Freezing plant load day-ahead regulation capability intelligent assessment method considering multiple factors

Abstract

The invention discloses a freezing plant load day-ahead adjustment capability intelligent evaluation method considering multiple factors. The method comprises the following steps: firstly, proposing a day-ahead load prediction model based on MoE; according to the model, different influence factors are modeled through a multi-expert mechanism, expert output fusion is achieved through a gating network, and nonlinear response characteristics of compressor loads to external variables are accurately captured. Secondly, in order to improve the self-adaptive modeling capability of a condenser fan to environment change, a day-ahead load prediction model based on AdaMix is provided; according to the method, a sparse activation mechanism is introduced on the basis of a MoE framework, key experts are dynamically selected and weighted fusion is carried out, and the robustness to extreme working conditions such as high temperature and high humidity is enhanced. Thirdly, in consideration of a complex reasoning path between evaporator fan operation and cold storage microenvironment change, a ToT-based day-ahead load prediction model is provided, and implicit influence characteristics of temperature, humidity and operation frequency are gradually extracted through a multi-level thinking reasoning process, so that the prediction fineness and generalization ability are improved.

Description

Technical Field

[0001] The present invention belongs to the field of power systems, and in particular relates to an intelligent evaluation method for the day-ahead load regulation capability of a refrigeration plant taking multiple factors into consideration. Background Art

[0002] Assessing the day-ahead scalability of refrigeration plant loads helps identify their potential for grid peak regulation and load response, providing a reliable basis for day-ahead dispatch in the power market. By clarifying their ability to adjust upwards and downwards during the day-ahead period, it is possible to effectively balance supply and demand forecast errors, improve system operational flexibility and economic efficiency, and facilitate the precise utilization of refrigeration plant load resources in scenarios such as low-carbon dispatch and green power consumption. However, existing methods for assessing the day-ahead scalability of refrigeration plants still face the following key limitations in practical applications, limiting their effectiveness in exploring flexible loads, grid dispatch response, and green power consumption. 1) Load forecasting accuracy is insufficient and cannot adapt to diverse operating conditions. Current mainstream methods often use simple linear regression, support vector regression, or shallow neural networks to model refrigeration plant loads. These methods ignore the nonlinear effects of factors such as ambient temperature, humidity, operating frequency, and cargo inflow and outflow on cooling loads. This significantly increases prediction errors when faced with sudden load changes or extreme weather conditions, causing the scalability assessment to deviate from the actual operating potential. 2) The modeling capabilities for multi-device coupling are weak, lacking a structured prediction mechanism. There is a clear physical and thermal chain between the refrigeration compressor, condenser fan, and evaporator fan in a refrigeration plant. However, existing models generally treat the equipment as independent entities and fail to capture the energy efficiency linkage characteristics of the "compression-heat dissipation-evaporation" link, affecting the overall consistency and physical rationality of load forecasting and regulation capacity assessment. 3) Regulation capacity assessment methods are mostly rule-driven or static models, which are difficult to dynamically respond to changes in operating conditions on the previous day. Some existing assessment strategies simply estimate the adjustable space based on the equipment's rated power range or operating range, without considering dynamic factors such as actual load evolution trends, regulation response rate, and continuous equipment operation time. This results in overestimation or underestimation of the adjustable capacity, affecting the grid's accurate scheduling of refrigeration plants as flexible load resources. Summary of the Invention

[0003] Purpose of the Invention: This invention aims to provide a multi-factor intelligent assessment method for the day-ahead load regulation capability of a refrigeration plant. By incorporating advanced frameworks such as MoE, AdaMix, and ToT, this method accurately models the nonlinear load characteristics of the refrigeration compressor, condenser fan, and evaporator fan under different operating conditions. Furthermore, combined with the DPO model, the prediction results from each device are systematically integrated to quantify the overall refrigeration plant's regulation capability, providing refined support for low-carbon grid scheduling and market responsiveness.

[0004] Technical solution: The present invention provides an intelligent evaluation method for the load regulation capability of a refrigeration plant taking into account multiple factors, comprising the following steps:

[0005] Step 1: Build a day-ahead load forecasting model for refrigeration compressors in refrigeration plants based on a hybrid expert model (MoE). Multiple expert sub-models are used to model load sensitivity to temperature, humidity, and operating frequency. Ambient temperature, humidity, and operating frequency are used as unified inputs, and weights are assigned to each expert via a gating network. Each expert independently predicts the sensitivity coefficient. The gating weights are then used to weightedly fuse the outputs of all experts to generate the final temperature, humidity, and operating frequency sensitivity coefficients, resulting in the day-ahead load of the refrigeration compressor.

[0006] Step 2: Build a day-ahead load forecasting model for condenser fans in refrigeration plants based on the hybrid adaptive enhancement algorithm AdaMix. A dynamic sparse activation mechanism is introduced into the hybrid expert model (MoE) framework. Temperature, humidity, and operating frequency are used as inputs. The AdaMix gating network adaptively selects a small number of key experts for activation based on the input features and outputs a sparse weight distribution. The activated experts each predict the corresponding sensitivity coefficients, and finally, a sparse weighted fusion is used to obtain the prediction results, which are used to determine the day-ahead load of the condenser fans.

[0007] Step 3: Build a day-ahead load forecasting model for the evaporator fan in the refrigeration plant based on the ToT (Thinking Tree) model. Through a multi-level reasoning process, the sensitivity assessment of temperature, humidity, and operating frequency is gradually refined. First, a preliminary thinking vector is generated using ambient temperature, humidity, and operating frequency as input. Then, through the first layer of branching reasoning, a more detailed feature representation is extracted. Further reasoning is then performed at the second layer to form an inference result. Through linear mapping, the inference vector is converted into the sensitivity coefficient of the evaporator fan load to each input factor, resulting in the evaporator fan day-ahead load.

