A time sequence simulation method for new energy cloud energy storage operation mechanism

Abstract

The present application relates to the technical field of power system, particularly relates to a kind of time sequence simulation method of new energy cloud energy storage operation mechanism.The present application includes: collecting the output data of new energy power system, the data collected are cleaned, standardized and formatted processing.User behavior characteristic analysis, the user electricity consumption behavior collected is classified;A multi-time phase generation network model is constructed, the prediction data of future power system state is learned from historical data and is predicted;According to the prediction data of future power system state and real-time data predicted, dynamic bandwidth allocation strategy is constructed, and the charge-discharge plan of energy storage device is formulated;Real-time data and charge-discharge plan generated are used for identification, and the operation state of power system is classified;A multi-dimensional time sequence simulation model is constructed.The various factors influencing cloud energy storage operation are considered, more accurate simulation prediction is provided, and the adjustment and response capability of power grid to fluctuating energy is improved.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a time-series simulation method for the operation mechanism of new energy cloud energy storage. Background Technology

[0002] New energy power systems, such as wind and solar power, are becoming increasingly important because they provide a clean and renewable energy source. However, the volatility and unpredictability of these energy sources pose challenges to the stable operation of the power grid. Existing technologies often lack effective strategies to manage the power supply security of these new energy power systems, especially during peak and off-peak demand periods.

[0003] For example, Chinese patent application number 2021110623418 discloses a method and system for establishing a time-series simulation model of cloud energy storage operation mechanism. The method includes: dividing demand-side users into five categories based on demand-side load conditions and providing different power supply combination modes for each category; obtaining real-time electricity prices during user electricity consumption; and obtaining the power supply relationship between users and cloud energy storage based on the real-time electricity prices and the cloud energy storage time-series charging and discharging strategy; and constructing a time-series simulation model of the cloud energy storage operation mechanism based on the power supply combination mode, the cloud energy storage time-series charging and discharging strategy, and the power supply relationship. This invention addresses different types of users by discussing specific scenarios and arranging suitable combinations of multiple energy storage devices to reduce electricity costs while ensuring power balance; it considers important influencing factors such as user electricity prices and load volume in designing the cloud energy storage time-series simulation model; and it clarifies the charging and discharging conditions of multiple types of cloud energy storage under different operating conditions, obtaining the optimal charging and discharging strategy for cloud energy storage devices.

[0004] The main drawbacks of the above-mentioned technology are:

[0005] Insufficient analysis of user electricity consumption behavior: Existing technologies do not fully consider the differences in electricity consumption behavior among different types of users, and lack in-depth analysis of user behavior characteristics and targeted power supply strategies.

[0006] The combination of energy storage devices is too simple: the advantages of multiple energy storage devices are not fully utilized, and there is a lack of optimization strategies for combining multiple devices.

[0007] Lack of real-time electricity price response: Existing technologies fail to dynamically adjust charging and discharging strategies in accordance with real-time electricity price changes.

[0008] Limitations of the simulation model: It may not have established a time-series simulation model that comprehensively considers multiple dimensions such as energy storage devices, user behavior, and power system status. Summary of the Invention

[0009] In view of the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a time-series simulation method and system for the operation mechanism of cloud energy storage in new energy power systems. Taking into account factors such as user behavior and energy storage device combination, it provides a more comprehensive, flexible and efficient cloud energy storage operation mechanism, which significantly improves the stability of the power system.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A time-series simulation method for the operation mechanism of new energy cloud energy storage includes the following:

[0012] S1. Real-time monitoring of users' electricity consumption data via smart meters, and acquisition of weather information related to electricity consumption data using meteorological data APIs. Electricity consumption data includes:

[0013] Peak and off-peak electricity consumption periods: the specific times that distinguish between peak and off-peak electricity consumption periods;

[0014] Average daily electricity consumption: The total electricity consumption of a user per day;

[0015] Seasonal electricity consumption patterns: Users' electricity consumption habits in different seasons;

[0016] Types and quantities of electrical equipment: The types and quantities of electrical appliances used by the user.

[0017] S2. User behavior feature analysis: Apply hierarchical clustering algorithm to classify user electricity consumption behavior in S1 and identify the characteristics and electricity consumption patterns of different user groups.

[0018] S3. Construct a multi-temporal generative network model. Based on the results of the user behavior feature analysis in S2, construct a multi-temporal generative network model. The multi-temporal generative network model is used to learn from historical data and predict future power system states.

[0019] S4. Based on the predicted data of the future power system state in S3 and the real-time data obtained in S1, construct a dynamic bandwidth allocation strategy and formulate a charging and discharging plan for energy storage devices;

[0020] S5. Time-series signal classification and identification: The classification and identification method is used to identify the real-time data obtained in S1 and the charging and discharging plan generated in S4, integrate historical operating data, and classify the operating status of the power system.

[0021] S6. Construction of a time-series simulation model for cloud energy storage operation mechanism: Based on the results of S1 to S5, a time-series simulation model is constructed that includes multiple dimensions such as energy storage devices, user behavior, and power system status.

[0022] By utilizing the multi-dimensional time-series simulation models established from S1 to S6, the operating status of the cloud energy storage system is comprehensively reflected, providing a simulation environment for testing and optimizing cloud energy storage operation strategies, thereby achieving stable operation of the power system and efficient utilization of energy.

[0023] Furthermore, in S2, the hierarchical clustering algorithm includes:

[0024] S21. Select features from the user electricity consumption behavior collected in S1 that help distinguish different user behaviors, such as peak and off-peak electricity consumption periods, average daily electricity consumption, and seasonal electricity consumption patterns.

