energy conversion efficiency are sorted in ascending order, and the efficiency variance within the window is calculated using a sliding window. The location where the variance has a local maximum value is determined as the segmentation point to obtain multiple load intervals. A basic energy consumption characteristic function is constructed for each load interval by using a linear interpolation method, and energy consumption characteristic functions of adjacent intervals are subjected to smooth transition processing based on a smoothing factor to obtain a smoothed energy consumption characteristic function; The priority scoring vector is combined with the priority constraint matrix, wherein the elements of the constraint matrix are multiplied by the timing correlation coefficient to obtain the optimization weight, the smoothed energy consumption characteristic function is multiplied by the optimization weight and normalized, and the energy allocation ratio of each level of electrical equipment in different load intervals is calculated; The efficiency gradient of the smoothed energy consumption characteristic function is calculated. When the efficiency gradient exceeds a preset threshold, the energy allocation ratio is recalculated based on the current load power, otherwise the current allocation strategy is kept unchanged.

[0011] In an optional embodiment, Real-time detection of the power value of the power supply power, when the power value exceeds the target power threshold, allocating the excess power to the power device with the second highest energy conversion efficiency according to the power device priority strategy, and calculating the optimal operating power according to the power characteristic curve includes: Acquire real-time power data, extract the power value and power change rate in the real-time power data to construct a state vector, construct a state space equation according to the state vector, calculate the system error covariance matrix, obtain the optimal filter gain based on the error covariance matrix and the measured noise characteristics, iteratively update the optimal filter gain using the recursive least squares method, and use the updated filter gain to perform Kalman filtering on the power data; Establishing an exponentially weighted integration window for the filtered power supply value, calculating the deviation between the power supply value and the target power threshold within the exponentially weighted integration window, and extracting the excess power at the current moment when the deviation exceeds the decision threshold continuously; Collecting energy conversion efficiency parameters of each of the electric devices in real time, sorting the energy conversion efficiency parameters in descending order according to their numerical values, generating a priority list of electric devices, and selecting the electric device with the second highest energy conversion efficiency as the target electric device according to the priority list of electric devices; Constructing an optimization objective function including an energy conversion efficiency term and a power deviation term, taking the power characteristic curve and operating boundary conditions of the target electrical equipment as constraints, and taking the excess power as an optimization variable; Initialize the position and velocity of the particle swarm, normalize the position, calculate the fitness value of each particle, select the particle position with the best fitness as the global optimal solution, calculate the inertia weight based on the convergence characteristic of the optimization objective function, dynamically update the velocity component of the particle according to the inertia weight, and use the entropy method to calculate the position distribution characteristics and velocity distribution characteristics of the particle swarm to obtain the population diversity index; When the population diversity index is lower than a preset threshold, a Gaussian mutation operation is performed on the local optimal particle, the mutation probability of the Gaussian mutation operation is inversely proportional to the distance of the particle to the global optimal position, the fitness value of the mutated particle is recalculated and the global optimal solution is updated, and when the convergence condition is met, the global optimal solution is output as the optimal operating power.

[0012] In an optional embodiment, The optimal operating power is set as the updated target power threshold and a prediction control model is generated, an adaptive energy balance control instruction is output according to the prediction control model, and the operating state of the electrical device is adjusted according to the adaptive energy balance control instruction, including: The optimal operating power is set as the target power threshold, the optimal operating power, the actual operating power and the historical power sequence at the current moment are collected, and an autoregressive neural network with an input layer, a hidden layer and an output layer is constructed. The hidden layer adopts a hyperbolic tangent activation function, and the optimal operating power, the actual operating power and the historical power sequence are input into the autoregressive neural network to obtain an initial predicted power; The prediction error mean square error between the initial predicted power and the actual operating power is calculated using historical data, the network weights of the autoregressive neural network are learned online using a gradient descent method with momentum terms, a regularization term is introduced to suppress overfitting, and a corrected power prediction sequence is obtained, and a prediction control model including the prediction error mean square error is established according to the corrected power prediction sequence; The power tracking error is obtained by calculating the deviation between the corrected power prediction sequence and the target power threshold according to the prediction control model, an energy balance objective function including a power tracking error term and a control increment term is constructed, the error weight and the control weight in the energy balance objective function are dynamically allocated using fuzzy rules, and the energy balance objective function is solved to obtain an initial control sequence; Establishing a time delay predictor including a controlled object model, combining the output of the time delay predictor with an initial control sequence based on the predictive control model to obtain a compensation control sequence that eliminates the influence of the time delay; An adaptive energy balance control instruction including a target power value and an adjustment rate limit is generated according to the compensation control sequence, and the adaptive energy balance control instruction is sent to a control unit of the electrical equipment through a proportional-integral-differential controller to adjust the operating state of the electrical equipment.

[0013] A second aspect of an embodiment of the present invention provides an adaptive power supply energy-saving configuration system based on energy cascade utilization, comprising: The first unit is used to collect the operating parameters of the power-consuming equipment to establish a historical database and a real-time data stream, construct a dynamic state observation model based on the historical database, perform state estimation in combination with the real-time data stream to obtain initial energy loss characteristics, construct a multi-scale time-frequency analysis model based on the initial energy loss characteristics, perform feature decomposition on the multi-scale time-frequency analysis model in a collaborative iteration manner to obtain energy change characteristics, perform deep feature extraction on the energy change characteristics and equipment load characteristics to establish an initial power consumption model, and use recursive least squares method to adaptively correct power characteristic parameters, energy conversion parameters and temperature response parameters in the initial power consumption model to obtain a corrected power consumption model, wherein the corrected power consumption model includes a power characteristic curve, an energy conversion efficiency curve and a temperature response characteristic curve; The second unit is used to construct a multi-level energy allocation matrix according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, calculate the energy efficiency evaluation index of the multi-level energy allocation matrix, perform energy efficiency classification on the power consumption equipment to obtain the equipment classification result, and based on the equipment classification result and the pre-constructed load coupling evaluation index, use an adaptive fuzzy neural network to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold, and construct a running timing matrix according to the target power threshold and the power characteristic curve, use a bidirectional attention mechanism to extract the spatiotemporal features of the running timing matrix and combine it with the self-calibration network to construct an energy optimization model to generate a priority strategy for power consumption equipment, and calculate the energy allocation ratio of each level of power consumption equipment in combination with the energy conversion efficiency curve; The third unit is used to generate a pulse width modulation control signal according to the energy allocation ratio, adjust the power supply of electrical equipment at each level, detect the power value of the power supply in real time, and when the power value exceeds the target power threshold, allocate the excess power to the electrical equipment with the second highest energy conversion efficiency according to the electrical equipment priority strategy, and calculate the optimal operating power according to the power characteristic curve, set the optimal operating power as the updated target power threshold and generate a predictive control model, output an adaptive energy balance control instruction according to the predictive control model, and adjust the operating state of the electrical equipment according to the adaptive energy balance control instruction.

