Control method based on coal dressing intelligent filter pressing

Abstract

The invention discloses a control method based on coal dressing intelligent filter pressing, and belongs to the technical field of coal dressing, and the control method specifically comprises the following steps: pre-grading original coal slime through an intelligent multi-stage cyclone, collecting coal slime property data of each section in real time by using online detection equipment, and transmitting the coal slime property data to an intelligent control system; through a dual optimization mechanism, coal slime properties and filter pressing grades are matched, and optimal filter pressing parameters are generated; a multi-machine cooperative control strategy is adopted to drive a three-stage filter press to operate step by step, a virtual model of the filter press is constructed through real-time working state parameters of a sensor, and an abnormal state is dynamically corrected in combination with the real-time working state parameters and a fuzzy PID algorithm; carrying out resource processing on a filter cake generated by filter pressing, and meanwhile, carrying out grading treatment on filtrate; intelligent control over the whole process of coal slime grading filter pressing is achieved, and the coal slime treatment efficiency and the resource utilization rate are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of coal preparation, and particularly relates to a control method based on intelligent filter pressing for coal preparation. BACKGROUND

[0002] In the process of coal washing and processing, coal slime is one of the main by-products. Traditional coal slime treatment methods mainly include direct stacking, simple landfill or low-efficiency dehydration treatment and combustion, etc. These methods not only occupy a large amount of land resources and cause environmental pollution, but also fail to effectively utilize the large amount of resources contained in the coal slime, resulting in resource waste.

[0003] The existing coal slime filter pressing technology has problems, such as inaccurate control of the filter pressing process, difficulty in real-time adjustment of filter pressing parameters according to the properties of the coal slime, low filter pressing efficiency, high energy consumption, high water content of the coal slime after filter pressing, which is not conducive to subsequent resource utilization, improper treatment of the filtrate generated during the filter pressing process, which easily causes secondary pollution and cannot realize the recycling of useful substances in the filtrate. Therefore, there is an urgent need for a method capable of realizing efficient filter pressing of coal slime and effectively recycling the coal slime and the filtrate.

[0004] A Chinese patent application with publication number CN117902776A discloses a coal-containing scattered pile yard wastewater resource recycling system and method, which comprises: wastewater collecting device, sedimentation tank, treatment tank and recovery tank arranged in sequence. The wastewater collecting device is used for collecting coal-containing wastewater. The sedimentation tank is connected with the wastewater collecting device and is used for sedimentation of the wastewater. The treatment tank is connected with the sedimentation tank and is used for treatment of supernatant. The recovery tank is connected with the treatment tank and is used for recycling of water. The coal-containing wastewater is separated into coal blocks and wastewater by the wastewater collecting device, and then enters the sedimentation tank for sedimentation. The coal slime is deposited at the bottom of the sedimentation tank, and the supernatant enters the treatment tank for treatment. The treated water enters the recovery tank for recycling. The wastewater is collected by the wastewater collecting device, and then subjected to sedimentation and treatment, thereby saving the treatment time of the coal-containing wastewater and improving the treatment efficiency of the coal-containing wastewater in the railway coal-containing scattered pile yard.

[0005] The above existing technology has the following problems: focusing on the treatment of coal-containing wastewater, only realizing water resource recycling through sedimentation and separation, resulting in resource waste; lacking real-time monitoring and dynamic regulation and control of water quality and coal slime properties, which belongs to a single water treatment closed loop. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a control method based on intelligent filter pressing of coal preparation, which pre-classifies raw coal slime through an intelligent multi-stage cyclone, uses online detection equipment to collect coal slime property data of each section in real time and transmits the data to an intelligent control system, matches the coal slime properties and filter pressing grades through a double optimization mechanism, generates optimal filter pressing parameters, uses a multi-machine collaborative control strategy to drive a three-stage filter press to operate step by step, constructs a virtual model of the filter press through real-time working state parameters of a sensor, dynamically corrects abnormal states in combination with real-time working state parameters and a fuzzy PID algorithm, processes filter cake generated by filter pressing in a resourceful manner, and classifies filter liquor; the control method realizes intelligent control of the whole process of coal slime classification and filter pressing, and improves the coal slime treatment efficiency and resource utilization rate.

[0007] To achieve the above object, the present application provides the following technical scheme:

[0008] The control method based on intelligent filter pressing of coal preparation comprises:

[0009] S1: raw coal slime is pre-classified through an intelligent multi-stage cyclone, and online detection equipment is used to detect coal slime property data of each section of the pre-classified coal slime in real time and transmit the data to an intelligent control system; the online detection equipment integrates near-infrared spectroscopy, a laser particle size instrument and a density sensor;

[0010] S2: the intelligent control system receives coal slime property data, obtains optimal filter pressing parameters corresponding to the coal slime properties and filter pressing grades through a double optimization mechanism; the double optimization mechanism comprises parameter coarse tuning optimization based on coal slime properties and dynamic fine calibration optimization based on filter pressing grades;

[0011] S3: the intelligent control system controls a three-stage filter press to perform step-by-step filter pressing operation according to optimal filter pressing parameters through a multi-machine collaborative control strategy, real-time working state parameters of the three-stage filter press are monitored through a sensor, and the parameters are fed back to the intelligent control system, a virtual model of the filter press is constructed using digital twinning technology in combination with real-time working state parameters and a fuzzy PID algorithm, and abnormal states are dynamically corrected;

[0012] S4: after filter pressing, filter cake generated by the three-stage filter press is processed in a comprehensive resourceful manner, and filter liquor generated by the three-stage filter press is controlled in stages.

[0013] Specifically, the specific steps of S2 comprise:

[0014] S2.1: the intelligent control system inputs received coal slime property data into a preset filter pressing process model to perform preliminary filter pressing grade matching, obtains a preliminary filter pressing grade matching result, and obtains a classification filter pressing grade combination after secondary verification and correction;

[0015] The coal slime property data comprises particle size distribution, concentration and ash content.

