Fine identification method for industrial load model parameters

Abstract

The invention discloses a refined identification method for industrial load model parameters, and belongs to the technical field of industrial load model parameter identification, and the method comprises the steps: firstly analyzing the characteristics of an industrial load, carrying out the data collection and processing, carrying out the generation of an industrial load scene set through employing a Monte Carlo method, and carrying out the reduction of the scene set through employing k-means clustering; then, the industrial load type is analyzed according to the industrial load data, and a typical industrial load system including load characteristic curves of a steel production load, a cement production load and an electrolytic aluminum load and corresponding production data parameters is given; secondly, modeling is carried out on the typical industrial load, and parameters to be identified are given; thirdly, iteratively solving the parameters by adopting an improved tuna swarm optimization algorithm; and finally, substituting the parameters identified by the optimization algorithm into the model for load calculation, thereby realizing the construction of the refined identification method for the industrial load model parameters.

Description

Technical Field

[0001] The present invention relates to industrial load model parameter identification, and more particularly, to parameter identification between production capacity and power consumption of a typical industrial load in an industrial production process. Background Art

[0002] The technical background of industrial load model parameter identification is rooted in the rise of intelligent manufacturing, which emphasizes improving production efficiency and flexibility by integrating advanced information technologies such as the Internet of Things, big data analysis, and artificial intelligence. The development of this modeling technology has benefited from the significant increase in computing power and the advancement of algorithms, which enable complex production systems to be accurately simulated and optimized to cope with changing market demands and production conditions. The refined identification of industrial load model parameters achieves precise control and management of production loads by simulating and analyzing various factors in the production process. Through refined parameter identification, it is possible to more accurately predict resource requirements and time arrangements in the production process, thereby optimizing the production process, reducing waiting and idle time, and improving overall production efficiency; helping enterprises to allocate resources more reasonably, reduce waste, and accurately control the input of raw materials, energy, and manpower, effectively reducing production costs; refined production model parameter identification can provide accurate production and demand forecasts, help enterprises achieve more reasonable production management, reduce inventory backlogs, and reduce inventory costs; the data and analysis results provided by the model can serve as an important basis for enterprise production and energy efficiency management, helping management to make more scientific and reasonable decisions; the refined identification of load model parameters often needs to be combined with the latest information technology, such as the Internet of Things, big data, and artificial intelligence, which can not only improve production efficiency, but also provide impetus for the technological innovation of enterprises; by optimizing resource use and reducing waste, refined parameter identification helps to reduce the impact on the environment and achieve green production. The refined identification of industrial load model parameters is one of the key technologies for realizing intelligent manufacturing. It brings many benefits to enterprises by improving the accuracy and controllability of production, and is an indispensable part of modern manufacturing. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a refined identification method for industrial load model parameters to solve the problem that industrial load is difficult to fully predict and quantify due to uncertain factors in the production process, such as equipment failure, fluctuations in raw material quality, human operational errors, etc.

[0004] In order to achieve the above object, the present invention proposes a refined identification method for industrial load model parameters, which comprises the following specific steps:

[0005] Step 1: Collect data on industrial loads, industrial production materials and products, use the Monte Carlo method to generate a scenario set, and use K-means clustering to reduce the scenarios; obtain the load characteristic curve of the typical industrial load model and the corresponding production parameters.

[0006] Step 1.1: Collect historical industrial load data, including power consumption and time series data during industrial production

[0007] Step 1.2: Generate scenarios using the Monte Carlo method. First, determine the probability distribution: Determine the probability distribution of the load based on historical industrial load data. Commonly used distributions include normal distribution, Weibull distribution, and Beta distribution. For example, if the load follows a normal distribution, its probability density function (PDF) is:

[0008]

[0009] Among them, μ is the mean, σ 2 is the variance, and xp is the number of data.

[0010] Step 1.3: Generate random samples: Use a random number generator to generate random samples that conform to the above probability distribution. For example, to generate a normally distributed sample use the Box-Muller transformation or the central limit theorem.

[0011] Step 1.4: Apply scenario constraints: Screen or adjust the generated samples according to the physical and operational constraints of the industrial load to ensure the rationality of the samples.

