Federal learning-based new energy cluster consumption intelligent regulation method and device

Abstract

This invention provides a method and apparatus for intelligent regulation and control of renewable energy clusters based on federated learning. The method includes the following steps: collecting data from different renewable energy power generation equipment and power users to establish a multi-timescale index set, which includes power generation and consumption characteristic index sets at different time scales; evaluating the potential of adjustable resources to participate in response scheduling based on the multi-timescale index set, obtaining the potential evaluation result of the adjustable resources; and intelligently scheduling resources across the entire network based on federated learning, considering the potential evaluation result of the adjustable resources. The method proposed in this invention, while considering the influence of uncertain factors, realizes the cluster regulation and control potential evaluation of renewable energy equipment and users, and based on the federated learning concept, protects the data privacy of each region during subsequent resource regulation and control, thereby improving the overall security of intelligent regulation and control.

Description

Technical Field

[0001] This invention relates to the field of source-load resource regulation, specifically a method and device for intelligent regulation of new energy cluster consumption based on federated learning. Background Technology

[0002] Uncertain factors influencing the response potential of adjustable resources include temperature, meteorological environment, operating status of adjustable resources, response behavior, and response incentive prices. These uncertainties reduce the reliability of adjustable resource regulation output in grid balancing operations and exacerbate the volatility of adjustable resource response regulation output. Therefore, numerous uncertainties within adjustable resource clusters limit the accuracy of response potential assessment. Properly describing and handling various uncertainties is crucial for response potential assessment. Current research mainly considers the impact of market, environmental, and adjustable resource's own electricity consumption characteristics on response potential based on various types of adjustable resource aggregation response models, and also considers the volatility of coordinated response output across multiple adjustable resource clusters. However, research on the response potential of adjustable resource clusters primarily focuses on the response capacity of adjustable resources, without considering the dynamic process of internal resources over time during cluster participation in response, and without further analysis of the dynamic performance changes under complementary and coordinated operation of adjustable resource clusters. Therefore, it is necessary to consider the dynamic process of cluster participation in response and assess the response potential of large-scale regional adjustable resource clusters operating in a complementary and coordinated manner at different time scales.

[0003] When regulating source and load resources based on potential assessment results, different regions have their own data, which needs to be used for training, updating, and optimizing the actual model. However, the amount of data in a single region is insufficient, resulting in coarse model training. Integrating data from different regions for overall model training is subject to restrictions on user data privacy protection in practice, as regional centers cannot arbitrarily disclose user data to third parties. Distributed computing can be used to solve this problem. Federated learning is a new computational concept developed from distributed learning. Compared with distributed computing, federated learning has advantages in the following five aspects:

[0004] 1. In terms of control, in traditional distributed learning, the central server has absolute control over the workers in each region; in federated learning, users have absolute control over devices and data and can stop participating in computation at any time.

[0005] 2. Regarding node stability, distributed worker nodes are very stable with almost identical performance, which is too idealistic; in federated learning, worker nodes are unstable and vary from one another, just like mobile phones with different network speeds or different devices, resulting in different computing speeds and performance, which is more in line with actual devices and user situations.

[0006] 3. In terms of data properties, traditional distributed systems have similar data at each node, so random shuffling and recalculation have little impact; federated learning data is not independent and identically distributed, and the properties of the data at each node are different, taking into account the practical problem of different user habits.

[0007] 4. Regarding node data load, traditional distributed computing requires ensuring load balance; however, the data load of federated nodes can be unbalanced, with different units of power having different weights, allowing for flexible adjustment of computing speed based on node data.

[0008] 5. In terms of cost control, distributed computing mainly involves the computational cost of each worker node; while federated learning costs are mainly concentrated in the communication costs between worker nodes and server nodes. Summary of the Invention

[0009] To address the aforementioned shortcomings of existing technologies, this invention provides a method and apparatus for intelligent regulation and control of new energy clusters based on federated learning. By considering the influence of uncertain factors, it achieves the assessment of the cluster regulation and control potential of new energy equipment and users. Furthermore, based on the federated learning concept, it protects the data privacy of each region during subsequent resource regulation and control, thereby improving the overall security of intelligent regulation and control.

[0010] A smart regulation method for the consumption of new energy clusters based on federated learning includes the following steps:

[0011] Collect data on different new energy power generation equipment and power users, and establish a multi-time-scale indicator set, which includes power generation and power consumption characteristic indicator sets at different time scales;

[0012] Based on the multi-timescale index set, the potential of adjustable resources to participate in response scheduling is evaluated, and the potential evaluation results of adjustable resources are obtained.

[0013] Based on the potential assessment results of the adjustable resources, intelligent scheduling of resources across the entire network is performed using federated learning.

[0014] Furthermore, data on different new energy power generation equipment and electricity users are collected. Considering the uncertainty of source-load resource regulation, a multi-time-scale index set is established. The multi-time-scale index set includes power generation and consumption characteristic index sets at different time scales, specifically including the following steps:

[0015] Identify the external influencing factors that determine the scheduling potential of new energy power resources and collect daily data to establish a set of power generation characteristic indicators at different time scales;

[0016] The external influencing factors of the uncertainty of dispatch potential include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunny days, and snow. The power generation characteristic index set at different time scales is obtained by statistical calculation of daily power generation data. The electricity consumption characteristic index set establishment module is used to determine the internal influencing factors of the uncertainty of dispatch potential of user demand response resources, collect data, and establish electricity consumption characteristic index sets at different time scales.