[0014] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0016] In this embodiment, through the multi-scale time-frequency analysis model and the adaptively corrected power consumption model, the power characteristics, energy conversion efficiency and temperature response characteristics of the equipment can be accurately grasped, and energy efficiency classification and energy allocation ratio can be optimized based on this, thereby improving energy utilization efficiency. Based on the adaptive fuzzy neural network and the two-way attention mechanism, a variety of parameters are integrated to build an energy optimization model, and a power equipment priority strategy is generated to achieve intelligent control and management of power equipment and avoid energy waste. Through the predictive control model and the adaptive energy balance control instruction, the operating state of the power equipment can be dynamically adjusted according to the real-time power value and the target power threshold, to achieve dynamic balance of energy and enhance the stability and adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic flow chart of an adaptive power supply energy-saving configuration method based on energy cascade utilization according to an embodiment of the present invention; Figure 2 It is a structural schematic diagram of an adaptive power supply energy-saving configuration system based on energy cascade utilization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0020] Figure 1 FIG. 1 is a flow chart of an adaptive power supply energy saving configuration method based on energy cascade utilization according to an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Collect the operating parameters of the power-consuming equipment to establish a historical database and a real-time data stream, construct a dynamic state observation model based on the historical database, perform state estimation in combination with the real-time data stream to obtain initial energy loss characteristics, construct a multi-scale time-frequency analysis model based on the initial energy loss characteristics, use a collaborative iteration method to perform feature decomposition on the multi-scale time-frequency analysis model to obtain energy change characteristics, perform deep feature extraction on the energy change characteristics and equipment load characteristics to establish an initial power consumption model, use the recursive least squares method to adaptively correct the power characteristic parameters, energy conversion parameters and temperature response parameters in the initial power consumption model to obtain a corrected power consumption model, and the corrected power consumption model includes a power characteristic curve, an energy conversion efficiency curve and a temperature response characteristic curve; S102. Construct a multi-level energy allocation matrix based on the power characteristic curve and energy conversion efficiency curve in the corrected power consumption model, calculate the energy efficiency evaluation index of the multi-level energy allocation matrix to perform energy efficiency classification on the power consumption equipment to obtain the equipment classification result, and based on the equipment classification result and the pre-constructed load coupling evaluation index, use an adaptive fuzzy neural network to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold, and construct an operation timing matrix based on the target power threshold and the power characteristic curve, use a bidirectional attention mechanism to extract the spatiotemporal features of the operation timing matrix and combine it with the self-calibration network to construct an energy optimization model to generate a priority strategy for power consumption equipment, and calculate the energy allocation ratio of each level of power consumption equipment in combination with the energy conversion efficiency curve; S103. Generate a pulse width modulation control signal according to the energy allocation ratio, adjust the power supply of electrical equipment at each level, detect the power value of the power supply in real time, and when the power value exceeds the target power threshold, allocate the excess power to the electrical equipment with the second highest energy conversion efficiency according to the electrical equipment priority strategy, and calculate the optimal operating power according to the power characteristic curve, set the optimal operating power as the updated target power threshold and generate a predictive control model, output an adaptive energy balance control instruction according to the predictive control model, and adjust the operating state of the electrical equipment according to the adaptive energy balance control instruction.

[0021] The operating parameters of the equipment include power, energy consumption, load characteristics, temperature response and other information, through which the operating status of the equipment can be comprehensively evaluated. Through the collected data, a historical database and real-time data stream are established to provide necessary data support for subsequent dynamic state observation and energy efficiency analysis. The historical database contains the energy efficiency performance of the equipment in different time periods, while the real-time data stream reflects the status of the equipment under the current operating conditions. The combination of the two helps to build a more accurate operation model of the equipment.

[0022] The energy change characteristics include the power consumption characteristics of the device under different load conditions, and take into account the impact of ambient temperature changes on energy efficiency. These energy change characteristics and the load characteristics of the device are deeply extracted, and the model is adaptively corrected by combining the recursive least squares method to adjust the power characteristic parameters, energy conversion parameters and temperature response parameters, and finally an accurate power consumption model is obtained. This corrected power consumption model can accurately reflect the energy efficiency performance of the device under different conditions. The power characteristic curve, energy conversion efficiency curve and temperature response characteristic curve in the model respectively reflect the relationship between the power consumption of the device and the load change, the change of energy conversion efficiency with load and environmental changes, and the impact of device temperature on energy efficiency.

[0023] Next, a multi-level energy allocation matrix is ​​constructed based on the power characteristic curve and energy conversion efficiency curve in the corrected power consumption model. This matrix is ​​used to classify the energy efficiency of different levels of electrical equipment. By combining these classification results with the load coupling evaluation index, an adaptive fuzzy neural network is used to integrate the ambient temperature, operating parameters and load characteristics of the equipment to calculate the target power threshold. The target power threshold is a key indicator that is dynamically adjusted based on the energy efficiency performance and load requirements of the equipment.

[0024] The priority strategy ensures the priority of high-efficiency equipment in energy distribution, and calculates the energy allocation ratio of each level of electrical equipment to ensure that the overall energy efficiency of the system is maximized.

[0025] In an optional implementation, a dynamic state observation model is constructed according to the historical database, state estimation is performed in combination with the real-time data stream to obtain initial energy loss characteristics, a multi-scale time-frequency analysis model is constructed based on the initial energy loss characteristics, and a collaborative iteration method is used to perform feature decomposition on the multi-scale time-frequency analysis model to obtain energy change characteristics, including: An autoregressive moving average model is constructed according to the historical database to obtain a dynamic state observation model, wherein the dynamic state observation model includes a state transfer equation and an observation equation, and the order of the dynamic state observation model is determined by the Akaike information criterion based on the white noise sequence of the state transfer equation and the observation equation; Input the real-time data stream into the dynamic state observation model, use the particle filter algorithm based on importance sampling to perform state estimation, calculate the particle weight value according to the likelihood function of the state observation value, correct the particle weight value by the system resampling method to eliminate particle degradation, and use the weighted mean estimation method to obtain the initial energy loss characteristic sequence, the initial energy loss characteristic sequence includes power loss amount, energy conversion efficiency and temperature change amount; A multi-scale time-frequency analysis model of wavelet packet transform is constructed based on the initial energy loss feature sequence, an optimal wavelet basis function is selected to perform multi-layer decomposition on the initial energy loss feature sequence to obtain a wavelet packet coefficient matrix, an adaptive threshold function is constructed according to the noise standard deviation of each frequency band of the wavelet packet coefficient matrix, and the wavelet packet coefficient matrix is ​​optimized to obtain an optimized time-frequency feature matrix; Constructing a characteristic correlation matrix of the optimized time-frequency characteristic matrix, performing eigenvalue decomposition on the characteristic correlation matrix using a singular value decomposition method to obtain an eigenvalue sequence and an eigenvector group, and determining the number of principal components based on the energy contribution rate of the eigenvalue sequence; A eigenvector group corresponding to the number of principal components is selected to construct an initial multidimensional energy change feature, the initial multidimensional energy change feature is multiplied by the feature correlation matrix to obtain a feature projection amount, the sum of the squares of the differences between the feature projection amount and the corresponding eigenvalue is calculated to obtain an iterative objective function value, the iteration is terminated when the iterative objective function value is less than a preset threshold, the current eigenvector group is determined as the optimal eigenvector group, and a final energy change feature is constructed based on the optimal eigenvector group, the final energy change feature including a power characteristic change amount, an energy conversion efficiency change amount, and a temperature response change amount.

[0026] Exemplarily, first, a dynamic state observation model is established using a large amount of historical data. Specifically, an autoregressive moving average model is used to model historical data, which contains a state transfer equation and an observation equation to describe the dynamic changes in the system state and the relationship between the observed values. The order of the model is determined by the Akaike information criterion to balance the complexity and fitting accuracy of the model. For example, suppose we have power system operation data for the past year, including hourly power output, energy conversion efficiency, and temperature. By analyzing this data, we can determine a suitable order of an autoregressive moving average model, such as order 3, to describe the dynamic characteristics of the system.

[0027] Next, the real-time data stream is input into the established dynamic state observation model, and the particle filter algorithm based on importance sampling is used for state estimation. The particle filter algorithm approximates the posterior probability distribution of the system state through a series of weighted particles. The weight of each particle reflects its similarity to the actual state. In order to avoid particle degradation, the particle weights are corrected by the system resampling method to ensure that the particles are concentrated in the high likelihood area. Then, the weighted mean estimation method is used to calculate the initial energy loss feature sequence, which contains information such as power loss, energy conversion efficiency, and temperature change. For example, assuming that the real-time data stream contains information such as power output, energy conversion efficiency, and temperature at the current moment, this information is input into the dynamic state observation model. Through the particle filter algorithm, it can be estimated that the power loss at the current moment is 10kW, the energy conversion efficiency is 90%, and the temperature change is 2℃.

[0028] Then, a multi-scale time-frequency analysis model is constructed based on the initial energy loss feature sequence. Specifically, the wavelet packet transform is used to perform multi-layer decomposition on the initial energy loss feature sequence. The wavelet packet transform can decompose the signal into sub-signals of different frequency bands, thereby revealing the characteristics of the signal at different time scales and frequency scales. The optimal wavelet basis function is selected to obtain the best time-frequency resolution. An adaptive threshold function is constructed according to the noise standard deviation of each frequency band of the wavelet packet coefficient matrix, and the wavelet packet coefficient matrix is ​​optimized to remove noise interference and obtain the optimized time-frequency feature matrix. For example, the Daubechies wavelet is selected as the basis function, the initial energy loss feature sequence is decomposed into 8 frequency bands, and the threshold is set according to the noise standard deviation of each frequency band to denoise the wavelet packet coefficients.