[0025] S22. Determine the number of clusters by using silhouette coefficient analysis;

[0026] S23. Perform hierarchical clustering, using Euclidean distance to measure the similarity between samples. Based on the selected distance metric, each user is considered as an initial cluster, and then the most similar cluster pairs are merged step by step. The Euclidean distance formula is:

[0027]

[0028] In the formula: x and y are two samples, and n is the number of features;

[0029] S24. Clustering results evaluation: Use indicators such as silhouette coefficient and Davies-Bouldin index to evaluate the clustering effect and ensure that the clustering results are meaningful.

[0030] The formula for the profile coefficient is: a is the average distance between a sample and its nearest sample in the same cluster, and b is the average distance between a sample and its nearest samples in other clusters.

[0031] The formula for the Davies-Bouldin index is: S i and S j It is the sum of the internal distances between clusters i and j, d(C i C j ) is the distance between clusters i and j.

[0032] S25. Interpretation and Application of Results: Interpret the clustering results, identify the characteristics and power consumption patterns of each cluster, arrange the characteristics and power consumption patterns of each cluster in chronological order, and record the characteristics at each time point. Apply the clustering results to formulate targeted power supply strategies.

[0033] Furthermore, in S3, the multi-temporal generation network model includes layers of the following types:

[0034] Input layer: The input layer receives time series data, which is the data in S2 after user behavior feature analysis;

[0035] Recurrent layer: GRU is used as a recurrent layer to process time series data. The recurrent layer is used to process sequence data, capture dynamic features and long-term dependencies in the time series, and the final output of the recurrent layer is a feature sequence.

[0036] Fully connected layer: Use a fully connected layer to integrate the output of the recurrent layer, mapping the output of the recurrent layer to a dense high-dimensional space for classification or regression tasks, learning high-level feature representations of the data;

[0037] Generative layer: The generative layer generates new data points or sequences based on the output of the fully connected layer.

[0038] Furthermore, the GRU is a popular variant of the recurrent neural network (RNN), particularly suitable for processing and predicting time series data. The GRU includes:

[0039] Update gate: determines how much information is extracted from the current input x. t Inflow into hidden state h t ;

[0040] Reset Gate: Determines how to handle the hidden state h from the previous time step. t-1 Allows the network to reset or remember the previous state;

[0041] Candidate hidden states: These are candidates for the hidden states at the current time step, without considering previous hidden states;

[0042] Final hidden state: This is the final hidden state at the current time step. It is updated by combining the update gate, the reset gate, and the candidate hidden state, where:

[0043] Update Gate:

[0044] z t =σ(W z ·[h t-1 ,x t ]+b z );

[0045] By updating the gate formula, we know that: z t It is based on the hidden state h of the previous time step. t-1 and the current input x t The combination of these factors updates the gate's weight matrix W through the weight matrix. z Perform a linear transformation, then add a bias term to update the bias term b of the gate. z Finally, the output value z of the updated gate is obtained by performing a nonlinear transformation using the sinmoid function σ. tThis will determine the candidate hidden state at the current time step. The hidden state h of the previous time step t-1 Between these points, how much information will be updated, that is, how much new information will be passed to the next time step;

[0046] Reset Door:

[0047] r t =σ(W r ·[h t-1 ,x t ]+b r );

[0048] From the reset gate formula, we know that: r t It is based on the hidden state h of the previous time step. t-1 and the input data x at the current time step t The combination of these is achieved by resetting the gate weight matrix W. r Then add the offset term b for resetting the door. r Finally, the output value r of the reset gate is obtained through a nonlinear transformation using the sinmoid function σ. t Calculate the candidate hidden state at the current time step At that time, the hidden state h of the previous time step t-1 How much weight should be assigned to r? t Approaching 1, h t-1 In calculation The time will be retained if r t h is close to 0 t-1 This will be ignored, thus allowing the network to reset its state.

[0049] Candidate hidden state:

[0050]

[0051] The candidate hidden state formula shows that the candidate hidden state is based on the hidden state h from the previous time step. t-1 The output r is obtained by resetting the gate at the current time step. t Adjust, plus the current input x t The impact, and the weight matrices W and W x The candidate hidden state is obtained by a linear transformation of the bias term b, followed by a nonlinear transformation using the tanh function. Together with the update gate, it will determine the final hidden state of the current time step.

[0052] Final hidden state:

[0053]

[0054] Indicates the candidate hidden state Based on the output value z of the update gate t complement (1-z) t The weighted contribution to the final hidden state;

[0055] z t ·h t-1 h represents the hidden state at the previous time step. t-1 Based on the output value z of the update gate t Update the output value z of the gate t ;

[0056] The formula for the final hidden state shows that the final hidden state h at the current time step can be obtained. t It contains both the candidate hidden states at the current time step and... The information also retains the hidden state h from the previous time step. t-1 Information;

[0057] In the formula, z t The update gate output is the value at the current time step. σ is the sinmoid activation function, whose output value ranges from 0 to 1 and is commonly used in gating mechanisms. W z Update the gate weight matrix, W z h is used to perform a linear transformation on the input data and the hidden state of the previous time step. t-1 b is the hidden state of the previous time step. z It is the bias term of the updated gate, x t The input data for the current time step; r t W is the reset gate output for the current time step. r The weight matrix of the reset gate is used to linearly transform the input data and the hidden state of the previous time step, h. t-1 b is the hidden state of the previous time step. r This resets the door's bias.