[0012] Step 1.5: Construct scenario set: Combine the generated samples into different scenarios, each of which represents a possible load state, thereby constructing the industrial load scenario set.

[0013] Step 1.6: Use k-means clustering to reduce the scenario. First, choose the number of clusters: determine how many clusters the data should be divided into. Use the elbow rule and silhouette coefficient method to assist in the decision.

[0014] Step 1.7: Initialize the center point: Randomly select K data points as the initial cluster center (centroid).

[0015] Step 1.8: Assign data points to the nearest cluster center: For each data point, calculate its distance from each cluster center and assign it to the nearest cluster, using Euclidean distance as the distance metric:

[0016]

[0017] Among them, d(xi,cj) is the distance between the data point xi and the cluster center cj, n is the characteristic dimension of the data point, compare the d(xi,cj) of each data point xi, select the smallest d(xi,cj), and the corresponding cluster center cj is the cluster center to which it belongs;

[0018] Step 1.9: Recalculate the cluster center: For each cluster, calculate the mean of all points in the cluster and use the mean as the new cluster center:

[0019]

[0020] Among them, N j is the number of data points in cluster j, sj is the new cluster center obtained after calculation, xi is the data point, and cj is the cluster center to which xi belongs, that is, the cluster center.

[0021] Step 1.10: Repeat steps 1.8 and 1.9: Repeat steps 1.8 and 1.9 until the stopping condition is met, that is, the cluster center no longer changes significantly or the preset number of iterations is reached.

[0022] Step 1.11: Evaluate clustering effect: Use the silhouette coefficient indicator to evaluate the clustering effect, or evaluate according to the application scenario. The K-means algorithm is usually implemented by iteratively optimizing the position of the cluster center. However, the K-means algorithm has some limitations, such as being sensitive to the initial cluster center, being able to find only local optimal solutions, and assuming that clusters are convex and of similar size. In practical applications, it may be necessary to run the algorithm multiple times to avoid falling into the local optimum and to improve the clustering quality.

[0023] Step 2: Construct a typical industrial load model and provide the parameters to be identified.

[0024] Step 2.1: The steel production load model is as follows:

[0025] Psz.it=Ps.deal.i.t+Ps.eaf.it

[0026] Among them, i represents the i-th steel production enterprise; t represents time t, Pzs.it represents the total load of steel production by steel production enterprise i at time t; Ps.deal.it represents the power consumption of steel rolling; Ps.eaf.it represents the power consumption of electric arc furnace;

[0027] Ps.deal.it calculates this way:

[0028]

[0029] Wherein, k is the number of steel tires on the steel rolling production line per unit time (pieces). as.i is the coefficient of the quadratic term of the production consumption coefficient, and bs.i is the coefficient of the linear term of the production consumption coefficient. Psteel tire is the unit power consumed in the production of a single steel tire (Kw). Psk.deal.it is the fixed energy consumption of the steel rolling production line in the hot start state (Kw).

[0030] Ps.eaf.it is calculated as follows:

[0031]

[0032] Where Ps.eafs.it is the power output of the arc furnace during stable operation (Kw), ts is the start-up time of the arc furnace (h), and te is the stop time of the arc furnace (h).

[0033] Step 2.2: The cement production load model is as follows:

[0034]

[0035] Among them, Pconz.it is the motor power (kW), Q is the raw material input (tons / h), E is the efficiency of the crushing equipment (%), η is the motor efficiency (%), V is the voltage (V), and cosφ is the power factor.

[0036] Step 2.3: The electrolytic aluminum production load model is as follows:

[0037] M AL .it=[λa 1 +λa 2 (λa 3 +λa 4 P AL .it) 0.5 ]Δt

[0038] Among them, PAL.it is the power of the electrolytic cell (Kw), and MAL.it is the output of the electrolytic cell (g). a1 (g / h), λa 2 (unitless), λa3(g / h) 2 and λ a4 (g / Kw.h) 2 is the coefficient of relationship between electrolytic cell output and power.

[0039] Step 3: Improve the tuna school optimization algorithm.