[0008] Step 4: Construct a day-ahead regulation capacity assessment model for the refrigeration plant based on the direct optimization preference (DPO) mechanism. Based on the day-ahead loads of the refrigeration compressor, condenser fan, and evaporator fan, combined with the total load value and the optimal floating ratio, determine the maximum upward and maximum downward regulation capacity ranges of the refrigeration plant on the day-ahead time scale to complete the regulation capacity assessment.

[0009] Furthermore, step 1 is specifically as follows: constructing a day-ahead load assessment model for refrigeration compressors in refrigeration plants that considers multiple factors, as shown in formula (1);

[0010] P c (t)=a1T(t)+a2H(t)+a3F(t)+b1 (1)

[0011] Where: P c(t) is the refrigeration compressor load at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; a1, a2, and a3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the compressor load, respectively; b1 is the bias term, reflecting the base load level;

[0012] A MoE-based prediction method for the sensitivity coefficient of refrigeration compressors is proposed to predict the sensitivity coefficients of temperature, humidity, and operating frequency to the compressor load in formula (1).

[0013] First, define the input features:

[0014]

[0015] Where: x(t) is the three-dimensional input feature column vector, T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operation frequency at time t;

[0016] The expert weights are output through the gating network. The weight distribution of each expert is obtained through the gating network G(·), as shown in formula (3).

[0017]

[0018] Where: w(t) is the weight at time t; G(·) is the gated network, usually using Softmax output; K is the number of experts; w K (t) is the weight of the kth expert at time t, satisfying

[0019] Calculate the sensitivity coefficient of each expert's prediction; each expert E k (·) Output a set of sensitivity coefficient prediction values ​​based on the input feature x(t), as shown in formula (4);

[0020]

[0021] Where: E k (·) is the kth expert, a k (t) is the sensitivity coefficient vector predicted by the kth expert, a 1k (t) is the temperature sensitivity coefficient predicted by the kth expert, a 2k (t) is the humidity sensitivity coefficient predicted by the kth expert, a 3k (t) is the sensitivity coefficient of the operation frequency predicted by the kth expert;

[0022] The outputs of all experts are combined to obtain the final sensitivity coefficient prediction; the outputs of all experts are weighted and summed to obtain the final sensitivity coefficient prediction result, as shown in formula (5);

[0023]

[0024] Expand formula (5) as shown in formula (6);

[0025]

[0026] Among them, the calculation formulas of a1(t), a2(t), and a3(t) are shown in formula (7);

[0027]

[0028] Where a(t) is the final integrated sensitivity coefficient vector; a1(t) is the final temperature sensitivity coefficient; a2(t) is the final humidity sensitivity coefficient; a3(t) is the final operation frequency sensitivity coefficient; w K (t) is the weight of the kth expert at time t, satisfying

[0029] Substitute the predicted sensitivity coefficient into formula (1) to obtain the final load forecast, as shown in formula (8);

[0030]

[0031] Where: is the refrigeration compressor load predicted at time t (kW); a k (t) is the sensitivity coefficient vector predicted by the kth expert; a1(t) is the final temperature sensitivity coefficient; a2(t) is the final humidity sensitivity coefficient; a3(t) is the final operation frequency sensitivity coefficient; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operation frequency at time t (times / h); b1 is the basic load.

[0032] Furthermore, step 2 is specifically as follows: constructing a day-ahead load assessment model for condenser fans in refrigeration plants that considers multiple factors, as shown in formula (9);

[0033] P cond (t)=c1T(t)+c2H(t)+c3F(t)+b2 (9)

[0034] Where: P cond (t) is the condenser fan load at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; c1, c2, c3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the condenser load respectively; b2 is the condenser fan load;

[0035] A refrigeration compressor sensitivity coefficient prediction method based on AdaMix is ​​proposed to predict the sensitivity coefficient of temperature, humidity and operating frequency to the condenser fan in formula (9);

[0036] Define the input feature vector; package temperature, humidity, and operation frequency into an input vector, as shown in formula (10);

[0037]

[0038] Where: x(t) is the environmental input vector at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t;

[0039] AdaMix algorithm sparse dynamic gating output; through the adaptive gating network G Ada (·), outputs the sparse expert activation vector, as shown in formula (11);

[0040]

[0041] Where: γ(t) is the sparse gating weight vector; G Ada (·) is the adaptive gating network; γ k (t) is the kth expert weight, and to satisfy sparsity, there are only a few γ k (t)>0;

[0042] Each expert sub-network outputs a local sensitivity coefficient; the kth expert sub-network E k (·) According to the input and output local sensitivity coefficients, formula (12) is obtained;

[0043]

[0044] Where: c k (t) is the local sensitivity vector output by the k-th expert; E k (·) is the kth expert sub-network; c 1k (t) is the local sensitivity corresponding to temperature, c 2k (t) is the local sensitivity corresponding to humidity, c 3k (t) is the local sensitivity corresponding to the operating frequency;

[0045] The AdaMix algorithm sparsely weights the expert outputs; the activated expert outputs are sparsely weighted to obtain the final predicted sensitivity coefficient vector c(t), as shown in formula (13);

[0046]

[0047] Where: c(t) is the sensitivity coefficient vector of the final prediction; is the set of experts activated at time t; c k (t) is the local sensitivity vector output by the k-th expert;

[0048] Among them, the calculation formulas of c1(t), c2(t), and c3(t) are shown in formula (14);

[0049]

[0050] Substitute the load formula to complete the condenser fan load prediction; substitute the final predicted sensitivity coefficients c1(t), c2(t), c3(t) into formula (9) to obtain formula (15);

[0051]

[0052] Where: is the predicted condenser fan load at time t; c 1k (t) is the local sensitivity corresponding to temperature, c 2k (t) is the local sensitivity corresponding to humidity, c 3k (t) is the local sensitivity corresponding to the operating frequency; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; b2 is the condenser fan load.