[0029] Finally, the optimized time-frequency feature matrix is ​​subjected to eigendecomposition to extract the energy change features. First, the feature correlation matrix of the time-frequency feature matrix is ​​constructed. Then, the singular value decomposition method is used to perform eigenvalue decomposition on the feature correlation matrix to obtain eigenvalue sequences and eigenvector groups. The number of principal components is determined according to the energy contribution rate of the eigenvalue sequence, and the eigenvector groups corresponding to the principal components with the largest energy contribution rates are selected to construct the initial multidimensional energy change features. The initial multidimensional energy change features are multiplied by the feature correlation matrix to obtain the feature projection amount, and the sum of the squares of the differences between the feature projection amount and the corresponding eigenvalues ​​is calculated to obtain the iterative objective function value. Through iterative optimization, the iterative objective function value is made less than the preset threshold, and finally the optimal eigenvector group is obtained. The final energy change features are constructed based on the optimal eigenvector group, including the power characteristic change amount, the energy conversion efficiency change amount, and the temperature response change amount. For example, assuming that five principal components are obtained after eigenvalue decomposition, the eigenvectors corresponding to the first three principal components with the largest energy contribution rates are selected to construct the initial multidimensional energy change characteristics. After iterative optimization, the final energy change characteristics are obtained, such as a power characteristic change of 0.5%, an energy conversion efficiency change of -0.2%, and a temperature response change of 0.1°C / kW.

[0030] In this embodiment, by combining the dynamic state observation model and multi-scale time-frequency analysis, the dynamic change characteristics of energy loss can be captured more accurately, thereby improving the accuracy of feature extraction. The adaptive threshold function is used to optimize the wavelet packet coefficient matrix, which can effectively remove noise interference and enhance the robustness of the feature. This method can extract multi-dimensional energy change characteristics such as power characteristic changes, energy conversion efficiency changes, and temperature response changes, providing more comprehensive information for the state assessment of the energy system.

[0031] In an optional implementation, a multi-level energy allocation matrix is ​​constructed according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, and the energy efficiency evaluation index of the multi-level energy allocation matrix is ​​calculated to perform energy efficiency grading on the power consumption equipment to obtain the equipment grading result, including: Collecting the operating parameters of the electrical equipment to obtain the power characteristic curve and energy conversion efficiency curve of the electrical equipment, calculating the product of the physical layout parameters of the electrical equipment and the energy conversion efficiency curve to obtain the energy transfer function, performing time-series mapping on the energy transfer function and the power characteristic curve to obtain the energy transfer coefficient between the electrical equipment; mapping the energy transfer coefficient to the corresponding matrix position based on the physical connection topology relationship of the electrical equipment to construct a multi-level energy allocation matrix; Performing a ratio operation on the measured efficiency value of the electrical equipment and the industry benchmark efficiency value to obtain a static energy efficiency index, performing a time domain integration operation on the operating efficiency curve of the electrical equipment and the rated efficiency value to obtain a dynamic energy efficiency index, and normalizing the energy transfer coefficient in the multi-level energy allocation matrix to obtain a coupling energy efficiency index; performing a weighted combination operation on the static energy efficiency index, the dynamic energy efficiency index and the coupling energy efficiency index to obtain a comprehensive energy efficiency evaluation index of the electrical equipment; An initial fuzzy membership matrix of electrical equipment for different energy efficiency levels is constructed, the cluster center of each energy efficiency level is calculated based on the comprehensive energy efficiency evaluation index, the fuzzy membership matrix is ​​updated according to the Euclidean distance between the cluster center and the comprehensive energy efficiency evaluation index, and iteration is repeated until the membership difference between two adjacent iterations is less than a preset membership threshold, so as to obtain the energy efficiency grading result of the electrical equipment.

[0032] Exemplarily, the operating parameters of the electrical equipment are first collected to obtain the power characteristic curve and energy conversion efficiency curve. This step requires real-time monitoring of the target electrical equipment to collect its operating parameters such as voltage, current, power, efficiency, etc. For example, for a central air conditioner, it is necessary to collect the operating parameters of key components such as the compressor, fan, and control system. The collection frequency can be set according to the operating characteristics of the equipment, such as collecting once per second or once per minute. Through the collected data, the power characteristic curve of the equipment can be drawn to reflect the law of equipment power changes over time; and the energy conversion efficiency curve can reflect the law of equipment energy conversion efficiency changes over time.

[0033] Calculate the energy transfer function. Calculate the energy transfer function based on the physical layout parameters of the device, such as the distance between the devices, the connection method, etc., and the energy conversion efficiency curve obtained previously. Specifically, the physical layout parameters of the device can be multiplied by the energy conversion efficiency curve to obtain the energy transfer function. For example, if the distance between the two devices is far, the loss during energy transfer is large, and the energy transfer function value is low. Conversely, if the distance between the two devices is close, the loss during energy transfer is small, and the energy transfer function value is high.

[0034] Perform time series mapping to obtain the energy transfer coefficient. Perform time series mapping on the energy transfer function obtained in the previous step and the power characteristic curve to obtain the energy transfer coefficient between devices. Time series mapping refers to matching the energy transfer function values ​​at different time points with the power characteristic curve values, thereby reflecting the energy transfer efficiency at different time points. For example, at a certain time point, if the power of the device is high, the energy transfer efficiency may also be high, and the corresponding energy transfer coefficient is large.

[0035] Construct a multi-level energy distribution matrix. According to the physical connection topology of the devices, map the energy transfer coefficient to the corresponding matrix position to construct a multi-level energy distribution matrix. For example, if device A is connected to device B, fill the energy transfer coefficient from device A to device B in the position of row A and column B in the matrix. By constructing a multi-level energy distribution matrix, the network structure of energy transfer between devices can be clearly displayed.

[0036] Calculate the static energy efficiency index. Ratio the measured efficiency value of the equipment with the industry benchmark efficiency value to obtain the static energy efficiency index. For example, if the measured efficiency of a central air conditioner is 80% and the industry benchmark efficiency is 70%, then its static energy efficiency index is 80% / 70%=1.14.

[0037] Calculate the dynamic energy efficiency index. Perform time domain integration on the equipment's operating efficiency curve and rated efficiency value to obtain the dynamic energy efficiency index. Time domain integration refers to integrating the equipment's operating efficiency curve and rated efficiency value over a period of time to reflect the overall operating efficiency of the equipment over a period of time.

[0038] Calculate the coupling energy efficiency index. The energy transfer coefficient in the multi-level energy allocation matrix is ​​normalized to obtain the coupling energy efficiency index. Normalization refers to scaling the value of the energy transfer coefficient to between 0 and 1 to facilitate comparison of the energy transfer efficiency between different devices.

[0039] Calculate the comprehensive energy efficiency evaluation index. Perform weighted combination operations on the static energy efficiency index, dynamic energy efficiency index, and coupling energy efficiency index to obtain the comprehensive energy efficiency evaluation index of the equipment. For example, the weight of the static energy efficiency index can be set to 0.3, the weight of the dynamic energy efficiency index can be set to 0.5, and the weight of the coupling energy efficiency index can be set to 0.2, and then the weighted sum can be performed.

[0040] Construct an initial fuzzy membership matrix. According to the preset energy efficiency level, construct the initial fuzzy membership matrix of the equipment for different energy efficiency levels. For example, the energy efficiency level can be divided into four levels: excellent, good, qualified, and unqualified.

[0041] Calculate the cluster center and update the fuzzy membership matrix. Based on the comprehensive energy efficiency evaluation index, calculate the cluster center of each energy efficiency level. Then, update the fuzzy membership matrix according to the Euclidean distance between the cluster center and the comprehensive energy efficiency evaluation index. Euclidean distance refers to the distance between two vectors. Repeat the iteration until the difference between the membership of two consecutive iterations is less than the preset membership threshold.

[0042] The energy efficiency classification results of the equipment are obtained. The final fuzzy membership matrix reflects the possibility of the equipment belonging to different energy efficiency levels. For example, if a device has a membership of 0.8 in the "excellent" level and a membership of 0.2 in the "good" level, it can be classified as "excellent".

