CN-122015361-A - Permafrost refrigeration system control method based on cold load

CN122015361ACN 122015361 ACN122015361 ACN 122015361ACN-122015361-A

Abstract

The invention discloses a permafrost refrigeration system control method based on cold load, which belongs to the technical field of frozen soil protection measures and comprises the following steps of firstly, predicting the cold load of a permafrost protection work point; and secondly, constructing a refrigerating system control module based on the cold load. The intelligent control of the refrigerating system operation process is realized through the intelligent algorithm, and the double aims of multi-year frozen soil temperature control and refrigerating system energy saving are achieved. The intelligent multi-year frozen soil protection method has the advantages of being strong in generalization capability, energy-saving, efficient and the like, and can timely adjust the control strategy of the refrigerating system according to the cold load prediction and feature recognition results to generate control instructions, so that the intelligent multi-year frozen soil protection method with the environment self-adaptation function is realized.

Inventors

HU TIANFEI
CAI DECHAO
LIU TIANFANG
YUE LU
ZHANG YUNLONG
LIU BEI
YANG RUI
He Taofan

Assignees

石家庄铁道大学

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The permafrost refrigeration system control method based on the cold load is characterized by comprising the following steps of: Firstly, predicting the cold load of a permafrost foundation working point; the second step, construct the refrigerating system control module based on cold load, the step is as follows: S1, identifying the cold load characteristics of the permafrost subgrade according to the cold load prediction result; The cold load characteristics refer to the size and fluctuation rule of the cold load, and the load rate is calculated, wherein the load rate refers to the ratio of the cold load of a certain time or time section in each day to the peak value of the daily cold load, the daily cold load is divided into peak sections, flat sections and valley sections according to the load rate, the load rate of the peak sections is more than or equal to 80%, the load rate of the flat sections is more than or equal to 30%, and the load rate of the valley sections is less than or equal to 30%; S2, formulating a control strategy of the refrigeration system as follows: when the time belongs to the valley period, the refrigerating system is stopped; When the time belongs to the peak section, starting the refrigerating system, and setting the opening of an expansion valve of the refrigerating system to be 100%; when the time belongs to the flat section, starting the refrigerating system, and dynamically adjusting the opening of the expansion valve; S3, optimizing an operation control scheme of the refrigeration system; S4, determining an adaptive loop control method of the refrigeration system; S5, designing an intelligent control module of the refrigeration system; based on the first step and the second step, the design of the intelligent control module of the permafrost refrigeration system is completed by utilizing a Niagara software platform.
2. The method for controlling a permafrost refrigeration system based on a cooling load according to claim 1, wherein in the first step, the method for predicting the cooling load of the permafrost is as follows: S1, collecting multi-source data related to permafrost stability of a permafrost foundation site, wherein the multi-source data comprise historical record data of meteorological parameters and geological parameters; S2, preprocessing the multi-source data based on periodic registration; s3, establishing a perennial frozen soil cold load mechanism model, and calculating and acquiring perennial frozen soil cold loads corresponding to time by time all the year; s4, constructing a feature set of the cold load of the frozen soil, the meteorological parameters and the geological parameters for a plurality of years; preprocessing the acquired annual time-by-time cold load data based on periodic registration to obtain a multi-year frozen soil cold load data set, analyzing the correlations of different characteristic variables and cold load by adopting a Pearson correlation coefficient method by taking meteorological parameters and geological parameters as input characteristic variables, screening the characteristic variables with the Pearson correlation coefficient more than or equal to 0.3, and combining the characteristic variables with the cold load data set to obtain a multi-source data characteristic set for the subsequent steps; the calculation formula of the pearson correlation coefficient P is: (1) Wherein n is the number of data in each characteristic variable, X i is the ith characteristic variable value; y i is the ith cold load data; Is the average of the cold load data; s5, performing empirical mode decomposition of permafrost cold load; firstly, processing the multi-source data feature set obtained in the step S4 through a signal processing method, and decomposing an original data sequence of the frozen soil cold load for many years into a modal component imf and a residual res, wherein the modal component imf and the residual res are shown as a formula (2) to form modal data x (t) consisting of imf and res; (2) Wherein t is the moment, K is the total number of modal data components, imf i (t) is the ith inherent modal data component, namely the data sequence obtained by subtracting the average envelope curve from the original data sequence, res (t) is the residual signal obtained by subtracting imf from the original data, and the long-term variation trend of the cold load can be represented; S6, sample entropy pretreatment of the permafrost cold load modal data is carried out; s7, constructing a neural network model of the cold load of the permafrost; S8, checking and circulating training of a cold load prediction result of the frozen soil for a plurality of years; And S9, predicting the permafrost cold load, namely inputting corresponding characteristic variable values into a neural network prediction model according to the permafrost foundation working point conditions of different scenes to obtain a cold load predicted value.
3. The method of claim 2, wherein in the first step S1, the multisource data of the permafrost base station includes meteorological parameters including average annual temperature, average winter air temperature, average annual solar radiation amount, average annual wind speed and average annual rainfall, and geological parameters including vegetation coverage, permafrost ice content, permafrost upper limit, permafrost annual average temperature, engineering structure design size and engineering structure trend.