[0040] The following is the process of improving the tuna school optimization algorithm:

[0041] Step 3.1: Initialization: Randomly initialize the first generation population positions, with each tuna representing a candidate solution.

[0042] xint=rand·(ub-lb)+lb,i=1,2,…,NP

[0043] Among them, xint is the first generation population; ran is a random number; ub and lb are the upper and lower limits of variable changes; NP is the population size.

[0044] Step 3.2: Enter the iteration phase, and then use the adaptive mutation strategy to enhance the algorithm performance;

[0045]

[0046] Where p is the pth iteration, pmax is the maximum number of iterations, xvari.p is the mutated individual, xi.p is the i-th individual in the population at the pth iteration, xbest.p and xrand.p are the individual with the best fitness and the randomly generated individual at the pth iteration respectively; res is the scaling factor, which is calculated as follows:

[0047]

[0048] This adaptive mutation strategy can enhance the algorithm's search capability in the early stages of iteration and enhance the algorithm's optimization capability in the later stages of iteration. The mutated individuals also need to use a greedy strategy to determine whether to replace other individuals (if the fitness of the mutated individual is better than that of the original individual, the mutated individual will replace the original individual).

[0049] Step 3.3: Adaptive crossover strategy;

[0050]

[0051] Among them, xi.pj, xrand.pj and xbest.pj represent the j-th decision variables of the crossover individual, random individual and optimal individual during iteration; this mutation strategy can enhance the search ability by crossing with random individuals in the early stage of iteration, and can improve the optimization accuracy by crossing with the optimal individual in the later stage of iteration.

[0052] Step 3.4: Calculate the fitness of the new population after mutation and crossover, and determine the leader fish: Calculate the fitness of each individual, and select the tuna with the highest fitness as the leader fish, which is the optimal solution.

[0053] Step 3.5: Update position: There are three ways to update during the iteration:

[0054] Update position: There are three ways to update during the iteration process:

[0055] Method 1: Random update

[0056] When entering the random update, the update formula is the same as the update method initialized in step (1);

[0057] Method 2: Spiral Update

[0058] There are two strategies for spiral foraging updates:

[0059] The first one:

[0060]

[0061] Second type:

[0062]

[0063] Among them, xi.p+1 and xi.p represent the i-th individual in the population at the p+1th iteration and the pth iteration respectively; xbest.p represents the individual with the best fitness in the population at the pth iteration; xrand.p represents a randomly generated individual, which is calculated as follows:

[0064] x rand.p = rand·(ub-lb)+lb

[0065] α 1 , α 2 and β are calculated as follows:

[0066]

[0067]

[0068] β=el·cos(2π)

[0069]

[0070] Where a is a constant;

[0071] Different strategies are selected based on random numbers:

[0072] Strategy 1

[0073] Strategy 2

[0074] Method 3: Parabolic foraging: The update formula of parabolic foraging is as follows:

[0075]

[0076] Among them, TF is a random number with a value of -1 or 1; the value of w is as follows:

[0077]

[0078] The mechanism for selecting the three update methods is shown in the following pseudo code:

[0079]

[0080]

[0081] Step 3.6: Consider boundary conditions: Check if the updated tuna position is outside the predefined search space boundary, and if so, adjust it back to the boundary.

[0082] Step 3.7: Repeat steps 3.2 to 3.6 until the maximum number of iterations is reached or other stopping conditions are met.

[0083] Step 3.8: Output the optimal solution: When the maximum number of iterations is reached or the iteration accuracy meets the requirements, the iteration is completed and the optimal fitness is output.

[0084] The process of the enhanced tuna school optimization algorithm is as follows:

[0085]

[0086]

[0087] Step 4: Use the enhanced tuna school optimization algorithm to identify parameters and bring the optimal parameters into the system for verification.

[0088] Step 4.1: Determine input parameters

[0089] For steel production loads, the input data include:

[0090] Ps.deal.it Power consumption of steel rolling, Ps.eaf.it Power consumption of electric arc furnace, k Number of steel tires on the steel rolling production line per unit time (pieces), Psteel tire Unit power consumed in producing a single steel tire (Kw), Psk.deal.it Fixed energy consumption of the steel rolling production line in hot start state (Kw), Ps.eafs.it Power output of the electric arc furnace during stable operation (Kw).