[0053] Furthermore, step 3 is specifically as follows: constructing a day-ahead load assessment model for the evaporator fan of a refrigeration plant considering multiple factors, as shown in formula (16);

[0054] P evap (t)=d1T(t)+d2H(t)+d3F(t)+b3 (16)

[0055] Where: P evap (t) is the evaporator fan load at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; d1, d2, d3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the evaporator fan load, respectively; b3 is the basic evaporator fan load;

[0056] A prediction method for the sensitivity coefficient of evaporator fan based on ToT is proposed to predict the sensitivity coefficient of temperature, humidity and operating frequency to the evaporator fan load in formula (16);

[0057] Initial input feature processing: First, temperature, humidity, and operation frequency are input as root nodes to generate a preliminary thought vector:

[0058] z (0) (t) = f root (T(t),H(t),F(t)) (17)

[0059] Where z (0) (t) is the three-dimensional preliminary reasoning vector, f root (·) is the root node feature extraction function; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operation frequency at time t;

[0060] First level reasoning; based on the initial thinking vector z (0) (t), generates the first-level thinking branch, as shown in formula (18);

[0061]

[0062] Where z (1) (t) is the first-layer reasoning vector, f branch1 (·) is the first-level inference function, The characteristic representations of the factors after the first-level reasoning are the ambient temperature T(t), relative humidity H(t), and operating frequency F(t) at time t;

[0063] Second-level reasoning: Based on the first-level reasoning results, further reasoning is performed to obtain the final refined sensitivity vector, as shown in formula (19);

[0064]

[0065] Where z (1) (t) is the first-level reasoning vector, z (2) (t) is the second-level reasoning vector, f branch2 (·) is the second-level inference function, The characteristic representations of the factors after the first-level reasoning are the ambient temperature T(t), relative humidity H(t), and operating frequency F(t) at time t;

[0066] The final sensitivity coefficient is generated; the final inference result is mapped to the sensitivity coefficient through linear mapping, as shown in formula (20):

[0067]

[0068] Where: d1(t), d2(t), d3(t) are the sensitivity coefficients of temperature, humidity, and operating frequency to the evaporator fan load at time t; W out is the output weight matrix, b out is the output bias vector;

[0069] Substitute the load formula for prediction; substitute the predicted sensitivity coefficients d1(t), d2(t), d3(t) into formula (16) to obtain formula (21);

[0070]

[0071] Where: is the predicted evaporator fan load at time t; d1(t), d2(t), d3(t) are the sensitivity coefficients of temperature, humidity, and operating frequency at time t to the evaporator fan load respectively; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; b3 is the basic evaporator fan load.

[0072] Furthermore, step 4 is specifically as follows: a DPO-based refrigeration plant day-ahead regulation capacity assessment model is proposed. Based on the predicted loads of the refrigeration compressor, condenser fan, and evaporator fan of the refrigeration plant, the DPO algorithm is used to consider the influence of temperature, humidity, and operation frequency to predict the regulation capacity of the refrigeration plant on the day-ahead time scale;

[0073] Total load estimation: First, the load forecast of the refrigeration compressor, condenser fan, and evaporator fan is integrated to obtain the total load of the refrigeration plant, as shown in formula (22);

[0074]

[0075] Where: is the total load; is the refrigeration compressor load predicted at time t; is the predicted condenser fan load at time t; is the evaporator fan load predicted at time t;

[0076] Generate a candidate set of adjustment capabilities based on the total load; By floating up and down the proportion, multiple candidate intervals of adjustment capacity are generated, as shown in formula (23);

[0077]

[0078] Where: is the reduction capacity ratio coefficient of the i-th candidate, satisfying is the proportional coefficient of the upward adjustment capability of the i-th candidate, satisfying is the total load;

[0079] Calculate the preference score based on the total load and regulation capacity interval; for each set of candidate regulation intervals, calculate its preference score s (i) (t), as shown in formula (24);

[0080]

[0081] Where: f pref(·) is the preferred partition function, which can evaluate the pros and cons by combining the total load and the up and down floating ratio; s (i) (t) is the preference score of the i-th candidate interval; is the reduction capacity ratio coefficient of the i-th candidate; is the proportional coefficient of the upward adjustment capability of the i-th candidate; is the total load;

[0082] Select the regulatory ability interval with the highest preference score; select the highest score from all candidate intervals as the final prediction result, as shown in formula (25);

[0083]

[0084] Where: is the final selected reduction ratio, is the final selected increase ratio, s (i) (t) is the preference score of the i-th candidate interval;

[0085] Finally, the adjustment capacity range of the refrigeration plant is predicted; based on the selected optimal ratio, the specific power value is reversed, as shown in formula (26);

[0086]

[0087] Where: is the forecast range of the regulation capacity of the refrigeration plant on the day-ahead time scale; is the reduction capacity ratio coefficient of the i-th candidate; is the proportional coefficient of the upward adjustment capability of the i-th candidate; is the total load.

[0088] The present invention further discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0089] The present invention further discloses a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the method of the present invention when the computer program / instruction is executed by a processor.

[0090] The present invention further discloses a computer program product, comprising a computer program / instruction, which implements the steps of the method of the present invention when executed by a processor.

[0091] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0092] This paper introduces advanced architectures such as MoE, AdaMix and ToT to accurately model the nonlinear load characteristics of refrigeration compressors, condenser fans and evaporator fans under different operating conditions. It further combines the DPO model to systematically integrate the prediction results of various equipment, quantify the overall adjustable capacity of the refrigeration plant, and provide refined support for low-carbon scheduling and market response of the power grid.