[0043] In this embodiment, the method comprehensively considers the static efficiency, dynamic efficiency and energy coupling relationship between the equipment, and can evaluate the energy efficiency level of the equipment more comprehensively and accurately. By grading the energy efficiency of the equipment, more targeted energy-saving measures can be taken to achieve more refined energy efficiency management. This method can provide users with a more scientific basis for equipment selection, help users choose more efficient equipment, and reduce energy consumption.

[0044] In an optional implementation, based on the equipment classification result and the pre-constructed load coupling evaluation index, an adaptive fuzzy neural network is used to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold, an operation timing matrix is ​​constructed according to the target power threshold and the power characteristic curve, a bidirectional attention mechanism is used to extract spatiotemporal features of the operation timing matrix, and an energy optimization model is constructed in combination with a self-calibration network to generate a priority strategy for electrical equipment, and the energy allocation ratio of each level of electrical equipment is calculated in combination with the energy conversion efficiency curve, including: Collect physical layout data, process flow data and operation sequence data of electrical equipment, calculate the distance matrix between electrical equipment according to the physical layout data to obtain the physical space coupling coefficient, construct the upstream and downstream dependency diagram of electrical equipment according to the process flow data to obtain the functional coupling coefficient, and calculate the opening time overlap of electrical equipment according to the operation sequence data to obtain the timing coupling coefficient; perform product operation on the physical space coupling coefficient, the functional coupling coefficient and the timing coupling coefficient to obtain the load coupling evaluation index; The equipment classification result, the load coupling evaluation index, the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter are input into an adaptive fuzzy neural network, the adaptive fuzzy neural network performs fuzzification processing on each parameter to obtain a membership value, performs a neural network reasoning operation based on the membership value, and generates a target power threshold; Constructing a running time matrix according to the target power threshold and the power characteristic curve; applying a bidirectional attention mechanism to the running time matrix, calculating the weight coefficients of different time steps through the time dimension attention network to obtain time series features, calculating the correlation strength between different devices through the space dimension attention network to obtain space features, and performing tensor fusion of the time series features and the space features to obtain a multi-dimensional feature matrix; The multidimensional feature matrix is ​​input into a self-calibration network, the self-calibration network performs feature enhancement on the multidimensional feature matrix through a residual connection module, selectively filters the enhanced features through a gating unit, and outputs an optimized feature vector after batch normalization processing; The optimized feature vector is input into the feature dimension reduction network to obtain a priority scoring vector, a priority constraint matrix is ​​generated based on the functional coupling coefficient and the timing coupling coefficient, and the priority scoring vector and the priority constraint matrix are weightedly fused to construct an energy optimization model; the energy conversion efficiency curve is piecewise linearly fitted to obtain an equipment energy consumption characteristic function, and the energy optimization model is substituted into the equipment energy consumption characteristic function to calculate the energy allocation ratio of each level of electrical equipment in different load ranges.

[0045] For example, first, it is necessary to collect the physical layout data of the electrical equipment, such as the coordinate position of the equipment, the floor space, etc. Assume that there are two devices A and B, A is located at (1, 2), and B is located at (4, 5). At the same time, collect the process flow data of the equipment, such as the upstream and downstream relationship between the equipment, for example, A is the upstream equipment of B. Finally, collect the running sequence data of the equipment, such as the start time and duration of the equipment, for example, the running time of A is 8:00-12:00, and the running time of B is 10:00-14:00.

[0046] Calculate the load coupling evaluation index: Based on the collected data, calculate the degree of coupling between devices. First, calculate the distance according to the coordinate position of the device, for example, the distance between A and B is 5. Convert the distance into a physical space coupling coefficient. For example, the closer the distance, the larger the coefficient. Assume that the physical space coupling coefficient of A and B is 0.8. Then, build an upstream and downstream dependency graph based on the process flow data and convert it into a functional coupling coefficient. For example, the closer the upstream and downstream relationship, the larger the coefficient. Assume that the functional coupling coefficient of A and B is 0.9. Finally, count the overlap of device start-up time based on the running sequence data and convert it into a timing coupling coefficient. For example, the longer the overlap time, the larger the coefficient. Assume that the timing coupling coefficient of A and B is 0.7. Multiply the physical space coupling coefficient, the functional coupling coefficient and the timing coupling coefficient to obtain the load coupling evaluation index. For example, the load coupling evaluation index of A and B is 0.8×0.9×0.7=0.504.

[0047] Calculate the target power threshold: Input the equipment classification results, load coupling evaluation index, ambient temperature (e.g. 25°C), equipment operation parameters (e.g. A's rated power is 10kW, B's rated power is 5kW), and load characteristic parameters (e.g. A's load rate is 0.8, B's load rate is 0.6) into the adaptive fuzzy neural network. The network first fuzzifies these parameters, for example, fuzzifies the temperature of 25°C into "suitable temperature" and fuzzifies A's rated power of 10kW into "high power". Then, neural network reasoning operations are performed based on these fuzzy values ​​to finally generate the target power threshold, for example, A's target power threshold is 8kW, and B's target power threshold is 3kW.

[0048] Constructing a runtime matrix: Constructing a runtime matrix based on the target power threshold and the power characteristic curve (e.g., a curve showing power changes over time). For example, the power values ​​of A and B at each hour of a day are arranged into a matrix.

[0049] Extracting spatiotemporal features: A bidirectional attention mechanism is applied to the runtime matrix. The time-dimensional attention network calculates the weight coefficients of different time steps, such as a higher weight in the morning and a lower weight in the afternoon, thereby extracting temporal features. The spatial-dimensional attention network calculates the association strength between different devices, such as the higher association strength between A and B, thereby extracting spatial features. The temporal features and spatial features are tensor-fused to obtain a multi-dimensional feature matrix.

[0050] Constructing an energy optimization model: Input the multidimensional feature matrix into the self-calibration network. The network enhances features through the residual connection module and filters features through the gating unit, and finally outputs the optimized feature vector. Input the optimized feature vector into the feature dimension reduction network to obtain the priority score vector. Generate the priority constraint matrix based on the functional coupling coefficient and the temporal coupling coefficient. The priority score vector and the priority constraint matrix are weighted and fused to construct the energy optimization model.

[0051] Calculate the energy allocation ratio: Perform piecewise linear fitting on the energy conversion efficiency curve to obtain the equipment energy consumption characteristic function. Substitute the energy optimization model into the equipment energy consumption characteristic function to calculate the energy allocation ratio of each level of electrical equipment in different load ranges. For example, in the peak load range, the energy allocation ratio of A is 0.7, and the energy allocation ratio of B is 0.3.

[0052] In this embodiment, by prioritizing and optimizing energy distribution for electrical equipment, energy waste can be minimized and energy utilization efficiency can be improved. By optimizing energy use, the energy consumption cost of the enterprise can be reduced and economic benefits can be improved. By considering the coupling relationship and operating sequence between devices, device overload and conflict can be avoided, and the stability and reliability of the system can be improved.

[0053] In an optional implementation, performing piecewise linear fitting on the energy conversion efficiency curve to obtain a device energy consumption characteristic function, substituting the energy optimization model into the device energy consumption characteristic function to calculate the energy distribution ratio of each level of electrical equipment in different load intervals includes: According to the historical operation data of the equipment, the operating power and energy conversion efficiency are sorted in ascending order, and the efficiency variance within the window is calculated using a sliding window. The location where the variance has a local maximum value is determined as the segmentation point to obtain multiple load intervals. A basic energy consumption characteristic function is constructed for each load interval by using a linear interpolation method, and energy consumption characteristic functions of adjacent intervals are subjected to smooth transition processing based on a smoothing factor to obtain a smoothed energy consumption characteristic function; The priority scoring vector is combined with the priority constraint matrix, wherein the elements of the constraint matrix are multiplied by the timing correlation coefficient to obtain the optimization weight, the smoothed energy consumption characteristic function is multiplied by the optimization weight and normalized, and the energy allocation ratio of each level of electrical equipment in different load intervals is calculated; The efficiency gradient of the smoothed energy consumption characteristic function is calculated. When the efficiency gradient exceeds a preset threshold, the energy allocation ratio is recalculated based on the current load power, otherwise the current allocation strategy is kept unchanged.