[0091] The parameters to be identified are: as.i is the coefficient of the quadratic term of the production consumption coefficient, bs.i is the coefficient of the linear term of the production consumption coefficient, ts is the start time of the arc furnace (h), and te is the stop time of the arc furnace (h).

[0092] For cement production load, the input parameters are:

[0093] Pconz.it is the motor power (kW), Q is the raw material input (tons / h), V is the voltage (V), and cosφ is the power factor.

[0094] The parameters to be identified are: E efficiency of the crushing equipment (%), η motor efficiency (%).

[0095] For the electrolytic aluminum production load, the input parameters are: PAL.it electrolytic cell power (Kw), MAL.it electrolytic cell output (g).

[0096] Step 4.2: Determine the parameters to be identified. The parameters to be identified are: a1 (g / h), λ a2 (unitless), λ a3 (g / h) 2 and λ a4 (g / Kw.h) 2 A total of 5 electrolytic cell output and power relationship coefficients.

[0097] There are 12 sets of input data and 10 sets of parameters to be identified. In addition, the parameters to be identified should have a parameter variation range to ensure the efficiency of the parameter identification process. The variation range of the parameters to be identified is based on the operating characteristics of the load equipment, nameplate, and expert experience data.

[0098] Step 4.3: Objective function definition. The design objective function is the deviation sum of the load after parameter identification and the actual load. The formula of the objective function is as follows:

[0099] fz=f 1 +f 2 +f 3 +f 4

[0100] Among them, f 1 、f 2 、f 3 and f 4 They are respectively the steel rolling load fitting deviation, arc furnace load fitting deviation, cement load fitting deviation, and electrolytic aluminum load fitting deviation. The detailed calculation method is as follows:

[0101]

[0102] Among them, Psw.deal.it represents the fitted steel rolling load; Ps.deal.it represents the actual steel rolling load; Psw.eaf.it represents the fitted steel rolling load; Ps.eaf.it represents the actual steel rolling load; Pwconz.it represents the fitted steel rolling load; Pconz.it represents the actual steel rolling load; PwAL.it represents the fitted steel rolling load; PAL.it represents the actual steel rolling load.

[0103] Step 4.4: Definition of decision variables. The parameters to be identified are: as.i is the coefficient of the quadratic term of the production consumption coefficient, bs.i is the coefficient of the linear term of the production consumption coefficient. ts is the start time of the arc furnace (h), te is the stop time of the arc furnace (h), E is the efficiency of the crushing equipment (%), η is the motor efficiency (%), λ a1 (g / h), λ a2 (unitless), λ a3 (g / h) 2 and λ a4 (g / Kw.h) 2 As decision variables for enhancing the tuna school optimization algorithm.

[0104] Step 4.5: Definition of constraints. In the process of enhancing the optimization algorithm for identifying tuna schools, there are only upper and lower bounds of the parameters to be identified, but no balance constraints.

[0105] Step 4.6: Output the optimal parameter identification result. After the enhanced tuna school optimization algorithm is iterated, the optimal objective function value (fitness value) and the optimal decision variable are output.

[0106] Intentional effects:

[0107] By accurately modeling the production process of industrial loads and identifying production parameters, the relationship between industrial products and industrial production power loads can be established, thereby helping industrial enterprises to more accurately understand and predict their power demand, thereby optimizing production plans and energy use, and improving the intelligence and automation level of the production process. Reduce energy waste. It helps power companies to allocate and dispatch power resources more effectively. Therefore, the construction of a refined production simulation model for industrial loads can improve energy utilization efficiency, optimize power system operation, and support sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 The three-dimensional image of the Monte Carlo scene in Example 1;

[0109] Figure 2 It is a two-dimensional diagram of the Monte Carlo method scene in Example 1;

[0110] Figure 3 This is the clustering power diagram of Example 1;

[0111] Figure 4 This is the clustering power probability diagram of Example 1;

[0112] Figure 5 This is a graph showing the power consumption identification result of steel rolling in Example 1;