[0016] The preset filter pressing process model internally builds a coal slime property and filter pressing grade corresponding relationship table based on historical production data, and matches different stage coal slimes to the preliminary filter pressing grade by calculating the similarity of the input coal slime property data and the coal slime property data in the corresponding relationship table;

[0017] S2.2: Based on the determined hierarchical filter pressing grade combination, the working parameters of each stage filter press are cooperatively optimized to obtain efficiency optimal parameters;

[0018] S2.3: Based on the thermodynamic law and the equipment characteristics, the efficiency optimal parameters and the energy consumption optimal parameters are combined to obtain energy consumption optimal parameters;

[0019] S2.4: The efficiency optimal parameters and the energy consumption optimal parameters are linearly fused by using a weighted fusion strategy, a constraint condition filtering mechanism is introduced, and the best filter pressing parameters corresponding to the coal slime properties and the filter pressing grades are generated;

[0020] The constraint condition filtering mechanism adopts a two-stage strategy of clipping and compensation;

[0021] The best filter pressing parameters include optimal feeding pressure and optimal filter pressing time.

[0022] Specifically, the secondary verification and correction in S2.1 includes:

[0023] The coal slime property data and the preliminary filter pressing grade matching result are input as features into the preloaded improved BP neural network model, and the actual filter pressing effect data in the historical production is combined as a label to cross-verify the preliminary filter pressing grade matching result;

[0024] If the verification matching grade confidence is lower than a preset threshold, the filter pressing grade is recalculated and corrected, and finally the hierarchical filter pressing grade combination with the highest confidence is output.

[0025] Specifically, the process of obtaining the preliminary filter pressing grade matching result in S2.1 includes:

[0026] S2.11: The intelligent control system receives the coal slime property data transmitted by the online detection equipment through the communication interface and performs normalization processing;

[0027] S2.12: A pre-trained filter pressing process model is loaded; the filter pressing process model is configured based on a rule engine model, and the mapping relationship between the coal slime property and the filter pressing grade has been defined in advance;

[0028] S2.13: The normalized coal slime property data is input into the pre-trained filter pressing process model for preliminary filter pressing grade matching, and the matching rules include:

[0029] If R1=(G1>60)∧(C>500) is satisfied, the preliminary matching is the coarse particle high concentration level R1, and corresponds to the first-stage filter press model, wherein G1 represents the proportion of particle size greater than 0.25 mm in the coal slime particle size distribution, C represents the coal slime concentration, and ∧ represents a logical and relationship;

[0030] If R2=(G2>60)∧(A>40) is satisfied: the preliminary matching is the high-ash fine particle level R2, and corresponds to the third-stage filter press model, wherein G2 represents the proportion of particle size less than 0.074 mm in the coal slime particle size distribution, A represents the ash content of the coal slime, and represents a logical not relationship;

[0031] If R3=(G1>40)∧(C>30) is satisfied: the preliminary matching is the medium particle medium concentration level R3, and corresponds to the second-stage filter press model;

[0032] S2.14: Store the coarse particle high concentration level, the medium particle medium concentration level, and the high-ash fine particle level obtained by preliminary matching in the system database, and transmit them to the improved BP neural network model for secondary verification.

[0033] Specifically, the specific steps of S2.3 include:

[0034] S2.31: Establish a filter press energy consumption model based on the principle of energy conservation in the filter pressing process;

[0035] S2.32: Import the inherent parameters of the filter press, and correct the filter press energy consumption model; the inherent parameters include the mechanical efficiency of the filter press and the heat dissipation loss coefficient;

[0036] S2.33: Obtain the efficiency optimal parameters and the classified filter pressing level combination, and obtain real-time coal slime quantity and environmental temperature data through a sensor, and form comprehensive data of the running state of the filter pressing system by integration;

[0037] S2.34: Convert the corrected filter press energy consumption model into a multivariate energy consumption function, solve the minimum energy consumption solution that satisfies the classification constraints, obtain the energy consumption optimal parameters, and store the energy consumption optimal parameters, the corresponding coal slime property data, and the classified filter pressing level in a historical database; the classification constraints include pressure constraints, flow constraints, and pressure maintaining time constraints.

[0038] Specifically, the specific steps of S2.4 include:

[0039] S2.41: Obtain the efficiency optimal parameters and the energy consumption optimal parameters;

[0040] S2.42: Weighted fusion is performed on the efficiency optimal parameters and the energy consumption optimal parameters of each filter press to obtain the optimal feed pressure P i and the optimal filter pressing time T i after weighted fusion;

[0041] S2.43: The weighted fused optimal feed pressure and optimal pressure filtration time are clipped according to the equipment usage range to obtain the clipped optimal feed pressure P i clip and the clipped optimal pressure filtration time T i clip for each stage of the pressure filter.

[0042] S2.44: According to the preset compensation step, the gradient compensation is performed on P i clip and T i clip to obtain the compensated and corrected optimal feed pressure and optimal pressure filtration time.

[0043] Specifically, the specific steps of S3 include:

[0044] S3.1: The intelligent control system generates a cooperative control instruction sequence according to the optimal pressure filtration parameters output by S2 and the known process characteristics of the three-stage pressure filter.

[0045] S3.2: The working state parameters are collected in real time by the distributed sensor group installed on the three-stage pressure filter and transmitted to the intelligent control system in real time through the industrial Ethernet; the working state parameters include pressure value, filter cloth displacement, filter cake thickness, and equipment temperature.

[0046] S3.3: Based on the CAD drawings of the three-stage pressure filter, the fluid mechanics equations, and the historical working state parameters of the pressure filter, a virtual model of the pressure filter is constructed using digital twinning technology.

[0047] Specifically, the specific steps of S3 further include:

[0048] S3.4: The working state parameters collected in real time by the distributed sensor group are injected into the virtual model of the pressure filter in real time, and virtual-real synchronization is achieved through bidirectional data mapping. An improved time sequence comparison algorithm is used to compare the predicted values of the virtual model of the pressure filter and the working state parameters collected in real time, and a deviation sequence is calculated.

[0049] S3.5: According to the deviation sequence and in combination with the deviation threshold, an abnormal state is located; the abnormal state includes pressure fluctuation out of limit, filter cloth deviation, and feed blockage.