4. The method of claim 3, wherein in the step S2 of the first step, the time series data of each data source obtained in the step S1 are aligned to a uniform time period, and the data of different time scales or different time periods are synchronously aligned.
5. The method for controlling a permafrost refrigeration system based on cold load according to claim 4, wherein in the step S3 of the first step, a permafrost subgrade cold load mechanism model is built by using DeST software, and annual 8760h time-by-time cold load data of a target protection area of the permafrost subgrade is obtained.
6. The method for controlling a permafrost refrigeration system based on cold load according to claim 5, wherein in step S6 of the first step, sample entropy preprocessing is performed on modal data of cold load data after empirical mode decomposition, and the sample entropy preprocessing step is as follows: S61, assuming that the total number of data points included in components imf and res of the cold load data after empirical mode decomposition is N, X refers to a calculation vector corresponding to the sample entropy preprocessing tool, X is 1 calculation vector formed by continuous m data from the ith data, that is, m is the dimension of a sliding data window, then an expression of N-m+1 m-dimensional calculation vector X m ,X m can be constructed as shown in formula (3): (3) Wherein i represents the number of the data window X m , N represents the total number of data points included by components imf and res of the cold load data after empirical mode decomposition, m represents the dimension of a sliding data window, and u is time sequence data of the cold load; s62, calculating vector And (3) with Maximum distance between L m : (4) wherein k is more than or equal to 0 and less than or equal to m-1, j represents the number of another sliding data window, and k represents the element position index inside the sliding data window; Due to And (3) with The two sliding data windows are m-dimensional calculation vectors, so that the value range of k is from 0 to m-1, and the positions of 1 st to m elements in the corresponding windows; S63, setting a given threshold value The calculation satisfies And (3) with The number of (1) is respectively recorded as And The calculation methods of the matching rate of the m-dimension calculation vector and the m+1-dimension calculation vector are respectively formula (5) and formula (6); (5) where N-m is the total number of m+1-dimensional sliding data windows, N-m+1 is the total number of m-dimensional sliding data windows, The number of m-dimensional vectors with the distance smaller than r in the data sequence for the ith m-dimensional sliding data window; (6) In the formula, Means the number of m+1-dimensional vectors with a distance smaller than r in the data sequence for the ith m+1-dimensional sliding data window; S64, a sample entropy expression is: (7) And (3) calculating sample entropy SampEn (m, r) of the cold load modal data components imf and res according to the formula (7), judging the overall stability of the cold load data sequence according to the sample entropy SampEn (m, r), screening the cold load data with the fluctuation amplitude of the sample entropy SampEn (m, r) conforming to the normal rule, and carrying out neural network model construction in the next step S7.
7. The method for controlling a permafrost refrigeration system based on cold load of claim 6, wherein in step S7 of the first step, the neural network model of the permafrost cold load is trained and constructed by utilizing the cold load data screened in step S6 and the meteorological parameters and geological parameters in the multisource data feature set screened in step S4, so as to further predict the cold load of the permafrost in other different scenes; The neural network model of the frozen soil cold load for many years is a CNN-BiLSTM neural network model, the model structure comprises an input layer, a CNN characteristic extraction module, a BiLSTM modeling layer and an output layer, all of which are composed of neurons, the CNN characteristic extraction module comprises a convolution layer and a pooling layer, and the training and construction steps comprise: S71, carrying out characteristic convolution on meteorological parameters and geological parameters in characteristic sets of permafrost cold load data and multisource data by utilizing a convolution layer, sliding a convolution kernel on input data, and executing convolution operation, so as to generate a characteristic matrix of the input data, wherein a calculation formula of the characteristic matrix is as follows: (8) wherein: And The method comprises the steps of respectively inputting data of a convolution layer and outputting data in a feature matrix of the convolution layer, wherein a is the size of a convolution kernel, namely the number of data covered by convolution operation, q is the number of zero padding layers, is used for keeping the matching degree of the output length and the input length of the feature matrix and avoiding edge feature loss, s is a step length, and the convolution kernel slides each time; The pooling layer performs down sampling, namely maximum pooling treatment, on the feature matrix output by the convolution layer, and the method is that a sliding data window is used for sweeping the feature matrix, and the maximum value in the data window is taken as output data, so that the dimension compression of the feature matrix is realized, the calculated amount is reduced, and the robustness of the output data is enhanced; s72, continuously inputting one-dimensional data output by a CNN feature extraction module into a BiLSTM modeling layer, enabling the BiLSTM modeling layer to simultaneously start two branches of a forward LSTM neural network and a backward LSTM neural network, learning the influence of historical local features in a data sequence on the current moment along the direction from the early to the late of an input data sequence, capturing the association of the future local features to the current moment along the direction from the late to the early of the input data sequence by the backward LSTM neural network, splicing and fusing a forward hidden state and a backward hidden state corresponding to each input data by the BiLSTM modeling layer after the bidirectional LSTM neural network completes full sequence traversal, generating a high-dimensional feature matrix containing bidirectional time sequence dependence, inputting the high-dimensional feature matrix