[0113] Figure 6 This is a diagram of the arc furnace power identification result of Example 1;

[0114] Figure 7 This is the cement power identification result diagram of Example 1;

[0115] Figure 8 This is a diagram of the electrolytic aluminum power identification result of Example 1;

[0116] Fig. 9 The three-dimensional image of the Monte Carlo scene in Example 2;

[0117] Fig.10 The three-dimensional image of the Monte Carlo scene in Example 2;

[0118] Fig.11 This is the clustering power diagram of Example 2;

[0119] Fig.12 This is the clustering power probability diagram of Example 2;

[0120] Fig.13 This is a graph showing the power consumption identification result of steel rolling in Example 2;

[0121] Fig.14 This is a diagram showing the power consumption identification result of the arc furnace in Example 2;

[0122] Fig.15 This is the result diagram of cement power consumption identification in Example 2;

[0123] Fig.16 This is a diagram showing the power consumption identification result of electrolytic aluminum in Example 2;

[0124] Fig.17 This is a graph showing the power consumption identification result of steel rolling in Example 3;

[0125] Fig.18 This is a diagram showing the power consumption identification result of the arc furnace in Example 3;

[0126] Fig.19 This is the result diagram of cement power consumption identification in Example 3;

[0127] Fig. 20 This is a diagram showing the results of identifying the power consumption of electrolytic aluminum in Example 3;

[0128] Fig.21 Refined production simulation model flow chart for industrial loads;

[0129] Fig. 22 Flowchart of algorithm optimization for improved tuna schools;

[0130] Fig.23 Generate and reduce flow charts for industrial load scenarios. DETAILED DESCRIPTION

[0131] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical scheme in the present invention is clearly and completely described below in conjunction with the embodiments of the present invention and the accompanying drawings. Obviously, the present invention is not limited to the following embodiments, and the specific implementation method can be determined according to the technical scheme of the present invention and the actual situation. In order to avoid confusing the essence of the present invention, the known methods, processes and flows are not described in detail.

[0132] The embodiment of the present invention proposes a refined identification method for load model parameters in a typical industrial production process. The proposed method includes the following contents: firstly, the industrial load characteristics are analyzed, data is collected and processed, and based on a limited amount of industrial load data, the Monte Carlo method is used to generate an industrial load scenario set, and the k-means clustering is used to reduce the scenario set; then, the industrial load type is analyzed according to the industrial load data, and a typical industrial load system is given, including the load characteristic curves of steel production load, cement production load and electrolytic aluminum load and the corresponding means of production parameters; secondly, the above-mentioned typical industrial load is modeled and the parameters to be identified are given; thirdly, the improved tuna swarm optimization algorithm is used to iteratively solve the parameters; finally, the parameters identified by the optimization algorithm are brought into the model for load calculation, thereby realizing the construction of a refined identification method for industrial load model parameters.

[0133] A refined identification method for industrial load model parameters, the specific steps are:

[0134] Step S101: industrial load characteristic analysis and data collection, using the Monte Carlo method to generate industrial load scenario sets, and using k-means clustering to give a typical load characteristic curve;

[0135] Step S102: Count and classify industrial loads to obtain a typical industrial load system;

[0136] Step S103: Based on the load characteristic curve obtained by k-means, an improved tuna swarm optimization algorithm is used to perform parameter identification on a typical industrial load system;

[0137] Step S104: Bring the optimal identification parameters into the typical industrial load model to achieve accurate modeling of the industrial load.

[0138] Fig.21 The flow chart of the industrial load refined production simulation model is shown; Fig. 22 The flowchart of the improved tuna school optimization algorithm is shown; Fig.23 An industrial load scenario generation and reduction flow chart is shown.

[0139] Embodiment 1:

[0140] The experiment uses summer load data from an industrial area in Tangshan, and then generates Monte Carlo method and K-means clustering scenarios based on the industrial load data. The specific steps are as follows:

[0141] S01: Collect historical industrial load data, including: Ps.deal.it steel rolling power consumption, Ps.eaf.it electric arc furnace power consumption, k number of steel tires on the steel rolling production line per unit time (pieces), Psteel tire unit power consumed in producing a single steel tire (Kw), Psk.deal.it fixed energy consumption of the steel rolling production line in hot start state (Kw), Ps.eafs.it power output of the electric arc furnace during stable operation (Kw).