[0093] This paper proposes a day-ahead load forecasting model for refrigeration compressors in cold storage plants based on MoE (Mixture of Experts), which adapts to nonlinear characteristic distributions under various operating conditions and improves forecast stability. It also proposes an AdaMix-based day-ahead load forecasting model for condenser fans, which adaptively models external environmental conditions and equipment status. It also proposes a day-ahead load forecasting model for evaporator fans based on ToT (Token-to-Token Attention), which captures the complex relationship between the cold storage microenvironment and evaporative load at a fine-grained level. Finally, it proposes a day-ahead adjustment capacity assessment model for cold storage plants based on DPO (Direct Preference Optimization). This model integrates load forecast results for each device, constructs multiple candidate adjustment capacity intervals, and introduces a preference function for comprehensive evaluation to ultimately determine the optimal range for adjusting capacity up or down. This method comprehensively integrates environmental characteristics, load forecasting, and optimization decision-making mechanisms to achieve high-precision assessment of the day-ahead adjustable capacity of cold storage plants, providing strong support for grid scheduling optimization and low-carbon demand response. It systematically assesses the adjustable capacity of cold storage plants on a day-ahead timescale, supporting flexible grid scheduling and market transactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0095] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0096] like Figure 1 As shown, the present invention proposes an intelligent evaluation method for the day-ahead load regulation capability of a refrigeration plant taking into account multiple factors. The method comprises the following steps:

[0097] (1) A MoE-based day-ahead load prediction model for refrigeration compressors in refrigeration plants is proposed. The present invention first constructs a day-ahead load evaluation model for refrigeration compressors in refrigeration plants that takes multiple factors into consideration. The load of the refrigeration compressor is mainly affected by ambient temperature, air humidity, and operating frequency. As the ambient temperature rises, the system requires stronger refrigeration capacity to maintain the target temperature, resulting in an increase in the compressor load; the higher the air humidity, the greater the latent heat load in the air, which also increases the workload of the compressor; the operating frequency reflects the changes in the refrigeration load. For example, an increase in processing volume or frequent inbound and outbound storage will also lead to an increase in the compressor start-up frequency and load. Furthermore, a MoE (Mixture of Experts)-based refrigeration compressor sensitivity coefficient prediction method is proposed, which models the load sensitivity of temperature, humidity, and operating frequency through multiple expert sub-models. First, the ambient temperature, humidity, and operating frequency are used as unified inputs, and weights are assigned to each expert through a gating network; each expert independently predicts the sensitivity coefficient; then, the gating weights are used to weightedly fuse the outputs of all experts to generate the final temperature, humidity, and operating frequency sensitivity coefficients.

[0098] (2) A day-ahead load prediction model for condenser fans in refrigeration plants based on AdaMix is ​​proposed. This paper constructs a day-ahead load evaluation model for condenser fans in refrigeration plants that takes multiple factors into consideration. The load of condenser fans is mainly affected by the combined effects of ambient temperature, air humidity, and operating frequency. When the outside temperature rises, the condenser needs to dissipate heat more efficiently to maintain the refrigerant condensation effect, so the fan speed increases and the load increases; in a high humidity environment, the air heat exchange efficiency decreases, and the condenser needs to work for a longer time or at a higher wind speed to maintain the condensation pressure; an increase in operating frequency means an increase in the overall load of the refrigeration system, and accordingly, the load of the condenser fan also increases synchronously. Furthermore, a refrigeration compressor sensitivity coefficient prediction method based on AdaMix (Adaptive Mixture of Experts) is proposed, and a dynamic sparse activation mechanism is introduced on the traditional MoE framework. With temperature, humidity, and operating frequency as input, the AdaMix gating network adaptively selects a few key experts for activation based on the input features and outputs a sparse weight distribution; the activated experts predict the corresponding sensitivity coefficients respectively, and finally the final prediction result is obtained through sparse weighted fusion.

[0099] (3) A day-ahead load prediction model for evaporator fans in refrigeration plants based on ToT is proposed. This paper first constructs a day-ahead load assessment model for evaporator fans in refrigeration plants that takes multiple factors into consideration. The load of evaporator fans is mainly affected by ambient temperature, air humidity, and operating frequency. As the ambient temperature rises, the heat input inside the cold storage increases, and the evaporator fan needs to increase the air circulation rate to enhance the transfer of cold energy, thereby increasing the load. In a high-humidity environment, the amount of water vapor in the air increases, and the evaporator needs to bear the additional dehumidification load, and the fan operation time and intensity also increase accordingly. The increase in operating frequency, such as frequent door opening and closing and frequent personnel operations, will lead to the loss of cold energy, and the evaporator fan needs to operate more frequently to replenish the cold energy. Furthermore, a ToT (Tree of Thoughts)-based evaporator fan sensitivity coefficient prediction method is proposed, which gradually refines the sensitivity assessment of temperature, humidity, and operating frequency through a multi-level reasoning process. First, a preliminary thought vector is generated with ambient temperature, humidity, and operating frequency as input; then, a more detailed feature representation is extracted through the first-level branch reasoning; then, further reasoning is performed at the second level to form the final reasoning result. Finally, the inference vector is converted into the sensitivity coefficient of the evaporator fan load to each input factor through linear mapping. This method embodies the characteristics of layer-by-layer progressive and tree-branching reasoning, and can capture the complex nonlinear dependencies between different input features.

[0100] (4) A DPO-based day-ahead regulation capacity assessment model for refrigeration plants is proposed. First, temperature, humidity, and operating frequency are used as input features. The loads of key equipment in the refrigeration plant are predicted separately through the pre-model, including the refrigeration compressor load, condenser fan load, and evaporator fan load. Subsequently, the loads of each equipment are summed up to obtain the total load forecast of the refrigeration plant on the day-ahead time scale. Based on the total load forecast results, a series of upper and lower floating ratios are set to generate multiple candidate regulation capacity intervals. Each candidate interval is calculated by multiplying the total load by the corresponding downward adjustment ratio and upward adjustment ratio, forming a complete set of regulation capacity candidates. In order to evaluate the pros and cons of each candidate interval, the preference function is further used to calculate the score of each candidate regulation interval based on the current total load level and floating ratio, comprehensively considering the operating characteristics and regulation potential of the refrigeration plant equipment. Through the direct optimization preference (DPO) mechanism, the candidate interval with the highest score is selected as the final regulation capacity prediction result. Finally, the maximum upward adjustment capacity and maximum downward adjustment capacity interval of the refrigeration plant on the day-ahead time scale are determined by combining the total load value and the optimal floating ratio, completing the regulation capacity assessment.