[0050] S3.6: The located abnormal state is dynamically corrected using a fuzzy PID control algorithm.

[0051] Specifically, the improved time sequence comparison algorithm used in S3.4 to compare the predicted values of the virtual model of the pressure filter and the working state parameters collected in real time to calculate the deviation sequence includes:

[0052] S3.41: Predict the value x of the acquired filter press virtual model j and the real-time collected working state parameter y j Timestamp alignment, j represents the time point index;

[0053] S3.42: For each time point after alignment, calculate the absolute deviation of x j and y j , get the point state deviation value;

[0054] S3.43: Compare the time sequence characteristics of x j and y j , calculate the root mean square error of the feature parameter, get the feature deviation value;

[0055] S3.44: Combine the point state deviation value and the feature deviation value, and generate the final deviation sequence e j by weighted summation.

[0056] Compared with the prior art, the beneficial effects of the present application are:

[0057] 1. The present application proposes a control method based on intelligent coal preparation filter, in the pretreatment stage, through intelligent multi-stage cyclone and online detection equipment integrating multiple sensors, accurate coal slime pre-classification and property detection is completed; The intelligent control system determines the optimal filter parameters by using the double optimization mechanism, and drives the operation of the three-stage filter by using the multi-machine cooperative control strategy, combines digital twin, fuzzy PID algorithm to realize abnormal dynamic correction, effectively guarantees the efficient and stable operation of filter operation, reduces the risk of manual intervention and equipment failure.

[0058] 2. The present application proposes a control method based on intelligent coal preparation filter, in the aspect of resource utilization and environmental protection, the method carries out comprehensive resource processing on the filter cake after filter pressing, turns waste into treasure, and improves the added value of coal resources; At the same time, membrane separation and ion exchange technology are used for grading treatment of filtrate, realizing the recycling of water resources and the effective separation of pollutants, and improving the automation and intelligent level of coal slime treatment. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 is a schematic diagram of the control method based on intelligent coal preparation filter of the present application;

[0060] Figure 2 is a principle flow chart of the control method based on intelligent coal preparation filter of the present application;

[0061] Figure 3 is a flow chart for realizing the optimal filter parameters of the control method based on intelligent coal preparation filter of the present application;

[0062] Figure 4 is a flow chart for generating the deviation sequence of the control method based on intelligent coal preparation filter of the present application. DETAILED DESCRIPTION

[0063] Embodiment 1:

[0064] Referring to Figure 1 and Figure 2 , the present application provides an embodiment: a control method based on intelligent filter pressing of coal preparation, comprising the following steps:

[0065] S1: the raw slime is pre-classified by the intelligent multi-stage cyclone, and the slime property data of each section of the pre-classified slime is detected in real time by using an online detection device and transmitted to an intelligent control system; the online detection device fuses near-infrared spectroscopy, a laser particle size instrument and a density sensor;

[0066] Further, the specific steps of S1 include:

[0067] (1) the raw slime is transported to the feed inlet of the intelligent multi-stage cyclone by a belt conveyor or a slurry pump, to ensure stable feed flow, wherein a variable frequency speed control device is provided in the transportation process to control the flow rate;

[0068] (2) a vibrating screen or a hydrocyclone front filter screen is arranged before the feed inlet, to remove impurities such as wood chips and ironware with a particle size greater than 2 mm in the slime, to avoid clogging of the cyclone;

[0069] (3) the intelligent multi-stage cyclone is used for pre-classification, including primary cyclone classification and secondary cyclone classification;

[0070] Primary cyclone classification: the raw slime enters the primary cyclone, and is preliminarily separated according to particle size and density by using centrifugal force, to obtain underflow and overflow, wherein the underflow is coarse particle slime with a particle size greater than 0.5 mm, which has high ash content and enters the coarse particle level filter pressing channel; the overflow is fine particle slime with a particle size less than or equal to 0.5 mm, which enters the secondary cyclone for re-classification;

[0071] Secondary cyclone classification: after the primary overflow enters the secondary cyclone, the secondary cyclone is adjusted in terms of cone angle and pressure, to separate out medium-fine particle slime and fine particle slime, which enter corresponding filter pressing channels, respectively, wherein the medium-fine particle slime has a particle size in the range of 0.1 to 0.5 mm, and the fine particle slime has a particle size less than 0.1 mm;

[0072] (4) real-time detection is performed by using the online detection device, including:

[0073] a near-infrared spectroscopy probe is installed on the slime discharge pipeline of each section, to scan the slime sample in real time, obtain spectral data, and perform standardization processing;

[0074] The static laser scattering method is used to measure the particle size distribution of the slime, laser particle size data is obtained, and the particle size distribution is fitted, wherein the static laser scattering method is a prior art content in the field and is not the creative scheme of the present application, and will not be described here;

[0075] A vibrating densimeter is installed in the pipeline to measure the density of the slime slurry in real time, and the density data of the slime slurry is temperature-compensated and corrected;

[0076] (5) The Kalman filter is used to fuse the spectral data, the laser particle size data and the compensated slime slurry density data into a unified slime property parameter vector, and the fused slime property parameter vector is transmitted to the intelligent control system database in real time through an industrial Ethernet, wherein the Kalman filter is a prior art content in the field and is not the creative scheme of the present application, and will not be described here.

[0077] S2: The intelligent control system receives the slime property data, and obtains the optimal filter pressing parameters corresponding to the slime property and the filter pressing grade through a double optimization mechanism; the double optimization mechanism includes parameter coarse tuning optimization based on the slime property and dynamic fine calibration optimization based on the filter pressing grade;

[0078] S3: The intelligent control system controls the three-stage filter press to perform step-by-step filter pressing operation according to the optimal filter pressing parameters, and monitors the working state parameters of the three-stage filter press in real time through a sensor, and feeds back to the intelligent control system, and then uses the digital twin technology to construct a virtual model of the filter press, and combines the real-time collected working state parameters and the fuzzy PID algorithm to dynamically correct the located abnormal state.

[0079] S4: After the filter pressing is completed, the filter cake generated by the three-stage filter press is comprehensively processed, and the filtrate generated by the three-stage filter press is controlled in stages.