[0017] The internal influencing factors of the uncertainty in the scheduling potential include the equivalent heat capacity C, equivalent thermal resistance R, energy efficiency ratio η, and set temperature T of the polymer. set Allowable temperature adjustment amount ΔT, adjustment duration Δt, user water consumption (m³) n Ambient temperature T out The market electricity price (TOU) is calculated by statistically analyzing daily power generation data, based on the electricity consumption characteristic index set at different time scales.

[0018] Furthermore, based on the aforementioned multi-timescale index set and considering the uncertainty of source load resources, the potential of adjustable resources to participate in response scheduling is evaluated to obtain the potential evaluation results of adjustable resources. Specifically, this includes the following steps:

[0019] Establish a model for assessing the response potential of adjustable resources:

[0020]

[0021] Where r1, r2, and r3 are known deterministic model parameters; r4 is a normal distribution that follows a certain pattern, where the mean of this distribution is... Standard deviation is The parameters μ0 and σ0 are probability estimates obtained by point estimation based on the historical response dataset of r4;

[0022] By treating the obtained probability estimates as parameters of a normal distribution that r4 satisfies, we can obtain the potential assessment results for adjustable resources to participate in response scheduling:

[0023]

[0024] Among them, the mean benchmark value μ and the variance benchmark value σ 2 , Mean fuzzy value μ e and the variance fuzzy value σ e These four indicators are model metrics that describe the uncertainty parameters of user response potential. They comprehensively reflect the uncertainty of the adjustable potential of adjustable resources to participate in response business.

[0025] Furthermore, based on the potential assessment results of the adjustable resources, intelligent scheduling of network resources is performed using federated learning, specifically including:

[0026] To optimize the overall effect of new energy consumption, a particle swarm optimization model is established:

[0027] The specific optimization objectives are shown in formula (6), including reducing power fluctuations and reducing the curtailment of renewable energy. The constraints are shown in formula (8), including equipment output limits and power fluctuation rate limits.

[0028]

[0029]

[0030]

[0031] Where T new λ represents the number of time periods used to calculate power fluctuations. o,t For the purchase price of electricity, λ flu P is the cost factor for power fluctuation loss. G,t To make actual contributions to new energy, P Go,t For the power output capacity of new energy, P Gmax,t P Gmin,t δ represents the upper and lower limits of the output of new energy equipment during time period t. t δ represents the fluctuation in the output power of new energy sources during time period t. rise δ is the upper limit of the power increase rate. drop Δt represents the lower limit of the power reduction rate, and Δt is the time interval between the two time periods.

[0032] Furthermore, the central control center server uses federated learning to regulate network resources, specifically including:

[0033] Step (a) The central control center server transmits the parameters w of the fixed particle swarm optimization algorithm model to each sub-regional center;

[0034] Step (b) Based on the potential assessment results of the adjustable resources, the workers of each sub-regional center obtain the power data of local adjustable resources participating in demand response. Using the source-load power data information and the model parameters w transmitted from the central control center, the sub-regional center workers perform local conventional particle swarm optimization to calculate the adjustment amount of new energy output or user load energy consumption in each region.

[0035] Step (c) Each sub-regional center worker uses local data to perform multiple gradient descent operations locally to obtain the updated gradient g, and calculates the new parameters w to return to the central control center server.

[0036] w←w-α·g (8)

[0037] Where w represents the particle swarm optimization model parameters, α represents the specific weight coefficient update step size, and g represents the gradient;

[0038] Step (d) The central control center server collects the new parameters w returned by the workers in each region, performs a weighted average of the parameters, updates the particle swarm model parameters, and then sends them down to the workers in each sub-regional center.

[0039] Step (e) repeats steps (b)-(d) until the controllable source load resource scheduling in each region is completed.

[0040] A smart control device for the consumption of renewable energy clusters based on federated learning, comprising:

[0041] The feature index set establishment module is used to collect data from different new energy power generation equipment and power users, and establish a multi-time scale index set, which includes power generation and power consumption feature index sets at different time scales.

[0042] The potential assessment module is used to assess the potential of adjustable resources to participate in response scheduling based on the multi-timescale indicator set, and obtain the potential assessment results of adjustable resources.

[0043] The intelligent scheduling module is used to intelligently schedule resources across the entire network based on the potential assessment results of the adjustable resources and federated learning.

[0044] The feature index set establishment module includes a power generation feature index set establishment module and a power consumption feature index set establishment module.

[0045] The power generation characteristic index set establishment module is used to determine the external influencing factors of the uncertainty of the dispatch potential of new energy output resources, and to collect daily data to establish power generation characteristic index sets at different time scales.

[0046] The external influencing factors of the uncertainty of the dispatch potential include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunny days, and snow. The power generation characteristic index set at different time scales is obtained by statistical calculation of daily power generation data.

[0047] The electricity consumption characteristic index set establishment module is used to determine the internal influencing factors of the uncertainty of the scheduling potential of user demand response resources, collect data, and establish electricity consumption characteristic index sets at different time scales.

[0048] The internal influencing factors of the uncertainty in the scheduling potential include the equivalent heat capacity C, equivalent thermal resistance R, energy efficiency ratio η, and set temperature T of the polymer. set Allowable temperature adjustment amount ΔT, adjustment duration Δt, user water consumption (m³) n Ambient temperature Tout The market electricity price (TOU) is calculated by statistically analyzing daily power generation data, based on the electricity consumption characteristic index set at different time scales.