[0058] represents the candidate hidden state at the current time step; tanh is the hyperbolic tangent function, a non-linear activation function used to introduce non-linear characteristics, with an output value ranging from -1 to 1; W is the weight matrix, used to linearly transform the input data and the previous hidden state; b is the bias term, where b is a constant offset added to the candidate hidden state; W x For the current input x t The weight matrix, W x Used to process the input data at the current time step.

[0059] h t This is the final hidden state. Indicates the candidate hidden state Based on the output value z of the update gate t complement (1-z) t The weighted contribution to the final hidden state;

[0060] z t ·h t-1 h represents the hidden state at the previous time step. t-1 Based on the output value z of the update gate t Update the output value z of the gate t .

[0061] Furthermore, the calculation formula for the fully connected layer is as follows:

[0062] p = f(W) fc ·y+b fc );

[0063] y = F(H) = [h1, h2…h T ];

[0064] H = {h1, h2, ..., h} T};

[0065] In the formula: p is the output of the fully connected layer, typically used as a feature representation for classification or regression tasks; H is the output sequence of the recurrent layer; T is the length of the sequence; each h... t It is the final hidden state at the current time step t; y is the flattened vector of the recurrent layer output sequence H; W fc W is the weight matrix of the fully connected layer. fc The dimension of b depends on the dimension of the input vector y and the dimension of the fully connected layer output; fc b is the bias vector of the fully connected layer. fc Dimensions and W fc The output dimensions are the same, and f is the activation function.

[0066] Furthermore, the formula by which the generation layer converts the output of the fully connected layer into the desired data points or sequences is as follows:

[0067]

[0068] p is the output of the fully connected layer. W is the output of the generation layer, i.e., the predicted data points or sequences. g It is the weight matrix of the generation layer, b g It is the bias term of the generation layer.

[0069] Furthermore, in S4, a method for constructing a dynamic bandwidth allocation strategy and formulating a charging and discharging plan for energy storage devices based on predicted data of the future power system state and real-time data is described below.

[0070] S41. Using the multi-temporal generative network model in S3 to predict the generator layer output. Combine real-time data p r This is used to update the estimate of the power system state, and the estimation formula is:

[0071]

[0072] p e For the updated estimation of the power system state, For the output of the generation layer, p r For real-time data, ω is a weighting coefficient between 0 and 1, which is adjusted based on the reliability of the predicted data and the real-time data;

[0073] S42. Analyze electricity demand in different time periods to determine the differences in resource demand between peak and off-peak periods;

[0074] S43. Based on the resource demand analysis results, formulate a charging and discharging plan for the energy storage equipment, using the following formula:

[0075]

[0076] In the formula: S(t) represents the charging / discharging plan, which is determined by the power system state to either charge / discharge or remain unchanged; S c The charging plan will be implemented during periods when electricity demand is lower than off-peak hours; S d This is a discharge plan, executed when electricity demand is higher than during peak hours; p v For electricity demand during off-peak hours, P P For peak-hour electricity demand, p e For the estimation of the updated power system state;

[0077] S44. Convert the charging / discharging plan into bandwidth allocation, that is, adjust the charging / discharging rate of the energy storage device to match fluctuations in power demand. The conversion formula is:

[0078] B a =B b +(S c -S d )·δ s ;

[0079] In the formula, B a It is the allocated bandwidth, B b ΔB is the base bandwidth, ΔB is the bandwidth adjustment amount, and δ is the base bandwidth. s It is an adjustment factor determined according to the charge / discharge plan, δ s Between 0 and 1;

[0080] S45. Based on the established energy storage device charging and discharging plan and converting the charging and discharging plan into bandwidth allocation, implement specific charging and discharging operations to ensure the stable operation of the power system.

[0081] Furthermore, in S5, the specific method for classifying the operating status of the power system is as follows:

[0082] S51. The real-time monitored power system data is fused with the pre-defined charging and discharging plan to form a comprehensive dataset. Features that are helpful for classification are extracted from the comprehensive dataset. The comprehensive dataset is represented as follows:

[0083] F={x pc ,x ts ,x c / dr};

[0084] x pc x represents the change in power, reflecting fluctuations in electricity demand; ts This is a timestamp used to account for the impact of time factors on electricity demand; x c / dr The charge / discharge rate represents the current charge / discharge state of the energy storage device.

[0085] S52. Using historical data and known power system state labels, train a classification model, and apply the trained classification model to real-time data to classify the system state;

[0086] S53. The results of the S52 classification are evaluated using metrics such as precision and recall. The formulas for calculating precision and recall are as follows:

[0087]

[0088]

[0089] In the formula: A is the accuracy evaluation metric; R is the recall evaluation metric; P c K is the number of correctly classified samples; K is the total number of classes; G is the total number of samples that are actually in that state. G refers to the total number of samples in the entire dataset whose actual label is positive, including those correctly predicted by the model and those incorrectly predicted.

[0090] S54. Transform the classification results into state feedback of the power system to provide decision support for the cloud energy storage system, and adjust the charging and discharging plan according to the classification results to adapt to changes in power demand.

[0091] Furthermore, the time-series simulation model constructed by S6, which includes multiple dimensions such as energy storage devices, user behavior, and power system status, is as follows:

[0092] M t+1 =L(M t ,Pn D t U t W t ,S t B t );

[0093] L is the state update function, which calculates the state at the next time step based on the current state and the input variables;

[0094] M t The state of the power system at time t includes power supply, demand, and energy storage levels; P n Time t represents the output of the new energy power system; D t Load demand at time t, i.e., the electricity consumption of the power system, U t For user behavior features identified in S2 at time t; W t Let S be the weather conditions at time t; t For the charge / discharge schedule at time t, B t Bandwidth allocation for time t.