[0080] Further, the comprehensive resource processing of the filter cake includes:

[0081] (1) The mineral composition, moisture content, calorific value, and heavy metal content of the filter cake are determined through laboratory tests such as elemental analysis and calorific value determination, or online detection equipment;

[0082] (2) According to the resource utilization purposes such as fuel, building material raw material, and chemical filling, the filter cake is crushed to a target particle size by a crusher, and then classified by a vibrating screen, a drum screen and other equipment to remove impurities such as large gangue;

[0083] (3) Fuel processing path:

[0084] If the calorific value of the filter cake is greater than 15 MJ / kg, the filter cake is mixed with clean coal and coal gangue in a certain proportion to make briquettes or coal water slurry, which is used for industrial boiler or power plant combustion, wherein MJ represents megajoule.

[0085] If containing silicon and aluminum, it can be used as raw material for cement production to replace clay, or mixed with fly ash, lime, etc. to press into autoclaved bricks;

[0086] Extraction of rare metals such as germanium and gallium, or as an adsorbent for wastewater treatment after modification;

[0087] If it has no economic value, it can be used for mine filling or land reclamation after stabilization treatment, but it must meet environmental standards such as heavy metal leaching toxicity limits;

[0088] (4) The finished products after processing are stored according to their uses, such as fuel storage and building material storage, and are transported to the user end by belt conveyors and tank trucks.

[0089] Further, the filtrate is subjected to fractional treatment, which includes:

[0090] (1) On-line detection of the filtrate's suspended solids, turbidity, COD, heavy metal ions, and oil substances to determine its pollution level and reusability;

[0091] (2) Fractional treatment process:

[0092] Primary treatment: physical separation, mainly through sedimentation tanks, cyclones, or plate and frame filter presses to remove coarse particulate suspended solids in the filtrate and recover fine coal slurry;

[0093] Secondary treatment: chemical treatment, if the organic matter or heavy metals in the filtrate exceed the standard, coagulants, flocculants, or heavy metal capture agents are added to remove pollutants by reaction precipitation, and the supernatant is subjected to tertiary treatment;

[0094] Tertiary treatment: deep purification, using membrane separation, ion exchange, activated carbon adsorption, etc. to further remove dissolved pollutants and make the water quality meet the recycled water standard, for example, the circulating water quality requirement in a coal preparation plant is that the suspended solids are less than or equal to 50 mg / L and the turbidity is less than or equal to 20 NTU;

[0095] (3) The treated clear liquid is reused in the coal preparation process, such as the circulating water tank, to save water resources; if it cannot be reused, it needs to be discharged into the municipal pipeline or natural water bodies after meeting the standard.

[0096] Example 2:

[0097] Please refer to Figure 3 and Figure 4 The specific steps of S2 in this embodiment include:

[0098] S2.1: The intelligent control system inputs the received coal slime property data into the preset pressure filtration process model to perform preliminary pressure filtration grade matching, obtains a preliminary pressure filtration grade matching result, and obtains a fractional pressure filtration grade combination after secondary verification and correction.

[0099] The slime property data includes particle size distribution, concentration, ash content;

[0100] The preset filter pressing process model is built based on historical production data, and the correspondence table of slime property and filter pressing grade is built. By calculating the similarity of the input slime property data and the slime property data in the correspondence table, different section slimes are matched to the preliminary filter pressing grade;

[0101] S2.2: Based on the determined classification filter pressing grade combination, the working parameters of each stage filter press are optimized, and the efficiency optimal parameters are obtained;

[0102] Further, the specific steps of S2.2 include:

[0103] (1) Construct the core network of the deep deterministic policy gradient algorithm, including the policy network and the value network, wherein the policy network inputs the state information of the current filter pressing system and outputs the working parameter adjustment strategy of each stage filter press; the value network inputs the state information and the action output by the policy network, and evaluates the value of the action in the current state;

[0104] (2) Initialize the experience replay pool for storing experience data generated by the interaction between the agent and the environment, in the format of <state, action, reward, next state>;

[0105] (3) According to the classification filter pressing grade combination, set the initial parameters of the filter pressing machine working environment, including the initial values of the pressure, feed flow rate, pressure holding time of each stage filter press, and the safety boundary of the equipment operation, such as the maximum pressure, minimum feed flow rate;

[0106] (4) Integrate the slime property data, classification filter pressing grade combination, current working parameters of each stage filter press, and equipment operation state into a system state vector as the input state space of the deep deterministic policy gradient algorithm, wherein the current working parameters of each stage filter press include pressure, flow rate, and time, and the equipment operation state includes motor power, filter plate displacement, and other sensor data;

[0107] (5) Define the adjustment range of the working parameters of each stage filter press as the action space, for example, the pressure adjustment range is [3MPa, 12MPa], the feed flow rate adjustment range is [2m 3 / h,10m 3 / h], and the pressure holding time adjustment range is [10min, 30min], and the action output by the policy network needs to be within the adjustment range;

[0108] (6) Take the filter pressing efficiency as the core optimization target, and quantify it as a reward value, for example, if the current filter pressing efficiency is 10% higher than the historical average efficiency, give a positive reward; otherwise, give a negative reward;

[0109] (7) Introduce a penalty mechanism for actions that exceed the safety boundaries of the device, such as pressure exceeding 12 MPa, giving negative rewards;

[0110] (8) The agent controls the pressure filter according to the action output by the policy network, collects the system state changes after executing the action, the rewards obtained, and stores them in the experience replay pool. When the data volume of the experience replay pool reaches 1000, training begins;

[0111] (9) Randomly sample experience data in batches from the experience replay pool and input them into the policy network and the value network. The value network calculates Q values based on the sampled data to evaluate the value of the action. The policy network updates the policy parameters through gradient ascent algorithm based on the feedback from the value network to maximize the cumulative reward;

[0112] Further, the Q value represents the long-term cumulative reward that the agent expects to obtain after performing a certain action in a given state.