into a full-connection layer, converting the high-dimensional feature into a scalar result corresponding to cold load data through linear mapping, and finally obtaining a predicted value of a multi-year frozen soil cold load; s73, optimizing the super parameters of the CNN-BiLSTM neural network model by utilizing a GA/PSO optimization algorithm; The GA/PSO optimization algorithm optimizes the super parameters of the CNN-BiLSTM neural network model by adopting a mode of combining a particle swarm algorithm with a genetic algorithm, wherein the super parameters are configuration parameters of the CNN-BiLSTM neural network model before training, and comprise the neuron number, the learning rate and the like of a BiLSTM modeling layer, and the steps are as follows: 1) Initializing GA parameters, initializing hybridization probability , Refers to the probability of two different super-parameter combinations to execute the cross operation, and initializes the variation probability , Refers to the probability of a parameter in a single superparameter combination to perform a mutation operation; 2) Initializing PSO parameters, wherein the initial inertia weight w refers to the degree of searching direction before the control combination scheme continues, the initial learning factors c 1 and c 2 ,c 1 refer to the degree of learning the combination scheme to the self history optimal scheme, and c 2 refers to the degree of learning the combination scheme to the group global optimal scheme; 3) Initializing the number of neurons of a BiLSTM modeling layer, the learning rate and the CNN convolution kernel size; 4) And (3) taking each super-parameter combination as an optimized particle, searching global optimal particles and individual optimal particles through a PSO optimization algorithm, taking the inverse of Root Mean Square Error (RMSE) of a cold load predicted value as an objective function fitness, wherein the formula of the objective function is as follows: (9) Wherein n is the number of cold load samples; Is a predicted value of the cold load; The objective function fitness is the fitness value of the optimized particles, and the larger the fitness value is, the better the super-parameter combination is represented; 5) Dividing particles into two types of dominant particles and inferior particles according to the particle fitness, and reserving the dominant particles to enter the next generation; 6) Randomly selecting two particles from the dominant particles to cross, and updating the particle speed and position; 7) Randomly selecting two particles from the crossed particles to perform mutation; 8) Updating the fitness value, global optimum and individual optimum of the mutated particles; 9) Judging whether the requirements are met, outputting optimal parameters, and otherwise returning to the step 3).
8. The method for controlling a permafrost refrigeration system based on a cooling load according to claim 7, wherein in step S8 of the first step, the prediction result of step S7 is checked, and the index for evaluating the prediction accuracy is a determination coefficient R 2 and an average absolute percentage error E MAPE : (10) In the formula, Is the average value of the real value of the cold load; A true value representing the kth sample; representing a predicted value of the model for a kth sample; (11)。
9. The method of claim 2 to 8, wherein in step S3, the method of optimizing the operation control scheme of the refrigerating system comprises the steps of first specifying 1 objective function (equation 12) targeting a single day minimum power consumption minP and using the controlled operation state of the refrigerating system and the opening degree of the expansion valve as variables, then specifying 2 constraint condition functions, wherein equation (13) corresponds to the cooling capacity of the refrigerating system, equation (14) corresponds to the power consumption of the refrigerating system, and finally solving equation (12) to obtain the controlled operation state of the refrigerating system satisfying the constraint condition And the opening phi of the expansion valve is the optimized refrigerating system operation control scheme; s31, objective function The operation control of the refrigerating system takes the cold load of the permafrost roadbed as an essential task, takes the single-day minimum power consumption minP as a target, and the objective function is as follows: (12) Wherein P is the total power consumption of the refrigerating system in one day, kW.h, P is the power consumption rate of the refrigerating system in the period t, kW, phi is the opening of the expansion valve and is regulated between 0 and 100 percent, and for a refrigerating system of a fixed model, the power consumption rate of the refrigerating system depends on the opening phi of the expansion valve, t is the period h; The method is characterized in that the method is in a controlled running state of the refrigerating system in a t time period, and the refrigerating system is opened to 1 and closed to 0; S32, constraint conditions 1) Constraint condition of cooling capacity The cold load of frozen soil for many years corresponds to the cold quantity required to be output by the refrigerating system, and the relation is as follows: (13) Wherein COP is the refrigerating coefficient of a refrigerating system, and y (t) is the cold load of permafrost in a t period, kW; 2) Constraint on power consumption The power consumption of the refrigerating system cannot exceed the power supply capacity of the photovoltaic system, and the single-day minimum power consumption minP does not exceed the power supply capacity of the photovoltaic system: (14) In the formula, For a single day maximum power supply of a photovoltaic system, kW.h.
10. The method for controlling a permafrost refrigeration system based on a refrigeration load according to claim 9, wherein in step S4 of the second step, the adaptive loop control method of the refrigeration system comprises the following steps: Firstly, mining historical data of the operation of a refrigeration system by adopting a correlation rule mining method, and finding out the corresponding relation between the opening parameter of an expansion valve of the refrigeration system and the cold output power of the refrigeration system; and then, according to the objective function and constraint condition requirements in the step S3, the optimal control of the running start-stop state of the refrigerating system and the opening degree of the expansion valve is realized from the perspective of global optimization, and the self-adaptive loop control method of the permafrost refrigerating system is formed.