[0054] Exemplarily, first, historical operation data of the equipment is collected, including the operation power and energy conversion efficiency under different load powers. For example, the operation power and energy conversion efficiency data of a certain level of electrical equipment under 10%, 20%, ..., 100% load power are collected. Then, the collected data is sorted in the order of increasing energy conversion efficiency.

[0055] Next, use a sliding window to process the sorted data. Set a suitable window size, such as a window size of 3, and calculate the efficiency variance in each window in turn. For example, calculate the efficiency variance of the first three data points, the efficiency variance of the second to fourth data points, and so on. The location where the local maximum value of the efficiency variance appears is determined as the segmentation point, thereby dividing the entire load interval into multiple sub-intervals. For example, the efficiency variance has a local maximum value at 50% and 80% load power, and the entire load interval is divided into three sub-intervals: 0-50%, 50-80%, and 80-100%.

[0056] For each load sub-interval, a basic energy consumption characteristic function is constructed by linear interpolation. For example, for the 0-50% load interval, data points at both ends of the interval are selected, and the basic energy consumption characteristic function of the interval is constructed by linear interpolation. Then, the energy consumption characteristic functions of adjacent intervals are smoothly transitioned based on the smoothing factor to obtain a smoothed energy consumption characteristic function. For example, the energy consumption characteristic functions on the left and right sides of the 50% load power point are weighted averaged to ensure a smooth transition of the function at the segmentation point. The smoothing factor can be adjusted according to actual conditions.

[0057] Next, determine the priority scoring vector and constraint matrix for each level of power-consuming equipment. The priority scoring vector reflects the importance of different equipment. For example, the equipment can be scored based on its function, frequency of use, and other factors. The constraint matrix reflects the interdependence between different equipment and its correlation with time. Each element of the constraint matrix is ​​multiplied by a timing correlation coefficient to obtain the optimization weight. For example, if some equipment requires higher energy supply in a specific time period, this can be reflected by adjusting the timing correlation coefficient.

[0058] The priority score vector is combined with the priority constraint matrix after weight optimization, and the result is multiplied by the smoothed energy consumption characteristic function and normalized, and finally the energy allocation ratio of each level of electrical equipment in different load intervals is calculated. For example, it is calculated that a certain level of equipment allocates 30%, 50% and 20% of energy in the three load intervals of 0-50%, 50-80% and 80-100% respectively.

[0059] Finally, the efficiency gradient of the smoothed energy consumption characteristic function is calculated. When the efficiency gradient exceeds the preset threshold, it indicates that the current energy allocation strategy may not be optimal, and the energy allocation ratio needs to be recalculated based on the current load power. Otherwise, the current allocation strategy remains unchanged. For example, when the load power is 60%, if the calculated efficiency gradient exceeds the preset threshold, it is necessary to recalculate the energy allocation ratio of each level of equipment in the 50-80% load range.

[0060] In this embodiment, by accurately modeling and optimizing the energy consumption characteristics of the equipment, energy waste can be effectively reduced and energy utilization efficiency can be improved. By optimizing the energy allocation strategy, the operating energy consumption of the equipment can be reduced, thereby reducing the operating cost. By considering the interdependence and time correlation between devices, system failures caused by unreasonable energy allocation can be avoided and system stability can be enhanced.

[0061] In an optional implementation, the power value of the power supply is detected in real time, and when the power value exceeds the target power threshold, the excess power is allocated to the power device with the second highest energy conversion efficiency according to the power device priority strategy, and the optimal operating power is calculated according to the power characteristic curve, including: Acquire real-time power data, extract the power value and power change rate in the real-time power data to construct a state vector, construct a state space equation according to the state vector, calculate the system error covariance matrix, obtain the optimal filter gain based on the error covariance matrix and the measured noise characteristics, iteratively update the optimal filter gain using the recursive least squares method, and use the updated filter gain to perform Kalman filtering on the power data; Establishing an exponentially weighted integration window for the filtered power supply value, calculating the deviation between the power supply value and the target power threshold within the exponentially weighted integration window, and extracting the excess power at the current moment when the deviation exceeds the decision threshold continuously; Collecting energy conversion efficiency parameters of each of the electric devices in real time, sorting the energy conversion efficiency parameters in descending order according to their numerical values, generating a priority list of electric devices, and selecting the electric device with the second highest energy conversion efficiency as the target electric device according to the priority list of electric devices; Constructing an optimization objective function including an energy conversion efficiency term and a power deviation term, taking the power characteristic curve and operating boundary conditions of the target electrical equipment as constraints, and taking the excess power as an optimization variable; Initialize the position and velocity of the particle swarm, normalize the position, calculate the fitness value of each particle, select the particle position with the best fitness as the global optimal solution, calculate the inertia weight based on the convergence characteristic of the optimization objective function, dynamically update the velocity component of the particle according to the inertia weight, and use the entropy method to calculate the position distribution characteristics and velocity distribution characteristics of the particle swarm to obtain the population diversity index; When the population diversity index is lower than a preset threshold, a Gaussian mutation operation is performed on the local optimal particle, the mutation probability of the Gaussian mutation operation is inversely proportional to the distance of the particle to the global optimal position, the fitness value of the mutated particle is recalculated and the global optimal solution is updated, and when the convergence condition is met, the global optimal solution is output as the optimal operating power.

[0062] Exemplarily, first, real-time power data is collected. The data includes power value and power change rate. A state vector is constructed with the power value and power change rate. For example, if the power value collected at a certain moment is 10 kW and the power change rate is 0.1 kW / s, then the state vector is [10, 0.1].

[0063] Next, construct the state space equation based on the state vector. The state space equation describes the change of the system state over time. According to the measured noise characteristics and the system error covariance matrix, calculate the optimal filter gain. Assuming that the variance of the measured noise is 0.01 and the system error covariance matrix is ​​[[0.1, 0], [0, 0.01]], the optimal filter gain is calculated to be [0.8, 0.05].

[0064] Then, the optimal filter gain is iteratively updated using the recursive least squares method. For example, at the next moment, the filter gain is recalculated and updated to [0.85, 0.06] based on the new state vector and the measured noise. The power data is Kalman filtered using the updated filter gain to obtain a smoothed power value.

[0065] An exponentially weighted integration window is established for the filtered power supply value. For example, the window size is set to 10 sampling points and the attenuation factor is 0.9. The deviation between the power supply value and the target power threshold is calculated within the window. Assuming the target power threshold is 9kW and the power supply value at the current moment is 10kW, the deviation is 1kW. When the deviation exceeds the decision threshold continuously, the excess power at the current moment is extracted. Assuming the decision threshold is 0.5kW, and the deviation exceeds 0.5kW for 5 consecutive sampling points, the excess power at the current moment is extracted, for example, 1kW.

[0066] Collect the energy conversion efficiency parameters of each power-consuming device in real time. For example, the energy conversion efficiency of device A is 90%, device B is 85%, and device C is 80%. Sort these parameters in descending order according to the numerical value to generate a priority list of power-consuming devices: A>B>C. Based on this list, select the power-consuming device with the second highest energy conversion efficiency as the target power-consuming device, that is, device B.

[0067] Construct an optimization objective function that includes energy conversion efficiency and power deviation. Use the power characteristic curve and operating boundary conditions of the target power-consuming device (device B) as constraints. Use the excess power (1kW) as the optimization variable.

[0068] Initialize the position and speed of the particle swarm. For example, initialize 10 particles, the position of each particle represents the operating power of device B, and the speed represents the rate of change of power. Normalize the positions. Calculate the fitness value of each particle. The fitness value reflects the quality of the particle position and is calculated by the optimization objective function. Select the particle position with the best fitness as the global optimal solution.

[0069] The inertia weight is calculated based on the convergence characteristics of the optimization objective function. The inertia weight is used to balance the movement trend of the particle itself and the influence of the group. The velocity component of the particle is dynamically updated according to the inertia weight. The entropy method is used to calculate the position distribution characteristics and velocity distribution characteristics of the particle group to obtain the population diversity index.

[0070] When the population diversity index is lower than the preset threshold, a Gaussian mutation operation is performed on the local optimal particle. The mutation probability is inversely proportional to the distance of the particle to the global optimal position. The fitness value of the mutated particle is recalculated and the global optimal solution is updated. When the convergence conditions are met, such as reaching the maximum number of iterations or the fitness value no longer changes significantly, the global optimal solution is output as the optimal operating power. For example, if the final global optimal solution output is 0.8kW, 0.8kW of power is allocated to device B.