[0049] Furthermore, the potential assessment module includes a potential assessment model establishment module and a potential assessment result acquisition module; wherein,

[0050] The potential assessment model building module is used to build adjustable resource response potential assessment models.

[0051]

[0052] Where r1, r2, and r3 are known deterministic model parameters; r4 is a normal distribution that follows a certain pattern, where the mean of this distribution is... Standard deviation is The parameters μ0 and σ0 are probability estimates obtained by point estimation based on the historical response dataset of r4;

[0053] The potential assessment result acquisition module is used to treat the obtained probability estimates as parameters of a normal distribution that r4 satisfies, thereby obtaining the potential assessment results of adjustable resources participating in response scheduling:

[0054]

[0055] Among them, the mean benchmark value μ and the variance benchmark value σ 2 , Mean fuzzy value μ e and the variance fuzzy value σ e These four indicators are model metrics that describe the uncertainty parameters of user response potential. They comprehensively reflect the uncertainty of the adjustable potential of adjustable resources to participate in response business.

[0056] Furthermore, the intelligent scheduling module includes a particle swarm optimization model establishment module and a network-wide resource regulation module;

[0057] The particle swarm optimization model establishment module is used to establish a particle swarm optimization model with the goal of achieving the optimal overall effect of new energy consumption.

[0058] Furthermore, the network-wide resource control module includes a central control center (server) and regional control centers (workers);

[0059] The central control center server is used to perform network-wide resource control using the federated learning approach, specifically including:

[0060] Step (a) The central control center server transmits the parameters w of the fixed particle swarm optimization algorithm model to each sub-regional center;

[0061] Step (b) Based on the potential assessment results of the adjustable resources, the workers of each sub-regional center obtain the power data of local adjustable resources participating in demand response. Using the source-load power data information and the model parameters w transmitted from the central control center, the sub-regional center workers perform local conventional particle swarm optimization to calculate the adjustment amount of new energy output or user load energy consumption in each region.

[0062] Step (c) Each sub-regional center worker uses local data to perform multiple gradient descent operations locally to obtain the updated gradient g, and calculates the new parameters w to return to the central control center server.

[0063] w←w-α·g (9)

[0064] Where w represents the particle swarm optimization model parameters, α represents the specific weight coefficient update step size, and g represents the gradient;

[0065] Step (d) The central control center server collects the new parameters w returned by the workers in each region, performs a weighted average of the parameters, updates the particle swarm model parameters, and then sends them down to the workers in each sub-regional center.

[0066] Step (e) repeats steps (b)-(d) until the controllable source load resource scheduling in each region is completed.

[0067] Furthermore, the specific optimization objectives of the particle swarm optimization model are shown in formula (6), including reducing power fluctuations and reducing renewable energy curtailment. The constraints are shown in formula (8), including equipment output limits and power fluctuation rate limits.

[0068]

[0069]

[0070]

[0071] Where T new λ represents the number of time periods used to calculate power fluctuations. o,t For the purchase price of electricity, λ flu P is the cost factor for power fluctuation loss. G,t To make actual contributions to new energy, P Go,t For the power output capacity of new energy, P Gmax,t P Gmin,t δ represents the upper and lower limits of the output of new energy equipment during time period t. t δ represents the fluctuation in the output power of new energy sources during time period t. rise δ is the upper limit of the power increase rate. drop Δt represents the lower limit of the power reduction rate, and Δt is the time interval between the two time periods.

[0072] A new energy cluster consumption intelligent control system based on federated learning includes: a computer-readable storage medium and a processor;

[0073] The computer-readable storage medium is used to store executable instructions;

[0074] The processor is used to read executable instructions stored in the computer-readable storage medium and execute the intelligent control method for the consumption of new energy clusters based on federated learning.

[0075] A non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned intelligent control method for the consumption of new energy clusters based on federated learning.

[0076] The present invention has the following beneficial effects:

[0077] (1) The intelligent regulation and control method for new energy cluster consumption based on federated learning in this invention can establish a set of adjustable indicators at multiple time scales, and consider the impact of uncertain factors of source load resources to evaluate the regulation potential of cluster resources. Thus, energy companies can formulate orderly management strategies for source load resources of different clusters based on the controllable characteristics of different source load resource clusters, thereby improving energy management efficiency.

[0078] (2) The intelligent control method for new energy cluster consumption based on federated learning in this invention can realize the local utilization and protection of data without uploading it to other control centers, and at the same time realize the updating of model parameters. On the one hand, this method can effectively reduce the transmission process of actual user data, protect it locally, and improve the privacy and security of user data; on the other hand, this method only transmits parameters rather than direct data, and distributes the optimization calculation pressure of the control center to each branch center, which can reduce communication costs and improve the efficiency of updating optimization model parameters. Attached Figure Description

[0079] Figure 1 This is a graph showing the relationship between adjustable resource response and incentive intensity under the influence of uncertainties in this invention.

[0080] Figure 2 This is a diagram illustrating the parameter transfer process of the federated learning method used in this invention.