[0101] For step (1), a MoE-based day-ahead load prediction model for refrigeration compressors in refrigeration plants is proposed. First, a day-ahead load evaluation model for refrigeration compressors in refrigeration plants is constructed, which considers multiple factors, as shown in formula (1).

[0102] P c(t)=a1T(t)+a2H(t)+a3F(t)+b1 (1)

[0103] Where: P c (t) is the refrigeration compressor load at time t (kW); T(t) is the ambient temperature at time t (°C); H(t) is the relative humidity at time t (%); F(t) is the operating frequency at time t (times / h); a1, a2, and a3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the compressor load, respectively; b1 is the bias term, which can reflect the basic load level.

[0104] Then, a MoE-based prediction method for the sensitivity coefficient of refrigeration compressors is proposed to predict the sensitivity coefficients of temperature, humidity, and operating frequency to the compressor load in formula (1).

[0105] 1) Define input features

[0106]

[0107] Where x(t) is the three-dimensional input feature column vector, T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operation frequency at time t (times / h).

[0108] 2) Gating Network outputs expert weights

[0109] Through the gating network G(·), the weight distribution of each expert is obtained, as shown in formula (3).

[0110]

[0111] Where: w(t) is the weight at time t; G(·) is the gated network, usually using Softmax output; K is the number of experts; w K (t) is the weight of the kth expert at time t, satisfying

[0112] 3) Sensitivity coefficient of each expert prediction

[0113] Each expert E k (·) Outputs a set of sensitivity coefficient prediction values ​​based on the input feature x(t), as shown in formula (4).

[0114]

[0115] Where: E k (·) is the kth expert, a k (t) is the sensitivity coefficient vector predicted by the kth expert, a 1k (t) is the temperature sensitivity coefficient predicted by the kth expert, a2k (t) is the humidity sensitivity coefficient predicted by the kth expert, a 3k (t) is the sensitivity coefficient of the operation frequency predicted by the kth expert.

[0116] 4) Combining the outputs of various experts to obtain the final sensitivity coefficient prediction

[0117] The outputs of all experts are weighted and summed to obtain the final sensitivity coefficient prediction result, as shown in formula (5).

[0118]

[0119] Expand formula (5) to get formula (6).

[0120]

[0121] The calculation formulas of a1(t), a2(t), and a3(t) are shown in formula (7).

[0122]

[0123] Where a(t) is the final integrated sensitivity coefficient vector; a1(t) is the final temperature sensitivity coefficient; a2(t) is the final humidity sensitivity coefficient; a3(t) is the final operation frequency sensitivity coefficient; w K (t) is the weight of the kth expert at time t, satisfying

[0124] Substituting the predicted sensitivity coefficient into formula (1), the final load forecast is obtained, as shown in formula (8).

[0125]

[0126] Where: is the refrigeration compressor load predicted at time t (kW); a k (t) is the sensitivity coefficient vector predicted by the kth expert; a1(t) is the final temperature sensitivity coefficient; a2(t) is the final humidity sensitivity coefficient; a3(t) is the final operation frequency sensitivity coefficient; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operation frequency at time t (times / h); b1 is the basic load.

[0127] For step (2), a day-ahead load forecasting model for condenser fans in refrigeration plants based on AdaMix is ​​proposed. First, a day-ahead load assessment model for condenser fans in refrigeration plants is constructed, which considers multiple factors, as shown in formula (9).

[0128] P cond(t)=c1T(t)+c2H(t)+c3F(t)+b2 (9)

[0129] Where: P cond (t) is the condenser fan load at time t (kW); T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operating frequency at time t (times / h); c1, c2, and c3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the condenser load, respectively; b2 is the condenser fan load.

[0130] Then, a refrigeration compressor sensitivity coefficient prediction method based on AdaMix is ​​proposed to predict the sensitivity coefficient of temperature, humidity, and operating frequency to the condenser fan in formula (9).

[0131] 1) Define the input feature vector

[0132] Package temperature, humidity, and operation frequency into an input vector, as shown in formula (10).

[0133]

[0134] Where: x(t) is the environmental input vector at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operation frequency at time t (times / h).

[0135] 2) AdaMix algorithm sparse dynamic gating output

[0136] Through the adaptive gating network G Ada (·), and outputs the sparse expert activation vector, as shown in formula (11).

[0137]

[0138] Where: γ(t) is the sparse gating weight vector; G Ada (·) is the adaptive gating network; γ k (t) is the kth expert weight, and in order to meet the sparsity, there are only a few γ k (t)>0 (such as Top-2 activation).

[0139] 3) Each expert sub-network outputs a local sensitivity coefficient

[0140] The kth expert sub-network E k (·) According to the input and output local sensitivity coefficients, formula (12) is obtained.

[0141]

[0142] Where: c k(t) is the local sensitivity vector output by the k-th expert; E k (·) is the kth expert sub-network; c 1k (t) is the local sensitivity corresponding to temperature, c 2k (t) is the local sensitivity corresponding to humidity, c 3k (t) is the local sensitivity corresponding to the operating frequency.

[0143] 4) AdaMix algorithm sparse weighted combination expert output (dynamic combination)

[0144] The activated expert outputs are sparsely weighted to obtain the final predicted sensitivity coefficient vector c(t), as shown in formula (13).

[0145]

[0146] Where: c(t) is the sensitivity coefficient vector of the final prediction; is the set of experts activated at time t (Top-K experts); c k (t) is the local sensitivity vector output by the kth expert.

[0147] The calculation formulas for c1(t), c2(t), and c3(t) are shown in formula (14).

[0148]

[0149] 5) Substitute into the load formula to complete the condenser fan load prediction

[0150] Substitute the final predicted sensitivity coefficients c1(t), c2(t), and c3(t) into formula (9) to obtain formula (15).

[0151]

[0152] Where: is the predicted condenser fan load at time t (kW); c 1k (t) is the local sensitivity corresponding to temperature, c 2k (t) is the local sensitivity corresponding to humidity, c 3k (t) is the local sensitivity corresponding to the operating frequency; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operating frequency at time t (times / h); b2 is the condenser fan load.