[0071] In this embodiment, by allocating excess power to the device with the second highest energy conversion efficiency, energy is utilized to the maximum extent, energy waste is reduced, and overall energy utilization efficiency is improved. The method can monitor the power supply in real time and effectively control it when the power exceeds the threshold, thereby avoiding damage to the system caused by power overload and ensuring the stable operation of the system. Through the optimization algorithm based on the power characteristic curve and the operating boundary conditions, the optimal operating power of the device can be found, so that it can maintain an efficient and stable operating state while meeting the system requirements.

[0072] In an optional implementation, setting the optimal operating power as an updated target power threshold and generating a predictive control model, outputting an adaptive energy balance control instruction according to the predictive control model, and adjusting the operating state of the electrical device according to the adaptive energy balance control instruction includes: The optimal operating power is set as the target power threshold, the optimal operating power, the actual operating power and the historical power sequence at the current moment are collected, and an autoregressive neural network with an input layer, a hidden layer and an output layer is constructed. The hidden layer adopts a hyperbolic tangent activation function, and the optimal operating power, the actual operating power and the historical power sequence are input into the autoregressive neural network to obtain an initial predicted power; The prediction error mean square error between the initial predicted power and the actual operating power is calculated using historical data, the network weights of the autoregressive neural network are learned online using a gradient descent method with momentum terms, a regularization term is introduced to suppress overfitting, and a corrected power prediction sequence is obtained, and a prediction control model including the prediction error mean square error is established according to the corrected power prediction sequence; The power tracking error is obtained by calculating the deviation between the corrected power prediction sequence and the target power threshold according to the prediction control model, an energy balance objective function including a power tracking error term and a control increment term is constructed, the error weight and the control weight in the energy balance objective function are dynamically allocated using fuzzy rules, and the energy balance objective function is solved to obtain an initial control sequence; Establishing a time delay predictor including a controlled object model, combining the output of the time delay predictor with an initial control sequence based on the predictive control model to obtain a compensation control sequence that eliminates the influence of the time delay; An adaptive energy balance control instruction including a target power value and an adjustment rate limit is generated according to the compensation control sequence, and the adaptive energy balance control instruction is sent to a control unit of the electrical equipment through a proportional-integral-differential controller to adjust the operating state of the electrical equipment.

[0073] Exemplarily, first, the optimal operating power of the electrical equipment is set as the target power threshold, for example, the optimal operating power of an air conditioner is set to 1000 watts.

[0074] Then, the optimal operating power, actual operating power, and historical power sequence of the current electrical equipment are collected. For example, the optimal operating power of the air conditioner at the current moment is collected to be 1000 watts, the actual operating power is 900 watts, and the actual operating power data of each minute in the past hour, such as [910, 905, 900, ..., 920].

[0075] Next, construct an autoregressive neural network with an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is the same as the dimension of the input data, which is the length of the historical power sequence plus the optimal operating power and the actual operating power, for example, 62 neurons. The hidden layer can be set to one or more layers, for example, a layer containing 32 neurons and a hyperbolic tangent activation function. The output layer contains one neuron for outputting the predicted power. The collected optimal operating power, actual operating power, and historical power sequences are input into the constructed autoregressive neural network to obtain the initial predicted power, for example, 915 watts.

[0076] The historical data is used to calculate the mean square error of the prediction error between the initial prediction power and the actual operating power. For example, the square of the difference between the prediction power and the actual operating power for each minute in the past hour is calculated, and then the average is calculated to obtain the mean square error of the prediction error, such as 25.

[0077] The gradient descent method with momentum term is used to learn the network weights of the autoregressive neural network online, and the regularization term is introduced to suppress overfitting, so as to obtain the corrected power prediction sequence. For example, by continuously adjusting the weights of the neural network, the mean square error of the prediction error is gradually reduced, and finally the power prediction sequence for a period of time in the future is obtained.

[0078] According to the modified power prediction sequence, a predictive control model including the mean square error of prediction error is established. The model can describe the relationship between the power prediction sequence, control instructions and the mean square error of prediction error.

[0079] The deviation between the corrected power prediction sequence and the target power threshold is calculated according to the prediction control model to obtain the power tracking error. For example, the difference between the predicted power sequence and 1000 watts is calculated to obtain the power tracking error sequence.

[0080] Construct an energy balance objective function that includes a power tracking error term and a control increment term. For example, the objective function can be expressed as the sum of the square of the power tracking error and the square of the control increment change. Use fuzzy rules to dynamically assign error weights and control weights in the energy balance objective function. For example, when the power tracking error is large, a larger weight is assigned to the error term; when the control increment change is large, a larger weight is assigned to the control increment term. Solve the energy balance objective function and obtain the initial control sequence.

[0081] A delay predictor including a controlled object model is established. For example, the predictor can simulate the time delay between when the electrical equipment receives a control instruction and when it actually changes its operating state. Based on the predictive control model, the output of the delay predictor is combined with the initial control sequence to obtain a compensation control sequence that eliminates the effect of the delay. For example, the initial control sequence is advanced for a period of time to offset the response delay of the electrical equipment to obtain a compensation control sequence.

[0082] Generate an adaptive energy balance control instruction including a target power value and an adjustment rate limit according to the compensation control sequence. For example, convert the compensation control sequence into a specific control instruction, such as adjusting the amplitude of the voltage or current, and limiting the adjustment rate to avoid impacting the electrical equipment. Send the adaptive energy balance control instruction to the control unit of the electrical equipment through the proportional integral differential controller to adjust the operating state of the electrical equipment. For example, send the control instruction to the control unit of the air conditioner to gradually adjust the power output and eventually reach the target power threshold.

[0083] In this embodiment, through adaptive energy balance control, the actual operating power of the electrical equipment can be used to be closer to the optimal operating power, thereby improving energy utilization efficiency. Through predictive control and delay compensation, it is possible to effectively avoid oscillation and overshoot of the electrical equipment during the adjustment process, thereby enhancing the stability of the system. By limiting the adjustment rate and avoiding power fluctuations, the impact and wear on the electrical equipment can be reduced, thereby extending its service life.

[0084] Figure 2 FIG. 1 is a schematic diagram of a structure of an adaptive power supply energy-saving configuration system based on energy cascade utilization according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to collect the operating parameters of the power-consuming equipment to establish a historical database and a real-time data stream, construct a dynamic state observation model based on the historical database, perform state estimation in combination with the real-time data stream to obtain initial energy loss characteristics, construct a multi-scale time-frequency analysis model based on the initial energy loss characteristics, perform feature decomposition on the multi-scale time-frequency analysis model in a collaborative iteration manner to obtain energy change characteristics, perform deep feature extraction on the energy change characteristics and equipment load characteristics to establish an initial power consumption model, and use recursive least squares method to adaptively correct power characteristic parameters, energy conversion parameters and temperature response parameters in the initial power consumption model to obtain a corrected power consumption model, wherein the corrected power consumption model includes a power characteristic curve, an energy conversion efficiency curve and a temperature response characteristic curve; The second unit is used to construct a multi-level energy allocation matrix according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, calculate the energy efficiency evaluation index of the multi-level energy allocation matrix, perform energy efficiency classification on the power consumption equipment to obtain the equipment classification result, and based on the equipment classification result and the pre-constructed load coupling evaluation index, use an adaptive fuzzy neural network to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold, and construct a running timing matrix according to the target power threshold and the power characteristic curve, use a bidirectional attention mechanism to extract the spatiotemporal features of the running timing matrix and combine it with the self-calibration network to construct an energy optimization model to generate a priority strategy for power consumption equipment, and calculate the energy allocation ratio of each level of power consumption equipment in combination with the energy conversion efficiency curve; The third unit is used to generate a pulse width modulation control signal according to the energy allocation ratio, adjust the power supply of electrical equipment at each level, detect the power value of the power supply in real time, and when the power value exceeds the target power threshold, allocate the excess power to the electrical equipment with the second highest energy conversion efficiency according to the electrical equipment priority strategy, and calculate the optimal operating power according to the power characteristic curve, set the optimal operating power as the updated target power threshold and generate a predictive control model, output an adaptive energy balance control instruction according to the predictive control model, and adjust the operating state of the electrical equipment according to the adaptive energy balance control instruction.