[0081] Figure 3 This is a flowchart of one embodiment of the intelligent regulation and control method for new energy cluster consumption based on federated learning of the present invention. Detailed Implementation

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0083] like Figure 3 As shown, this embodiment of the invention provides a method for intelligent regulation and control of new energy clusters based on federated learning, including establishing a multi-timescale index set, assessing scheduling potential considering the uncertainty of source and load resources, and intelligent scheduling based on federated learning. Specifically, it includes the following steps:

[0084] Step 1: Collect data on different new energy power generation equipment and electricity users. Considering the uncertainty of source-load resource regulation, establish a multi-time-scale index set. The multi-time-scale index set includes power generation and consumption characteristic index sets at different time scales. Step 1 specifically includes the following steps:

[0085] Step 1.1: Identify the external influencing factors of the uncertainty in the dispatch potential of new energy power resources, collect daily data, and establish a set of power generation characteristic indicators at different time scales.

[0086] External factors influencing the intelligent regulation of new energy power generation resources mainly refer to weather factors, such as wind speed, solar radiation, temperature, and humidity. Weather conditions such as rain, sunshine, and snowfall also affect the power consumption of system equipment and the power generation of power plants. Detailed indicators are shown in Table 1, and these indicators can be calculated through daily power generation data statistics.

[0087] Table 1. Set of power generation characteristic indicators at different time scales

[0088]

[0089] Step 1.2: Identify the internal influencing factors of the uncertainty in the scheduling potential of user demand response resources, collect data, and establish a set of electricity consumption characteristic indicators at different time scales.

[0090] Uncertainty in load model parameters is a significant factor affecting response potential. For example, refined modeling of adjustable resources such as temperature-controlled loads and electric vehicles can accurately determine the response potential of user equipment, thereby enabling accurate differentiated control. However, in many cases, the equivalent parameters of user equipment and the charging model parameters of electric vehicles are difficult to measure directly, and these parameters will also change with changes in environmental, market, and user behavior factors.

[0091] Therefore, research on these load models should not be limited to static load models, but should also consider the uncertainties in model parameters caused by various influencing factors. For example, in the aggregation model of adjustable air conditioning resources, the model parameters such as the equivalent heat capacity C, equivalent thermal resistance R, and energy efficiency ratio η of the aggregate are affected by various uncertainties, and the equivalent model parameters of the adjustable resource cluster will also change at different time points. In addition, the set temperature T is also an influencing factor on the adjustable potential of temperature-controlled loads. set Allowable temperature adjustment amount ΔT, adjustment duration Δt, user water consumption (m³) n (Water heater load), ambient temperature T out Electricity price (TOU), etc.

[0092] Detailed indicators are shown in Table 2. These indicators can be obtained by statistical calculation of daily power generation data.

[0093] Table 2. Set of electricity consumption characteristic indicators at different time scales

[0094]

[0095]

[0096] Step 2: Based on the aforementioned multi-timescale index set, and considering the uncertainty of source load resources, assess the potential of adjustable resources to participate in response scheduling, and obtain the potential assessment results of adjustable resources. Step 2 includes the following steps:

[0097] Step 2.1: Considering the impact of the uncertainties described in Step 1, establish an adjustable resource response potential assessment model.

[0098] like Figure 1 The figure shows the relationship between adjustable resource response and incentive intensity under the influence of uncertainties. Considering the impact of various uncertainties in assessing the potential of adjustable resource response, the linear region is generally considered to be determined by... Figure 1 The area enclosed by the solid black line indicates that, under a certain stimulus intensity in the linear region, the user response rate is not at a single point, but varies within a possible range.

[0099] Ignoring the randomness of the adjustable resource's response in the dead and saturation regions, the modeling and solution primarily focus on the relevant parameters in the linear region, specifically the dead region inflection point r1, the saturation region inflection point r2, and its ordinate r3. Due to the influence of uncertainties, a linear function cannot be used to model the linear region; instead, the linear region response curve is represented by a quadratic function.

[0100] η=r4(δ-a)(δ-b) (1)

[0101] In the formula: r4, a, b are the relevant parameters of the quadratic model; η is the user response rate; δ is the incentive intensity.

[0102] Step 2.2: Based on the known key parameters of the power generation equipment and user response model, transform the adjustable resource response potential assessment model into a solvable form.

[0103] If the key parameters of a user's response model are known, namely the inflection points (r1, 0) and (r2, r3), then the user response load reduction rate can be obtained from the formula:

[0104]

[0105] The r4 parameter in the formula is a random parameter that considers the impact of uncertainty. The random variation of r4 can characterize the stochastic characteristics of adjustable resources participating in the response process. It is assumed that the deterministic parameters r1, r2, and r3 of users exhibit inter-individual differences, while the random parameter r4 has a similar impact on the response potential of similar types of adjustable resources. Therefore, the random parameter r4 can be analyzed for a cluster of adjustable resources with similar electricity consumption behaviors, thereby obtaining the cluster's adjustable resource response potential under the influence of uncertainties.

[0106] Step 2.3: By performing information mining on the daily data such as the electricity capacity and production electricity characteristics of users in the indicator set obtained in Step 1, the determined parameters r1, r2, and r3 in the adjustable resource response potential assessment model are obtained.

[0107] Considering the large number of indicators in the feature set, and the fact that some indicators contain repetitive features, principal component analysis is used to reduce the dimensionality of the indicator set obtained in step 1. Based on this, least squares fitting is used to mine the key parameters of the user's adjustable potential model, and the following relationship is finally obtained:

[0108]

[0109] In the formula: b 1j b 2j b 3j The coefficients of the principal components of the key parameter influencing factors; U j This represents the principal component of the electricity consumption characteristic index of the j-th user extracted by principal component analysis; a ij x is the coefficient of the i-th index in the j-th principal component; i Let be the value of the i-th indicator. The relationship in the formula represents the common characteristics of this type of user and can be directly used to determine the parameters for general user response characteristics.