[0142] S02: Determine the probability distribution: Determine the probability distribution of the load based on historical industrial load data. Use normal distribution and generate a probability density function.

[0143] S03: Generate random samples: Based on the probability density function in step S02, generate a power or capacity data sequence according to the corresponding probability.

[0144] S04: Constructing a scene set: repeatedly generating data sequences until the number of data sequences meets the requirement of the number of scene sets.

[0145] S05: Use k-means clustering to reduce the scene, first select 5 cluster centers.

[0146] S06: Initialize center point: Randomly select 5 data points as the initial cluster center (centroid).

[0147] S07: Calculate the distance from each scene to each centroid, and each scene selects the centroid closest to itself as a cluster.

[0148] S08: Recalculate cluster center: For each cluster, calculate the mean of all points in the cluster and use this mean as the new cluster center.

[0149] S09: Repeat step S07 and step S08 until all the centroids no longer change.

[0150] S10: Enter the parameter identification phase. First, initialize the population. Set the parameters to be identified: as.i, bs.i, ts, te, E, η, λ a1 ,λ a2 ,λ a3 ,λ a4 As the decision variable of the improved tuna school optimization algorithm. rand.t =rand·(ub-lb)+lb to initialize the population, where ub and lb are the upper and lower bounds of the decision variables.

[0151] S11: Enter the iteration phase and use adaptive mutation and crossover strategies to update the population.

[0152] S12: Calculate the fitness of the new population after mutation and crossover. K, Psteeltire, Psk.deal.it, Ps.eafs.it are the data obtained by clustering. Substitute all the above variables into the following formula for calculation:

[0153]

[0154]

[0155]

[0156] MAL.it=[λa1+λa2(λa3+λa4PwAL.it)0.5]Δt

[0157] Among them, Psw.deal.it represents the steel rolling load obtained by fitting; Psw.eaf.it represents the steel rolling load obtained by fitting; Pwconz.it represents the steel rolling load obtained by fitting; PwAL.it represents the steel rolling load obtained by fitting. Ps.deal.it represents the actual steel rolling load obtained by clustering; Ps.eaf.it represents the actual steel rolling load obtained by clustering; Pconz.it represents the actual steel rolling load obtained by clustering; PAL.it represents the actual steel rolling load obtained by clustering. The above data and the decision variables in the new population in step S12 are used to calculate the objective function value fz.

[0158]

[0159] f z =f 1 +f 2 +f 3 +f 4

[0160] S13: Use random update, spiral update, and parabolic foraging to update the population.

[0161] S14: Boundary condition check, determine whether each decision variable is within the range of lb~ub. If not, correct it. If it is greater than the upper limit of the variable, correct it to the upper limit value; if it is less than the lower limit of the variable, correct it to the lower limit value.

[0162] S15: Determine whether the number of iterations has been reached, if not, repeat steps S11 to S14. Otherwise, output the best decision variables and objective function values.

[0163] This enables industrial load parameter identification.

[0164] The results of scene generation and reduction are as follows Figures 1 to 4 The parameter identification results are shown in Figures 5 to 8 shown. Figures 1 to 4 The industrial load data scenario set generated by the Monte Carlo method is shown. The industrial load data here is the total load of steel, electrolytic aluminum, and cement production. The cluster power diagram is the typical industrial load power obtained after K-meas clustering. Table 1 and the cluster power probability diagram show the probability of the corresponding typical industrial load power, and the sum is 1.

[0165] Table 1 Probability table of scene selection

[0166] Number of scenes Scenario 1 Scenario 2 Scene 3 Scene 4 Scene 5 Corresponding probability 0.181 0.215 0.199 0.301 0.104

[0167] Figures 5 to 8 The parameter identification results of steel rolling power consumption, arc furnace power consumption, cement power, and electrolytic aluminum power are shown. Table 2 shows the identification results of various loads obtained by the improved tuna swarm optimization algorithm and other algorithms:

[0168] Table 2 fz (objective function value) of five algorithms

[0169]

[0170]

[0171] It can be seen from Table 2 that the proposed ETSO algorithm has the best effect in fitting various industrial load models. Figure 9-12 It can be seen that ETSO can better track the industrial load trend and realize the refined identification of industrial load model parameters.