[0095] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are as follows:

[0096] The time-series simulation model for the operation mechanism of cloud energy storage in new energy power systems adopts a user-customized power supply strategy. Through detailed user classification and power supply mode provision, it increases adaptability to user needs and the flexibility of the power supply strategy. Simultaneously, by optimizing charging and discharging strategies and energy storage device combinations, the stability and reliability of power supply are improved. This optimized strategy helps to better integrate and utilize new energy sources such as wind power and photovoltaics.

[0097] The method for establishing a time-series simulation model of the cloud energy storage operation mechanism of new energy power systems improves the utilization efficiency of energy storage devices and the regulation capability of the power grid by combining multiple devices.

[0098] The method for establishing a time-series simulation model of the operation mechanism of cloud energy storage in new energy power systems comprehensively considers various factors affecting the operation of cloud energy storage, and provides more accurate simulation predictions. Through the cloud energy storage time-series simulation model, the power grid's regulation and response capabilities to fluctuating energy are improved. Attached Figure Description

[0099] Figure 1 This is a flowchart of a time-series simulation method for a new energy cloud energy storage operation mechanism. Detailed Implementation

[0100] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0101] like Figure 1 As shown, a time-series simulation method for the operation mechanism of new energy cloud energy storage includes the following:

[0102] S1. Real-time monitoring of users' electricity consumption data via smart meters, and acquisition of weather information related to electricity consumption data using meteorological data APIs. Electricity consumption data includes:

[0103] Peak and off-peak electricity consumption periods: the specific times that distinguish between peak and off-peak electricity consumption periods;

[0104] Average daily electricity consumption: The total electricity consumption of a user per day;

[0105] Seasonal electricity consumption patterns: Users' electricity consumption habits in different seasons;

[0106] Types and quantities of electrical equipment: The types and quantities of electrical appliances used by the user.

[0107] S1. Real-time monitoring of users' electricity consumption data via smart meters, and acquisition of weather information related to electricity consumption data using meteorological data APIs. Electricity consumption data includes:

[0108] Peak and off-peak electricity consumption periods: the specific times that distinguish between peak and off-peak electricity consumption periods;

[0109] Average daily electricity consumption: The total electricity consumption of a user per day;

[0110] Seasonal electricity consumption patterns: Users' electricity consumption habits in different seasons;

[0111] Types and quantities of electrical equipment: The types and quantities of electrical appliances used by the user.

[0112] S2. User behavior feature analysis: Apply hierarchical clustering algorithm to classify user electricity consumption behavior in S1 and identify the characteristics and electricity consumption patterns of different user groups.

[0113] Cluster analysis can identify user groups with similar electricity consumption behaviors, providing accurate user behavior characteristics for subsequent power supply strategy formulation, and helping to achieve more personalized and efficient energy allocation.

[0114] In S2, the hierarchical clustering algorithm includes the following:

[0115] S21. Select features from the electricity consumption behavior data obtained from S1 that help distinguish different user behaviors, such as peak and off-peak electricity consumption periods, average daily electricity consumption, and seasonal electricity consumption patterns.

[0116] S22. Determine the number of clusters, use the hierarchical clustering algorithm, and calculate the silhouette coefficient under different numbers of clusters (κ value). Plot the relationship between the silhouette coefficient and the number of clusters (κ value), and select the κ value with the largest silhouette coefficient as the optimal number of clusters.

[0117] S23. Perform hierarchical clustering, using Euclidean distance to measure the similarity between samples. The smaller the distance, the more similar the electricity consumption behavior between samples. Based on the selected distance metric, each user is regarded as an initial cluster, and then the most similar cluster pairs are gradually merged. In the analysis of user electricity consumption behavior, each sample represents a user's electricity consumption record within a certain time range, which can be the electricity consumption per hour.

[0118] The Euclidean distance formula is:

[0119]

[0120] In the formula: x and y are two samples, and n is the number of features;

[0121] S24. Clustering results evaluation: Use indicators such as silhouette coefficient and Davies-Bouldin index to evaluate the clustering effect and ensure that the clustering results are meaningful.

[0122] The formula for the profile coefficient is: a is the average distance between a sample and its nearest sample in the same cluster, b is the average distance between a sample and its nearest samples in other clusters, and a high silhouette coefficient indicates high similarity between samples within a cluster;

[0123] The formula for the Davies-Bouldin index is: S i and S j It is the sum of the internal distances between clusters i and j, d(C i C j The distance between clusters i and j is the distance between clusters i and j. A low Davies-Bouldin index indicates a high degree of separation between clusters.

[0124] S25. Interpretation and application of results: Interpret the clustering results, identify the characteristics and power consumption patterns of each cluster, arrange the characteristics and power consumption patterns of each cluster in chronological order, and record the characteristics at each time point. Apply the clustering results to formulate targeted power supply strategies.

[0125] Analyze the characteristics of each cluster, such as peak electricity consumption periods and average daily electricity consumption, and formulate differentiated power supply strategies for different user groups based on the different clustering results. For example, for user groups with large electricity consumption during peak hours, demand response incentives can be provided to encourage them to increase electricity consumption during off-peak hours; for users with significant seasonal changes in electricity consumption, seasonal electricity price discounts can be provided.

[0126] S3. Construct a multi-temporal generative network model. Based on the results of the user behavior feature analysis in S2, construct a multi-temporal generative network model. The multi-temporal generative network model is used to learn from historical data and predict future power system states.

[0127] Multi-phase generation networks can capture the temporal characteristics of power system state changes and provide more accurate prediction results, thereby helping cloud energy storage systems to make more reasonable charging and discharging plans in advance and ensure that there are appropriate resources to cope with peak or trough power demand.