[0085] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0086] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0087] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive power supply energy-saving configuration method based on energy cascade utilization, characterized in that: include: Collect the operating parameters of the power-consuming equipment to establish a historical database and a real-time data stream, build a dynamic state observation model based on the historical database, perform state estimation in combination with the real-time data stream to obtain initial energy loss characteristics, build a multi-scale time-frequency analysis model based on the initial energy loss characteristics, use a collaborative iteration method to perform feature decomposition on the multi-scale time-frequency analysis model to obtain energy change characteristics, perform deep feature extraction on the energy change characteristics and equipment load characteristics to establish an initial power consumption model, use a recursive least squares method to adaptively correct the power characteristic parameters, energy conversion parameters and temperature response parameters in the initial power consumption model to obtain a corrected power consumption model, and the corrected power consumption model includes a power characteristic curve, an energy conversion efficiency curve and a temperature response characteristic curve; A multi-level energy allocation matrix is ​​constructed according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, and the energy efficiency evaluation index of the multi-level energy allocation matrix is ​​calculated to perform energy efficiency grading on the power consumption equipment to obtain the equipment grading result. Based on the equipment grading result and the pre-constructed load coupling evaluation index, an adaptive fuzzy neural network is used to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold. According to the target power threshold and the power characteristic curve, an operation timing matrix is ​​constructed, and a bidirectional attention mechanism is used to extract spatiotemporal features of the operation timing matrix and an energy optimization model is constructed in combination with a self-calibration network to generate a priority strategy for power consumption equipment. The energy allocation ratio of each level of power consumption equipment is calculated in combination with the energy conversion efficiency curve. A pulse width modulation control signal is generated according to the energy allocation ratio, the power supply of each level of electrical equipment is adjusted, the power value of the power supply is detected in real time, when the power value exceeds the target power threshold, the excess power is allocated to the electrical equipment with the second highest energy conversion efficiency according to the electrical equipment priority strategy, and the optimal operating power is calculated according to the power characteristic curve, the optimal operating power is set as the updated target power threshold and a predictive control model is generated, an adaptive energy balance control instruction is output according to the predictive control model, and the operating state of the electrical equipment is adjusted according to the adaptive energy balance control instruction.

2. The method according to claim 1, characterized in that: A dynamic state observation model is constructed according to the historical database, and the state estimation is performed in combination with the real-time data stream to obtain the initial energy loss characteristics. A multi-scale time-frequency analysis model is constructed based on the initial energy loss characteristics, and the energy change characteristics are obtained by feature decomposing the multi-scale time-frequency analysis model in a collaborative iteration manner, including: An autoregressive moving average model is constructed according to the historical database to obtain a dynamic state observation model, wherein the dynamic state observation model includes a state transfer equation and an observation equation, and the order of the dynamic state observation model is determined by the Akaike information criterion based on the white noise sequence of the state transfer equation and the observation equation; Input the real-time data stream into the dynamic state observation model, use the particle filter algorithm based on importance sampling to perform state estimation, calculate the particle weight value according to the likelihood function of the state observation value, correct the particle weight value by the system resampling method to eliminate particle degradation, and use the weighted mean estimation method to obtain the initial energy loss characteristic sequence, the initial energy loss characteristic sequence includes power loss amount, energy conversion efficiency and temperature change amount; A multi-scale time-frequency analysis model of wavelet packet transform is constructed based on the initial energy loss feature sequence, an optimal wavelet basis function is selected to perform multi-layer decomposition on the initial energy loss feature sequence to obtain a wavelet packet coefficient matrix, an adaptive threshold function is constructed according to the noise standard deviation of each frequency band of the wavelet packet coefficient matrix, and the wavelet packet coefficient matrix is ​​optimized to obtain an optimized time-frequency feature matrix; Constructing a characteristic correlation matrix of the optimized time-frequency characteristic matrix, performing eigenvalue decomposition on the characteristic correlation matrix using a singular value decomposition method to obtain an eigenvalue sequence and an eigenvector group, and determining the number of principal components based on the energy contribution rate of the eigenvalue sequence; A eigenvector group corresponding to the number of principal components is selected to construct an initial multidimensional energy change feature, the initial multidimensional energy change feature is multiplied by the feature correlation matrix to obtain a feature projection amount, the sum of the squares of the differences between the feature projection amount and the corresponding eigenvalue is calculated to obtain an iterative objective function value, the iteration is terminated when the iterative objective function value is less than a preset threshold, the current eigenvector group is determined as the optimal eigenvector group, and a final energy change feature is constructed based on the optimal eigenvector group, the final energy change feature including a power characteristic change amount, an energy conversion efficiency change amount, and a temperature response change amount.

3. The method according to claim 1, characterized in that Constructing a multi-level energy allocation matrix according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, calculating the energy efficiency evaluation index of the multi-level energy allocation matrix, and performing energy efficiency grading on power consumption equipment to obtain equipment grading results include: Collecting the operating parameters of the electrical equipment to obtain the power characteristic curve and energy conversion efficiency curve of the electrical equipment, calculating the product of the physical layout parameters of the electrical equipment and the energy conversion efficiency curve to obtain the energy transfer function, performing time-series mapping on the energy transfer function and the power characteristic curve to obtain the energy transfer coefficient between the electrical equipment; mapping the energy transfer coefficient to the corresponding matrix position based on the physical connection topology relationship of the electrical equipment to construct a multi-level energy allocation matrix; Performing a ratio operation on the measured efficiency value of the electrical equipment and the industry benchmark efficiency value to obtain a static energy efficiency index, performing a time domain integration operation on the operating efficiency curve of the electrical equipment and the rated efficiency value to obtain a dynamic energy efficiency index, and normalizing the energy transfer coefficient in the multi-level energy allocation matrix to obtain a coupling energy efficiency index; performing a weighted combination operation on the static energy efficiency index, the dynamic energy efficiency index and the coupling energy efficiency index to obtain a comprehensive energy efficiency evaluation index of the electrical equipment; An initial fuzzy membership matrix of electrical equipment for different energy efficiency levels is constructed, the cluster center of each energy efficiency level is calculated based on the comprehensive energy efficiency evaluation index, the fuzzy membership matrix is ​​updated according to the Euclidean distance between the cluster center and the comprehensive energy efficiency evaluation index, and iteration is repeated until the membership difference between two adjacent iterations is less than a preset membership threshold, so as to obtain the energy efficiency grading result of the electrical equipment.

4. The method according to claim 1, characterized in that Based on the equipment classification result and the pre-constructed load coupling evaluation index, an adaptive fuzzy neural network is used to fuse the ambient temperature parameters, the equipment operation parameters and the load characteristic parameters to calculate the target power threshold, and an operation timing matrix is ​​constructed according to the target power threshold and the power characteristic curve. A bidirectional attention mechanism is used to extract the spatiotemporal features of the operation timing matrix and an energy optimization model is constructed in combination with a self-calibration network to generate a priority strategy for electrical equipment. The energy allocation ratio of each level of electrical equipment is calculated in combination with the energy conversion efficiency curve, including: Collect physical layout data, process flow data and operation sequence data of electrical equipment, calculate the distance matrix between electrical equipment according to the physical layout data to obtain the physical space coupling coefficient, construct the upstream and downstream dependency diagram of electrical equipment according to the process flow data to obtain the functional coupling coefficient, and calculate the opening time overlap of electrical equipment according to the operation sequence data to obtain the timing coupling coefficient; perform product operation on the physical space coupling coefficient, the functional coupling coefficient and the timing coupling coefficient to obtain the load coupling evaluation index; The equipment classification result, the load coupling evaluation index, the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter are input into an adaptive fuzzy neural network, the adaptive fuzzy neural network performs fuzzification processing on each parameter to obtain a membership value, performs a neural network reasoning operation based on the membership value, and generates a target power threshold; Constructing a running time matrix according to the target power threshold and the power characteristic curve; applying a bidirectional attention mechanism to the running time matrix, calculating the weight coefficients of different time steps through the time dimension attention network to obtain time series features, calculating the correlation strength between different devices through the space dimension attention network to obtain space features, and performing tensor fusion of the time series features and the space features to obtain a multi-dimensional feature matrix; The multidimensional feature matrix is ​​input into a self-calibration network, the self-calibration network performs feature enhancement on the multidimensional feature matrix through a residual connection module, selectively filters the enhanced features through a gating unit, and outputs an optimized feature vector after batch normalization processing; The optimized feature vector is input into the feature dimension reduction network to obtain a priority scoring vector, a priority constraint matrix is ​​generated based on the functional coupling coefficient and the timing coupling coefficient, and the priority scoring vector and the priority constraint matrix are weightedly fused to construct an energy optimization model; the energy conversion efficiency curve is piecewise linearly fitted to obtain an equipment energy consumption characteristic function, and the energy optimization model is substituted into the equipment energy consumption characteristic function to calculate the energy allocation ratio of each level of electrical equipment in different load ranges.