[0153] For step (3), a day-ahead load prediction model for evaporator fans in refrigeration plants based on ToT is proposed. First, a day-ahead load evaluation model for evaporator fans in refrigeration plants is constructed considering multiple factors, as shown in formula (16).

[0154] P evap (t)=d1T(t)+d2H(t)+d3F(t)+b3 (16)

[0155] Where: P evap (t) is the evaporator fan load at time t (kW); T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operating frequency at time t (times / h); d1, d2, and d3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the evaporator fan load, respectively; b3 is the basic evaporator fan load.

[0156] Then, a prediction method for the evaporator fan sensitivity coefficient based on ToT is proposed to predict the sensitivity coefficient of temperature, humidity and operating frequency to the evaporator fan load in formula (16).

[0157] 1) Initial input feature processing

[0158] First, we input temperature, humidity, and operating frequency as root nodes to generate a preliminary thinking vector:

[0159] z (0) (t) = f root (T(t),H(t),F(t)) (17)

[0160] Where z (0) (t) is the three-dimensional preliminary reasoning vector (root node), f root (·) is the root node feature extraction function (which can be an MLP, linear layer, etc.); T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operation frequency at time t (times / h).

[0161] 2) First level reasoning (branch expansion)

[0162] Based on the initial thinking vector z (0) (t), generates the first level of thinking branches (refinement), as shown in formula (18).

[0163]

[0164] Where z (1) (t) is the first-level reasoning vector (first branch), f branch1 (·) is the first layer inference function (can be different sub-networks), The characteristic representation of each factor after the first-level reasoning are the ambient temperature T(t), relative humidity (%) H(t), and operation frequency (times / h) F(t) at time t.

[0165] 3) Second level reasoning (further refinement)

[0166] Based on the first-level reasoning results, further reasoning is performed to obtain the final refined sensitivity vector, as shown in formula (19).

[0167]

[0168] Where z (1) (t) is the first-level reasoning vector, z (2) (t) is the second-level reasoning vector, f branch2 (·) is the second-level inference function, The characteristic representation of each factor after the first-level reasoning are the ambient temperature T(t), relative humidity (%) H(t), and operation frequency (times / h) F(t) at time t.

[0169] 4) Final sensitivity coefficient generation

[0170] The final inference result is mapped to the sensitivity coefficient through linear mapping, as shown in formula (20):

[0171]

[0172] Where: d1(t), d2(t), d3(t) are the sensitivity coefficients of temperature, humidity, and operating frequency to the evaporator fan load at time t; W out is the output weight matrix (3×3), b out is the output bias vector (3×1).

[0173] 5) Substitute into the load formula for prediction

[0174] Substitute the predicted sensitivity coefficients d1(t), d2(t), and d3(t) into formula (16) to obtain formula (21).

[0175]

[0176] Where: is the predicted evaporator fan load at time t (kW); d1(t), d2(t), d3(t) are the sensitivity coefficients of temperature, humidity, and operating frequency at time t to the evaporator fan load, respectively; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t (%); F(t) is the operating frequency at time t (times / h); b3 is the basic evaporator fan load.

[0177] For step (4), a DPO-based refrigeration plant day-ahead regulation capacity assessment model is proposed. Based on the previously predicted loads of the refrigeration compressor, condenser fan, and evaporator fan of the refrigeration plant, and based on the DPO algorithm, the regulation capacity of the refrigeration plant on the day-ahead time scale is predicted, taking into account the influence of temperature, humidity, and operation frequency.

[0178] 1) Total load estimation

[0179] First, the load forecast of the refrigeration compressor, condenser fan, and evaporator fan is integrated to obtain the total load of the refrigeration plant, as shown in formula (22).

[0180]

[0181] Where: is the total load; is the refrigeration compressor load predicted at time t; is the predicted condenser fan load at time t; is the evaporator fan load predicted at time t

[0182] 2) Generate a candidate set of regulation capabilities based on the total load

[0183] According to the total load By floating up and down in a certain proportion, multiple candidate intervals of adjustment capacity are generated, as shown in formula (23).

[0184]

[0185] Where: is the reduction capacity ratio coefficient of the i-th candidate, satisfying is the proportional coefficient of the upward adjustment capability of the i-th candidate, satisfying is the total load.

[0186] 3) Preference score calculation (based on total load and regulation capacity range)

[0187] For each set of candidate adjustment intervals, calculate its preference score s (i) (t), as shown in formula (24).

[0188]

[0189] Where: f pref (·) is the preferred partition function, which can evaluate the pros and cons by combining the total load and the up and down floating ratio; s (i) (t) is the preference score of the i-th candidate interval; is the reduction capacity ratio coefficient of the i-th candidate; is the proportional coefficient of the upward adjustment capability of the i-th candidate; is the total load.

[0190] 4) Select the regulatory ability interval with the highest preference score

[0191] The one with the highest score among all candidate intervals is selected as the final prediction result, as shown in formula (25).

[0192]

[0193] Where: is the final selected reduction ratio, is the final selected increase ratio, s (i) (t) is the preference score of the i-th candidate interval.

[0194] 5) Final prediction of the refrigeration plant's adjustment capacity range

[0195] According to the selected optimal ratio, the specific power value is inversely deduced, as shown in formula (26).

[0196]

[0197] Where: is the forecast range of the regulation capacity of the refrigeration plant on the day-ahead time scale; is the reduction capacity ratio coefficient of the i-th candidate; is the proportional coefficient of the upward adjustment capability of the i-th candidate; is the total load.