[0113] (10) Softly update the parameters of the policy network and the value network to the corresponding target network after every 100 training iterations to avoid falling into local optima;

[0114] (11) Set the training termination conditions, such as the cumulative reward value reaching stability for 10 consecutive times and meeting the preset threshold, or the number of training iterations reaching 10000, considering that the deep deterministic policy gradient algorithm converges;

[0115] (12) After the deep deterministic policy gradient algorithm converges, the parameter combination of each stage of the pressure filter generated by the current policy network is output, which is the efficiency-optimal parameter under the current hierarchical pressure filtration level combination.

[0116] S2.3: Based on the laws of thermodynamics and the characteristics of the device, combined with the determined hierarchical pressure filtration level combination and efficiency-optimal parameters, the energy-optimal parameters are obtained;

[0117] S2.4: Adopt a weighted fusion strategy to linearly fuse the efficiency-optimal parameters and the energy-optimal parameters, introduce a constraint condition filter mechanism, and generate the best pressure filtration parameters corresponding to the coal slime properties and the pressure filtration level;

[0118] The constraint condition filter mechanism adopts a two-stage strategy of clipping and compensation;

[0119] The best pressure filtration parameters include optimal feed pressure and optimal pressure filtration time.

[0120] The secondary verification and correction in S2.1 includes:

[0121] The coal slime property data and the preliminary filter pressing grade matching result are taken as characteristic inputs into the preloaded improved BP neural network model, and the actual filter pressing effect data in the historical production is taken as a label to cross-verify the preliminary filter pressing grade matching result;

[0122] If the verification matching grade confidence is lower than the preset threshold, the filter pressing grade is recalculated and corrected, and the final output is the classification filter pressing grade combination with the highest confidence.

[0123] Further, the specific steps of the secondary verification and correction include:

[0124] (1) obtaining the preliminary filter pressing grade matching result output by S2.1 and the normalized coal slime property data to construct characteristic data;

[0125] (2) constructing an improved BP neural network model, including:

[0126] input layer: the number of neurons is equal to the dimension of the characteristic, such as the sum of the dimension of the coal slime property data and the dimension of the preliminary filter pressing grade matching result;

[0127] hidden layer: at least two fully connected layers are adopted, and the number of neurons in each layer is determined through cross-verification;

[0128] output layer: the number of neurons is equal to the number of filter pressing grade categories, and the confidence score of each grade is output;

[0129] Adagrad optimizer is used to automatically adjust the learning rate;

[0130] a Dropout layer is introduced to prevent overfitting;

[0131] the hidden layer uses a ReLU activation function, and the output layer uses a Softmax function as an activation function for classification;

[0132] (3) using historical coal slime data and corresponding actual filter pressing effect as a training set, wherein each sample contains coal slime property data, preliminary filter pressing grade matching result and actual optimal filter pressing grade, and a trained improved BP neural network model is obtained through training;

[0133] (4) inputting the characteristic data into the trained improved BP neural network model to output the confidence score of each filter pressing grade, selecting the filter pressing grade with the highest confidence score as the correction result, and if the highest confidence score is greater than or equal to the preset threshold, accepting the correction result, and if the confidence score is lower than the preset threshold, triggering manual intervention.

[0134] The obtaining process of the preliminary filter pressing grade matching result in S2.1 includes:

[0135] S2.11: The intelligent control system receives the coal slime property data transmitted by the online detection device through the communication interface and performs normalization processing;

[0136] S2.12: A pre-trained filter pressing process model is loaded; the filter pressing process model is configured based on a rule engine model, and the mapping relationship between the coal slime properties and the filter pressing grades has been defined in advance;

[0137] S2.13: The normalized coal slime property data is input into the pre-trained filter pressing process model for preliminary filter pressing grade matching, and the matching rules include:

[0138] If R1=(G1>60)∧(C>500) is satisfied, the preliminary matching is coarse particle high concentration grade R1, and corresponds to a first-stage filter press model, wherein G1 represents the proportion of particle size greater than 0.25 mm in the coal slime particle size distribution, C represents the coal slime concentration, and ∧ represents the logical and relationship;

[0139] If R2=(G2>60)∧(A>30) is satisfied: The preliminary matching is high-ash fine particle grade R2, and corresponds to a third-stage filter press model, wherein G2 represents the proportion of particle size less than 0.074 mm in the coal slime particle size distribution, A represents the coal slime ash content, and represents the logical not relationship.

[0140] If R3=(G1>30)∧(C>30) is satisfied: The preliminary matching is medium particle medium concentration grade R3, and corresponds to a second-stage filter press model.

[0141] S2.14: The coarse particle high concentration grade, medium particle medium concentration grade, and high-ash fine particle grade obtained by preliminary matching are stored in the system database and transmitted to the improved BP neural network model for secondary verification.

[0142] The specific steps of S2.3 include:

[0143] S2.31: Based on the energy conservation principle of the filter pressing process, a filter press energy consumption model is established, and the formula is: E=F×Δp×t+m×a×ΔT, wherein E represents the total energy consumption, F represents the feed flow rate, t represents the pressure holding time, m represents the coal slime mass, a represents the coal slime specific heat capacity, Δp represents the filter pressing pressure difference, and ΔT represents the filter pressing process temperature rise.

[0144] ​Further, based on the law of conservation of energy, the energy conversion mechanism in the filter pressing process is analyzed, wherein the input electric energy of the filter press is mainly converted into three types of energy: one is the effective mechanical energy to overcome the filter resistance of coal slime, the second is the heat loss in the equipment operation, such as motor heating and hydraulic oil heating, and the third is the invalid energy consumption such as mechanical friction and hydraulic leakage. Based on the energy distribution relationship, the basic framework of the filter press energy consumption model is constructed, and the quantitative relationship expression of energy input and various energy output in the model is determined to provide an initial structure for model correction.

[0145] S2.32: import the inherent parameters of the filter press, and correct the filter press energy consumption model; the inherent parameters include the mechanical efficiency of the filter press and the heat loss coefficient;

[0146] wherein the mechanical efficiency of the filter press reflects the proportion of energy loss in the mechanical transmission process; the heat loss coefficient quantifies the heat dissipation amount per unit time.