Description

Permafrost refrigeration system control method based on cold load Technical Field The invention belongs to the technical field of frozen soil protection measures, and particularly relates to a permafrost refrigeration system control method based on a cold load. Background As global climate warms, permafrost is challenged by warming degradation, which can cause a series of engineering geological problems such as surface hot melting sedimentation, slumping and the like. The deformation and damage of the surface structures caused by the degradation of permafrost seriously affect the normal use of various infrastructures. Therefore, the protection of permafrost is a primary task, and in view of the lack of timeliness of the traditional measures such as a stone block air-cooling structure, a hot rod and the like, the field of engineering in recent years starts to introduce a refrigeration technology to perform forced cold transmission protection on the permafrost foundation. The invention patent of application number CN201711190185.7 discloses a compression type refrigerating system for preventing permafrost degradation, belongs to a refrigerating system driven by solar energy and wind energy in a combined mode, and the invention patent of application number CN202311743133.3 discloses a permafrost region full Ji Re rod refrigerating device and a construction method, and belongs to a compression type refrigerating system which is driven by solar energy and has a plurality of refrigerating sections, wherein the refrigerating sections are uniformly arranged on the outer side of a structure foundation according to a certain interval, and the permafrost between the refrigerating sections can be ensured to obtain cold energy. The existing refrigeration system mostly adopts a vapor compression type thermodynamic cycle principle, a closed loop is formed by four main components of a compressor, a condenser, an expansion valve and an evaporator, refrigerant is filled in the closed loop, and heat absorption, transfer and release are realized through the cyclic change of the phase state (gas-liquid state) of the refrigerant. The evaporator is an endothermic functional component, the condenser is an exothermic functional component, the refrigeration cycle is driven by the compressor and is dynamically regulated by the expansion valve, and the details are well known and will not be described here. The structural form of the evaporator in the refrigerating system adopts a columnar structure, and is implanted into a plurality of frozen soil layers through mechanical drilling. The other components (compressor, condenser, expansion valve) are integrated and installed in a case and are arranged beside various structures such as railway roadbed. The main problems in the above disclosed patent documents are that the refrigeration system needs external electric energy for driving, if the traditional control strategies of industries and building fields such as continuous operation, timing intermittent start-stop or PID setting temperature are blindly adopted without according to the actual demands of permafrost, the problems of insufficient or excessive cooling of the permafrost, excessively high energy consumption and the like can be caused. Particularly, a photovoltaic energy storage system needs to be installed under the condition of no power grid power supply, and the power supply cost can be increased sharply due to the fact that energy consumption is too high. How to consider the effectiveness of frozen soil protection and energy conservation depends on the relative relation between the refrigerating system refrigerating output and the perennial frozen soil cold load. However, due to complexity of permafrost geological conditions and variability of meteorological environments, the traditional control strategy of the refrigerating system lacks parameter self-adaptation capability, can not meet the requirements of permafrost cooling protection under severe fluctuation meteorological environment conditions of a plateau, and the phenomena of no-load, light load and overload of the refrigerating system are difficult to identify and avoid. In view of the fact that no technology exists at present for accurately quantitatively expressing the frozen soil cold load and the operation mode of the refrigerating system, the refrigerating system cannot realize optimal operation. Therefore, in the application of permafrost refrigeration protection technology, how to balance the relationship between the performance and the energy consumption of the refrigeration system is needed to be solved. Disclosure of Invention The invention provides a control method of a permafrost refrigeration system based on cold load, and aims to solve the technical problems that the conventional permafrost refrigeration protection technology cannot quantify the amount of cold required by permafrost, the operation scheme is insufficient in basis and the energy consumpt