Claims

1. An intelligent evaluation method for the load regulation capability of a refrigeration plant considering multiple factors, characterized in that: The steps include: Step 1: Build a day-ahead load forecasting model for refrigeration compressors in refrigeration plants based on a hybrid expert model (MoE). Multiple expert sub-models are used to model load sensitivity to temperature, humidity, and operating frequency. Ambient temperature, humidity, and operating frequency are used as unified inputs, and weights are assigned to each expert via a gating network. Each expert independently predicts the sensitivity coefficient. The gating weights are then used to weightedly fuse the outputs of all experts to generate the final temperature, humidity, and operating frequency sensitivity coefficients, resulting in the day-ahead load of the refrigeration compressor. Step 2: Build a day-ahead load forecasting model for condenser fans in refrigeration plants based on the hybrid adaptive enhancement algorithm AdaMix. A dynamic sparse activation mechanism is introduced into the hybrid expert model (MoE) framework. Temperature, humidity, and operating frequency are used as inputs. The AdaMix gating network adaptively selects a small number of key experts for activation based on the input features and outputs a sparse weight distribution. The activated experts each predict the corresponding sensitivity coefficients, and finally, a sparse weighted fusion is used to obtain the prediction results, which are used to determine the day-ahead load of the condenser fans. Step 3: Build a day-ahead load forecasting model for the evaporator fan in the refrigeration plant based on the ToT (Thinking Tree) model. Through a multi-level reasoning process, the sensitivity assessment of temperature, humidity, and operating frequency is gradually refined. First, a preliminary thinking vector is generated using ambient temperature, humidity, and operating frequency as input. Then, through the first layer of branching reasoning, a more detailed feature representation is extracted. Further reasoning is then performed at the second layer to form an inference result. Through linear mapping, the inference vector is converted into the sensitivity coefficient of the evaporator fan load to each input factor, resulting in the evaporator fan day-ahead load. Step 4: Construct a day-ahead regulation capacity assessment model for the refrigeration plant based on the direct optimization preference (DPO) mechanism. Based on the day-ahead loads of the refrigeration compressor, condenser fan, and evaporator fan, combined with the total load value and the optimal floating ratio, determine the maximum upward and maximum downward regulation capacity ranges of the refrigeration plant on the day-ahead time scale to complete the regulation capacity assessment.

2. The intelligent evaluation method for the load regulation capability of a refrigeration plant considering multiple factors according to claim 1 is characterized in that: Step 1 is as follows: construct a day-ahead load assessment model for refrigeration compressors in refrigeration plants that considers multiple factors, as shown in formula (1); P c (t)=a1T(t)+a2H(t)+a3F(t)+b1 (1) Where: P c (t) is the refrigeration compressor load at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; a1, a2, and a3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the compressor load, respectively; b1 is the bias term, reflecting the base load level; A MoE-based prediction method for the sensitivity coefficient of refrigeration compressors is proposed to predict the sensitivity coefficients of temperature, humidity, and operating frequency to the compressor load in formula (1). First, define the input features: Where: x(t) is the three-dimensional input feature column vector, T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operation frequency at time t; The expert weights are output through the gating network. The weight distribution of each expert is obtained through the gating network G(·), as shown in formula (3). Where: w(t) is the weight at time t; G(·) is the gated network, usually using Softmax output; K is the number of experts; w K (t) is the weight of the kth expert at time t, satisfying Calculate the sensitivity coefficient of each expert's prediction; each expert E k (·) Output a set of sensitivity coefficient prediction values ​​based on the input feature x(t), as shown in formula (4); Where: E k (·) is the kth expert, a k (t) is the sensitivity coefficient vector predicted by the kth expert, a 1k (t) is the temperature sensitivity coefficient predicted by the kth expert, a 2k (t) is the humidity sensitivity coefficient predicted by the kth expert, a 3k (t) is the sensitivity coefficient of the operation frequency predicted by the kth expert; The outputs of all experts are combined to obtain the final sensitivity coefficient prediction; the outputs of all experts are weighted and summed to obtain the final sensitivity coefficient prediction result, as shown in formula (5); Expand formula (5) as shown in formula (6); Among them, the calculation formulas of a1(t), a2(t), and a3(t) are shown in formula (7); Where a(t) is the final integrated sensitivity coefficient vector; a1(t) is the final temperature sensitivity coefficient; a2(t) is the final humidity sensitivity coefficient; a3(t) is the final operation frequency sensitivity coefficient; w K (t) is the weight of the kth expert at time t, satisfying Substitute the predicted sensitivity coefficient into formula (1) to obtain the final load forecast, as shown in formula (8); Where: is the load of the refrigeration compressor predicted at time t; a k (t) is the sensitivity coefficient vector predicted by the kth expert; a1(t) is the final temperature sensitivity coefficient; a2(t) is the final humidity sensitivity coefficient; a3(t) is the final operation frequency sensitivity coefficient; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operation frequency at time t; b1 is the basic load.

3. The intelligent evaluation method for the load regulation capability of a refrigeration plant taking into account multiple factors according to claim 1 is characterized in that: Step 2 is as follows: construct a day-ahead load assessment model for condenser fans in refrigeration plants that considers multiple factors, as shown in formula (9); P cond (t)=c1T(t)+c2H(t)+c3F(t)+b2 (9) Where: P cond (t) is the condenser fan load at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; c1, c2, c3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the condenser load respectively; b2 is the condenser fan load; A refrigeration compressor sensitivity coefficient prediction method based on AdaMix is ​​proposed to predict the sensitivity coefficient of temperature, humidity and operating frequency to the condenser fan in formula (9); Define the input feature vector; package temperature, humidity, and operation frequency into an input vector, as shown in formula (10); Where: x(t) is the environmental input vector at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; AdaMix algorithm sparse dynamic gating output; through the adaptive gating network G Ada (·), outputs the sparse expert activation vector, as shown in formula (11); Where: γ(t) is the sparse gating weight vector; G Ada (·) is the adaptive gating network; γ k (t) is the kth expert weight, and to satisfy sparsity, there are only a few γ k (t)>0; Each expert sub-network outputs a local sensitivity coefficient; the kth expert sub-network E k (·) According to the input and output local sensitivity coefficients, formula (12) is obtained; Where: c k (t) is the local sensitivity vector output by the k-th expert; E k (·) is the kth expert sub-network; c 1k (t) is the local sensitivity corresponding to temperature, c 2k (t) is the local sensitivity corresponding to humidity, c 3k (t) is the local sensitivity corresponding to the operating frequency; The AdaMix algorithm sparsely weights the expert outputs; the activated expert outputs are sparsely weighted to obtain the final predicted sensitivity coefficient vector c(t), as shown in formula (13); Where: c(t) is the sensitivity coefficient vector of the final prediction; is the set of experts activated at time t; c k (t) is the local sensitivity vector output by the k-th expert; Among them, the calculation formulas of c1(t), c2(t), and c3(t) are shown in formula (14); Substitute the load formula to complete the condenser fan load prediction; substitute the final predicted sensitivity coefficients c1(t), c2(t), c3(t) into formula (9) to obtain formula (15); Where: is the predicted condenser fan load at time t; c 1k (t) is the local sensitivity corresponding to temperature, c 2k (t) is the local sensitivity corresponding to humidity, c 3k (t) is the local sensitivity corresponding to the operating frequency; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; b2 is the condenser fan load.