[0147] wherein the correction of the filter press energy consumption model is to take the mechanical efficiency of the filter press as the denominator and add the heat loss coefficient as a bias term to the filter press energy consumption model.

[0148] S2.33: obtain the hierarchical filter pressing level combination and efficiency optimal parameters, and obtain the real-time coal slime amount and environmental temperature data through sensors, and integrate to form the comprehensive data of the filter pressing system running state;

[0149] Further, the hierarchical filter pressing level combination and efficiency optimal parameters determined from the early optimization results of the intelligent control system are extracted, i.e. the filter pressure, time, flow and other parameter combinations that can maximize the coal slime treatment amount per unit time, and at the same time, the real-time coal slime amount data in the current filter pressing process, such as instantaneous flow, cumulative treatment amount and environmental temperature data, are collected through distributed sensors. The hierarchical filter pressing level combination, efficiency optimal parameters, real-time coal slime amount data and environmental temperature data are integrated to form comprehensive data of the filter pressing system running state containing process parameters, equipment state and environmental factors, which provides input basis for dynamic calculation of the energy consumption model.

[0150] S2.34: convert the corrected filter press energy consumption model into an energy consumption multivariate function Solve the minimum energy consumption solution that satisfies the hierarchical constraint to obtain the energy consumption optimal parameters, and store the energy consumption optimal parameters and corresponding coal slime property data and hierarchical filter pressing level in the historical database; the hierarchical constraint includes pressure constraint, flow constraint and pressure maintaining time constraint, wherein f(·) represents a linear multivariate function, which describes the mapping rule between p, F, t, β and ε, and the function is to quantify the physical relationship between energy consumption and different influencing factors, p represents the filter pressing pressure, β represents the coal slime density, and ε represents the filter cloth resistance coefficient related to the coal slime particle size.

[0151] It should be noted that, because the specific heat capacity a and the coal slime mass m are constants, and the independent variables in the energy consumption multivariate function are usually adjustable operating parameters or physical quantities that vary with working conditions, therefore, the specific heat capacity a and the coal slime mass m are not included in the independent variables, but are integrated into the coefficients of the function expression as constants.

[0152] Further, the conversion process of the filter press energy consumption model into the energy consumption multivariate function includes:

[0153] Because the coal slime mass satisfies m = βV, and the volume V satisfies V = Ft, therefore, m = βFt; and the temperature change, i.e. the filter press temperature rise ΔT, is related to the filter press heat generation, and the filter cloth resistance coefficient is introduced to obtain ΔT = εpt.

[0154] Further, the hierarchical constraints include:

[0155] If the filter press level is a coarse particle high concentration level, and satisfies 3 ≤ p < 5 and 8 < F ≤ 10, then the energy consumption optimization goal is to save energy under the priority of efficiency;

[0156] If the filter press level is a medium particle medium concentration level, and satisfies 5 ≤ p ≤ 8 and 5 ≤ F ≤ 8, then the energy consumption optimization goal is to balance efficiency and energy consumption;

[0157] If the filter press level is a high ash fine particle level, and satisfies 8 < p ≤ 12 and 3 ≤ F < 5, then the energy consumption optimization goal is to minimize unit energy consumption.

[0158] The specific steps of S2.4 include:

[0159] S2.41: Obtain the efficiency optimal parameters X eff = [P eff,1 ,T eff,1 ,P eff,2 ,T eff,2 ,P eff,3 ,T eff,3 ] and the energy consumption optimal parameters X energy = [P energy,1 ,T energy,1 ,P energy,2 ,T energy,2 ,P energy,3 ,T energy,3 ], wherein P eff,1 , P eff,2 , P eff,3 respectively represent the efficiency optimal feed pressure of the first filter press, the second filter press, and the third filter press, and P eff,1For controlling the extrusion strength of the primary filter press on coarse particle slime, priority is given to ensure the dewatering speed and quality of filter cake, while medium particle slime needs medium pressure to avoid filter cloth blockage while ensuring dewatering efficiency, fine particles need low pressure to prevent filter cloth from solidifying, and too high pressure can easily lead to particle penetration of filter liquor, T eff,1 、T eff,2 、T eff,3 respectively represent the optimal pressure filtration time of the primary filter press, the secondary filter press and the tertiary filter press, P energy,1 、P energy,2 、P energy,3 respectively represent the optimal feed pressure of the primary filter press, the secondary filter press and the tertiary filter press, T energy,1 、T energy,2 、T energy,3 respectively represent the optimal pressure filtration time of the primary filter press, the secondary filter press and the tertiary filter press in terms of energy consumption;

[0160] S2.42: Fuse the parameters of each filter press, and the formula is:

[0161]

[0162] wherein P i and T i respectively represent the optimal feed pressure and the optimal pressure filtration time of each filter press after weighted fusion, w eff represents the weight of the efficiency optimal parameter, and w energy represents the weight of the energy consumption optimal parameter, and i represents the index of the filter press;

[0163] S2.43: Clip P i and T i to the range of device usage, to obtain the clipped optimal feed pressure P i clip and the clipped optimal pressure filtration time T i clip of each filter press, and the formula is: wherein P max respectively represent the upper limit of the safe pressure of the device, P min represents the minimum working pressure allowed by the device, T min represents the shortest effective pressure filtration time, T max represents the maximum allowable time for a single pressure filtration cycle, and min(·) represents the minimum value function and max(·) represents the maximum value function;

[0164] S2.44: Use the formula to compensate P i clip and T i clip , to obtain the compensated and corrected optimal feed pressure P iadj and the optimal filter pressing time T i adj wherein γ represents the compensation step size, and respectively represent the energy consumption multivariate function to the rate of change of P i and T i .

[0165] The specific steps of the S3 include:

[0166] S3.1: The intelligent control system generates a collaborative control instruction sequence according to the parameter distribution according to the known process characteristics of the three-stage filter press according to the optimal filter pressing parameters output by S2.

[0167] S3.2: Real-time acquisition of working state parameters through a distributed sensor group installed on the three-stage filter press, and real-time transmission to the intelligent control system through an industrial Ethernet; the working state parameters include pressure value, filter cloth displacement, filter cake thickness and equipment temperature.