[0172] Example 2

[0173] The experiment uses spring load data from an industrial zone in Tangshan, and the specific steps are similar to those in Example 1. Figures 9 to 12 The industrial load parameter identification results are shown in Figures 13-16 As shown:

[0174] Example 3

[0175] The experiment uses spring load data from an industrial area in Taiyuan, and the specific steps are similar to those in Example 1. The difference is that K-means is directly performed on the data set, and the centroid (typical load data) obtained after clustering is no longer the sum of the scenes belonging to the same centroid and then averaged, but the centroid is calculated first, and then the scene closest to the centroid is selected as the new centroid. The significance of this step is that the scenes after reduction are real scenes, which are more in line with production reality.

[0176] The parameter identification process of Example 3 is the same as that of Example 1 and Example 2. The results are as follows: Figures 17-20 As shown, Fig.17 The power consumption identification result diagram of steel rolling in Example 3 is shown; Fig.18 The electric arc furnace power consumption identification result diagram of Example 3 is shown; Fig.19 The cement power consumption identification result diagram of Example 3 is shown; Fig. 20 The figure shows the result of identifying the power consumption of electrolytic aluminum in Example 3.

[0177] In summary, the method proposed in the present invention and the construction of a refined production simulation model of industrial load can improve energy utilization efficiency, optimize power system operation, and support sustainable development.

[0178] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A refined identification method for industrial load model parameters, characterized in that: include: Data collection of industrial loads, industrial production materials and products, using the Monte Carlo method to generate industrial load data scenario sets and using k-means clustering to reduce the scenarios to give load characteristic curves of typical industrial load models and corresponding production parameters; Construct a typical industrial load model and provide the parameters to be identified; The improved tuna school optimization algorithm is used for parameter identification; Finally, the parameters identified by the improved tuna school optimization algorithm are brought into the typical industrial load model for load calculation, thereby realizing the construction of a refined identification method for industrial load model parameters.

2. The method for fine identification of industrial load model parameters according to claim 1 is characterized in that: The steps of the Monte Carlo scenario generation method are as follows: (1) Collect data: Collect historical industrial load data, including power consumption patterns and time series data during industrial production; (2) Determine the probability distribution: Determine the probability distribution of loads based on historical industrial load data; (3) Generate random samples: Use a random number generator to generate random samples that conform to the above probability distribution; (4) Sample screening: Screen or adjust the generated samples according to the physical and operational constraints of the industrial load to ensure the rationality of the samples; (5) Constructing scene sets: The generated samples are combined into different scenarios, each of which represents a load state, thereby realizing the construction of an industrial load scenario set.

3. The method for fine identification of industrial load model parameters according to claim 2 is characterized in that: The steps of k-means clustering are as follows: (1) Select the number of clusters: determine the number of clusters the data is divided into; (2) Initialize the center point: randomly select K data points as the initial cluster center; (3) Assign data points to the nearest cluster center: For each data point, calculate its distance to each cluster center and assign it to the nearest cluster. Here, Euclidean distance is used as the distance metric: Among them, d(xi,cj) is the distance between the data point xi and the cluster center cj, n is the characteristic dimension of the data point, compare the d(xi,cj) of each data point xi, select the smallest d(xi,cj), and the corresponding cluster center cj is the cluster center to which it belongs; (4) Recalculate the cluster center: For each cluster, calculate the mean of all points in the cluster and use this mean as the new cluster center: Where Nj is the number of data points in cluster j, s j is the new cluster center obtained after calculation, xi is the data point, and cj is the cluster center to which xi belongs, that is, the cluster center; (5) Repeat steps 3 and 4 until the stopping condition is met, that is, the cluster center no longer changes significantly or the preset number of iterations is reached; (6) Evaluate clustering effect: Use the silhouette coefficient indicator to evaluate the clustering effect, or evaluate according to the application scenario.