4. The intelligent evaluation method for the load regulation capability of a refrigeration plant considering multiple factors according to claim 1 is characterized in that: Step 3 is as follows: construct a day-ahead load assessment model for the evaporator fan in the refrigeration plant considering multiple factors, as shown in formula (16); P evap (t)=d1T(t)+d2H(t)+d3F(t)+b3 (16) Where: P evap (t) is the evaporator fan load at time t; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; d1, d2, d3 are the sensitivity coefficients of temperature, humidity, and operating frequency to the evaporator fan load, respectively; b3 is the basic evaporator fan load; A prediction method for the sensitivity coefficient of evaporator fan based on ToT is proposed to predict the sensitivity coefficient of temperature, humidity and operating frequency to the evaporator fan load in formula (16); Initial input feature processing: First, temperature, humidity, and operation frequency are input as root nodes to generate a preliminary thought vector: z (0) (t)=f root (T(t),H(t),F(t)) (17) Where z (0) (t) is the three-dimensional preliminary reasoning vector, f root (·) is the root node feature extraction function; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operation frequency at time t; First level reasoning; based on the initial thinking vector z (0) (t), generates the first-level thinking branch, as shown in formula (18); Where z (1) (t) is the first-layer reasoning vector, f branch1 (·) is the first-level inference function, The characteristic representations of the factors after the first-level reasoning are the ambient temperature T(t), relative humidity H(t), and operating frequency F(t) at time t; Second-level reasoning: Based on the first-level reasoning results, further reasoning is performed to obtain the final refined sensitivity vector, as shown in formula (19); Where z (1) (t) is the first-level reasoning vector, z (2) (t) is the second-level reasoning vector, f branch2 (·) is the second-level inference function, The characteristic representations of the factors after the first-level reasoning are the ambient temperature T(t), relative humidity H(t), and operating frequency F(t) at time t; The final sensitivity coefficient is generated; the final inference result is mapped to the sensitivity coefficient through linear mapping, as shown in formula (20): Where: d1(t), d2(t), d3(t) are the sensitivity coefficients of temperature, humidity, and operating frequency to the evaporator fan load at time t; W out is the output weight matrix, b out is the output bias vector; Substitute the load formula for prediction; substitute the predicted sensitivity coefficients d1(t), d2(t), d3(t) into formula (16) to obtain formula (21); Where: is the predicted evaporator fan load at time t; d1(t), d2(t), d3(t) are the sensitivity coefficients of temperature, humidity, and operating frequency at time t to the evaporator fan load respectively; T(t) is the ambient temperature at time t; H(t) is the relative humidity at time t; F(t) is the operating frequency at time t; b3 is the basic evaporator fan load.

5. The intelligent evaluation method for the load regulation capability of a refrigeration plant considering multiple factors according to claim 1 is characterized in that: Step 4 specifically involves: proposing a DPO-based day-ahead regulation capacity assessment model for the refrigeration plant. Based on the predicted loads of the refrigeration compressor, condenser fan, and evaporator fan of the refrigeration plant, the DPO algorithm is used to predict the regulation capacity of the refrigeration plant on a day-ahead time scale, taking into account the influence of temperature, humidity, and operation frequency. Total load estimation: First, the load forecast of the refrigeration compressor, condenser fan, and evaporator fan is integrated to obtain the total load of the refrigeration plant, as shown in formula (22); Where: is the total load; is the refrigeration compressor load predicted at time t; is the predicted condenser fan load at time t; is the evaporator fan load predicted at time t; Generate a candidate set of adjustment capabilities based on the total load; By floating up and down the proportion, multiple candidate intervals of adjustment capacity are generated, as shown in formula (23); Where: is the reduction capacity ratio coefficient of the i-th candidate, satisfying is the proportional coefficient of the upward adjustment capability of the i-th candidate, satisfying is the total load; Calculate the preference score based on the total load and regulation capacity interval; for each set of candidate regulation intervals, calculate its preference score s (i) (t), as shown in formula (24); Where: f pref (·) is the preference function, which can evaluate the pros and cons by combining the total load and the floating ratio; s(i)(t) is the preference score of the i-th candidate interval; is the reduction capacity ratio coefficient of the i-th candidate; is the proportional coefficient of the upward adjustment capability of the i-th candidate; is the total load; Select the regulatory ability interval with the highest preference score; select the highest score from all candidate intervals as the final prediction result, as shown in formula (25); Where: is the final selected reduction ratio, is the final selected increase ratio, s (i) (t) is the preference score of the i-th candidate interval; Finally, the adjustment capacity range of the refrigeration plant is predicted; based on the selected optimal ratio, the specific power value is reversed, as shown in formula (26); Where: is the forecast range of the regulation capacity of the refrigeration plant on the day-ahead time scale; is the reduction capacity ratio coefficient of the i-th candidate; is the proportional coefficient of the upward adjustment capability of the i-th candidate; is the total load.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to claim 1.

7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images