[0128] In S3, the multi-temporal generation network model includes the following types of layers:

[0129] Input layer: The input layer receives time series data, which is the data in S2 arranged in chronological order after user behavior feature analysis;

[0130] Recurrent layer: GRU is used as a recurrent layer to process time series data. The recurrent layer is used to process sequence data, capture dynamic features and long-term dependencies in the time series, and the final output of the recurrent layer is a feature sequence.

[0131] Fully connected layer: Use a fully connected layer to integrate the output of the recurrent layer, mapping the output of the recurrent layer to a dense high-dimensional space for classification or regression tasks, learning high-level feature representations of the data;

[0132] Generative layer: The generative layer generates new data points or sequences based on the output of the fully connected layer.

[0133] The GRU is a popular variant of the recurrent neural network (RNN), particularly suitable for processing and predicting time series data. The GRU includes:

[0134] Update gate: determines how much information is extracted from the current input x. t Inflow into hidden state h t ;

[0135] Reset Gate: Determines how to handle the hidden state h from the previous time step. t-1 Allows the network to reset or remember the previous state;

[0136] Candidate hidden states: These are candidates for the hidden states at the current time step, without considering previous hidden states;

[0137] Final hidden state: This is the final hidden state at the current time step. It is updated by combining the update gate, the reset gate, and the candidate hidden state, where:

[0138] Update Gate:

[0139] zt =σ(W z ·[h t-1 ,x t ]+b z );

[0140] By updating the gate formula, we know that: z t It is based on the hidden state h of the previous time step. t-1 and the current input x t The combination of these factors updates the gate's weight matrix W through the weight matrix. z Perform a linear transformation, then add a bias term to update the bias term b of the gate. z Finally, the output value z of the updated gate is obtained by performing a nonlinear transformation using the sinmoid function σ. t This will determine the candidate hidden state at the current time step. The hidden state h of the previous time step t-1 Between these points, how much information will be updated, that is, how much new information will be passed to the next time step;

[0141] Reset Door:

[0142] r t =σ(W r ·[h t-1 ,x t ]+b r );

[0143] From the reset gate formula, we know that: r t It is based on the hidden state h of the previous time step. t-1 and the input data x at the current time step t The combination of these is achieved by resetting the gate weight matrix W. r Then add the offset term b for resetting the door. r Finally, the output value r of the reset gate is obtained through a nonlinear transformation using the sinmoid function σ. t Calculate the candidate hidden state at the current time step At that time, the hidden state h of the previous time step t-1 How much weight should be assigned to r? t Approaching 1, h t-1 In calculation The time will be retained if r t h is close to 0 t-1 This will be ignored, thus allowing the network to reset its state.

[0144] Candidate hidden state:

[0145]

[0146] The candidate hidden state formula shows that the candidate hidden state is based on the hidden state h from the previous time step. t-1 The output r is obtained by resetting the gate at the current time step. t Adjust, plus the current input x t The impact, and the weight matrices W and W x The candidate hidden state is obtained by a linear transformation of the bias term b, followed by a nonlinear transformation using the tanh function. Together with the update gate, it will determine the final hidden state of the current time step.

[0147] Final hidden state:

[0148]

[0149] Indicates the candidate hidden state Based on the output value z of the update gate t complement (1-z) t The weighted contribution to the final hidden state;

[0150] z t ·h t-1 h represents the hidden state at the previous time step. t-1 Based on the output value z of the update gate t Update the output value z of the gate t ;

[0151] The formula for the final hidden state shows that the final hidden state h at the current time step can be obtained. t It contains both the candidate hidden states at the current time step and... The information also retains the hidden state h from the previous time step. t-1 Information.

[0152] In the formula, z t The update gate output is the value at the current time step. σ is the sinmoid activation function, whose output value ranges from 0 to 1 and is commonly used in gating mechanisms. W z Update the gate weight matrix, W z h is used to perform a linear transformation on the input data and the hidden state of the previous time step. t-1 b is the hidden state of the previous time step. z It is the bias term of the updated gate, x t The input data for the current time step; r t W is the reset gate output for the current time step. r The weight matrix of the reset gate is used to linearly transform the input data and the hidden state of the previous time step, h. t-1 b is the hidden state of the previous time step. rThis resets the door's bias.

[0153] represents the candidate hidden state at the current time step; tanh is the hyperbolic tangent function, a non-linear activation function used to introduce non-linear characteristics, with an output value ranging from -1 to 1; W is the weight matrix, used to linearly transform the input data and the previous hidden state; b is the bias term, where b is a constant offset added to the candidate hidden state; W x For the current input x t The weight matrix, W x Used to process the input data at the current time step.

[0154] h t This is the final hidden state. Indicates the candidate hidden state Based on the output value z of the update gate t complement (1-z) t The weighted contribution to the final hidden state;

[0155] z t ·h t-1 h represents the hidden state at the previous time step. t-1 Based on the output value z of the update gate t Update the output value z of the gate t .

[0156] The calculation formula for the fully connected layer is as follows:

[0157] p = f(W) fc ·y+b fc );

[0158] y = F(H) = [h1, h2…h T ];

[0159] H = {h1, h2, ..., h} T};

[0160] In the formula: p is the output of the fully connected layer, typically used as a feature representation for classification or regression tasks; H is the output sequence of the recurrent layer; T is the length of the sequence; each h... t It is the final hidden state at the current time step t; y is the flattened vector of the recurrent layer output sequence H; W fc W is the weight matrix of the fully connected layer. fc The dimension of b depends on the dimension of the input vector y and the dimension of the fully connected layer output; fc b is the bias vector of the fully connected layer. fc Dimensions and W fc The output dimensions are the same, and f is the activation function.