[0110] Step 2.4: Based on the historical response data of the user's adjustable resources, substitute it into formula (2) to calculate the value of the corresponding random parameter r4, thereby forming a historical dataset of random parameters based on the user's historical response data.

[0111] Assuming the random parameter r4 follows a normal distribution within a certain range, the box method can be used to determine the fuzzy set of the random parameter:

[0112]

[0113] Where r1, r2, and r3 are known deterministic model parameters; r4 follows a normal distribution with a certain regularity, where the mean of this distribution is... Standard deviation is The parameters μ0 and σ0 are estimated values ​​obtained by point estimation based on the historical response dataset of r4.

[0114] Step 2.5: Treat the obtained estimated value as the normal distribution parameter that r4 satisfies, thereby obtaining the potential assessment result of adjustable resources participating in response scheduling.

[0115]

[0116] Among them, the mean benchmark value μ and the variance benchmark value σ 2 , Mean fuzzy value μ e and the variance fuzzy value σ e These four indicators are model metrics that describe the uncertainty parameters of user response potential. They comprehensively reflect the uncertainty of the adjustable potential of adjustable resources to participate in response business.

[0117] Step 3: Based on the potential assessment results of the adjustable resources, intelligent scheduling of all network resources is performed using federated learning. Step 3 specifically includes the following steps:

[0118] Step 3.1: Establish a particle swarm optimization model with the goal of achieving the best overall effect of new energy consumption.

[0119] The specific optimization objectives are shown in formula (6), including reducing power fluctuations and reducing the abandonment of new energy sources. The constraints are shown in (8), including equipment output limits and power fluctuation rate limits.

[0120]

[0121]

[0122]

[0123] Among them, T new This represents the number of time periods used to calculate power fluctuations. For time period t, λ o,t For the purchase price of electricity, λ flu P represents the cost coefficient for power fluctuation losses. G,t Contributing practically to new energy. Go,t This represents the power output capability of new energy sources. Gmax,t PGmin,t This represents the upper and lower bounds of the output of new energy equipment during time period t. (Used by δ) t δ represents the fluctuation in renewable energy output power during time period t. rise δ is the upper limit of the power increase rate. drop This represents the lower limit of the power reduction rate. Δt is the time interval between the two periods.

[0124] Step 3.2: The central control center server uses federated learning to regulate network resources. The specific implementation steps are as follows:

[0125] Step (a) The central control center server transmits the parameters w of the fixed particle swarm optimization algorithm model to each sub-regional center.

[0126] Step (b) is based on the probability assessment of the potential for adjustable resources to participate in demand response scheduling obtained in Step 2. Each regional center worker obtains local power data on the participation of adjustable resources in demand response. Using this source-load power data and the model parameters w transmitted from the central control center, the regional center workers perform local conventional particle swarm optimization to calculate the adjustment amount of new energy output or user load energy consumption in each region.

[0127] Step (c): Each sub-regional center worker uses local data to perform multiple gradient descent operations locally to obtain the updated gradient g, and calculates the new parameters w to return to the central control center server.

[0128] w←w-α·g (8)

[0129] Where w represents the parameters of the particle swarm optimization model, α represents the specific weight coefficient update step size, and g represents the gradient.

[0130] Step (d): The central control center server collects the new parameters w (e.g., ...) returned by the workers in each region. Figure 2 As shown in the figure, the parameters are weighted and averaged to update the particle swarm model parameters, which are then sent to the workers in each sub-region center.

[0131] Step (e): Repeat steps (b)-(d) until controllable source load resource scheduling is completed in each region.

[0132] This invention also provides an intelligent control device for the consumption of new energy clusters based on federated learning, comprising:

[0133] The feature index set establishment module is used to collect data on different new energy power generation equipment and power users, and to establish a multi-time scale index set considering the uncertainty of source and load resource regulation. The multi-time scale index set includes power generation and power consumption feature index sets at different time scales.

[0134] The potential assessment module is used to assess the potential of adjustable resources to participate in response scheduling based on the multi-timescale index set and considering the uncertainty of source load resources, so as to obtain the potential assessment result of adjustable resources.

[0135] The intelligent scheduling module is used to intelligently schedule resources across the entire network based on the potential assessment results of the adjustable resources and federated learning.

[0136] The feature index set establishment module includes a power generation feature index set establishment module and a power consumption feature index set establishment module.

[0137] The power generation characteristic index set establishment module is used to determine the external influencing factors of the uncertainty of the dispatch potential of new energy output resources, and to collect daily data to establish power generation characteristic index sets at different time scales.

[0138] The external influencing factors of the uncertainty in dispatch potential include weather and climate factors. Weather factors include wind speed, solar radiation, temperature, and humidity, while climate factors include rain, sunny days, and snow. The set of power generation characteristic indicators at different time scales is shown in Table 1, which is obtained through statistical calculation of daily power generation data.

[0139] Table 1. Set of power generation characteristic indicators at different time scales

[0140]

[0141] The electricity consumption characteristic index set establishment module is used to determine the internal influencing factors of the uncertainty of the scheduling potential of user demand response resources, collect data, and establish electricity consumption characteristic index sets at different time scales.