4. The method for fine identification of industrial load model parameters according to claim 3 is characterized in that: Typical industrial load models include: steel production load, cement production load and electrolytic aluminum load; The steel production load model is as follows: Psz.it=Ps.deal.i.t+Ps.eaf.it Among them, i represents the i-th steel production enterprise; t represents time t, Pzs.it represents the total load of steel production by steel production enterprise i at time t; Ps.deal.it represents the power consumption of steel rolling; Ps.eaf.it represents the power consumption of electric arc furnace; Ps.deal.it calculates this way: Among them, k is the number of steel tires on the steel rolling production line per unit time; as.i is the quadratic coefficient of the production consumption coefficient, bs.i is the linear coefficient of the production consumption coefficient; Psteel tire is the unit power consumed in the production of a single steel tire; Psk.deal.it is the fixed energy consumption of the steel rolling production line in the hot start state; Ps.eaf.it is calculated as follows: Where Ps.eafs.it is the power output of the arc furnace during stable operation, ts is the start time of the arc furnace, and te is the stop time of the arc furnace; The cement production load model is as follows: Among them, Pconz.it is the motor power, Q is the raw material input, E is the efficiency of the crushing equipment, η is the motor efficiency, V is the voltage, and cosφ is the power factor; The electrolytic aluminum production load model is as follows: MAL.i.t ＝ [λa1+λa2(λa3+λa4PAL.i.t ) 0.5]Δt Among them, PAL.it is the electrolytic cell power, MAL.it is the electrolytic cell output, λ a1 ,λa2 and λa3,λ a4 is the coefficient of relationship between electrolytic cell output and power.

5. The method for fine identification of industrial load model parameters according to claim 4 is characterized in that: Improved tuna school optimization algorithm process: (1) Initialization: Randomly initialize the first generation population position, and each tuna represents a candidate solution; xint=rand·(ub-lb)+lb,i=1,2,…,NP Among them, xint is the first generation population; ran is a random number; ub and lb are the upper and lower limits of variable changes; NP is the population size; (2) Enter the iteration phase and then use the adaptive mutation strategy to enhance the algorithm performance; Where p is the pth iteration, pmax is the maximum number of iterations, xvari.p is the mutated individual, xi.p is the i-th individual in the population at the pth iteration, xbest.p and xrand.p are the individual with the best fitness and the randomly generated individual at the pth iteration respectively; res is the scaling factor, which is calculated as follows: (3) Adaptive crossover strategy; Among them, xi.pj, xrand.pj and xbest.pj represent the j-th decision variables of the crossover individual, random individual and optimal individual during iteration; (4) Calculate the fitness of the new population after mutation and crossover, and determine the leader fish: Calculate the fitness of each individual and select the tuna with the highest fitness as the leader fish, which is the optimal solution; (5) Update position: There are three ways to update during the iteration process: Method 1: Random update When entering the random update, the update formula is the same as the update method initialized in step (1); Method 2: Spiral Update There are two strategies for spiral foraging updates: The first one: Second type: Among them, xi.p+1 and xi.p represent the i-th individual in the population at the p+1th iteration and the pth iteration respectively; xbest.p represents the individual with the best fitness in the population at the pth iteration; xrand.p represents a randomly generated individual, which is calculated as follows: x rand.p =rand·(ub-lb)+lb α1, α2, and β are calculated as follows: β =el·cos(2 π ) Where a is a constant; Different strategies are selected based on random numbers: Strategy 1 Strategy 2 Method 3: Parabolic foraging: The update formula of parabolic foraging is as follows: Among them, TF is a random number with a value of -1 or 1; the value of w is as follows: The mechanism for selecting the three update methods is shown in the following pseudo code: (6) Consider boundary conditions: check whether the updated tuna position exceeds the predetermined search space boundary, and if so, adjust it back to the boundary; (7) Repeat steps (2)-(6) until the maximum number of iterations is reached or other stopping conditions are met; (8) Output the optimal solution: When the maximum number of iterations is reached or the iteration accuracy meets the requirements, the iteration is completed and the optimal fitness is output.

Images