[0168] S3.3: Based on the CAD drawings of the three-stage filter press, the fluid mechanics equation and the historical working state parameters of the filter press, a virtual model of the filter press is constructed by using digital twinning technology, wherein the digital twinning technology is a prior art content in the field and is not the inventive scheme of the present application, and will not be described here.

[0169] S3.4: Real-time injection of the working state parameters collected by the distributed sensor group into the virtual model of the filter press, realizing virtual-real synchronization through bidirectional data mapping, and comparing the predicted values of the virtual model of the filter press and the real-time collected working state parameters by using an improved time sequence comparison algorithm to calculate the deviation sequence.

[0170] Further, the bidirectional data mapping process includes:

[0171] (1) Entity to virtual model injection: real-time data is written into the virtual model input port through the OPC UA Pub / Sub mode, and the device coordinate system is converted into the simulation coordinate system through coordinate transformation;

[0172] (2) Virtual to entity feedback: simulation results are published to the programmable logic controller through the data distribution service.

[0173] S3.5: According to the deviation sequence, combined with the deviation threshold, the abnormal state is located; the abnormal state includes pressure fluctuation out of limit, filter cloth deviation, and feed blockage.

[0174] Further, the specific steps of S3.5 include:

[0175] (1) Extract features from the deviation sequence;

[0176] (2) Normal working condition statistical history deviation 3δ range, and set special threshold for different abnormal types, as shown in Table 1;

[0177] Table 1 Abnormal type and threshold table

[0178] Abnormal type Pressure fluctuation threshold Filter cloth runout threshold Feed blockage threshold Trigger condition P > 0.15 MPa ​ Slope persistence > 0.1 / min Energy ratio > 2.0

[0179] If the deviation sequence vector value of the continuous 5 sampling points is greater than 0.2 times the optimal feed pressure value, and δ P > 0.15 MPa, it is judged that the PID parameter is out of adjustment or the hydraulic leakage, wherein δ P represents the standard deviation of the pressure data;

[0180] And the filter cloth deviation trend is detected by the edge camera to detect the actual position of the filter cloth, and the OpenCV image recognition method is adopted;

[0181] If the extracted feature satisfies the energy ratio > 2.0, it is judged that the feed is blocked.

[0182] S3.6: For the positioning abnormal state, a fuzzy PID control algorithm is used for dynamic correction, wherein the fuzzy PID control algorithm is a prior art content in the art, and is not the inventive scheme of the present application, and will not be repeated here.

[0183] The improved time sequence comparison algorithm in S3.4 is used to compare the virtual model prediction value of the pressure filter and the real-time collected working state parameters, and the deviation sequence is calculated, including:

[0184] S3.41: The acquired virtual model prediction value x j and the real-time collected working state parameters y j are time stamped, and j represents the time point index;

[0185] S3.42: For each time point after alignment, the absolute deviation of x j and y j is calculated to obtain the point state deviation value;

[0186] S3.43: The time sequence characteristics of x j and y j are compared, the root mean square error of the feature parameters is calculated, and the feature deviation value is obtained;

[0187] S3.44: The point state deviation value and the feature deviation value are combined to generate the final deviation sequence e j .

[0188] The embodiments of the present application are described above with reference to the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, but not restrictive, and a person of ordinary skill in the art can make changes, modifications, replacements and variations to the above-described embodiments without departing from the purpose of the present application and the protected scope, and these are all within the protection of the present application.

Claims

1. A control method based on intelligent filter pressing of coal selection, characterized in that, The application relates to a coal slime intelligent pressure filtration method and system. S1: original coal slime is pre-classified through an intelligent multi-stage cyclone, and online detection equipment is used to detect the coal slime property data of the pre-classified coal slime in real time and transmit the coal slime property data to an intelligent control system; the online detection equipment combines near-infrared spectroscopy, a laser particle size instrument and a density sensor; S2: the intelligent control system receives the coal slime property data, obtains optimal pressure filtration parameters under corresponding coal slime properties and pressure filtration grades through a double optimization mechanism; the double optimization mechanism comprises parameter coarse adjustment optimization based on coal slime properties and dynamic fine calibration optimization based on pressure filtration grades; S3: the intelligent control system controls three-stage pressure filters to perform step-by-step pressure filtration operation according to the optimal pressure filtration parameters through a multi-machine collaborative control strategy, monitors the working state parameters of the three-stage pressure filters in real time through sensors, feeds back the working state parameters to the intelligent control system, constructs a pressure filter virtual model by using a digital twin technology, and dynamically corrects the located abnormal state by combining the real-time collected working state parameters and a fuzzy PID algorithm; S4: after pressure filtration, comprehensive resource processing is performed on filter cakes generated by the three-stage pressure filters, and the filter liquid generated by the three-stage pressure filters is controlled in stages.

2. The control method based on intelligent filter pressing of coal selection according to claim 1, characterized in that, The specific steps of S2 comprise: S2.1: the intelligent control system inputs the received coal slime property data into a preset pressure filtration process model to perform preliminary pressure filtration grade matching, obtains a preliminary pressure filtration grade matching result, and obtains a classified pressure filtration grade combination after secondary verification and correction; The coal slime property data comprises particle size distribution, concentration and ash content; The preset pressure filtration process model is internally provided with a coal slime property-pressure filtration grade corresponding relation table constructed based on historical production data, the similarity of the input coal slime property data and the coal slime property data in the corresponding relation table is compared, and different sections of coal slime are matched to preliminary pressure filtration grades; S2.2: based on the determined classified pressure filtration grade combination, the working parameters of the pressure filters are collaboratively optimized to obtain efficiency optimal parameters; S2.3: based on the law of thermodynamics and equipment characteristics, the classified pressure filtration grade combination and the efficiency optimal parameters are combined to obtain energy consumption optimal parameters; S2.4: a weighted fusion strategy is adopted to linearly fuse the efficiency optimal parameters and the energy consumption optimal parameters, a constraint condition filtering mechanism is introduced, and the optimal pressure filtration parameters under corresponding coal slime properties and pressure filtration grades are generated; The constraint condition filtering mechanism adopts a clipping-compensation two-stage strategy; The optimal pressure filtration parameters comprise optimal feeding pressure and optimal pressure filtration time.