[0142] The internal influencing factors of the uncertainty in the scheduling potential include the equivalent heat capacity C, equivalent thermal resistance R, energy efficiency ratio η, and set temperature T of the polymer. set Allowable temperature adjustment amount ΔT, adjustment duration Δt, user water consumption (m³) n Ambient temperature T out The market electricity price (TOU) and the set of electricity consumption characteristic indicators at different time scales are shown in Table 2, which are obtained through statistical calculation of daily power generation data.

[0143] Table 2. Set of electricity consumption characteristic indicators at different time scales

[0144]

[0145] The potential assessment module includes a potential assessment model establishment module and a potential assessment result acquisition module; wherein...

[0146] The potential assessment model building module is used to build an adjustable resource response potential assessment model. Considering the impact of various uncertainties in the assessment of adjustable resource response potential, the user response rate is not a unique point under a certain incentive intensity in the linear region, but varies within a possible range.

[0147] Ignoring the randomness of the adjustable resource response in the dead and saturation regions, we model and solve for the relevant parameters in the linear region, specifically including the dead region inflection point r1, the saturation region inflection point r2 and its ordinate r3, and represent the linear region response curve as a quadratic function:

[0148] η=r4(δ-a)(δ-b) (1)

[0149] In the formula: r4, a, b are the relevant parameters of the quadratic model; η is the user response rate; δ is the incentive intensity;

[0150] The potential assessment model conversion module is used to convert the adjustable resource response potential assessment model into a solvable form based on the known key parameters of the power generation equipment and user response model.

[0151]

[0152] The r4 parameter in the formula is a random parameter that takes into account the influence of uncertainty. The randomness of r4 can be used to characterize the random characteristics of adjustable resources participating in the response process. It is assumed that the deterministic parameters r1, r2, and r3 of users have individual differences, while the random parameter r4 has a similar influence on the response potential of the same type of adjustable resources. By analyzing the random parameter r4 of a group of adjustable resources with similar electricity consumption behavior, the response potential of the group of adjustable resources under the influence of uncertain factors can be obtained.

[0153] Information mining was performed on the daily data from the multi-timescale indicator set to obtain the definite parameters r1, r2, and r3 in the adjustable resource response potential assessment model:

[0154] Principal component analysis was used to reduce the dimensionality of the multi-timescale index set. Based on this, least squares fitting was used to mine the key parameters of the user adjustable potential model, and the following relationship was finally obtained:

[0155]

[0156] In the formula: b 1j b 2j b 3j The coefficients of the principal components of the key parameter influencing factors; U j This represents the principal component of the electricity consumption characteristic index of the j-th user extracted by principal component analysis; a ij x is the coefficient of the i-th index in the j-th principal component; iLet be the value of the i-th indicator;

[0157] Based on the historical response data of user-adjustable resources, the corresponding r4 value is calculated by substituting it into formula (2), thus forming a historical dataset of random parameters based on the user's historical response data; assuming that the random parameters satisfy a normal distribution within a certain range, the box method is used to determine the fuzzy set of the random parameters:

[0158]

[0159] Where r1, r2, and r3 are known deterministic model parameters; r4 is a normal distribution that follows a certain pattern, where the mean of this distribution is... Standard deviation is The parameters μ0 and σ0 are probability estimates obtained by point estimation based on the historical response dataset of r4;

[0160] The potential assessment result acquisition module is used to treat the obtained probability estimates as parameters of a normal distribution that r4 satisfies, thereby obtaining the potential assessment results of adjustable resources participating in response scheduling:

[0161]

[0162] Among them, the mean benchmark value μ and the variance benchmark value σ 2 , Mean fuzzy value μ e and the variance fuzzy value σ e These four indicators are model metrics that describe the uncertainty parameters of user response potential. They comprehensively reflect the uncertainty of the adjustable potential of adjustable resources to participate in response business.

[0163] The intelligent scheduling module includes a particle swarm optimization model establishment module and a network-wide resource regulation module;

[0164] The particle swarm optimization model building module is used to establish a particle swarm optimization model with the goal of achieving the optimal overall effect of new energy consumption.

[0165] The specific optimization objectives are shown in formula (6), including reducing power fluctuations and reducing the curtailment of renewable energy. The constraints are shown in formula (8), including equipment output limits and power fluctuation rate limits.

[0166]

[0167]

[0168]

[0169] Where T new λ represents the number of time periods used to calculate power fluctuations. o,t For the purchase price of electricity, λflu P is the cost factor for power fluctuation loss. G,t To make actual contributions to new energy, P Go,t For the power output capacity of new energy, P Gmax,t P Gmin,t δ represents the upper and lower limits of the output of new energy equipment during time period t. t δ represents the fluctuation in the output power of new energy sources during time period t. rise δ is the upper limit of the power increase rate. drop This represents the lower limit of the power reduction rate, where Δt is the time interval between the two periods.

[0170] The network-wide resource control module includes a central control center (server) and regional control centers (workers).

[0171] The central control center server is used to perform network-wide resource control using the federated learning approach, specifically including:

[0172] Step (a) The central control center server transmits the parameters w of the fixed particle swarm optimization algorithm model to each sub-regional center;

[0173] Step (b) Based on the potential assessment results of the adjustable resources, the workers of each sub-regional center obtain the power data of local adjustable resources participating in demand response. Using the source-load power data information and the model parameters w transmitted from the central control center, the sub-regional center workers perform local conventional particle swarm optimization to calculate the adjustment amount of new energy output or user load energy consumption in each region.