[0161] The formula by which the generator layer transforms the output of the fully connected layer into the desired data points or sequences is:

[0162]

[0163] p is the output of the fully connected layer. W is the output of the generation layer, i.e., the predicted data points or sequences. g It is the weight matrix of the generation layer, b g It is the bias term of the generation layer.

[0164] S4. Based on the predicted data of the future power system state in S3 and the real-time data obtained in S1, construct a dynamic bandwidth allocation strategy and formulate a charging and discharging plan for energy storage devices;

[0165] A method for developing charging and discharging plans for energy storage devices by constructing dynamic bandwidth allocation strategies based on predicted and real-time data of the future power system state:

[0166] S41. Using the multi-temporal generative network model in S3 to predict the generator layer output. Combine real-time data p r This is used to update the estimate of the power system state, and the estimation formula is:

[0167]

[0168] p e For the updated estimation of the power system state, For the output of the generation layer, p r For real-time data, ω is a weighting coefficient between 0 and 1, which is adjusted based on the reliability of the predicted data and the real-time data;

[0169] S42. Analyze electricity demand in different time periods to determine the differences in resource demand between peak and off-peak periods;

[0170] S43. Based on the resource demand analysis results, formulate a charging and discharging plan for the energy storage equipment, using the following formula:

[0171]

[0172] In the formula: S(t) represents the charging / discharging plan, which is determined by the power system state to either charge / discharge or remain unchanged; S c The charging plan will be implemented during periods when electricity demand is lower than off-peak hours; S d This is a discharge plan, executed when electricity demand is higher than during peak hours; p v For electricity demand during off-peak hours, P P For peak-hour electricity demand, p e For the estimation of the updated power system state;

[0173] S44. Convert the charging / discharging plan into bandwidth allocation, that is, adjust the charging / discharging rate of the energy storage device to match fluctuations in power demand. The conversion formula is:

[0174] B a =B b +(S c -S d )·δ s ;

[0175] In the formula, B a It is the allocated bandwidth, B b ΔB is the base bandwidth, ΔB is the bandwidth adjustment amount, and δ is the base bandwidth. s It is an adjustment factor determined according to the charge / discharge plan, δ s Between 0 and 1;

[0176] S45. Based on the established energy storage device charging and discharging plan and converting the charging and discharging plan into bandwidth allocation, implement specific charging and discharging operations to ensure the stable operation of the power system.

[0177] S5. Time-series signal classification and identification: The classification and identification method is used to identify the real-time data obtained in S1 and the charging and discharging plan generated in S4, integrate historical operating data, and classify the operating status of the power system.

[0178] S51. The real-time monitored power system data is fused with the pre-defined charging and discharging plan to form a comprehensive dataset. Features that are helpful for classification are extracted from the comprehensive dataset. The comprehensive dataset is represented as follows:

[0179] F={x pc ,x ts ,x c / dr};

[0180] x pc x represents the change in power, reflecting fluctuations in electricity demand; ts This is a timestamp used to account for the impact of time factors on electricity demand; x c / dr The charge / discharge rate represents the current charge / discharge state of the energy storage device.

[0181] S52. Using historical data and known power system state labels, train a classification model, and apply the trained classification model to real-time data to classify the system state;

[0182] S53. The results of the S52 classification are evaluated using metrics such as precision and recall. The formulas for calculating precision and recall are as follows:

[0183]

[0184]

[0185] In the formula: A is the accuracy evaluation metric; R is the recall evaluation metric; P c K is the number of correctly classified samples; K is the total number of classes; G is the total number of samples that are actually in that state. G refers to the total number of samples in the entire dataset whose actual label is positive, including those correctly predicted by the model and those incorrectly predicted.

[0186] S54. Transform the classification results into power system status feedback to provide decision support for cloud energy storage systems, and adjust charging and discharging plans according to the classification results to adapt to changes in power demand;

[0187] By classifying and identifying the S51 to S54 time-series signals, the current state of the power system can be quickly and accurately determined, providing timely status feedback and decision support for the operation of the cloud energy storage system.

[0188] S6. Construction of a time-series simulation model for cloud energy storage operation mechanism: Based on the results from S1 to S5, a time-series simulation model is constructed that includes multiple dimensions such as energy storage devices, user behavior, and power system status; the specific model is represented as follows:

[0189] M t+1 =L(M t ,P n D t U t W t ,S t B t );

[0190] L is the state update function, which calculates the state at the next time step based on the current state and the input variables;

[0191] M t The state of the power system at time t includes power supply, demand, and energy storage levels; P n Time t represents the output of the new energy power system; D t Load demand at time t, i.e., the electricity consumption of the power system, U t For user behavior features identified in S2 at time t; W t Let S be the weather conditions at time t; t For the charge / discharge schedule at time t, B t Bandwidth allocation for time t.

[0192] By utilizing the multi-dimensional time-series simulation models established from S1 to S6, the operating status of the cloud energy storage system is comprehensively reflected, providing a simulation environment for testing and optimizing cloud energy storage operation strategies, thereby achieving stable operation of the power system and efficient utilization of energy.

[0193] This time-series simulation model can comprehensively reflect the operating status of cloud energy storage systems, providing a simulation environment for testing and optimizing cloud energy storage operation strategies, and helping to achieve stable operation of the power system and efficient use of energy.