5. The method according to claim 4, characterized in that Performing piecewise linear fitting on the energy conversion efficiency curve to obtain a device energy consumption characteristic function, substituting the energy optimization model into the device energy consumption characteristic function to calculate the energy distribution ratio of each level of electrical equipment in different load intervals includes: According to the historical operation data of the equipment, the operating power and energy conversion efficiency are sorted in ascending order, and the efficiency variance within the window is calculated using a sliding window. The location where the variance has a local maximum value is determined as the segmentation point to obtain multiple load intervals. A basic energy consumption characteristic function is constructed for each load interval by using a linear interpolation method, and energy consumption characteristic functions of adjacent intervals are subjected to smooth transition processing based on a smoothing factor to obtain a smoothed energy consumption characteristic function; The priority scoring vector is combined with the priority constraint matrix, wherein the elements of the constraint matrix are multiplied by the timing correlation coefficient to obtain the optimization weight, the smoothed energy consumption characteristic function is multiplied by the optimization weight and normalized, and the energy allocation ratio of each level of electrical equipment in different load intervals is calculated; The efficiency gradient of the smoothed energy consumption characteristic function is calculated. When the efficiency gradient exceeds a preset threshold, the energy allocation ratio is recalculated based on the current load power, otherwise the current allocation strategy is kept unchanged.

6. The method according to claim 1, characterized in that Real-time detection of the power value of the power supply power, when the power value exceeds the target power threshold, allocating the excess power to the power device with the second highest energy conversion efficiency according to the power device priority strategy, and calculating the optimal operating power according to the power characteristic curve includes: Acquire real-time power data, extract the power value and power change rate in the real-time power data to construct a state vector, construct a state space equation according to the state vector, calculate the system error covariance matrix, obtain the optimal filter gain based on the error covariance matrix and the measured noise characteristics, iteratively update the optimal filter gain using the recursive least squares method, and use the updated filter gain to perform Kalman filtering on the power data; Establishing an exponentially weighted integration window for the filtered power supply value, calculating the deviation between the power supply value and the target power threshold within the exponentially weighted integration window, and extracting the excess power at the current moment when the deviation exceeds the decision threshold continuously; Collecting energy conversion efficiency parameters of each of the electric devices in real time, sorting the energy conversion efficiency parameters in descending order according to their numerical values, generating a priority list of electric devices, and selecting the electric device with the second highest energy conversion efficiency as the target electric device according to the priority list of electric devices; Constructing an optimization objective function including an energy conversion efficiency term and a power deviation term, taking the power characteristic curve and operating boundary conditions of the target electrical equipment as constraints, and taking the excess power as an optimization variable; Initialize the position and velocity of the particle swarm, normalize the position, calculate the fitness value of each particle, select the particle position with the best fitness as the global optimal solution, calculate the inertia weight based on the convergence characteristic of the optimization objective function, dynamically update the velocity component of the particle according to the inertia weight, and use the entropy method to calculate the position distribution characteristics and velocity distribution characteristics of the particle swarm to obtain the population diversity index; When the population diversity index is lower than a preset threshold, a Gaussian mutation operation is performed on the local optimal particle, the mutation probability of the Gaussian mutation operation is inversely proportional to the distance of the particle to the global optimal position, the fitness value of the mutated particle is recalculated and the global optimal solution is updated, and when the convergence condition is met, the global optimal solution is output as the optimal operating power.

7. The method according to claim 1, characterized in that The optimal operating power is set as the updated target power threshold and a prediction control model is generated, an adaptive energy balance control instruction is output according to the prediction control model, and the operating state of the electrical device is adjusted according to the adaptive energy balance control instruction, including: The optimal operating power is set as the target power threshold, the optimal operating power, the actual operating power and the historical power sequence at the current moment are collected, and an autoregressive neural network with an input layer, a hidden layer and an output layer is constructed. The hidden layer adopts a hyperbolic tangent activation function, and the optimal operating power, the actual operating power and the historical power sequence are input into the autoregressive neural network to obtain an initial predicted power; The prediction error mean square error between the initial predicted power and the actual operating power is calculated using historical data, the network weights of the autoregressive neural network are learned online using a gradient descent method with momentum terms, a regularization term is introduced to suppress overfitting, and a corrected power prediction sequence is obtained, and a prediction control model including the prediction error mean square error is established according to the corrected power prediction sequence; The power tracking error is obtained by calculating the deviation between the corrected power prediction sequence and the target power threshold according to the prediction control model, an energy balance objective function including a power tracking error term and a control increment term is constructed, the error weight and the control weight in the energy balance objective function are dynamically allocated using fuzzy rules, and the energy balance objective function is solved to obtain an initial control sequence; Establishing a time delay predictor including a controlled object model, combining the output of the time delay predictor with an initial control sequence based on the predictive control model to obtain a compensation control sequence that eliminates the influence of the time delay; An adaptive energy balance control instruction including a target power value and an adjustment rate limit is generated according to the compensation control sequence, and the adaptive energy balance control instruction is sent to a control unit of the electrical equipment through a proportional-integral-differential controller to adjust the operating state of the electrical equipment.

8. An adaptive power supply energy-saving configuration system based on energy cascade utilization, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect the operating parameters of the power-consuming equipment to establish a historical database and a real-time data stream, construct a dynamic state observation model based on the historical database, perform state estimation in combination with the real-time data stream to obtain initial energy loss characteristics, construct a multi-scale time-frequency analysis model based on the initial energy loss characteristics, perform feature decomposition on the multi-scale time-frequency analysis model in a collaborative iteration manner to obtain energy change characteristics, perform deep feature extraction on the energy change characteristics and equipment load characteristics to establish an initial power consumption model, and use recursive least squares method to adaptively correct power characteristic parameters, energy conversion parameters and temperature response parameters in the initial power consumption model to obtain a corrected power consumption model, wherein the corrected power consumption model includes a power characteristic curve, an energy conversion efficiency curve and a temperature response characteristic curve; The second unit is used to construct a multi-level energy allocation matrix according to the power characteristic curve and the energy conversion efficiency curve in the corrected power consumption model, calculate the energy efficiency evaluation index of the multi-level energy allocation matrix, perform energy efficiency classification on the power consumption equipment to obtain the equipment classification result, and based on the equipment classification result and the pre-constructed load coupling evaluation index, use an adaptive fuzzy neural network to fuse the ambient temperature parameter, the equipment operation parameter and the load characteristic parameter to calculate the target power threshold, and construct a running timing matrix according to the target power threshold and the power characteristic curve, use a bidirectional attention mechanism to extract the spatiotemporal features of the running timing matrix and combine it with the self-calibration network to construct an energy optimization model to generate a priority strategy for power consumption equipment, and calculate the energy allocation ratio of each level of power consumption equipment in combination with the energy conversion efficiency curve; The third unit is used to generate a pulse width modulation control signal according to the energy allocation ratio, adjust the power supply of electrical equipment at each level, detect the power value of the power supply in real time, and when the power value exceeds the target power threshold, allocate the excess power to the electrical equipment with the second highest energy conversion efficiency according to the electrical equipment priority strategy, and calculate the optimal operating power according to the power characteristic curve, set the optimal operating power as the updated target power threshold and generate a predictive control model, output an adaptive energy balance control instruction according to the predictive control model, and adjust the operating state of the electrical equipment according to the adaptive energy balance control instruction.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images