[0174] Step (c) Each sub-regional center worker uses local data to perform multiple gradient descent operations locally to obtain the updated gradient g, and calculates the new parameters w to return to the central control center server.

[0175] w←w-α·g (9)

[0176] Where w represents the particle swarm optimization model parameters, α represents the specific weight coefficient update step size, and g represents the gradient;

[0177] Step (d) The central control center server collects the new parameters w returned by the workers in each region, performs a weighted average of the parameters, updates the particle swarm model parameters, and then sends them down to the workers in each sub-regional center.

[0178] Step (e) repeats steps (b)-(d) until the controllable source load resource scheduling in each region is completed.

[0179] This invention, taking into account the influence of uncertain factors, realizes the cluster control potential assessment of new energy equipment and users, and based on the concept of federated learning, protects the data privacy of each region in the subsequent resource control process, thereby improving the security of overall intelligent control.

[0180] Another aspect of the present invention provides a new energy cluster consumption intelligent control system based on federated learning, comprising: a computer-readable storage medium and a processor;

[0181] The computer-readable storage medium is used to store executable instructions;

[0182] The processor is used to read executable instructions stored in the computer-readable storage medium and execute the intelligent control method for new energy cluster consumption based on federated learning described in the first aspect.

[0183] In another aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent control method for the consumption of new energy clusters based on federated learning as described in the first aspect.

[0184] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0185] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0186] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0187] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0188] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for intelligent regulation and control of new energy cluster consumption based on federated learning, characterized in that: Includes the following steps: Collect data on different new energy power generation equipment and power users, and establish a multi-time-scale indicator set, which includes power generation and power consumption characteristic indicator sets at different time scales; Based on the multi-timescale index set, the potential of adjustable resources to participate in response scheduling is evaluated, and the potential evaluation results of adjustable resources are obtained. Based on the potential assessment results of the adjustable resources, intelligent scheduling of resources across the entire network is performed using federated learning. Based on the aforementioned multi-timescale index set, the potential of adjustable resources to participate in response scheduling is evaluated to obtain the potential evaluation results of adjustable resources. Specifically, this includes the following steps: Establish a model for assessing the response potential of adjustable resources: （4）； Where r1, r2, and r3 are known deterministic model parameters; r4 is a normal distribution that follows a certain pattern, where the mean of this distribution is... Standard deviation is ;parameter and These are probability estimates obtained by point estimation based on the historical response dataset of r4; By treating the obtained probability estimates as parameters of a normal distribution that r4 satisfies, we can obtain the potential assessment results for adjustable resources to participate in response scheduling: （5）； Among them, the mean benchmark value Variance baseline value fuzzy mean value and variance fuzzy value These are model metrics that describe the uncertainty parameters of user response potential. These four metrics comprehensively reflect the uncertainty of the adjustable potential of adjustable resources to participate in response business. Based on the potential assessment results of the adjustable resources, intelligent scheduling of network resources is performed using federated learning, specifically including: To optimize the overall effect of new energy consumption, a particle swarm optimization model is established: The specific optimization objectives are shown in formula (6), including reducing power fluctuations and reducing the curtailment of renewable energy. The constraints are shown in formula (8), including equipment output limits and power fluctuation rate limits. （6）； （7）； （8）； in This indicates the number of time periods used to calculate power fluctuations. For the purchase price of electricity, This is the cost coefficient for power fluctuation loss. To make practical contributions to new energy, For the power output capacity of new energy sources, , This represents the upper and lower limits of the output of new energy equipment during time period t. This represents the fluctuation in the output power of new energy sources during time period t. This is the upper limit of the power increase rate. This represents the lower limit of the power reduction rate. This represents the time interval between two time periods.

2. The intelligent regulation and control method for new energy cluster consumption based on federated learning according to claim 1, characterized in that: Data on different new energy power generation equipment and electricity users are collected to establish a multi-time-scale indicator set. The multi-time-scale indicator set includes power generation and electricity consumption characteristic indicator sets at different time scales, specifically including the following steps: Identify the external influencing factors that determine the scheduling potential of new energy power resources and collect daily data to establish a set of power generation characteristic indicators at different time scales; The external influencing factors of the uncertainty of the dispatch potential include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunny days, and snow. The power generation characteristic index set at different time scales is obtained by statistical calculation of daily power generation data. Identify the internal influencing factors of the uncertainty in the scheduling potential of user demand response resources, collect data, and establish a set of electricity consumption characteristic indicators at different time scales; The internal influencing factors of the uncertainty in the scheduling potential include the equivalent heat capacity C, equivalent thermal resistance R, and energy efficiency ratio of the polymer. Set temperature Permissible temperature adjustment range Duration of regulation User water consumption Ambient temperature The market electricity price (TOU) is calculated by statistically analyzing daily power generation data, based on the electricity consumption characteristic index set at different time scales.