[0194] The above description is a preferred embodiment of the invention and is not intended to limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A time sequence simulation method for new energy cloud energy storage operation mechanism, comprising the following steps: S1. Real-time monitoring of user electricity consumption behavior data through a smart meter, and obtaining weather information related to electricity consumption behavior using a meteorological data API. The electricity consumption behavior data includes: Peak-valley electricity consumption period: distinguishing the specific time of electricity consumption peak and valley; Daily average electricity consumption: the total electricity consumption of the user per day; Seasonal electricity consumption pattern: the electricity consumption habits of the user in different seasons; Electricity consumption equipment type and quantity: the type and quantity of electrical appliances used by the user; S2. User behavior feature analysis, applying a hierarchical clustering algorithm to classify the user electricity consumption behavior in S1, and identifying the characteristics and electricity consumption patterns of different user groups; S3. Constructing a multi-time phase generation network model, according to the results of the user behavior feature analysis in S2, a multi-time phase generation network model is constructed, which is used to learn from historical data and predict future power system state prediction data; S4. Constructing a dynamic bandwidth allocation strategy according to the prediction data of the future power system state predicted by S3 and the real-time data obtained by S1, and formulating a charging and discharging plan for the energy storage device; S5. Time sequence signal classification and identification, using classification and identification methods to identify the real-time data obtained by S1 and the charging and discharging plan generated in S4, integrating historical operation data, and classifying the power system operation state; S6. Construction of a time sequence simulation model for cloud energy storage operation mechanism, integrating the results of S1 to S5, and constructing a time sequence simulation model containing multiple dimensions of energy storage devices, user behavior, and power system state; The time sequence simulation model containing multiple dimensions of energy storage devices, user behavior, and power system state constructed in S6 is: ； is a state update function that computes the state at the next time step from the current state and the input variable; for time a power system state, including power supply, demand, energy storage level; for time a new energy power system output; for time a load demand, i.e. an amount of power consumption of the power system, for time a user behavior feature identified in S2; for time a weather condition; for time a charging and discharging plan, for time a bandwidth allocation; The multi-dimensional time sequence simulation model established by S1 to S6 comprehensively reflects the operation state of the cloud energy storage system, provides a simulation environment for testing and optimizing the cloud energy storage operation strategy, and realizes stable operation of the power system and efficient use of energy.

2. The time sequence simulation method of a new energy cloud energy storage operation mechanism according to claim 1, characterized in that, In S2, the hierarchical clustering algorithm includes: S21. Selecting features from the user electricity consumption behavior collected in S1 that are helpful in distinguishing different user behaviors, including peak-valley electricity consumption period, daily average electricity consumption, and seasonal electricity consumption pattern; S22. Determine the number of clusters by profile coefficient analysis; S23. Perform hierarchical clustering, use Euclidean distance to measure the similarity between samples, and according to the selected distance measure, consider each user as an initial cluster, then gradually merge the most similar cluster pairs, the Euclidean distance formula is: ； In the formula: and are two samples, is the number of features; S24. Cluster result evaluation, use profile coefficient and Davies-Bouldin index to evaluate the clustering effect, ensure that the clustering result is meaningful; The silhouette coefficient formula is: , is the average distance of the sample from the nearest samples in its own cluster, is the average distance of the sample from the nearest samples in other clusters. The Davies-Bouldin index formula is: , and is the sum of the internal distances of the clusters and , is the distance between the clusters and ; S25. Result interpretation and application, interpret the clustering results, identify the characteristics and electricity consumption patterns of each cluster, arrange the characteristics and electricity consumption patterns of each cluster in chronological order, and record the characteristics at each time point, and apply the clustering results to develop targeted power supply strategies.

3. The time sequence simulation method of a new energy cloud energy storage operation mechanism according to claim 1, characterized in that, In S3, the multi-time phase generation network model includes the following types of layers: Input layer: the input layer receives time series data, which is the data analyzed by user behavior characteristics in S2; Cycle layer: use As a cycle layer to process time series data, the cycle layer is used to process sequence data, capture dynamic features and long-term dependencies in time series, and the final output of the cycle layer is a feature sequence; Fully connected layer: The fully connected layer is used to integrate the output of the recurrent layer, mapping the output of the recurrent layer to a dense high-dimensional space for classification or regression tasks, learning high-level feature representations of data; Generation layer: The generation layer generates new data points or sequences based on the output of the fully connected layer.