3. The control method based on intelligent filter pressing of coal selection according to claim 2, characterized in that, The secondary verification and correction in S2.1 comprise: The coal slime property data and the preliminary pressure filtration grade matching result are input into a preloaded improved BP neural network model as features, actual pressure filtration effect data in historical production are combined as labels, and the preliminary pressure filtration grade matching result is cross-verified; If the matching grade confidence is lower than a preset threshold, the pressure filtration grade is recalculated and corrected, and finally the classified pressure filtration grade combination with the highest confidence is output.

4. The control method based on intelligent filter pressing of coal selection according to claim 3, characterized in that, The obtaining process of the preliminary pressure filtration grade matching result in S2.1 comprises: S2.11: the intelligent control system receives the coal slime property data transmitted by the online detection equipment through a communication interface and performs normalization processing; S2.12: Load a pre-trained filter pressing process model; the filter pressing process model is configured based on a rule engine model, and the mapping relationship between coal slime properties and filter pressing grades has been pre-defined; S2.13: Input the normalized coal slime property data into the pre-trained filter pressing process model for preliminary filter pressing grade matching, and the matching rules include: If R1=(G1>60)∧(C>500) is satisfied, the preliminary matching is coarse-grained high-concentration grade R1, and the corresponding first-stage filter press model is determined, wherein G1 represents the proportion of particle size greater than 0.25 mm in the coal slime particle size distribution, C represents the coal slime concentration, and ∧ represents the logical AND relationship; If the following condition is met: then the preliminary matching is a high-ash fine particle grade R2, and corresponds to a three-stage filter press model, wherein G2 represents the proportion of particle size distribution of coal slime less than 0.074 mm, A represents the ash content of coal slime, represents a logical NOT relationship; If the following is satisfied: then the preliminary match is the medium grain medium concentration grade R3, and the corresponding secondary filter press model is; S2.14: Store the coarse-grained high-concentration grade, medium-grained medium-concentration grade, and high-ash fine-grained grade obtained by preliminary matching in the system database, and transmit them to the improved BP neural network model for secondary verification.

5. The control method based on intelligent filter pressing of coal selection according to claim 4, characterized in that, The specific steps of S2.3 include: S2.31: Based on the energy conservation principle of filter pressing process, establish a filter press energy consumption model; S2.32: Import the inherent parameters of the filter press, and correct the filter press energy consumption model; the inherent parameters include the mechanical efficiency and heat loss coefficient of the filter press; S2.33: Obtain the efficiency-optimal parameters of the classified filter pressing grade combination, and obtain the real-time coal slime quantity and environmental temperature data through sensors, and integrate them to form comprehensive data of the running state of the filter pressing system; S2.34: Convert the corrected filter press energy consumption model into a multi-element function of energy consumption, solve the minimum energy consumption solution that satisfies the classification constraints, obtain the energy consumption optimal parameters, and store the energy consumption optimal parameters, corresponding coal slime property data, and classified filter pressing grade in the historical database; the classification constraints include pressure constraints, flow constraints, and pressure maintenance time constraints.

6. The control method based on intelligent filter pressing of coal selection according to claim 5, characterized in that, The specific steps of S2.4 include: S2.41: Obtain the efficiency-optimal parameters and energy consumption-optimal parameters; S2.42: the efficiency optimal parameters and the energy consumption optimal parameters of each stage filter press are weighted and fused to obtain the weighted and fused optimal feed pressure P i and the optimal filter pressing time T i ; S2.43: The weighted fused optimal feed pressure and optimal pressure filtration time are cut according to the equipment usage range to obtain the cut optimal feed pressure P of each stage of pressure filter i clip and the cut optimal pressure filtration time T i clip ; S2.44: According to the preset compensation step, P i clip and T i clip Gradient compensation is performed to obtain the compensated and corrected optimal feed pressure and optimal pressure filtration time.

7. The control method based on intelligent filter pressing of coal selection according to claim 6, characterized in that, The specific steps of S3 include: S3.1: The intelligent control system generates a collaborative control instruction sequence according to the optimal filter pressing parameters output by S2 and the known process characteristics of the three-stage filter press; S3.2: Real-time acquisition of working state parameters through the distributed sensor group installed on the three-stage filter press, and real-time transmission to the intelligent control system through industrial Ethernet; the working state parameters include pressure value, filter cloth displacement, filter cake thickness, and equipment temperature; S3.3: Based on the CAD drawings of the three-stage filter press, the fluid mechanics equation, and the historical working state parameters of the filter press, a virtual model of the filter press is constructed using digital twinning technology.

8. The control method based on intelligent filter pressing of coal selection according to claim 7, characterized in that, The specific steps of S3 also include: S3.4: Real-time injection of the working state parameters collected by the distributed sensor group into the filter press virtual model, realizing virtual-real synchronization through bidirectional data mapping, and comparing the predicted values of the filter press virtual model with the real-time collected working state parameters using an improved time sequence comparison algorithm to calculate the deviation sequence; S3.5: According to the deviation sequence, combined with the deviation threshold, locate the abnormal state; the abnormal state includes pressure fluctuation overrun, filter cloth deviation, and feed blockage; S3.6: Dynamically correct the located abnormal state using fuzzy PID control algorithm.

9. The control method based on intelligent filter pressing of coal separation according to claim 8, characterized in that, The improved time sequence comparison algorithm in S3.4 is used to compare the virtual model prediction value of the pressure filter and the real-time collected working state parameters, and a deviation sequence is calculated, including: S3.41: Predicted values x of the acquired filter press virtual model j and real-time collected working condition parameters y j Timestamp alignment, j denotes the time point index; S3.42: For each time point after alignment, calculate the absolute deviation of x j and y j , to get the point state deviation value; S3.43: contrast x j and y j the timing characteristics, calculate the root mean square error of the characteristic parameters, and obtain the characteristic deviation value; S3.44: Combine the point-wise bias values and the feature bias values to generate the final bias sequence e by weighted summation j .

Images