3. The intelligent regulation and control method for new energy cluster consumption based on federated learning according to claim 1, characterized in that: The central control center server uses federated learning to regulate network resources, specifically including: Step (a) The central control center server transmits the parameters w of the fixed particle swarm optimization algorithm model to each sub-region center; Step (b) Based on the potential assessment results of the adjustable resources, the workers of each sub-regional center obtain the power data of the local adjustable resources participating in demand response. Using the source-load power data information and the model parameters w transmitted from the central control center, the sub-regional center workers perform local conventional particle swarm optimization to calculate the adjustment amount of new energy output or user load energy consumption in each region. Step (c) Each sub-regional center worker performs multiple gradient descent operations locally using local data to obtain the updated gradient g, and calculates the new parameters w to return to the central control center server. w← w-α·g (9; Where w represents the particle swarm optimization model parameters, α represents the specific weight coefficient update step size, and g represents the gradient; Step (d) The central control center server collects the new parameters w returned by the workers in each region, performs a weighted average of the parameters, updates the particle swarm model parameters, and then sends them down to the workers in each sub-regional center. Step (e) repeats steps (b)-(d) until the controllable source load resource scheduling in each region is completed.

4. A smart control device for the consumption of new energy clusters based on federated learning, characterized in that, include: The feature index set establishment module is used to collect data from different new energy power generation equipment and power users, and establish a multi-time scale index set, which includes power generation and power consumption feature index sets at different time scales. The potential assessment module is used to assess the potential of adjustable resources to participate in response scheduling based on the multi-timescale indicator set, and obtain the potential assessment results of adjustable resources. The intelligent scheduling module is used to intelligently schedule resources across the entire network based on the potential assessment results of the adjustable resources and federated learning. The potential assessment module includes a potential assessment model establishment module and a potential assessment result acquisition module; wherein... The potential assessment model building module is used to build adjustable resource response potential assessment models. （4）； Where r1, r2, and r3 are known deterministic model parameters; r4 is a normal distribution that follows a certain pattern, where the mean of this distribution is... Standard deviation is ;parameter and These are probability estimates obtained by point estimation based on the historical response dataset of r4; The potential assessment result acquisition module is used to treat the obtained probability estimates as parameters of a normal distribution that r4 satisfies, thereby obtaining the potential assessment results of adjustable resources participating in response scheduling: （5）； Among them, the mean benchmark value Variance baseline value fuzzy mean value and variance fuzzy value These are model metrics that describe the uncertainty parameters of user response potential. These four metrics comprehensively reflect the uncertainty of the adjustable potential of adjustable resources to participate in response business. The intelligent scheduling module includes a particle swarm optimization model establishment module and a network-wide resource regulation module; The particle swarm optimization model establishment module is used to establish a particle swarm optimization model with the goal of achieving the optimal overall effect of new energy consumption. The specific optimization objectives of the particle swarm optimization model are shown in formula (6), including reducing power fluctuations and reducing the curtailment of new energy sources. The constraints are shown in formula (8), including equipment output limits and power fluctuation rate limits. （6）； （7）； （8）； in This indicates the number of time periods used to calculate power fluctuations. For the purchase price of electricity, This is the cost coefficient for power fluctuation loss. To make practical contributions to new energy, For the power output capacity of new energy sources, , This represents the upper and lower limits of the output of new energy equipment during time period t. This represents the fluctuation in the output power of new energy sources during time period t. This is the upper limit of the power increase rate. This represents the lower limit of the power reduction rate. This represents the time interval between two time periods.

5. The intelligent control device for new energy cluster consumption based on federated learning as described in claim 4, characterized in that, The feature index set establishment module includes a power generation feature index set establishment module and a power consumption feature index set establishment module. The power generation characteristic index set establishment module is used to determine the external influencing factors of the uncertainty of the dispatch potential of new energy output resources, and to collect daily data to establish power generation characteristic index sets at different time scales. The external influencing factors of the uncertainty of the dispatch potential include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunny days, and snow. The power generation characteristic index set at different time scales is obtained by statistical calculation of daily power generation data. The electricity consumption characteristic index set establishment module is used to determine the internal influencing factors of the uncertainty of the scheduling potential of user demand response resources, and to collect data to establish electricity consumption characteristic index sets at different time scales. The internal influencing factors of the uncertainty in the scheduling potential include the equivalent heat capacity C, equivalent thermal resistance R, and energy efficiency ratio of the polymer. Set temperature Permissible temperature adjustment range Duration of regulation User water consumption Ambient temperature The market electricity price (TOU) is calculated by statistically analyzing daily power generation data, based on the electricity consumption characteristic index set at different time scales.

6. The intelligent control device for new energy cluster consumption based on federated learning as described in claim 4, characterized in that, The network-wide resource control module includes a central control center (server) and regional control centers (workers). The central control center server is used to perform network-wide resource control using the federated learning approach, specifically including: Step (a) The central control center server transmits the parameters w of the fixed particle swarm optimization algorithm model to each sub-region center; Step (b) Based on the potential assessment results of the adjustable resources, the workers of each sub-regional center obtain the power data of the local adjustable resources participating in demand response. Using the source-load power data information and the model parameters w transmitted from the central control center, the sub-regional center workers perform local conventional particle swarm optimization to calculate the adjustment amount of new energy output or user load energy consumption in each region. Step (c) Each sub-regional center worker performs multiple gradient descent operations locally using local data to obtain the updated gradient g, and calculates the new parameters w to return to the central control center server. w← w-α·g (9; Where w represents the particle swarm optimization model parameters, α represents the specific weight coefficient update step size, and g represents the gradient; Step (d) The central control center server collects the new parameters w returned by the workers in each region, performs a weighted average of the parameters, updates the particle swarm model parameters, and then sends them down to the workers in each sub-regional center. Step (e) repeats steps (b)-(d) until the controllable source load resource scheduling in each region is completed.

Images