4. The time sequence simulation method of a new energy cloud energy storage operation mechanism according to claim 3, characterized in that, The described is a popular recurrent neural network Variants suitable for processing and predicting time series data, comprises: Update gate: decides how much information from the current input Inflow hidden state ; reset gate: determines how to treat the hidden state of the previous time step , allowing the network to reset or remember the previous state; Candidate hidden state: It is the candidate hidden state of the current time step, without considering the previous hidden states; Final hidden state: It is the final hidden state of the current time step, updated by combining the update gate, reset gate and candidate hidden state, where: Update gate: ； By updating the gate formula, we can see that: It is based on the hidden state of the previous time step. and current input The combination of these factors updates the gate's weight matrix through the weight matrix. Perform a linear transformation, then add a bias term to update the gate's bias term. Finally passed function The output value of the update gate is obtained by performing a nonlinear transformation. This will determine the candidate hidden state at the current time step. Hidden state compared to the previous time step Between these points, how much information will be updated, that is, how much new information will be passed to the next time step; Reset gate: ； The reset gate formula is: is the combination of the hidden state of the previous time step and the input data of the current time step , through the weight matrix of the reset gate , then add the bias term of the reset gate , and finally through the function for nonlinear transformation; the output value of the reset gate When calculating the candidate hidden state of the current time step , how much weight should be given to the hidden state of the previous time step , if is close to 1, will be retained when calculating , if is close to 0, will be ignored, thus allowing the network to reset its state; Candidate hidden state: ； From the candidate hidden state formula, it can be seen that the candidate hidden state is based on the hidden state of the previous time step , adjusted by the output of the reset gate of the current time step , plus the influence of the current input , and linear transformation of the weight matrix and and the bias term , and finally nonlinear transformation by the function, the candidate hidden state will determine the final hidden state of the current time step together with the update gate. Final hidden state: ； representing a candidate hidden state according to an output value of the update gate complement of weighted contribution to the final hidden state; denotes the hidden state of the previous time step; From the final hidden state formula, we can see that the final hidden state at the current time step is obtained by combining the information from the candidate hidden state at the current time step and the information from the hidden state at the previous time step . wherein, is the update gate output for the current time step, is the hidden state for the previous time step, is an activation function whose output value ranges between 0 and 1, commonly used in gating mechanisms, is the weight matrix of the update gate, is used to linearly transform the input data and the hidden state of the previous time step, is the hidden state for the previous time step, is the bias term of the update gate, is the input data for the current time step; is the reset gate output for the current time step, is the weight matrix of the reset gate used to linearly transform the input data and the hidden state of the previous time step, is the hidden state for the previous time step, is the bias term of the reset gate. is the candidate hidden state for the current time step; is the hyperbolic tangent function, a nonlinear activation function that introduces nonlinearity into the model, with output values ranging between -1 and 1; is the weight matrix, used to linearly transform the input data and previous hidden states; is the bias term, adds a constant offset to the candidate hidden state; is the current input is the weight matrix, is used to process the input data for the current time step; for the final hidden state, representing a candidate hidden state according to the output value of the update gate complement of weighted contribution to the final hidden state; denotes the hidden state of the previous time step.

5. The time sequence simulation method of a new energy cloud energy storage operation mechanism according to claim 3, characterized in that, The calculation formula of the fully connected layer is: ； ； ； In the formula: the output of the fully connected layer, typically used as a feature representation for a classification or regression task; is the output sequence of the recurrent layer; is the length of the sequence; each is the final hidden state of the recurrent layer at the current time step ; is the output sequence of the recurrent layer after flattening; is the weight matrix of the fully connected layer, whose dimensions depend on the dimensions of the input vector and the dimensions of the output of the fully connected layer; is the bias vector of the fully connected layer, whose dimensions are the same as the output dimensions of ; is the activation function.

6. The time sequence simulation method of a new energy cloud energy storage operation mechanism according to claim 3, characterized in that, The formula for the generation layer to convert the output of the fully connected layer into the required data points or sequences is: ; is the output of the fully connected layer, is the output of the generation layer, i.e. the predicted data point or sequence, is the weight matrix of the generation layer, is the bias term of the generation layer.

7. The time sequence simulation method of a new energy cloud energy storage operation mechanism according to claim 1, characterized in that, In S4, a dynamic bandwidth allocation strategy is constructed based on the predicted data and real-time data of the predicted future power system state, and a method for formulating the charge and discharge plan of the energy storage device is developed: S41. Using the multi-time step generated network model prediction of the generation layer output from S3 , in combination with real-time data , to update the estimate of the state of the power system, with the estimate formula being: ； for the updated estimate of the power system state, for generating the output of the layer, for the real-time data, is a weight coefficient between 0 and 1 that is adjusted according to the reliability of the predicted data and the real-time data; S42. Analyze the power demand in different time periods to determine the resource demand difference between peak and off-peak periods; S43. According to the resource demand analysis results, formulate the charge and discharge plan of the energy storage device, the formula is: ； In the formulae: is the charge or discharge plan, determined from the state of the power system to charge or discharge or remain unchanged; is the charge plan, executed when the power demand is below the low period; is the discharge plan, executed when the power demand is above the high period; is the low period power demand, is the high period power demand, is the updated estimate of the state of the power system; S44. Convert the charge and discharge plan to bandwidth allocation, i.e. adjust the charge and discharge rate of the energy storage device to match the power demand fluctuations, the conversion formula is: ； In the formula, is the allocated bandwidth, is the base bandwidth, is the bandwidth adjustment amount, is an adjustment coefficient determined according to the charging and discharging plan, is between 0 and 1; S45. According to the formulated charge and discharge plan of the energy storage device and the conversion of the charge and discharge plan to bandwidth allocation, implement specific charge and discharge operations to ensure stable operation of the power system.

8. The time sequence simulation method of new energy cloud energy storage operation mechanism according to claim 1, characterized in that, In S5, the specific method for classifying the operating state of the power system is: S51. Fuse the real-time monitored power system data with the pre-formulated charge and discharge plan to form a comprehensive data set, extract features that contribute to classification from the comprehensive data set, and the comprehensive data set is represented as: ； is a power variation amount, reflecting fluctuations in power demand; is a time stamp, used to consider the influence of time factors on power demand; is a charge / discharge rate, the current charge / discharge state of the energy storage device; S52. Use historical data and known power system state labels to train a classification model, and apply the trained classification model to real-time data to classify the system state; S53. Evaluate the results of S52 classification using accuracy and recall, where the calculation formulas for accuracy and recall are: ； ； In the formula: is the accuracy evaluation index; is the recall evaluation index; is the number of correct classifications; is the total number of classifications; is the total number of actual states that are the state, refers to the total number of samples with actual labels as positive classes in the entire data set, including those correctly predicted by the model and those incorrectly predicted; S54. Convert the classification results into state feedback of the power system to provide decision support for the cloud energy storage system, and adjust the charge and discharge plan according to the classification results to adapt to changes in power demand.

Images